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 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
267 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
269 r = kvm_mmu_create(vcpu);
271 goto fail_free_pio_data;
276 free_page((unsigned long)vcpu->pio_data);
278 free_page((unsigned long)vcpu->run);
282 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
284 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
286 kvm_mmu_destroy(vcpu);
287 free_page((unsigned long)vcpu->pio_data);
288 free_page((unsigned long)vcpu->run);
290 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
292 static struct kvm *kvm_create_vm(void)
294 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
297 return ERR_PTR(-ENOMEM);
299 kvm_io_bus_init(&kvm->pio_bus);
300 mutex_init(&kvm->lock);
301 INIT_LIST_HEAD(&kvm->active_mmu_pages);
302 kvm_io_bus_init(&kvm->mmio_bus);
303 spin_lock(&kvm_lock);
304 list_add(&kvm->vm_list, &vm_list);
305 spin_unlock(&kvm_lock);
309 static int kvm_dev_open(struct inode *inode, struct file *filp)
315 * Free any memory in @free but not in @dont.
317 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
318 struct kvm_memory_slot *dont)
322 if (!dont || free->phys_mem != dont->phys_mem)
323 if (free->phys_mem) {
324 for (i = 0; i < free->npages; ++i)
325 if (free->phys_mem[i])
326 __free_page(free->phys_mem[i]);
327 vfree(free->phys_mem);
330 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
331 vfree(free->dirty_bitmap);
333 free->phys_mem = NULL;
335 free->dirty_bitmap = NULL;
338 static void kvm_free_physmem(struct kvm *kvm)
342 for (i = 0; i < kvm->nmemslots; ++i)
343 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
346 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
350 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
351 if (vcpu->pio.guest_pages[i]) {
352 __free_page(vcpu->pio.guest_pages[i]);
353 vcpu->pio.guest_pages[i] = NULL;
357 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
360 kvm_mmu_unload(vcpu);
364 static void kvm_free_vcpus(struct kvm *kvm)
369 * Unpin any mmu pages first.
371 for (i = 0; i < KVM_MAX_VCPUS; ++i)
373 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
374 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
376 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
377 kvm->vcpus[i] = NULL;
383 static int kvm_dev_release(struct inode *inode, struct file *filp)
388 static void kvm_destroy_vm(struct kvm *kvm)
390 spin_lock(&kvm_lock);
391 list_del(&kvm->vm_list);
392 spin_unlock(&kvm_lock);
393 kvm_io_bus_destroy(&kvm->pio_bus);
394 kvm_io_bus_destroy(&kvm->mmio_bus);
396 kvm_free_physmem(kvm);
400 static int kvm_vm_release(struct inode *inode, struct file *filp)
402 struct kvm *kvm = filp->private_data;
408 static void inject_gp(struct kvm_vcpu *vcpu)
410 kvm_arch_ops->inject_gp(vcpu, 0);
414 * Load the pae pdptrs. Return true is they are all valid.
416 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
418 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
419 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
424 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
426 mutex_lock(&vcpu->kvm->lock);
427 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
433 pdpt = kmap_atomic(page, KM_USER0);
434 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
435 kunmap_atomic(pdpt, KM_USER0);
437 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
438 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
445 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
447 mutex_unlock(&vcpu->kvm->lock);
452 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
454 if (cr0 & CR0_RESERVED_BITS) {
455 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
461 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
462 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
467 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
468 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
469 "and a clear PE flag\n");
474 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
476 if ((vcpu->shadow_efer & EFER_LME)) {
480 printk(KERN_DEBUG "set_cr0: #GP, start paging "
481 "in long mode while PAE is disabled\n");
485 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
487 printk(KERN_DEBUG "set_cr0: #GP, start paging "
488 "in long mode while CS.L == 1\n");
495 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
496 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
504 kvm_arch_ops->set_cr0(vcpu, cr0);
507 mutex_lock(&vcpu->kvm->lock);
508 kvm_mmu_reset_context(vcpu);
509 mutex_unlock(&vcpu->kvm->lock);
512 EXPORT_SYMBOL_GPL(set_cr0);
514 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
516 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
518 EXPORT_SYMBOL_GPL(lmsw);
520 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
522 if (cr4 & CR4_RESERVED_BITS) {
523 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
528 if (is_long_mode(vcpu)) {
529 if (!(cr4 & X86_CR4_PAE)) {
530 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
535 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
536 && !load_pdptrs(vcpu, vcpu->cr3)) {
537 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
542 if (cr4 & X86_CR4_VMXE) {
543 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
547 kvm_arch_ops->set_cr4(vcpu, cr4);
548 mutex_lock(&vcpu->kvm->lock);
549 kvm_mmu_reset_context(vcpu);
550 mutex_unlock(&vcpu->kvm->lock);
552 EXPORT_SYMBOL_GPL(set_cr4);
554 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
556 if (is_long_mode(vcpu)) {
557 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
558 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
564 if (cr3 & CR3_PAE_RESERVED_BITS) {
566 "set_cr3: #GP, reserved bits\n");
570 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
571 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
577 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
579 "set_cr3: #GP, reserved bits\n");
587 mutex_lock(&vcpu->kvm->lock);
589 * Does the new cr3 value map to physical memory? (Note, we
590 * catch an invalid cr3 even in real-mode, because it would
591 * cause trouble later on when we turn on paging anyway.)
593 * A real CPU would silently accept an invalid cr3 and would
594 * attempt to use it - with largely undefined (and often hard
595 * to debug) behavior on the guest side.
597 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
600 vcpu->mmu.new_cr3(vcpu);
601 mutex_unlock(&vcpu->kvm->lock);
603 EXPORT_SYMBOL_GPL(set_cr3);
605 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
607 if (cr8 & CR8_RESERVED_BITS) {
608 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
614 EXPORT_SYMBOL_GPL(set_cr8);
616 void fx_init(struct kvm_vcpu *vcpu)
618 struct __attribute__ ((__packed__)) fx_image_s {
624 u64 operand;// fpu dp
630 /* Initialize guest FPU by resetting ours and saving into guest's */
632 fx_save(vcpu->host_fx_image);
634 fx_save(vcpu->guest_fx_image);
635 fx_restore(vcpu->host_fx_image);
638 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
639 fx_image->mxcsr = 0x1f80;
640 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
641 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
643 EXPORT_SYMBOL_GPL(fx_init);
646 * Allocate some memory and give it an address in the guest physical address
649 * Discontiguous memory is allowed, mostly for framebuffers.
