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.
20 #include <linux/kvm.h>
21 #include <linux/module.h>
22 #include <linux/errno.h>
23 #include <linux/magic.h>
24 #include <asm/processor.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <asm/uaccess.h>
32 #include <linux/reboot.h>
34 #include <linux/debugfs.h>
35 #include <linux/highmem.h>
36 #include <linux/file.h>
38 #include <linux/sysdev.h>
39 #include <linux/cpu.h>
40 #include <linux/file.h>
42 #include <linux/mount.h>
43 #include <linux/sched.h>
45 #include "x86_emulate.h"
46 #include "segment_descriptor.h"
48 MODULE_AUTHOR("Qumranet");
49 MODULE_LICENSE("GPL");
51 static DEFINE_SPINLOCK(kvm_lock);
52 static LIST_HEAD(vm_list);
54 struct kvm_arch_ops *kvm_arch_ops;
56 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
58 static struct kvm_stats_debugfs_item {
61 struct dentry *dentry;
62 } debugfs_entries[] = {
63 { "pf_fixed", STAT_OFFSET(pf_fixed) },
64 { "pf_guest", STAT_OFFSET(pf_guest) },
65 { "tlb_flush", STAT_OFFSET(tlb_flush) },
66 { "invlpg", STAT_OFFSET(invlpg) },
67 { "exits", STAT_OFFSET(exits) },
68 { "io_exits", STAT_OFFSET(io_exits) },
69 { "mmio_exits", STAT_OFFSET(mmio_exits) },
70 { "signal_exits", STAT_OFFSET(signal_exits) },
71 { "irq_window", STAT_OFFSET(irq_window_exits) },
72 { "halt_exits", STAT_OFFSET(halt_exits) },
73 { "request_irq", STAT_OFFSET(request_irq_exits) },
74 { "irq_exits", STAT_OFFSET(irq_exits) },
78 static struct dentry *debugfs_dir;
80 struct vfsmount *kvmfs_mnt;
82 #define MAX_IO_MSRS 256
84 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
85 #define LMSW_GUEST_MASK 0x0eULL
86 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
87 #define CR8_RESEVED_BITS (~0x0fULL)
88 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
91 // LDT or TSS descriptor in the GDT. 16 bytes.
92 struct segment_descriptor_64 {
93 struct segment_descriptor s;
100 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
103 static struct inode *kvmfs_inode(struct file_operations *fops)
106 struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
114 * Mark the inode dirty from the very beginning,
115 * that way it will never be moved to the dirty
116 * list because mark_inode_dirty() will think
117 * that it already _is_ on the dirty list.
119 inode->i_state = I_DIRTY;
120 inode->i_mode = S_IRUSR | S_IWUSR;
121 inode->i_uid = current->fsuid;
122 inode->i_gid = current->fsgid;
123 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
127 return ERR_PTR(error);
130 static struct file *kvmfs_file(struct inode *inode, void *private_data)
132 struct file *file = get_empty_filp();
135 return ERR_PTR(-ENFILE);
137 file->f_path.mnt = mntget(kvmfs_mnt);
138 file->f_path.dentry = d_alloc_anon(inode);
139 if (!file->f_path.dentry)
140 return ERR_PTR(-ENOMEM);
141 file->f_mapping = inode->i_mapping;
144 file->f_flags = O_RDWR;
145 file->f_op = inode->i_fop;
146 file->f_mode = FMODE_READ | FMODE_WRITE;
148 file->private_data = private_data;
152 unsigned long segment_base(u16 selector)
154 struct descriptor_table gdt;
155 struct segment_descriptor *d;
156 unsigned long table_base;
157 typedef unsigned long ul;
163 asm ("sgdt %0" : "=m"(gdt));
164 table_base = gdt.base;
166 if (selector & 4) { /* from ldt */
169 asm ("sldt %0" : "=g"(ldt_selector));
170 table_base = segment_base(ldt_selector);
172 d = (struct segment_descriptor *)(table_base + (selector & ~7));
173 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
176 && (d->type == 2 || d->type == 9 || d->type == 11))
177 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
181 EXPORT_SYMBOL_GPL(segment_base);
183 static inline int valid_vcpu(int n)
185 return likely(n >= 0 && n < KVM_MAX_VCPUS);
188 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
191 unsigned char *host_buf = dest;
192 unsigned long req_size = size;
200 paddr = gva_to_hpa(vcpu, addr);
202 if (is_error_hpa(paddr))
205 guest_buf = (hva_t)kmap_atomic(
206 pfn_to_page(paddr >> PAGE_SHIFT),
208 offset = addr & ~PAGE_MASK;
210 now = min(size, PAGE_SIZE - offset);
211 memcpy(host_buf, (void*)guest_buf, now);
215 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
217 return req_size - size;
219 EXPORT_SYMBOL_GPL(kvm_read_guest);
221 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
224 unsigned char *host_buf = data;
225 unsigned long req_size = size;
234 paddr = gva_to_hpa(vcpu, addr);
236 if (is_error_hpa(paddr))
239 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
240 mark_page_dirty(vcpu->kvm, gfn);
241 guest_buf = (hva_t)kmap_atomic(
242 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
243 offset = addr & ~PAGE_MASK;
245 now = min(size, PAGE_SIZE - offset);
246 memcpy((void*)guest_buf, host_buf, now);
250 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
252 return req_size - size;
254 EXPORT_SYMBOL_GPL(kvm_write_guest);
256 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
258 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
261 vcpu->guest_fpu_loaded = 1;
262 fx_save(vcpu->host_fx_image);
263 fx_restore(vcpu->guest_fx_image);
265 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
267 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
269 if (!vcpu->guest_fpu_loaded)
272 vcpu->guest_fpu_loaded = 0;
273 fx_save(vcpu->guest_fx_image);
274 fx_restore(vcpu->host_fx_image);
276 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
279 * Switches to specified vcpu, until a matching vcpu_put()
281 static void vcpu_load(struct kvm_vcpu *vcpu)
283 mutex_lock(&vcpu->mutex);
284 kvm_arch_ops->vcpu_load(vcpu);
288 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
289 * if the slot is not populated.
291 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
293 struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
295 mutex_lock(&vcpu->mutex);
297 mutex_unlock(&vcpu->mutex);
300 kvm_arch_ops->vcpu_load(vcpu);
304 static void vcpu_put(struct kvm_vcpu *vcpu)
306 kvm_arch_ops->vcpu_put(vcpu);
307 mutex_unlock(&vcpu->mutex);
310 static struct kvm *kvm_create_vm(void)
312 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
316 return ERR_PTR(-ENOMEM);
318 spin_lock_init(&kvm->lock);
319 INIT_LIST_HEAD(&kvm->active_mmu_pages);
320 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
321 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
323 mutex_init(&vcpu->mutex);
326 vcpu->mmu.root_hpa = INVALID_PAGE;
327 INIT_LIST_HEAD(&vcpu->free_pages);
328 spin_lock(&kvm_lock);
329 list_add(&kvm->vm_list, &vm_list);
330 spin_unlock(&kvm_lock);
335 static int kvm_dev_open(struct inode *inode, struct file *filp)
341 * Free any memory in @free but not in @dont.
343 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
344 struct kvm_memory_slot *dont)
348 if (!dont || free->phys_mem != dont->phys_mem)
349 if (free->phys_mem) {
350 for (i = 0; i < free->npages; ++i)
351 if (free->phys_mem[i])
352 __free_page(free->phys_mem[i]);
353 vfree(free->phys_mem);
356 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
357 vfree(free->dirty_bitmap);
359 free->phys_mem = NULL;
361 free->dirty_bitmap = NULL;
364 static void kvm_free_physmem(struct kvm *kvm)
368 for (i = 0; i < kvm->nmemslots; ++i)
369 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
372 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
376 for (i = 0; i < 2; ++i)
377 if (vcpu->pio.guest_pages[i]) {
378 __free_page(vcpu->pio.guest_pages[i]);
379 vcpu->pio.guest_pages[i] = NULL;
383 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
389 kvm_mmu_destroy(vcpu);
391 kvm_arch_ops->vcpu_free(vcpu);
392 free_page((unsigned long)vcpu->run);
394 free_page((unsigned long)vcpu->pio_data);
395 vcpu->pio_data = NULL;
396 free_pio_guest_pages(vcpu);
399 static void kvm_free_vcpus(struct kvm *kvm)
403 for (i = 0; i < KVM_MAX_VCPUS; ++i)
404 kvm_free_vcpu(&kvm->vcpus[i]);
407 static int kvm_dev_release(struct inode *inode, struct file *filp)
412 static void kvm_destroy_vm(struct kvm *kvm)
414 spin_lock(&kvm_lock);
415 list_del(&kvm->vm_list);
416 spin_unlock(&kvm_lock);
418 kvm_free_physmem(kvm);
422 static int kvm_vm_release(struct inode *inode, struct file *filp)
424 struct kvm *kvm = filp->private_data;
430 static void inject_gp(struct kvm_vcpu *vcpu)
432 kvm_arch_ops->inject_gp(vcpu, 0);
436 * Load the pae pdptrs. Return true is they are all valid.
438 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
440 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
441 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
448 spin_lock(&vcpu->kvm->lock);
449 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
450 /* FIXME: !page - emulate? 0xff? */
451 pdpt = kmap_atomic(page, KM_USER0);
454 for (i = 0; i < 4; ++i) {
455 pdpte = pdpt[offset + i];
456 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
462 for (i = 0; i < 4; ++i)
463 vcpu->pdptrs[i] = pdpt[offset + i];
466 kunmap_atomic(pdpt, KM_USER0);
467 spin_unlock(&vcpu->kvm->lock);
472 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
474 if (cr0 & CR0_RESEVED_BITS) {
475 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
481 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
482 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
487 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
488 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
489 "and a clear PE flag\n");
494 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
496 if ((vcpu->shadow_efer & EFER_LME)) {
500 printk(KERN_DEBUG "set_cr0: #GP, start paging "
501 "in long mode while PAE is disabled\n");
505 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
507 printk(KERN_DEBUG "set_cr0: #GP, start paging "
508 "in long mode while CS.L == 1\n");
515 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
516 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
524 kvm_arch_ops->set_cr0(vcpu, cr0);
527 spin_lock(&vcpu->kvm->lock);
528 kvm_mmu_reset_context(vcpu);
529 spin_unlock(&vcpu->kvm->lock);
532 EXPORT_SYMBOL_GPL(set_cr0);
534 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
536 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
538 EXPORT_SYMBOL_GPL(lmsw);
540 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
542 if (cr4 & CR4_RESEVED_BITS) {
543 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
548 if (is_long_mode(vcpu)) {
549 if (!(cr4 & CR4_PAE_MASK)) {
550 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
555 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
556 && !load_pdptrs(vcpu, vcpu->cr3)) {
557 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
561 if (cr4 & CR4_VMXE_MASK) {
562 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
566 kvm_arch_ops->set_cr4(vcpu, cr4);
567 spin_lock(&vcpu->kvm->lock);
568 kvm_mmu_reset_context(vcpu);
569 spin_unlock(&vcpu->kvm->lock);
571 EXPORT_SYMBOL_GPL(set_cr4);
573 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
575 if (is_long_mode(vcpu)) {
576 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
577 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
582 if (cr3 & CR3_RESEVED_BITS) {
583 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
587 if (is_paging(vcpu) && is_pae(vcpu) &&
588 !load_pdptrs(vcpu, cr3)) {
589 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
597 spin_lock(&vcpu->kvm->lock);
599 * Does the new cr3 value map to physical memory? (Note, we
600 * catch an invalid cr3 even in real-mode, because it would
601 * cause trouble later on when we turn on paging anyway.)
603 * A real CPU would silently accept an invalid cr3 and would
604 * attempt to use it - with largely undefined (and often hard
605 * to debug) behavior on the guest side.
607 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
610 vcpu->mmu.new_cr3(vcpu);
611 spin_unlock(&vcpu->kvm->lock);
613 EXPORT_SYMBOL_GPL(set_cr3);
615 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
617 if ( cr8 & CR8_RESEVED_BITS) {
618 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
624 EXPORT_SYMBOL_GPL(set_cr8);
626 void fx_init(struct kvm_vcpu *vcpu)
628 struct __attribute__ ((__packed__)) fx_image_s {
634 u64 operand;// fpu dp
640 fx_save(vcpu->host_fx_image);
642 fx_save(vcpu->guest_fx_image);
643 fx_restore(vcpu->host_fx_image);
645 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
646 fx_image->mxcsr = 0x1f80;
647 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
648 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
650 EXPORT_SYMBOL_GPL(fx_init);
652 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
654 spin_lock(&vcpu->kvm->lock);
655 kvm_mmu_slot_remove_write_access(vcpu, slot);
656 spin_unlock(&vcpu->kvm->lock);
660 * Allocate some memory and give it an address in the guest physical address
663 * Discontiguous memory is allowed, mostly for framebuffers.