651 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
652 struct kvm_memory_region *mem)
656 unsigned long npages;
658 struct kvm_memory_slot *memslot;
659 struct kvm_memory_slot old, new;
660 int memory_config_version;
663 /* General sanity checks */
664 if (mem->memory_size & (PAGE_SIZE - 1))
666 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
668 if (mem->slot >= KVM_MEMORY_SLOTS)
670 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
673 memslot = &kvm->memslots[mem->slot];
674 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
675 npages = mem->memory_size >> PAGE_SHIFT;
678 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
681 mutex_lock(&kvm->lock);
683 memory_config_version = kvm->memory_config_version;
684 new = old = *memslot;
686 new.base_gfn = base_gfn;
688 new.flags = mem->flags;
690 /* Disallow changing a memory slot's size. */
692 if (npages && old.npages && npages != old.npages)
695 /* Check for overlaps */
697 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
698 struct kvm_memory_slot *s = &kvm->memslots[i];
702 if (!((base_gfn + npages <= s->base_gfn) ||
703 (base_gfn >= s->base_gfn + s->npages)))
707 * Do memory allocations outside lock. memory_config_version will
710 mutex_unlock(&kvm->lock);
712 /* Deallocate if slot is being removed */
716 /* Free page dirty bitmap if unneeded */
717 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
718 new.dirty_bitmap = NULL;
722 /* Allocate if a slot is being created */
723 if (npages && !new.phys_mem) {
724 new.phys_mem = vmalloc(npages * sizeof(struct page *));
729 memset(new.phys_mem, 0, npages * sizeof(struct page *));
730 for (i = 0; i < npages; ++i) {
731 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
733 if (!new.phys_mem[i])
735 set_page_private(new.phys_mem[i],0);
739 /* Allocate page dirty bitmap if needed */
740 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
741 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
743 new.dirty_bitmap = vmalloc(dirty_bytes);
744 if (!new.dirty_bitmap)
746 memset(new.dirty_bitmap, 0, dirty_bytes);
749 mutex_lock(&kvm->lock);
751 if (memory_config_version != kvm->memory_config_version) {
752 mutex_unlock(&kvm->lock);
753 kvm_free_physmem_slot(&new, &old);
761 if (mem->slot >= kvm->nmemslots)
762 kvm->nmemslots = mem->slot + 1;
765 ++kvm->memory_config_version;
767 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
768 kvm_flush_remote_tlbs(kvm);
770 mutex_unlock(&kvm->lock);
772 kvm_free_physmem_slot(&old, &new);
776 mutex_unlock(&kvm->lock);
778 kvm_free_physmem_slot(&new, &old);
784 * Get (and clear) the dirty memory log for a memory slot.
786 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
787 struct kvm_dirty_log *log)
789 struct kvm_memory_slot *memslot;
792 unsigned long any = 0;
794 mutex_lock(&kvm->lock);
797 * Prevent changes to guest memory configuration even while the lock
801 mutex_unlock(&kvm->lock);
803 if (log->slot >= KVM_MEMORY_SLOTS)
806 memslot = &kvm->memslots[log->slot];
808 if (!memslot->dirty_bitmap)
811 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
813 for (i = 0; !any && i < n/sizeof(long); ++i)
814 any = memslot->dirty_bitmap[i];
817 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
820 mutex_lock(&kvm->lock);
821 kvm_mmu_slot_remove_write_access(kvm, log->slot);
822 kvm_flush_remote_tlbs(kvm);
823 memset(memslot->dirty_bitmap, 0, n);
824 mutex_unlock(&kvm->lock);
829 mutex_lock(&kvm->lock);
831 mutex_unlock(&kvm->lock);
836 * Set a new alias region. Aliases map a portion of physical memory into
837 * another portion. This is useful for memory windows, for example the PC
840 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
841 struct kvm_memory_alias *alias)
844 struct kvm_mem_alias *p;
847 /* General sanity checks */
848 if (alias->memory_size & (PAGE_SIZE - 1))
850 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
852 if (alias->slot >= KVM_ALIAS_SLOTS)
854 if (alias->guest_phys_addr + alias->memory_size
855 < alias->guest_phys_addr)
857 if (alias->target_phys_addr + alias->memory_size
858 < alias->target_phys_addr)
861 mutex_lock(&kvm->lock);
863 p = &kvm->aliases[alias->slot];
864 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
865 p->npages = alias->memory_size >> PAGE_SHIFT;
866 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
868 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
869 if (kvm->aliases[n - 1].npages)
873 kvm_mmu_zap_all(kvm);
875 mutex_unlock(&kvm->lock);
883 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
886 struct kvm_mem_alias *alias;
888 for (i = 0; i < kvm->naliases; ++i) {
889 alias = &kvm->aliases[i];
890 if (gfn >= alias->base_gfn
891 && gfn < alias->base_gfn + alias->npages)
892 return alias->target_gfn + gfn - alias->base_gfn;
897 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
901 for (i = 0; i < kvm->nmemslots; ++i) {
902 struct kvm_memory_slot *memslot = &kvm->memslots[i];
904 if (gfn >= memslot->base_gfn
905 && gfn < memslot->base_gfn + memslot->npages)
911 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
913 gfn = unalias_gfn(kvm, gfn);
914 return __gfn_to_memslot(kvm, gfn);
917 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
919 struct kvm_memory_slot *slot;
921 gfn = unalias_gfn(kvm, gfn);
922 slot = __gfn_to_memslot(kvm, gfn);
925 return slot->phys_mem[gfn - slot->base_gfn];
927 EXPORT_SYMBOL_GPL(gfn_to_page);
929 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
932 struct kvm_memory_slot *memslot;
933 unsigned long rel_gfn;
935 for (i = 0; i < kvm->nmemslots; ++i) {
936 memslot = &kvm->memslots[i];
938 if (gfn >= memslot->base_gfn
939 && gfn < memslot->base_gfn + memslot->npages) {
941 if (!memslot->dirty_bitmap)
944 rel_gfn = gfn - memslot->base_gfn;
947 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
948 set_bit(rel_gfn, memslot->dirty_bitmap);
954 int emulator_read_std(unsigned long addr,
957 struct kvm_vcpu *vcpu)
962 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
963 unsigned offset = addr & (PAGE_SIZE-1);
964 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
969 if (gpa == UNMAPPED_GVA)
970 return X86EMUL_PROPAGATE_FAULT;
971 pfn = gpa >> PAGE_SHIFT;
972 page = gfn_to_page(vcpu->kvm, pfn);
974 return X86EMUL_UNHANDLEABLE;
975 page_virt = kmap_atomic(page, KM_USER0);
977 memcpy(data, page_virt + offset, tocopy);
979 kunmap_atomic(page_virt, KM_USER0);
986 return X86EMUL_CONTINUE;
988 EXPORT_SYMBOL_GPL(emulator_read_std);
990 static int emulator_write_std(unsigned long addr,
993 struct kvm_vcpu *vcpu)
995 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
997 return X86EMUL_UNHANDLEABLE;
1000 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1004 * Note that its important to have this wrapper function because
1005 * in the very near future we will be checking for MMIOs against
1006 * the LAPIC as well as the general MMIO bus
1008 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1011 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1014 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1017 static int emulator_read_emulated(unsigned long addr,
1020 struct kvm_vcpu *vcpu)
1022 struct kvm_io_device *mmio_dev;
1025 if (vcpu->mmio_read_completed) {
1026 memcpy(val, vcpu->mmio_data, bytes);
1027 vcpu->mmio_read_completed = 0;
1028 return X86EMUL_CONTINUE;
1029 } else if (emulator_read_std(addr, val, bytes, vcpu)
1030 == X86EMUL_CONTINUE)
1031 return X86EMUL_CONTINUE;
1033 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1034 if (gpa == UNMAPPED_GVA)
1035 return X86EMUL_PROPAGATE_FAULT;
1038 * Is this MMIO handled locally?