665 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
666 struct kvm_memory_region *mem)
670 unsigned long npages;
672 struct kvm_memory_slot *memslot;
673 struct kvm_memory_slot old, new;
674 int memory_config_version;
677 /* General sanity checks */
678 if (mem->memory_size & (PAGE_SIZE - 1))
680 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
682 if (mem->slot >= KVM_MEMORY_SLOTS)
684 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
687 memslot = &kvm->memslots[mem->slot];
688 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
689 npages = mem->memory_size >> PAGE_SHIFT;
692 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
695 spin_lock(&kvm->lock);
697 memory_config_version = kvm->memory_config_version;
698 new = old = *memslot;
700 new.base_gfn = base_gfn;
702 new.flags = mem->flags;
704 /* Disallow changing a memory slot's size. */
706 if (npages && old.npages && npages != old.npages)
709 /* Check for overlaps */
711 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
712 struct kvm_memory_slot *s = &kvm->memslots[i];
716 if (!((base_gfn + npages <= s->base_gfn) ||
717 (base_gfn >= s->base_gfn + s->npages)))
721 * Do memory allocations outside lock. memory_config_version will
724 spin_unlock(&kvm->lock);
726 /* Deallocate if slot is being removed */
730 /* Free page dirty bitmap if unneeded */
731 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
732 new.dirty_bitmap = NULL;
736 /* Allocate if a slot is being created */
737 if (npages && !new.phys_mem) {
738 new.phys_mem = vmalloc(npages * sizeof(struct page *));
743 memset(new.phys_mem, 0, npages * sizeof(struct page *));
744 for (i = 0; i < npages; ++i) {
745 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
747 if (!new.phys_mem[i])
749 set_page_private(new.phys_mem[i],0);
753 /* Allocate page dirty bitmap if needed */
754 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
755 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
757 new.dirty_bitmap = vmalloc(dirty_bytes);
758 if (!new.dirty_bitmap)
760 memset(new.dirty_bitmap, 0, dirty_bytes);
763 spin_lock(&kvm->lock);
765 if (memory_config_version != kvm->memory_config_version) {
766 spin_unlock(&kvm->lock);
767 kvm_free_physmem_slot(&new, &old);
775 if (mem->slot >= kvm->nmemslots)
776 kvm->nmemslots = mem->slot + 1;
779 ++kvm->memory_config_version;
781 spin_unlock(&kvm->lock);
783 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
784 struct kvm_vcpu *vcpu;
786 vcpu = vcpu_load_slot(kvm, i);
789 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
790 do_remove_write_access(vcpu, mem->slot);
791 kvm_mmu_reset_context(vcpu);
795 kvm_free_physmem_slot(&old, &new);
799 spin_unlock(&kvm->lock);
801 kvm_free_physmem_slot(&new, &old);
807 * Get (and clear) the dirty memory log for a memory slot.
809 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
810 struct kvm_dirty_log *log)
812 struct kvm_memory_slot *memslot;
816 unsigned long any = 0;
818 spin_lock(&kvm->lock);
821 * Prevent changes to guest memory configuration even while the lock
825 spin_unlock(&kvm->lock);
827 if (log->slot >= KVM_MEMORY_SLOTS)
830 memslot = &kvm->memslots[log->slot];
832 if (!memslot->dirty_bitmap)
835 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
837 for (i = 0; !any && i < n/sizeof(long); ++i)
838 any = memslot->dirty_bitmap[i];
841 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
846 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
847 struct kvm_vcpu *vcpu;
849 vcpu = vcpu_load_slot(kvm, i);
853 do_remove_write_access(vcpu, log->slot);
854 memset(memslot->dirty_bitmap, 0, n);
857 kvm_arch_ops->tlb_flush(vcpu);
865 spin_lock(&kvm->lock);
867 spin_unlock(&kvm->lock);
872 * Set a new alias region. Aliases map a portion of physical memory into
873 * another portion. This is useful for memory windows, for example the PC
876 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
877 struct kvm_memory_alias *alias)
880 struct kvm_mem_alias *p;
883 /* General sanity checks */
884 if (alias->memory_size & (PAGE_SIZE - 1))
886 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
888 if (alias->slot >= KVM_ALIAS_SLOTS)
890 if (alias->guest_phys_addr + alias->memory_size
891 < alias->guest_phys_addr)
893 if (alias->target_phys_addr + alias->memory_size
894 < alias->target_phys_addr)
897 spin_lock(&kvm->lock);
899 p = &kvm->aliases[alias->slot];
900 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
901 p->npages = alias->memory_size >> PAGE_SHIFT;
902 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
904 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
905 if (kvm->aliases[n - 1].npages)
909 spin_unlock(&kvm->lock);
911 vcpu_load(&kvm->vcpus[0]);
912 spin_lock(&kvm->lock);
913 kvm_mmu_zap_all(&kvm->vcpus[0]);
914 spin_unlock(&kvm->lock);
915 vcpu_put(&kvm->vcpus[0]);
923 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
926 struct kvm_mem_alias *alias;
928 for (i = 0; i < kvm->naliases; ++i) {
929 alias = &kvm->aliases[i];
930 if (gfn >= alias->base_gfn
931 && gfn < alias->base_gfn + alias->npages)
932 return alias->target_gfn + gfn - alias->base_gfn;
937 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
941 for (i = 0; i < kvm->nmemslots; ++i) {
942 struct kvm_memory_slot *memslot = &kvm->memslots[i];
944 if (gfn >= memslot->base_gfn
945 && gfn < memslot->base_gfn + memslot->npages)
951 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
953 gfn = unalias_gfn(kvm, gfn);
954 return __gfn_to_memslot(kvm, gfn);
957 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
959 struct kvm_memory_slot *slot;
961 gfn = unalias_gfn(kvm, gfn);
962 slot = __gfn_to_memslot(kvm, gfn);
965 return slot->phys_mem[gfn - slot->base_gfn];
967 EXPORT_SYMBOL_GPL(gfn_to_page);
969 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
972 struct kvm_memory_slot *memslot = NULL;
973 unsigned long rel_gfn;
975 for (i = 0; i < kvm->nmemslots; ++i) {
976 memslot = &kvm->memslots[i];
978 if (gfn >= memslot->base_gfn
979 && gfn < memslot->base_gfn + memslot->npages) {
981 if (!memslot || !memslot->dirty_bitmap)
984 rel_gfn = gfn - memslot->base_gfn;
987 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
988 set_bit(rel_gfn, memslot->dirty_bitmap);
994 static int emulator_read_std(unsigned long addr,
997 struct x86_emulate_ctxt *ctxt)
999 struct kvm_vcpu *vcpu = ctxt->vcpu;
1003 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1004 unsigned offset = addr & (PAGE_SIZE-1);
1005 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1010 if (gpa == UNMAPPED_GVA)
1011 return X86EMUL_PROPAGATE_FAULT;
1012 pfn = gpa >> PAGE_SHIFT;
1013 page = gfn_to_page(vcpu->kvm, pfn);
1015 return X86EMUL_UNHANDLEABLE;
1016 page_virt = kmap_atomic(page, KM_USER0);
1018 memcpy(data, page_virt + offset, tocopy);
1020 kunmap_atomic(page_virt, KM_USER0);
1027 return X86EMUL_CONTINUE;
1030 static int emulator_write_std(unsigned long addr,
1033 struct x86_emulate_ctxt *ctxt)
1035 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1037 return X86EMUL_UNHANDLEABLE;
1040 static int emulator_read_emulated(unsigned long addr,
1043 struct x86_emulate_ctxt *ctxt)
1045 struct kvm_vcpu *vcpu = ctxt->vcpu;
1047 if (vcpu->mmio_read_completed) {
1048 memcpy(val, vcpu->mmio_data, bytes);
1049 vcpu->mmio_read_completed = 0;
1050 return X86EMUL_CONTINUE;
1051 } else if (emulator_read_std(addr, val, bytes, ctxt)
1052 == X86EMUL_CONTINUE)
1053 return X86EMUL_CONTINUE;
1055 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1057 if (gpa == UNMAPPED_GVA)
1058 return X86EMUL_PROPAGATE_FAULT;
1059 vcpu->mmio_needed = 1;
1060 vcpu->mmio_phys_addr = gpa;
1061 vcpu->mmio_size = bytes;
1062 vcpu->mmio_is_write = 0;
1064 return X86EMUL_UNHANDLEABLE;
1068 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1069 const void *val, int bytes)
1074 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1076 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1079 kvm_mmu_pre_write(vcpu, gpa, bytes);
1080 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1081 virt = kmap_atomic(page, KM_USER0);
1082 memcpy(virt + offset_in_page(gpa), val, bytes);
1083 kunmap_atomic(virt, KM_USER0);
1084 kvm_mmu_post_write(vcpu, gpa, bytes);
1088 static int emulator_write_emulated(unsigned long addr,
1091 struct x86_emulate_ctxt *ctxt)
1093 struct kvm_vcpu *vcpu = ctxt->vcpu;
1094 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1096 if (gpa == UNMAPPED_GVA) {
1097 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1098 return X86EMUL_PROPAGATE_FAULT;
1101 if (emulator_write_phys(vcpu, gpa, val, bytes))
1102 return X86EMUL_CONTINUE;
1104 vcpu->mmio_needed = 1;
1105 vcpu->mmio_phys_addr = gpa;
1106 vcpu->mmio_size = bytes;
1107 vcpu->mmio_is_write = 1;
1108 memcpy(vcpu->mmio_data, val, bytes);
1110 return X86EMUL_CONTINUE;
1113 static int emulator_cmpxchg_emulated(unsigned long addr,
1117 struct x86_emulate_ctxt *ctxt)
1119 static int reported;
1123 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1125 return emulator_write_emulated(addr, new, bytes, ctxt);
1128 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1130 return kvm_arch_ops->get_segment_base(vcpu, seg);
1133 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1135 return X86EMUL_CONTINUE;
1138 int emulate_clts(struct kvm_vcpu *vcpu)
1142 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1143 kvm_arch_ops->set_cr0(vcpu, cr0);
1144 return X86EMUL_CONTINUE;
1147 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1149 struct kvm_vcpu *vcpu = ctxt->vcpu;
1153 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1154 return X86EMUL_CONTINUE;
1156 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1158 return X86EMUL_UNHANDLEABLE;
1162 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1164 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1167 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1169 /* FIXME: better handling */
1170 return X86EMUL_UNHANDLEABLE;
1172 return X86EMUL_CONTINUE;
1175 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1177 static int reported;
1179 unsigned long rip = ctxt->vcpu->rip;
1180 unsigned long rip_linear;
1182 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1187 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1189 printk(KERN_ERR "emulation failed but !mmio_needed?"
1190 " rip %lx %02x %02x %02x %02x\n",
1191 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1195 struct x86_emulate_ops emulate_ops = {
1196 .read_std = emulator_read_std,
1197 .write_std = emulator_write_std,
1198 .read_emulated = emulator_read_emulated,
1199 .write_emulated = emulator_write_emulated,
1200 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1203 int emulate_instruction(struct kvm_vcpu *vcpu,
1204 struct kvm_run *run,
1208 struct x86_emulate_ctxt emulate_ctxt;
1212 vcpu->mmio_fault_cr2 = cr2;
1213 kvm_arch_ops->cache_regs(vcpu);
1215 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1217 emulate_ctxt.vcpu = vcpu;
1218 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1219 emulate_ctxt.cr2 = cr2;
1220 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1221 ? X86EMUL_MODE_REAL : cs_l
1222 ? X86EMUL_MODE_PROT64 : cs_db
1223 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1225 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1226 emulate_ctxt.cs_base = 0;
1227 emulate_ctxt.ds_base = 0;
1228 emulate_ctxt.es_base = 0;
1229 emulate_ctxt.ss_base = 0;
1231 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1232 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1233 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1234 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1237 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1238 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1240 vcpu->mmio_is_write = 0;
1241 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1243 if ((r || vcpu->mmio_is_write) && run) {
1244 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1245 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1246 run->mmio.len = vcpu->mmio_size;
1247 run->mmio.is_write = vcpu->mmio_is_write;
1251 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1252 return EMULATE_DONE;
1253 if (!vcpu->mmio_needed) {
1254 report_emulation_failure(&emulate_ctxt);
1255 return EMULATE_FAIL;
1257 return EMULATE_DO_MMIO;
1260 kvm_arch_ops->decache_regs(vcpu);
1261 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1263 if (vcpu->mmio_is_write) {
1264 vcpu->mmio_needed = 0;
1265 return EMULATE_DO_MMIO;
1268 return EMULATE_DONE;
1270 EXPORT_SYMBOL_GPL(emulate_instruction);
1272 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1274 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1276 kvm_arch_ops->cache_regs(vcpu);
1278 #ifdef CONFIG_X86_64
1279 if (is_long_mode(vcpu)) {
1280 nr = vcpu->regs[VCPU_REGS_RAX];
1281 a0 = vcpu->regs[VCPU_REGS_RDI];
1282 a1 = vcpu->regs[VCPU_REGS_RSI];
1283 a2 = vcpu->regs[VCPU_REGS_RDX];
1284 a3 = vcpu->regs[VCPU_REGS_RCX];
1285 a4 = vcpu->regs[VCPU_REGS_R8];
1286 a5 = vcpu->regs[VCPU_REGS_R9];
1290 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1291 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1292 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1293 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1294 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1295 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1296 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1300 run->hypercall.args[0] = a0;
1301 run->hypercall.args[1] = a1;
1302 run->hypercall.args[2] = a2;
1303 run->hypercall.args[3] = a3;
1304 run->hypercall.args[4] = a4;
1305 run->hypercall.args[5] = a5;
1306 run->hypercall.ret = ret;
1307 run->hypercall.longmode = is_long_mode(vcpu);
1308 kvm_arch_ops->decache_regs(vcpu);
1311 vcpu->regs[VCPU_REGS_RAX] = ret;
1312 kvm_arch_ops->decache_regs(vcpu);
1315 EXPORT_SYMBOL_GPL(kvm_hypercall);
1317 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1319 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1322 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1324 struct descriptor_table dt = { limit, base };
1326 kvm_arch_ops->set_gdt(vcpu, &dt);
1329 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1331 struct descriptor_table dt = { limit, base };
1333 kvm_arch_ops->set_idt(vcpu, &dt);
1336 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1337 unsigned long *rflags)
1340 *rflags = kvm_arch_ops->get_rflags(vcpu);
1343 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1345 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1356 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1361 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1362 unsigned long *rflags)