1040 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1042 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1043 return X86EMUL_CONTINUE;
1046 vcpu->mmio_needed = 1;
1047 vcpu->mmio_phys_addr = gpa;
1048 vcpu->mmio_size = bytes;
1049 vcpu->mmio_is_write = 0;
1051 return X86EMUL_UNHANDLEABLE;
1054 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1055 const void *val, int bytes)
1060 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1062 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1065 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1066 virt = kmap_atomic(page, KM_USER0);
1067 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1068 memcpy(virt + offset_in_page(gpa), val, bytes);
1069 kunmap_atomic(virt, KM_USER0);
1073 static int emulator_write_emulated_onepage(unsigned long addr,
1076 struct kvm_vcpu *vcpu)
1078 struct kvm_io_device *mmio_dev;
1079 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1081 if (gpa == UNMAPPED_GVA) {
1082 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1083 return X86EMUL_PROPAGATE_FAULT;
1086 if (emulator_write_phys(vcpu, gpa, val, bytes))
1087 return X86EMUL_CONTINUE;
1090 * Is this MMIO handled locally?
1092 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1094 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1095 return X86EMUL_CONTINUE;
1098 vcpu->mmio_needed = 1;
1099 vcpu->mmio_phys_addr = gpa;
1100 vcpu->mmio_size = bytes;
1101 vcpu->mmio_is_write = 1;
1102 memcpy(vcpu->mmio_data, val, bytes);
1104 return X86EMUL_CONTINUE;
1107 int emulator_write_emulated(unsigned long addr,
1110 struct kvm_vcpu *vcpu)
1112 /* Crossing a page boundary? */
1113 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1116 now = -addr & ~PAGE_MASK;
1117 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1118 if (rc != X86EMUL_CONTINUE)
1124 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1126 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1128 static int emulator_cmpxchg_emulated(unsigned long addr,
1132 struct kvm_vcpu *vcpu)
1134 static int reported;
1138 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1140 return emulator_write_emulated(addr, new, bytes, vcpu);
1143 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1145 return kvm_arch_ops->get_segment_base(vcpu, seg);
1148 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1150 return X86EMUL_CONTINUE;
1153 int emulate_clts(struct kvm_vcpu *vcpu)
1157 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1158 kvm_arch_ops->set_cr0(vcpu, cr0);
1159 return X86EMUL_CONTINUE;
1162 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1164 struct kvm_vcpu *vcpu = ctxt->vcpu;
1168 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1169 return X86EMUL_CONTINUE;
1171 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1173 return X86EMUL_UNHANDLEABLE;
1177 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1179 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1182 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1184 /* FIXME: better handling */
1185 return X86EMUL_UNHANDLEABLE;
1187 return X86EMUL_CONTINUE;
1190 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1192 static int reported;
1194 unsigned long rip = ctxt->vcpu->rip;
1195 unsigned long rip_linear;
1197 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1202 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1204 printk(KERN_ERR "emulation failed but !mmio_needed?"
1205 " rip %lx %02x %02x %02x %02x\n",
1206 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1210 struct x86_emulate_ops emulate_ops = {
1211 .read_std = emulator_read_std,
1212 .write_std = emulator_write_std,
1213 .read_emulated = emulator_read_emulated,
1214 .write_emulated = emulator_write_emulated,
1215 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1218 int emulate_instruction(struct kvm_vcpu *vcpu,
1219 struct kvm_run *run,
1223 struct x86_emulate_ctxt emulate_ctxt;
1227 vcpu->mmio_fault_cr2 = cr2;
1228 kvm_arch_ops->cache_regs(vcpu);
1230 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1232 emulate_ctxt.vcpu = vcpu;
1233 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1234 emulate_ctxt.cr2 = cr2;
1235 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1236 ? X86EMUL_MODE_REAL : cs_l
1237 ? X86EMUL_MODE_PROT64 : cs_db
1238 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1240 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1241 emulate_ctxt.cs_base = 0;
1242 emulate_ctxt.ds_base = 0;
1243 emulate_ctxt.es_base = 0;
1244 emulate_ctxt.ss_base = 0;
1246 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1247 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1248 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1249 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1252 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1253 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1255 vcpu->mmio_is_write = 0;
1256 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1258 if ((r || vcpu->mmio_is_write) && run) {
1259 run->exit_reason = KVM_EXIT_MMIO;
1260 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1261 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1262 run->mmio.len = vcpu->mmio_size;
1263 run->mmio.is_write = vcpu->mmio_is_write;
1267 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1268 return EMULATE_DONE;
1269 if (!vcpu->mmio_needed) {
1270 report_emulation_failure(&emulate_ctxt);
1271 return EMULATE_FAIL;
1273 return EMULATE_DO_MMIO;
1276 kvm_arch_ops->decache_regs(vcpu);
1277 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1279 if (vcpu->mmio_is_write) {
1280 vcpu->mmio_needed = 0;
1281 return EMULATE_DO_MMIO;
1284 return EMULATE_DONE;
1286 EXPORT_SYMBOL_GPL(emulate_instruction);
1288 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1290 if (vcpu->irq_summary)
1293 vcpu->run->exit_reason = KVM_EXIT_HLT;
1294 ++vcpu->stat.halt_exits;
1297 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1299 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1301 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1303 kvm_arch_ops->cache_regs(vcpu);
1305 #ifdef CONFIG_X86_64
1306 if (is_long_mode(vcpu)) {
1307 nr = vcpu->regs[VCPU_REGS_RAX];
1308 a0 = vcpu->regs[VCPU_REGS_RDI];
1309 a1 = vcpu->regs[VCPU_REGS_RSI];
1310 a2 = vcpu->regs[VCPU_REGS_RDX];
1311 a3 = vcpu->regs[VCPU_REGS_RCX];
1312 a4 = vcpu->regs[VCPU_REGS_R8];
1313 a5 = vcpu->regs[VCPU_REGS_R9];
1317 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1318 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1319 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1320 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1321 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1322 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1323 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1327 run->hypercall.nr = nr;