1366 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1367 *rflags = kvm_arch_ops->get_rflags(vcpu);
1376 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1379 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1384 * Register the para guest with the host:
1386 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1388 struct kvm_vcpu_para_state *para_state;
1389 hpa_t para_state_hpa, hypercall_hpa;
1390 struct page *para_state_page;
1391 unsigned char *hypercall;
1392 gpa_t hypercall_gpa;
1394 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1395 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1398 * Needs to be page aligned:
1400 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1403 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1404 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1405 if (is_error_hpa(para_state_hpa))
1408 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1409 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1410 para_state = kmap_atomic(para_state_page, KM_USER0);
1412 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1413 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1415 para_state->host_version = KVM_PARA_API_VERSION;
1417 * We cannot support guests that try to register themselves
1418 * with a newer API version than the host supports:
1420 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1421 para_state->ret = -KVM_EINVAL;
1422 goto err_kunmap_skip;
1425 hypercall_gpa = para_state->hypercall_gpa;
1426 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1427 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1428 if (is_error_hpa(hypercall_hpa)) {
1429 para_state->ret = -KVM_EINVAL;
1430 goto err_kunmap_skip;
1433 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1434 vcpu->para_state_page = para_state_page;
1435 vcpu->para_state_gpa = para_state_gpa;
1436 vcpu->hypercall_gpa = hypercall_gpa;
1438 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1439 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1440 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1441 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1442 kunmap_atomic(hypercall, KM_USER1);
1444 para_state->ret = 0;
1446 kunmap_atomic(para_state, KM_USER0);
1452 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1457 case 0xc0010010: /* SYSCFG */
1458 case 0xc0010015: /* HWCR */
1459 case MSR_IA32_PLATFORM_ID:
1460 case MSR_IA32_P5_MC_ADDR:
1461 case MSR_IA32_P5_MC_TYPE:
1462 case MSR_IA32_MC0_CTL:
1463 case MSR_IA32_MCG_STATUS:
1464 case MSR_IA32_MCG_CAP:
1465 case MSR_IA32_MC0_MISC:
1466 case MSR_IA32_MC0_MISC+4:
1467 case MSR_IA32_MC0_MISC+8:
1468 case MSR_IA32_MC0_MISC+12:
1469 case MSR_IA32_MC0_MISC+16:
1470 case MSR_IA32_UCODE_REV:
1471 case MSR_IA32_PERF_STATUS:
1472 /* MTRR registers */
1474 case 0x200 ... 0x2ff:
1477 case 0xcd: /* fsb frequency */
1480 case MSR_IA32_APICBASE:
1481 data = vcpu->apic_base;
1483 case MSR_IA32_MISC_ENABLE:
1484 data = vcpu->ia32_misc_enable_msr;
1486 #ifdef CONFIG_X86_64
1488 data = vcpu->shadow_efer;
1492 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1498 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1501 * Reads an msr value (of 'msr_index') into 'pdata'.
1502 * Returns 0 on success, non-0 otherwise.
1503 * Assumes vcpu_load() was already called.
1505 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1507 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1510 #ifdef CONFIG_X86_64
1512 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1514 if (efer & EFER_RESERVED_BITS) {
1515 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1522 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1523 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1528 kvm_arch_ops->set_efer(vcpu, efer);
1531 efer |= vcpu->shadow_efer & EFER_LMA;
1533 vcpu->shadow_efer = efer;
1538 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1541 #ifdef CONFIG_X86_64
1543 set_efer(vcpu, data);
1546 case MSR_IA32_MC0_STATUS:
1547 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1548 __FUNCTION__, data);
1550 case MSR_IA32_MCG_STATUS:
1551 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1552 __FUNCTION__, data);
1554 case MSR_IA32_UCODE_REV:
1555 case MSR_IA32_UCODE_WRITE:
1556 case 0x200 ... 0x2ff: /* MTRRs */
1558 case MSR_IA32_APICBASE:
1559 vcpu->apic_base = data;
1561 case MSR_IA32_MISC_ENABLE:
1562 vcpu->ia32_misc_enable_msr = data;
1565 * This is the 'probe whether the host is KVM' logic:
1567 case MSR_KVM_API_MAGIC:
1568 return vcpu_register_para(vcpu, data);
1571 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1576 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1579 * Writes msr value into into the appropriate "register".
1580 * Returns 0 on success, non-0 otherwise.
1581 * Assumes vcpu_load() was already called.
1583 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1585 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1588 void kvm_resched(struct kvm_vcpu *vcpu)
1590 if (!need_resched())
1596 EXPORT_SYMBOL_GPL(kvm_resched);
1598 void load_msrs(struct vmx_msr_entry *e, int n)
1602 for (i = 0; i < n; ++i)
1603 wrmsrl(e[i].index, e[i].data);
1605 EXPORT_SYMBOL_GPL(load_msrs);
1607 void save_msrs(struct vmx_msr_entry *e, int n)
1611 for (i = 0; i < n; ++i)
1612 rdmsrl(e[i].index, e[i].data);
1614 EXPORT_SYMBOL_GPL(save_msrs);
1616 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1620 struct kvm_cpuid_entry *e, *best;
1622 kvm_arch_ops->cache_regs(vcpu);
1623 function = vcpu->regs[VCPU_REGS_RAX];
1624 vcpu->regs[VCPU_REGS_RAX] = 0;
1625 vcpu->regs[VCPU_REGS_RBX] = 0;
1626 vcpu->regs[VCPU_REGS_RCX] = 0;
1627 vcpu->regs[VCPU_REGS_RDX] = 0;
1629 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1630 e = &vcpu->cpuid_entries[i];
1631 if (e->function == function) {
1636 * Both basic or both extended?
1638 if (((e->function ^ function) & 0x80000000) == 0)
1639 if (!best || e->function > best->function)
1643 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1644 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1645 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1646 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1648 kvm_arch_ops->decache_regs(vcpu);
1649 kvm_arch_ops->skip_emulated_instruction(vcpu);
1651 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1653 static int pio_copy_data(struct kvm_vcpu *vcpu)
1655 void *p = vcpu->pio_data;
1658 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1660 kvm_arch_ops->vcpu_put(vcpu);
1661 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1664 kvm_arch_ops->vcpu_load(vcpu);
1665 free_pio_guest_pages(vcpu);
1668 q += vcpu->pio.guest_page_offset;
1669 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1671 memcpy(q, p, bytes);
1673 memcpy(p, q, bytes);
1674 q -= vcpu->pio.guest_page_offset;
1676 kvm_arch_ops->vcpu_load(vcpu);
1677 free_pio_guest_pages(vcpu);
1681 static int complete_pio(struct kvm_vcpu *vcpu)
1683 struct kvm_pio_request *io = &vcpu->pio;
1687 kvm_arch_ops->cache_regs(vcpu);
1691 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1695 r = pio_copy_data(vcpu);
1697 kvm_arch_ops->cache_regs(vcpu);
1704 delta *= io->cur_count;
1706 * The size of the register should really depend on
1707 * current address size.