1328 run->hypercall.args[0] = a0;
1329 run->hypercall.args[1] = a1;
1330 run->hypercall.args[2] = a2;
1331 run->hypercall.args[3] = a3;
1332 run->hypercall.args[4] = a4;
1333 run->hypercall.args[5] = a5;
1334 run->hypercall.ret = ret;
1335 run->hypercall.longmode = is_long_mode(vcpu);
1336 kvm_arch_ops->decache_regs(vcpu);
1339 vcpu->regs[VCPU_REGS_RAX] = ret;
1340 kvm_arch_ops->decache_regs(vcpu);
1343 EXPORT_SYMBOL_GPL(kvm_hypercall);
1345 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1347 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1350 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1352 struct descriptor_table dt = { limit, base };
1354 kvm_arch_ops->set_gdt(vcpu, &dt);
1357 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1359 struct descriptor_table dt = { limit, base };
1361 kvm_arch_ops->set_idt(vcpu, &dt);
1364 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1365 unsigned long *rflags)
1368 *rflags = kvm_arch_ops->get_rflags(vcpu);
1371 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1373 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1384 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1389 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1390 unsigned long *rflags)
1394 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1395 *rflags = kvm_arch_ops->get_rflags(vcpu);
1404 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1407 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1412 * Register the para guest with the host:
1414 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1416 struct kvm_vcpu_para_state *para_state;
1417 hpa_t para_state_hpa, hypercall_hpa;
1418 struct page *para_state_page;
1419 unsigned char *hypercall;
1420 gpa_t hypercall_gpa;
1422 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1423 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1426 * Needs to be page aligned:
1428 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1431 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1432 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1433 if (is_error_hpa(para_state_hpa))
1436 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1437 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1438 para_state = kmap(para_state_page);
1440 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1441 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1443 para_state->host_version = KVM_PARA_API_VERSION;
1445 * We cannot support guests that try to register themselves
1446 * with a newer API version than the host supports:
1448 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1449 para_state->ret = -KVM_EINVAL;
1450 goto err_kunmap_skip;
1453 hypercall_gpa = para_state->hypercall_gpa;
1454 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1455 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1456 if (is_error_hpa(hypercall_hpa)) {
1457 para_state->ret = -KVM_EINVAL;
1458 goto err_kunmap_skip;
1461 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1462 vcpu->para_state_page = para_state_page;
1463 vcpu->para_state_gpa = para_state_gpa;
1464 vcpu->hypercall_gpa = hypercall_gpa;
1466 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1467 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1468 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1469 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1470 kunmap_atomic(hypercall, KM_USER1);
1472 para_state->ret = 0;
1474 kunmap(para_state_page);
1480 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1485 case 0xc0010010: /* SYSCFG */
1486 case 0xc0010015: /* HWCR */
1487 case MSR_IA32_PLATFORM_ID:
1488 case MSR_IA32_P5_MC_ADDR:
1489 case MSR_IA32_P5_MC_TYPE:
1490 case MSR_IA32_MC0_CTL:
1491 case MSR_IA32_MCG_STATUS:
1492 case MSR_IA32_MCG_CAP:
1493 case MSR_IA32_MC0_MISC:
1494 case MSR_IA32_MC0_MISC+4:
1495 case MSR_IA32_MC0_MISC+8:
1496 case MSR_IA32_MC0_MISC+12:
1497 case MSR_IA32_MC0_MISC+16:
1498 case MSR_IA32_UCODE_REV:
1499 case MSR_IA32_PERF_STATUS:
1500 case MSR_IA32_EBL_CR_POWERON:
1501 /* MTRR registers */
1503 case 0x200 ... 0x2ff:
1506 case 0xcd: /* fsb frequency */
1509 case MSR_IA32_APICBASE:
1510 data = vcpu->apic_base;
1512 case MSR_IA32_MISC_ENABLE:
1513 data = vcpu->ia32_misc_enable_msr;
1515 #ifdef CONFIG_X86_64
1517 data = vcpu->shadow_efer;
1521 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1527 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1530 * Reads an msr value (of 'msr_index') into 'pdata'.
1531 * Returns 0 on success, non-0 otherwise.
1532 * Assumes vcpu_load() was already called.
1534 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1536 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1539 #ifdef CONFIG_X86_64
1541 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1543 if (efer & EFER_RESERVED_BITS) {
1544 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1551 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1552 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1557 kvm_arch_ops->set_efer(vcpu, efer);
1560 efer |= vcpu->shadow_efer & EFER_LMA;
1562 vcpu->shadow_efer = efer;
1567 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1570 #ifdef CONFIG_X86_64
1572 set_efer(vcpu, data);
1575 case MSR_IA32_MC0_STATUS:
1576 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1577 __FUNCTION__, data);
1579 case MSR_IA32_MCG_STATUS:
1580 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1581 __FUNCTION__, data);
1583 case MSR_IA32_UCODE_REV:
1584 case MSR_IA32_UCODE_WRITE:
1585 case 0x200 ... 0x2ff: /* MTRRs */
1587 case MSR_IA32_APICBASE:
1588 vcpu->apic_base = data;
1590 case MSR_IA32_MISC_ENABLE:
1591 vcpu->ia32_misc_enable_msr = data;
1594 * This is the 'probe whether the host is KVM' logic:
1596 case MSR_KVM_API_MAGIC:
1597 return vcpu_register_para(vcpu, data);
1600 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1605 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1608 * Writes msr value into into the appropriate "register".
1609 * Returns 0 on success, non-0 otherwise.
1610 * Assumes vcpu_load() was already called.
1612 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1614 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1617 void kvm_resched(struct kvm_vcpu *vcpu)
1619 if (!need_resched())
1623 EXPORT_SYMBOL_GPL(kvm_resched);
1625 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1629 struct kvm_cpuid_entry *e, *best;
1631 kvm_arch_ops->cache_regs(vcpu);
1632 function = vcpu->regs[VCPU_REGS_RAX];
1633 vcpu->regs[VCPU_REGS_RAX] = 0;
1634 vcpu->regs[VCPU_REGS_RBX] = 0;
1635 vcpu->regs[VCPU_REGS_RCX] = 0;
1636 vcpu->regs[VCPU_REGS_RDX] = 0;
1638 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1639 e = &vcpu->cpuid_entries[i];
1640 if (e->function == function) {
1645 * Both basic or both extended?