1709 vcpu->regs[VCPU_REGS_RCX] -= delta;
1715 vcpu->regs[VCPU_REGS_RDI] += delta;
1717 vcpu->regs[VCPU_REGS_RSI] += delta;
1720 kvm_arch_ops->decache_regs(vcpu);
1722 io->count -= io->cur_count;
1726 kvm_arch_ops->skip_emulated_instruction(vcpu);
1730 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1731 int size, unsigned long count, int string, int down,
1732 gva_t address, int rep, unsigned port)
1734 unsigned now, in_page;
1739 vcpu->run->exit_reason = KVM_EXIT_IO;
1740 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1741 vcpu->run->io.size = size;
1742 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1743 vcpu->run->io.count = count;
1744 vcpu->run->io.port = port;
1745 vcpu->pio.count = count;
1746 vcpu->pio.cur_count = count;
1747 vcpu->pio.size = size;
1749 vcpu->pio.string = string;
1750 vcpu->pio.down = down;
1751 vcpu->pio.guest_page_offset = offset_in_page(address);
1752 vcpu->pio.rep = rep;
1755 kvm_arch_ops->cache_regs(vcpu);
1756 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1757 kvm_arch_ops->decache_regs(vcpu);
1762 kvm_arch_ops->skip_emulated_instruction(vcpu);
1766 now = min(count, PAGE_SIZE / size);
1769 in_page = PAGE_SIZE - offset_in_page(address);
1771 in_page = offset_in_page(address) + size;
1772 now = min(count, (unsigned long)in_page / size);
1775 * String I/O straddles page boundary. Pin two guest pages
1776 * so that we satisfy atomicity constraints. Do just one
1777 * transaction to avoid complexity.
1784 * String I/O in reverse. Yuck. Kill the guest, fix later.
1786 printk(KERN_ERR "kvm: guest string pio down\n");
1790 vcpu->run->io.count = now;
1791 vcpu->pio.cur_count = now;
1793 for (i = 0; i < nr_pages; ++i) {
1794 spin_lock(&vcpu->kvm->lock);
1795 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1798 vcpu->pio.guest_pages[i] = page;
1799 spin_unlock(&vcpu->kvm->lock);
1802 free_pio_guest_pages(vcpu);
1808 return pio_copy_data(vcpu);
1811 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1813 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1820 if (vcpu->sigset_active)
1821 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1823 /* re-sync apic's tpr */
1824 vcpu->cr8 = kvm_run->cr8;
1826 if (vcpu->pio.cur_count) {
1827 r = complete_pio(vcpu);
1832 if (vcpu->mmio_needed) {
1833 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1834 vcpu->mmio_read_completed = 1;
1835 vcpu->mmio_needed = 0;
1836 r = emulate_instruction(vcpu, kvm_run,
1837 vcpu->mmio_fault_cr2, 0);
1838 if (r == EMULATE_DO_MMIO) {
1840 * Read-modify-write. Back to userspace.
1842 kvm_run->exit_reason = KVM_EXIT_MMIO;
1848 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1849 kvm_arch_ops->cache_regs(vcpu);
1850 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1851 kvm_arch_ops->decache_regs(vcpu);
1854 r = kvm_arch_ops->run(vcpu, kvm_run);
1857 if (vcpu->sigset_active)
1858 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1864 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1865 struct kvm_regs *regs)
1869 kvm_arch_ops->cache_regs(vcpu);
1871 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1872 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1873 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1874 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1875 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1876 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1877 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1878 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1879 #ifdef CONFIG_X86_64
1880 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1881 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1882 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1883 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1884 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1885 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1886 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1887 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1890 regs->rip = vcpu->rip;
1891 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1894 * Don't leak debug flags in case they were set for guest debugging
1896 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1897 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1904 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1905 struct kvm_regs *regs)
1909 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1910 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1911 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1912 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1913 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1914 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1915 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1916 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1917 #ifdef CONFIG_X86_64
1918 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1919 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1920 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1921 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1922 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1923 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1924 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1925 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1928 vcpu->rip = regs->rip;
1929 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1931 kvm_arch_ops->decache_regs(vcpu);
1938 static void get_segment(struct kvm_vcpu *vcpu,
1939 struct kvm_segment *var, int seg)
1941 return kvm_arch_ops->get_segment(vcpu, var, seg);
1944 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1945 struct kvm_sregs *sregs)
1947 struct descriptor_table dt;
1951 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1952 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1953 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1954 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1955 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1956 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1958 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1959 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1961 kvm_arch_ops->get_idt(vcpu, &dt);
1962 sregs->idt.limit = dt.limit;
1963 sregs->idt.base = dt.base;
1964 kvm_arch_ops->get_gdt(vcpu, &dt);
1965 sregs->gdt.limit = dt.limit;
1966 sregs->gdt.base = dt.base;
1968 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1969 sregs->cr0 = vcpu->cr0;
1970 sregs->cr2 = vcpu->cr2;
1971 sregs->cr3 = vcpu->cr3;
1972 sregs->cr4 = vcpu->cr4;
1973 sregs->cr8 = vcpu->cr8;
1974 sregs->efer = vcpu->shadow_efer;
1975 sregs->apic_base = vcpu->apic_base;
1977 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1978 sizeof sregs->interrupt_bitmap);
1985 static void set_segment(struct kvm_vcpu *vcpu,
1986 struct kvm_segment *var, int seg)
1988 return kvm_arch_ops->set_segment(vcpu, var, seg);
1991 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1992 struct kvm_sregs *sregs)
1994 int mmu_reset_needed = 0;
1996 struct descriptor_table dt;
2000 dt.limit = sregs->idt.limit;
2001 dt.base = sregs->idt.base;
2002 kvm_arch_ops->set_idt(vcpu, &dt);
2003 dt.limit = sregs->gdt.limit;
2004 dt.base = sregs->gdt.base;
2005 kvm_arch_ops->set_gdt(vcpu, &dt);
2007 vcpu->cr2 = sregs->cr2;
2008 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2009 vcpu->cr3 = sregs->cr3;
2011 vcpu->cr8 = sregs->cr8;
2013 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2014 #ifdef CONFIG_X86_64
2015 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2017 vcpu->apic_base = sregs->apic_base;
2019 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2021 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2022 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2024 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2025 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2026 if (!is_long_mode(vcpu) && is_pae(vcpu))
2027 load_pdptrs(vcpu, vcpu->cr3);
2029 if (mmu_reset_needed)
2030 kvm_mmu_reset_context(vcpu);
2032 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2033 sizeof vcpu->irq_pending);
2034 vcpu->irq_summary = 0;
2035 for (i = 0; i < NR_IRQ_WORDS; ++i)
2036 if (vcpu->irq_pending[i])
2037 __set_bit(i, &vcpu->irq_summary);
2039 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2040 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2041 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2042 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2043 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2044 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2046 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2047 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2055 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2056 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2058 * This list is modified at module load time to reflect the
2059 * capabilities of the host cpu.
2061 static u32 msrs_to_save[] = {
2062 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2064 #ifdef CONFIG_X86_64
2065 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2067 MSR_IA32_TIME_STAMP_COUNTER,
2070 static unsigned num_msrs_to_save;
2072 static u32 emulated_msrs[] = {
2073 MSR_IA32_MISC_ENABLE,
2076 static __init void kvm_init_msr_list(void)
2081 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2082 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2085 msrs_to_save[j] = msrs_to_save[i];
2088 num_msrs_to_save = j;
2092 * Adapt set_msr() to msr_io()'s calling convention
2094 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2096 return set_msr(vcpu, index, *data);
2100 * Read or write a bunch of msrs. All parameters are kernel addresses.
2102 * @return number of msrs set successfully.
2104 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2105 struct kvm_msr_entry *entries,
2106 int (*do_msr)(struct kvm_vcpu *vcpu,
2107 unsigned index, u64 *data))
2113 for (i = 0; i < msrs->nmsrs; ++i)
2114 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2123 * Read or write a bunch of msrs. Parameters are user addresses.
2125 * @return number of msrs set successfully.
2127 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2128 int (*do_msr)(struct kvm_vcpu *vcpu,
2129 unsigned index, u64 *data),
2132 struct kvm_msrs msrs;
2133 struct kvm_msr_entry *entries;
2138 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2142 if (msrs.nmsrs >= MAX_IO_MSRS)
2146 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2147 entries = vmalloc(size);
2152 if (copy_from_user(entries, user_msrs->entries, size))
2155 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2160 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2172 * Translate a guest virtual address to a guest physical address.
2174 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2175 struct kvm_translation *tr)
2177 unsigned long vaddr = tr->linear_address;
2181 spin_lock(&vcpu->kvm->lock);
2182 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2183 tr->physical_address = gpa;
2184 tr->valid = gpa != UNMAPPED_GVA;
2187 spin_unlock(&vcpu->kvm->lock);
2193 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2194 struct kvm_interrupt *irq)
2196 if (irq->irq < 0 || irq->irq >= 256)
2200 set_bit(irq->irq, vcpu->irq_pending);
2201 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2208 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2209 struct kvm_debug_guest *dbg)
2215 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2222 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2223 unsigned long address,
2226 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2227 unsigned long pgoff;
2230 *type = VM_FAULT_MINOR;
2231 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2233 page = virt_to_page(vcpu->run);
2234 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2235 page = virt_to_page(vcpu->pio_data);
2237 return NOPAGE_SIGBUS;
2242 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2243 .nopage = kvm_vcpu_nopage,
2246 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2248 vma->vm_ops = &kvm_vcpu_vm_ops;
2252 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2254 struct kvm_vcpu *vcpu = filp->private_data;
2256 fput(vcpu->kvm->filp);
2260 static struct file_operations kvm_vcpu_fops = {
2261 .release = kvm_vcpu_release,
2262 .unlocked_ioctl = kvm_vcpu_ioctl,
2263 .compat_ioctl = kvm_vcpu_ioctl,
2264 .mmap = kvm_vcpu_mmap,
2268 * Allocates an inode for the vcpu.
2270 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2273 struct inode *inode;
2276 atomic_inc(&vcpu->kvm->filp->f_count);
2277 inode = kvmfs_inode(&kvm_vcpu_fops);
2278 if (IS_ERR(inode)) {
2283 file = kvmfs_file(inode, vcpu);
2289 r = get_unused_fd();
2293 fd_install(fd, file);
2302 fput(vcpu->kvm->filp);
2307 * Creates some virtual cpus. Good luck creating more than one.
2309 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2312 struct kvm_vcpu *vcpu;
2319 vcpu = &kvm->vcpus[n];
2321 mutex_lock(&vcpu->mutex);
2324 mutex_unlock(&vcpu->mutex);
2328 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2332 vcpu->run = page_address(page);
2334 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2338 vcpu->pio_data = page_address(page);
2340 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2342 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2345 r = kvm_arch_ops->vcpu_create(vcpu);
2347 goto out_free_vcpus;
2349 r = kvm_mmu_create(vcpu);
2351 goto out_free_vcpus;
2353 kvm_arch_ops->vcpu_load(vcpu);
2354 r = kvm_mmu_setup(vcpu);
2356 r = kvm_arch_ops->vcpu_setup(vcpu);
2360 goto out_free_vcpus;
2362 r = create_vcpu_fd(vcpu);
2364 goto out_free_vcpus;
2369 kvm_free_vcpu(vcpu);
2371 free_page((unsigned long)vcpu->run);
2374 mutex_unlock(&vcpu->mutex);
2379 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2380 struct kvm_cpuid *cpuid,
2381 struct kvm_cpuid_entry __user *entries)
2386 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2389 if (copy_from_user(&vcpu->cpuid_entries, entries,
2390 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2392 vcpu->cpuid_nent = cpuid->nent;
2399 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2402 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2403 vcpu->sigset_active = 1;
2404 vcpu->sigset = *sigset;
2406 vcpu->sigset_active = 0;
2411 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2412 * we have asm/x86/processor.h
2423 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2424 #ifdef CONFIG_X86_64
2425 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2427 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2431 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2433 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2437 memcpy(fpu->fpr, fxsave->st_space, 128);
2438 fpu->fcw = fxsave->cwd;
2439 fpu->fsw = fxsave->swd;
2440 fpu->ftwx = fxsave->twd;
2441 fpu->last_opcode = fxsave->fop;
2442 fpu->last_ip = fxsave->rip;
2443 fpu->last_dp = fxsave->rdp;
2444 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2451 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2453 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2457 memcpy(fxsave->st_space, fpu->fpr, 128);
2458 fxsave->cwd = fpu->fcw;
2459 fxsave->swd = fpu->fsw;
2460 fxsave->twd = fpu->ftwx;
2461 fxsave->fop = fpu->last_opcode;
2462 fxsave->rip = fpu->last_ip;
2463 fxsave->rdp = fpu->last_dp;
2464 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2471 static long kvm_vcpu_ioctl(struct file *filp,
2472 unsigned int ioctl, unsigned long arg)
2474 struct kvm_vcpu *vcpu = filp->private_data;
2475 void __user *argp = (void __user *)arg;
2483 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2485 case KVM_GET_REGS: {
2486 struct kvm_regs kvm_regs;
2488 memset(&kvm_regs, 0, sizeof kvm_regs);
2489 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2493 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2498 case KVM_SET_REGS: {
2499 struct kvm_regs kvm_regs;
2502 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2504 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2510 case KVM_GET_SREGS: {
2511 struct kvm_sregs kvm_sregs;
2513 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2514 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2518 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2523 case KVM_SET_SREGS: {
2524 struct kvm_sregs kvm_sregs;
2527 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2529 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2535 case KVM_TRANSLATE: {
2536 struct kvm_translation tr;
2539 if (copy_from_user(&tr, argp, sizeof tr))
2541 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2545 if (copy_to_user(argp, &tr, sizeof tr))
2550 case KVM_INTERRUPT: {
2551 struct kvm_interrupt irq;
2554 if (copy_from_user(&irq, argp, sizeof irq))
2556 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2562 case KVM_DEBUG_GUEST: {
2563 struct kvm_debug_guest dbg;
2566 if (copy_from_user(&dbg, argp, sizeof dbg))
2568 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2575 r = msr_io(vcpu, argp, get_msr, 1);