1647 if (((e->function ^ function) & 0x80000000) == 0)
1648 if (!best || e->function > best->function)
1652 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1653 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1654 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1655 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1657 kvm_arch_ops->decache_regs(vcpu);
1658 kvm_arch_ops->skip_emulated_instruction(vcpu);
1660 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1662 static int pio_copy_data(struct kvm_vcpu *vcpu)
1664 void *p = vcpu->pio_data;
1667 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1669 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1672 free_pio_guest_pages(vcpu);
1675 q += vcpu->pio.guest_page_offset;
1676 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1678 memcpy(q, p, bytes);
1680 memcpy(p, q, bytes);
1681 q -= vcpu->pio.guest_page_offset;
1683 free_pio_guest_pages(vcpu);
1687 static int complete_pio(struct kvm_vcpu *vcpu)
1689 struct kvm_pio_request *io = &vcpu->pio;
1693 kvm_arch_ops->cache_regs(vcpu);
1697 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1701 r = pio_copy_data(vcpu);
1703 kvm_arch_ops->cache_regs(vcpu);
1710 delta *= io->cur_count;
1712 * The size of the register should really depend on
1713 * current address size.
1715 vcpu->regs[VCPU_REGS_RCX] -= delta;
1721 vcpu->regs[VCPU_REGS_RDI] += delta;
1723 vcpu->regs[VCPU_REGS_RSI] += delta;
1726 kvm_arch_ops->decache_regs(vcpu);
1728 io->count -= io->cur_count;
1732 kvm_arch_ops->skip_emulated_instruction(vcpu);
1736 static void kernel_pio(struct kvm_io_device *pio_dev,
1737 struct kvm_vcpu *vcpu,
1740 /* TODO: String I/O for in kernel device */
1743 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1747 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1752 static void pio_string_write(struct kvm_io_device *pio_dev,
1753 struct kvm_vcpu *vcpu)
1755 struct kvm_pio_request *io = &vcpu->pio;
1756 void *pd = vcpu->pio_data;
1759 for (i = 0; i < io->cur_count; i++) {
1760 kvm_iodevice_write(pio_dev, io->port,
1767 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1768 int size, unsigned long count, int string, int down,
1769 gva_t address, int rep, unsigned port)
1771 unsigned now, in_page;
1775 struct kvm_io_device *pio_dev;
1777 vcpu->run->exit_reason = KVM_EXIT_IO;
1778 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1779 vcpu->run->io.size = size;
1780 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1781 vcpu->run->io.count = count;
1782 vcpu->run->io.port = port;
1783 vcpu->pio.count = count;
1784 vcpu->pio.cur_count = count;
1785 vcpu->pio.size = size;
1787 vcpu->pio.port = port;
1788 vcpu->pio.string = string;
1789 vcpu->pio.down = down;
1790 vcpu->pio.guest_page_offset = offset_in_page(address);
1791 vcpu->pio.rep = rep;
1793 pio_dev = vcpu_find_pio_dev(vcpu, port);
1795 kvm_arch_ops->cache_regs(vcpu);
1796 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1797 kvm_arch_ops->decache_regs(vcpu);
1799 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1807 kvm_arch_ops->skip_emulated_instruction(vcpu);
1811 now = min(count, PAGE_SIZE / size);
1814 in_page = PAGE_SIZE - offset_in_page(address);
1816 in_page = offset_in_page(address) + size;
1817 now = min(count, (unsigned long)in_page / size);
1820 * String I/O straddles page boundary. Pin two guest pages
1821 * so that we satisfy atomicity constraints. Do just one
1822 * transaction to avoid complexity.
1829 * String I/O in reverse. Yuck. Kill the guest, fix later.
1831 printk(KERN_ERR "kvm: guest string pio down\n");
1835 vcpu->run->io.count = now;
1836 vcpu->pio.cur_count = now;
1838 for (i = 0; i < nr_pages; ++i) {
1839 mutex_lock(&vcpu->kvm->lock);
1840 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1843 vcpu->pio.guest_pages[i] = page;
1844 mutex_unlock(&vcpu->kvm->lock);
1847 free_pio_guest_pages(vcpu);
1852 if (!vcpu->pio.in) {
1853 /* string PIO write */
1854 ret = pio_copy_data(vcpu);
1855 if (ret >= 0 && pio_dev) {
1856 pio_string_write(pio_dev, vcpu);
1858 if (vcpu->pio.count == 0)
1862 printk(KERN_ERR "no string pio read support yet, "
1863 "port %x size %d count %ld\n",
1868 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1870 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1877 if (vcpu->sigset_active)
1878 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1880 /* re-sync apic's tpr */
1881 vcpu->cr8 = kvm_run->cr8;
1883 if (vcpu->pio.cur_count) {
1884 r = complete_pio(vcpu);
1889 if (vcpu->mmio_needed) {
1890 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1891 vcpu->mmio_read_completed = 1;
1892 vcpu->mmio_needed = 0;
1893 r = emulate_instruction(vcpu, kvm_run,
1894 vcpu->mmio_fault_cr2, 0);
1895 if (r == EMULATE_DO_MMIO) {
1897 * Read-modify-write. Back to userspace.
1904 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1905 kvm_arch_ops->cache_regs(vcpu);
1906 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1907 kvm_arch_ops->decache_regs(vcpu);
1910 r = kvm_arch_ops->run(vcpu, kvm_run);
1913 if (vcpu->sigset_active)
1914 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1920 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1921 struct kvm_regs *regs)
1925 kvm_arch_ops->cache_regs(vcpu);
1927 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1928 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1929 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1930 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1931 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1932 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1933 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1934 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1935 #ifdef CONFIG_X86_64
1936 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1937 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1938 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1939 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1940 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1941 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1942 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1943 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1946 regs->rip = vcpu->rip;
1947 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1950 * Don't leak debug flags in case they were set for guest debugging
1952 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1953 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1960 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1961 struct kvm_regs *regs)
1965 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1966 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1967 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1968 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1969 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1970 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1971 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1972 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1973 #ifdef CONFIG_X86_64
1974 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1975 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1976 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1977 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1978 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1979 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1980 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1981 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1984 vcpu->rip = regs->rip;
1985 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1987 kvm_arch_ops->decache_regs(vcpu);
1994 static void get_segment(struct kvm_vcpu *vcpu,
1995 struct kvm_segment *var, int seg)
1997 return kvm_arch_ops->get_segment(vcpu, var, seg);
2000 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2001 struct kvm_sregs *sregs)
2003 struct descriptor_table dt;
2007 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2008 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2009 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2010 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2011 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2012 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2014 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2015 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2017 kvm_arch_ops->get_idt(vcpu, &dt);
2018 sregs->idt.limit = dt.limit;
2019 sregs->idt.base = dt.base;
2020 kvm_arch_ops->get_gdt(vcpu, &dt);
2021 sregs->gdt.limit = dt.limit;
2022 sregs->gdt.base = dt.base;
2024 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2025 sregs->cr0 = vcpu->cr0;
2026 sregs->cr2 = vcpu->cr2;
2027 sregs->cr3 = vcpu->cr3;
2028 sregs->cr4 = vcpu->cr4;
2029 sregs->cr8 = vcpu->cr8;
2030 sregs->efer = vcpu->shadow_efer;
2031 sregs->apic_base = vcpu->apic_base;
2033 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2034 sizeof sregs->interrupt_bitmap);
2041 static void set_segment(struct kvm_vcpu *vcpu,
2042 struct kvm_segment *var, int seg)
2044 return kvm_arch_ops->set_segment(vcpu, var, seg);
2047 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2048 struct kvm_sregs *sregs)
2050 int mmu_reset_needed = 0;
2052 struct descriptor_table dt;
2056 dt.limit = sregs->idt.limit;
2057 dt.base = sregs->idt.base;
2058 kvm_arch_ops->set_idt(vcpu, &dt);
2059 dt.limit = sregs->gdt.limit;
2060 dt.base = sregs->gdt.base;
2061 kvm_arch_ops->set_gdt(vcpu, &dt);
2063 vcpu->cr2 = sregs->cr2;
2064 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2065 vcpu->cr3 = sregs->cr3;
2067 vcpu->cr8 = sregs->cr8;
2069 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2070 #ifdef CONFIG_X86_64
2071 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2073 vcpu->apic_base = sregs->apic_base;
2075 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2077 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2078 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2080 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2081 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2082 if (!is_long_mode(vcpu) && is_pae(vcpu))
2083 load_pdptrs(vcpu, vcpu->cr3);
2085 if (mmu_reset_needed)
2086 kvm_mmu_reset_context(vcpu);
2088 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2089 sizeof vcpu->irq_pending);
2090 vcpu->irq_summary = 0;
2091 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2092 if (vcpu->irq_pending[i])
2093 __set_bit(i, &vcpu->irq_summary);
2095 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2096 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2097 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2098 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2099 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2100 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2102 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2103 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2111 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2112 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2114 * This list is modified at module load time to reflect the
2115 * capabilities of the host cpu.
2117 static u32 msrs_to_save[] = {
2118 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2120 #ifdef CONFIG_X86_64
2121 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2123 MSR_IA32_TIME_STAMP_COUNTER,
2126 static unsigned num_msrs_to_save;
2128 static u32 emulated_msrs[] = {
2129 MSR_IA32_MISC_ENABLE,
2132 static __init void kvm_init_msr_list(void)
2137 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2138 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2141 msrs_to_save[j] = msrs_to_save[i];
2144 num_msrs_to_save = j;
2148 * Adapt set_msr() to msr_io()'s calling convention
2150 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2152 return kvm_set_msr(vcpu, index, *data);
2156 * Read or write a bunch of msrs. All parameters are kernel addresses.
2158 * @return number of msrs set successfully.
2160 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2161 struct kvm_msr_entry *entries,
2162 int (*do_msr)(struct kvm_vcpu *vcpu,
2163 unsigned index, u64 *data))
2169 for (i = 0; i < msrs->nmsrs; ++i)
2170 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2179 * Read or write a bunch of msrs. Parameters are user addresses.
2181 * @return number of msrs set successfully.
2183 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2184 int (*do_msr)(struct kvm_vcpu *vcpu,
2185 unsigned index, u64 *data),
2188 struct kvm_msrs msrs;
2189 struct kvm_msr_entry *entries;
2194 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2198 if (msrs.nmsrs >= MAX_IO_MSRS)
2202 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2203 entries = vmalloc(size);
2208 if (copy_from_user(entries, user_msrs->entries, size))
2211 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2216 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2228 * Translate a guest virtual address to a guest physical address.
2230 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2231 struct kvm_translation *tr)
2233 unsigned long vaddr = tr->linear_address;
2237 mutex_lock(&vcpu->kvm->lock);
2238 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2239 tr->physical_address = gpa;
2240 tr->valid = gpa != UNMAPPED_GVA;
2243 mutex_unlock(&vcpu->kvm->lock);
2249 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2250 struct kvm_interrupt *irq)
2252 if (irq->irq < 0 || irq->irq >= 256)
2256 set_bit(irq->irq, vcpu->irq_pending);
2257 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2264 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2265 struct kvm_debug_guest *dbg)
2271 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2278 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2279 unsigned long address,
2282 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2283 unsigned long pgoff;
2286 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2288 page = virt_to_page(vcpu->run);
2289 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2290 page = virt_to_page(vcpu->pio_data);
2292 return NOPAGE_SIGBUS;
2295 *type = VM_FAULT_MINOR;
2300 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2301 .nopage = kvm_vcpu_nopage,
2304 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2306 vma->vm_ops = &kvm_vcpu_vm_ops;
2310 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2312 struct kvm_vcpu *vcpu = filp->private_data;
2314 fput(vcpu->kvm->filp);
2318 static struct file_operations kvm_vcpu_fops = {
2319 .release = kvm_vcpu_release,
2320 .unlocked_ioctl = kvm_vcpu_ioctl,
2321 .compat_ioctl = kvm_vcpu_ioctl,
2322 .mmap = kvm_vcpu_mmap,
2326 * Allocates an inode for the vcpu.
2328 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2331 struct inode *inode;
2334 r = anon_inode_getfd(&fd, &inode, &file,
2335 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2338 atomic_inc(&vcpu->kvm->filp->f_count);
2343 * Creates some virtual cpus. Good luck creating more than one.
2345 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2348 struct kvm_vcpu *vcpu;
2353 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2355 return PTR_ERR(vcpu);
2357 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2360 r = kvm_mmu_setup(vcpu);
2365 mutex_lock(&kvm->lock);
2366 if (kvm->vcpus[n]) {
2368 mutex_unlock(&kvm->lock);
2371 kvm->vcpus[n] = vcpu;
2372 mutex_unlock(&kvm->lock);
2374 /* Now it's all set up, let userspace reach it */
2375 r = create_vcpu_fd(vcpu);
2381 mutex_lock(&kvm->lock);
2382 kvm->vcpus[n] = NULL;
2383 mutex_unlock(&kvm->lock);
2387 kvm_mmu_unload(vcpu);
2391 kvm_arch_ops->vcpu_free(vcpu);
2395 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2399 struct kvm_cpuid_entry *e, *entry;
2401 rdmsrl(MSR_EFER, efer);
2403 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2404 e = &vcpu->cpuid_entries[i];
2405 if (e->function == 0x80000001) {
2410 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2411 entry->edx &= ~(1 << 20);
2412 printk(KERN_INFO "kvm: guest NX capability removed\n");
2416 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2417 struct kvm_cpuid *cpuid,
2418 struct kvm_cpuid_entry __user *entries)
2423 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2426 if (copy_from_user(&vcpu->cpuid_entries, entries,
2427 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2429 vcpu->cpuid_nent = cpuid->nent;
2430 cpuid_fix_nx_cap(vcpu);
2437 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2440 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2441 vcpu->sigset_active = 1;
2442 vcpu->sigset = *sigset;
2444 vcpu->sigset_active = 0;
2449 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2450 * we have asm/x86/processor.h
2461 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2462 #ifdef CONFIG_X86_64
2463 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2465 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2469 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2471 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2475 memcpy(fpu->fpr, fxsave->st_space, 128);
2476 fpu->fcw = fxsave->cwd;
2477 fpu->fsw = fxsave->swd;
2478 fpu->ftwx = fxsave->twd;
2479 fpu->last_opcode = fxsave->fop;
2480 fpu->last_ip = fxsave->rip;
2481 fpu->last_dp = fxsave->rdp;
2482 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2489 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2491 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2495 memcpy(fxsave->st_space, fpu->fpr, 128);
2496 fxsave->cwd = fpu->fcw;
2497 fxsave->swd = fpu->fsw;
2498 fxsave->twd = fpu->ftwx;
2499 fxsave->fop = fpu->last_opcode;
2500 fxsave->rip = fpu->last_ip;
2501 fxsave->rdp = fpu->last_dp;
2502 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2509 static long kvm_vcpu_ioctl(struct file *filp,
2510 unsigned int ioctl, unsigned long arg)
2512 struct kvm_vcpu *vcpu = filp->private_data;
2513 void __user *argp = (void __user *)arg;
2521 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2523 case KVM_GET_REGS: {
2524 struct kvm_regs kvm_regs;
2526 memset(&kvm_regs, 0, sizeof kvm_regs);
2527 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2531 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2536 case KVM_SET_REGS: {
2537 struct kvm_regs kvm_regs;
2540 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2542 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2548 case KVM_GET_SREGS: {
2549 struct kvm_sregs kvm_sregs;
2551 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2552 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2556 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2561 case KVM_SET_SREGS: {
2562 struct kvm_sregs kvm_sregs;
2565 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2567 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2573 case KVM_TRANSLATE: {
2574 struct kvm_translation tr;
2577 if (copy_from_user(&tr, argp, sizeof tr))
2579 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2583 if (copy_to_user(argp, &tr, sizeof tr))
2588 case KVM_INTERRUPT: {
2589 struct kvm_interrupt irq;
2592 if (copy_from_user(&irq, argp, sizeof irq))
2594 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2600 case KVM_DEBUG_GUEST: {
2601 struct kvm_debug_guest dbg;
2604 if (copy_from_user(&dbg, argp, sizeof dbg))
2606 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2613 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2616 r = msr_io(vcpu, argp, do_set_msr, 0);
2618 case KVM_SET_CPUID: {
2619 struct kvm_cpuid __user *cpuid_arg = argp;
2620 struct kvm_cpuid cpuid;
2623 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2625 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2630 case KVM_SET_SIGNAL_MASK: {
2631 struct kvm_signal_mask __user *sigmask_arg = argp;
2632 struct kvm_signal_mask kvm_sigmask;
2633 sigset_t sigset, *p;
2638 if (copy_from_user(&kvm_sigmask, argp,
2639 sizeof kvm_sigmask))
2642 if (kvm_sigmask.len != sizeof sigset)
2645 if (copy_from_user(&sigset, sigmask_arg->sigset,
2650 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2656 memset(&fpu, 0, sizeof fpu);
2657 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2661 if (copy_to_user(argp, &fpu, sizeof fpu))
2670 if (copy_from_user(&fpu, argp, sizeof fpu))
2672 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2685 static long kvm_vm_ioctl(struct file *filp,
2686 unsigned int ioctl, unsigned long arg)
2688 struct kvm *kvm = filp->private_data;
2689 void __user *argp = (void __user *)arg;
2693 case KVM_CREATE_VCPU:
2694 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2698 case KVM_SET_MEMORY_REGION: {
2699 struct kvm_memory_region kvm_mem;
2702 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2704 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2709 case KVM_GET_DIRTY_LOG: {
2710 struct kvm_dirty_log log;
2713 if (copy_from_user(&log, argp, sizeof log))
2715 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2720 case KVM_SET_MEMORY_ALIAS: {
2721 struct kvm_memory_alias alias;
2724 if (copy_from_user(&alias, argp, sizeof alias))
2726 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2738 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2739 unsigned long address,
2742 struct kvm *kvm = vma->vm_file->private_data;
2743 unsigned long pgoff;
2746 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2747 page = gfn_to_page(kvm, pgoff);
2749 return NOPAGE_SIGBUS;
2752 *type = VM_FAULT_MINOR;
2757 static struct vm_operations_struct kvm_vm_vm_ops = {
2758 .nopage = kvm_vm_nopage,
2761 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2763 vma->vm_ops = &kvm_vm_vm_ops;
2767 static struct file_operations kvm_vm_fops = {
2768 .release = kvm_vm_release,
2769 .unlocked_ioctl = kvm_vm_ioctl,
2770 .compat_ioctl = kvm_vm_ioctl,
2771 .mmap = kvm_vm_mmap,
2774 static int kvm_dev_ioctl_create_vm(void)
2777 struct inode *inode;
2781 kvm = kvm_create_vm();
2783 return PTR_ERR(kvm);
2784 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2786 kvm_destroy_vm(kvm);
2795 static long kvm_dev_ioctl(struct file *filp,
2796 unsigned int ioctl, unsigned long arg)
2798 void __user *argp = (void __user *)arg;
2802 case KVM_GET_API_VERSION:
2806 r = KVM_API_VERSION;
2812 r = kvm_dev_ioctl_create_vm();
2814 case KVM_GET_MSR_INDEX_LIST: {
2815 struct kvm_msr_list __user *user_msr_list = argp;
2816 struct kvm_msr_list msr_list;
2820 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2823 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2824 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2827 if (n < num_msrs_to_save)
2830 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2831 num_msrs_to_save * sizeof(u32)))
2833 if (copy_to_user(user_msr_list->indices
2834 + num_msrs_to_save * sizeof(u32),
2836 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2841 case KVM_CHECK_EXTENSION:
2843 * No extensions defined at present.
2847 case KVM_GET_VCPU_MMAP_SIZE:
2860 static struct file_operations kvm_chardev_ops = {
2861 .open = kvm_dev_open,
2862 .release = kvm_dev_release,
2863 .unlocked_ioctl = kvm_dev_ioctl,
2864 .compat_ioctl = kvm_dev_ioctl,
2867 static struct miscdevice kvm_dev = {
2874 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2877 static void decache_vcpus_on_cpu(int cpu)
2880 struct kvm_vcpu *vcpu;
2883 spin_lock(&kvm_lock);
2884 list_for_each_entry(vm, &vm_list, vm_list)
2885 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2886 vcpu = vm->vcpus[i];
2890 * If the vcpu is locked, then it is running on some
2891 * other cpu and therefore it is not cached on the
2894 * If it's not locked, check the last cpu it executed
2897 if (mutex_trylock(&vcpu->mutex)) {
2898 if (vcpu->cpu == cpu) {
2899 kvm_arch_ops->vcpu_decache(vcpu);
2902 mutex_unlock(&vcpu->mutex);
2905 spin_unlock(&kvm_lock);
2908 static void hardware_enable(void *junk)
2910 int cpu = raw_smp_processor_id();
2912 if (cpu_isset(cpu, cpus_hardware_enabled))
2914 cpu_set(cpu, cpus_hardware_enabled);
2915 kvm_arch_ops->hardware_enable(NULL);
2918 static void hardware_disable(void *junk)
2920 int cpu = raw_smp_processor_id();
2922 if (!cpu_isset(cpu, cpus_hardware_enabled))
2924 cpu_clear(cpu, cpus_hardware_enabled);
2925 decache_vcpus_on_cpu(cpu);
2926 kvm_arch_ops->hardware_disable(NULL);
2929 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2936 case CPU_DYING_FROZEN:
2937 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2939 hardware_disable(NULL);
2941 case CPU_UP_CANCELED:
2942 case CPU_UP_CANCELED_FROZEN:
2943 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2945 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2948 case CPU_ONLINE_FROZEN:
2949 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2951 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2957 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2960 if (val == SYS_RESTART) {
2962 * Some (well, at least mine) BIOSes hang on reboot if
2965 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2966 on_each_cpu(hardware_disable, NULL, 0, 1);
2971 static struct notifier_block kvm_reboot_notifier = {
2972 .notifier_call = kvm_reboot,
2976 void kvm_io_bus_init(struct kvm_io_bus *bus)
2978 memset(bus, 0, sizeof(*bus));
2981 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2985 for (i = 0; i < bus->dev_count; i++) {
2986 struct kvm_io_device *pos = bus->devs[i];
2988 kvm_iodevice_destructor(pos);
2992 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2996 for (i = 0; i < bus->dev_count; i++) {
2997 struct kvm_io_device *pos = bus->devs[i];
2999 if (pos->in_range(pos, addr))
3006 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3008 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3010 bus->devs[bus->dev_count++] = dev;
3013 static struct notifier_block kvm_cpu_notifier = {
3014 .notifier_call = kvm_cpu_hotplug,
3015 .priority = 20, /* must be > scheduler priority */
3018 static u64 stat_get(void *_offset)
3020 unsigned offset = (long)_offset;
3023 struct kvm_vcpu *vcpu;
3026 spin_lock(&kvm_lock);
3027 list_for_each_entry(kvm, &vm_list, vm_list)
3028 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3029 vcpu = kvm->vcpus[i];
3031 total += *(u32 *)((void *)vcpu + offset);
3033 spin_unlock(&kvm_lock);
3037 static void stat_set(void *offset, u64 val)
3041 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3043 static __init void kvm_init_debug(void)
3045 struct kvm_stats_debugfs_item *p;
3047 debugfs_dir = debugfs_create_dir("kvm", NULL);
3048 for (p = debugfs_entries; p->name; ++p)
3049 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3050 (void *)(long)p->offset,
3054 static void kvm_exit_debug(void)
3056 struct kvm_stats_debugfs_item *p;
3058 for (p = debugfs_entries; p->name; ++p)
3059 debugfs_remove(p->dentry);
3060 debugfs_remove(debugfs_dir);
3063 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3065 hardware_disable(NULL);
3069 static int kvm_resume(struct sys_device *dev)
3071 hardware_enable(NULL);
3075 static struct sysdev_class kvm_sysdev_class = {
3076 set_kset_name("kvm"),
3077 .suspend = kvm_suspend,
3078 .resume = kvm_resume,
3081 static struct sys_device kvm_sysdev = {
3083 .cls = &kvm_sysdev_class,
3086 hpa_t bad_page_address;
3089 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3091 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3094 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3096 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3098 kvm_arch_ops->vcpu_load(vcpu, cpu);
3101 static void kvm_sched_out(struct preempt_notifier *pn,
3102 struct task_struct *next)
3104 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3106 kvm_arch_ops->vcpu_put(vcpu);
3109 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3110 struct module *module)
3115 printk(KERN_ERR "kvm: already loaded the other module\n");
3119 if (!ops->cpu_has_kvm_support()) {
3120 printk(KERN_ERR "kvm: no hardware support\n");
3123 if (ops->disabled_by_bios()) {
3124 printk(KERN_ERR "kvm: disabled by bios\n");
3130 r = kvm_arch_ops->hardware_setup();
3134 on_each_cpu(hardware_enable, NULL, 0, 1);
3135 r = register_cpu_notifier(&kvm_cpu_notifier);
3138 register_reboot_notifier(&kvm_reboot_notifier);
3140 r = sysdev_class_register(&kvm_sysdev_class);
3144 r = sysdev_register(&kvm_sysdev);
3148 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3149 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3150 __alignof__(struct kvm_vcpu), 0, 0);
3151 if (!kvm_vcpu_cache) {
3156 kvm_chardev_ops.owner = module;
3158 r = misc_register(&kvm_dev);
3160 printk (KERN_ERR "kvm: misc device register failed\n");
3164 kvm_preempt_ops.sched_in = kvm_sched_in;
3165 kvm_preempt_ops.sched_out = kvm_sched_out;
3170 kmem_cache_destroy(kvm_vcpu_cache);
3172 sysdev_unregister(&kvm_sysdev);
3174 sysdev_class_unregister(&kvm_sysdev_class);
3176 unregister_reboot_notifier(&kvm_reboot_notifier);
3177 unregister_cpu_notifier(&kvm_cpu_notifier);
3179 on_each_cpu(hardware_disable, NULL, 0, 1);
3180 kvm_arch_ops->hardware_unsetup();
3182 kvm_arch_ops = NULL;
3186 void kvm_exit_arch(void)
3188 misc_deregister(&kvm_dev);
3189 kmem_cache_destroy(kvm_vcpu_cache);
3190 sysdev_unregister(&kvm_sysdev);
3191 sysdev_class_unregister(&kvm_sysdev_class);
3192 unregister_reboot_notifier(&kvm_reboot_notifier);
3193 unregister_cpu_notifier(&kvm_cpu_notifier);
3194 on_each_cpu(hardware_disable, NULL, 0, 1);
3195 kvm_arch_ops->hardware_unsetup();
3196 kvm_arch_ops = NULL;
3199 static __init int kvm_init(void)
3201 static struct page *bad_page;
3204 r = kvm_mmu_module_init();
3210 kvm_init_msr_list();
3212 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3217 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3218 memset(__va(bad_page_address), 0, PAGE_SIZE);
3224 kvm_mmu_module_exit();
3229 static __exit void kvm_exit(void)
3232 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3233 kvm_mmu_module_exit();
3236 module_init(kvm_init)
3237 module_exit(kvm_exit)
3239 EXPORT_SYMBOL_GPL(kvm_init_arch);
3240 EXPORT_SYMBOL_GPL(kvm_exit_arch);