2578 r = msr_io(vcpu, argp, do_set_msr, 0);
2580 case KVM_SET_CPUID: {
2581 struct kvm_cpuid __user *cpuid_arg = argp;
2582 struct kvm_cpuid cpuid;
2585 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2587 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2592 case KVM_SET_SIGNAL_MASK: {
2593 struct kvm_signal_mask __user *sigmask_arg = argp;
2594 struct kvm_signal_mask kvm_sigmask;
2595 sigset_t sigset, *p;
2600 if (copy_from_user(&kvm_sigmask, argp,
2601 sizeof kvm_sigmask))
2604 if (kvm_sigmask.len != sizeof sigset)
2607 if (copy_from_user(&sigset, sigmask_arg->sigset,
2612 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2618 memset(&fpu, 0, sizeof fpu);
2619 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2623 if (copy_to_user(argp, &fpu, sizeof fpu))
2632 if (copy_from_user(&fpu, argp, sizeof fpu))
2634 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2647 static long kvm_vm_ioctl(struct file *filp,
2648 unsigned int ioctl, unsigned long arg)
2650 struct kvm *kvm = filp->private_data;
2651 void __user *argp = (void __user *)arg;
2655 case KVM_CREATE_VCPU:
2656 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2660 case KVM_SET_MEMORY_REGION: {
2661 struct kvm_memory_region kvm_mem;
2664 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2666 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2671 case KVM_GET_DIRTY_LOG: {
2672 struct kvm_dirty_log log;
2675 if (copy_from_user(&log, argp, sizeof log))
2677 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2682 case KVM_SET_MEMORY_ALIAS: {
2683 struct kvm_memory_alias alias;
2686 if (copy_from_user(&alias, argp, sizeof alias))
2688 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2700 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2701 unsigned long address,
2704 struct kvm *kvm = vma->vm_file->private_data;
2705 unsigned long pgoff;
2708 *type = VM_FAULT_MINOR;
2709 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2710 page = gfn_to_page(kvm, pgoff);
2712 return NOPAGE_SIGBUS;
2717 static struct vm_operations_struct kvm_vm_vm_ops = {
2718 .nopage = kvm_vm_nopage,
2721 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2723 vma->vm_ops = &kvm_vm_vm_ops;
2727 static struct file_operations kvm_vm_fops = {
2728 .release = kvm_vm_release,
2729 .unlocked_ioctl = kvm_vm_ioctl,
2730 .compat_ioctl = kvm_vm_ioctl,
2731 .mmap = kvm_vm_mmap,
2734 static int kvm_dev_ioctl_create_vm(void)
2737 struct inode *inode;
2741 inode = kvmfs_inode(&kvm_vm_fops);
2742 if (IS_ERR(inode)) {
2747 kvm = kvm_create_vm();
2753 file = kvmfs_file(inode, kvm);
2760 r = get_unused_fd();
2764 fd_install(fd, file);
2771 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 = {
2856 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2859 if (val == SYS_RESTART) {
2861 * Some (well, at least mine) BIOSes hang on reboot if
2864 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2865 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2870 static struct notifier_block kvm_reboot_notifier = {
2871 .notifier_call = kvm_reboot,
2876 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2879 static void decache_vcpus_on_cpu(int cpu)
2882 struct kvm_vcpu *vcpu;
2885 spin_lock(&kvm_lock);
2886 list_for_each_entry(vm, &vm_list, vm_list)
2887 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2888 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 int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2914 case CPU_DOWN_PREPARE:
2915 case CPU_DOWN_PREPARE_FROZEN:
2916 case CPU_UP_CANCELED:
2917 case CPU_UP_CANCELED_FROZEN:
2918 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2920 decache_vcpus_on_cpu(cpu);
2921 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2925 case CPU_ONLINE_FROZEN:
2926 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2928 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2935 static struct notifier_block kvm_cpu_notifier = {
2936 .notifier_call = kvm_cpu_hotplug,
2937 .priority = 20, /* must be > scheduler priority */
2940 static u64 stat_get(void *_offset)
2942 unsigned offset = (long)_offset;
2945 struct kvm_vcpu *vcpu;
2948 spin_lock(&kvm_lock);
2949 list_for_each_entry(kvm, &vm_list, vm_list)
2950 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2951 vcpu = &kvm->vcpus[i];
2952 total += *(u32 *)((void *)vcpu + offset);
2954 spin_unlock(&kvm_lock);
2958 static void stat_set(void *offset, u64 val)
2962 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
2964 static __init void kvm_init_debug(void)
2966 struct kvm_stats_debugfs_item *p;
2968 debugfs_dir = debugfs_create_dir("kvm", NULL);
2969 for (p = debugfs_entries; p->name; ++p)
2970 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
2971 (void *)(long)p->offset,
2975 static void kvm_exit_debug(void)
2977 struct kvm_stats_debugfs_item *p;
2979 for (p = debugfs_entries; p->name; ++p)
2980 debugfs_remove(p->dentry);
2981 debugfs_remove(debugfs_dir);
2984 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2986 decache_vcpus_on_cpu(raw_smp_processor_id());
2987 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2991 static int kvm_resume(struct sys_device *dev)
2993 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2997 static struct sysdev_class kvm_sysdev_class = {
2998 set_kset_name("kvm"),
2999 .suspend = kvm_suspend,
3000 .resume = kvm_resume,
3003 static struct sys_device kvm_sysdev = {
3005 .cls = &kvm_sysdev_class,
3008 hpa_t bad_page_address;
3010 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
3011 const char *dev_name, void *data, struct vfsmount *mnt)
3013 return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
3016 static struct file_system_type kvm_fs_type = {
3018 .get_sb = kvmfs_get_sb,
3019 .kill_sb = kill_anon_super,
3022 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3027 printk(KERN_ERR "kvm: already loaded the other module\n");
3031 if (!ops->cpu_has_kvm_support()) {
3032 printk(KERN_ERR "kvm: no hardware support\n");
3035 if (ops->disabled_by_bios()) {
3036 printk(KERN_ERR "kvm: disabled by bios\n");
3042 r = kvm_arch_ops->hardware_setup();
3046 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3047 r = register_cpu_notifier(&kvm_cpu_notifier);
3050 register_reboot_notifier(&kvm_reboot_notifier);
3052 r = sysdev_class_register(&kvm_sysdev_class);
3056 r = sysdev_register(&kvm_sysdev);
3060 kvm_chardev_ops.owner = module;
3062 r = misc_register(&kvm_dev);
3064 printk (KERN_ERR "kvm: misc device register failed\n");
3071 sysdev_unregister(&kvm_sysdev);
3073 sysdev_class_unregister(&kvm_sysdev_class);
3075 unregister_reboot_notifier(&kvm_reboot_notifier);
3076 unregister_cpu_notifier(&kvm_cpu_notifier);
3078 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3079 kvm_arch_ops->hardware_unsetup();
3081 kvm_arch_ops = NULL;
3085 void kvm_exit_arch(void)
3087 misc_deregister(&kvm_dev);
3088 sysdev_unregister(&kvm_sysdev);
3089 sysdev_class_unregister(&kvm_sysdev_class);
3090 unregister_reboot_notifier(&kvm_reboot_notifier);
3091 unregister_cpu_notifier(&kvm_cpu_notifier);
3092 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3093 kvm_arch_ops->hardware_unsetup();
3094 kvm_arch_ops = NULL;
3097 static __init int kvm_init(void)
3099 static struct page *bad_page;
3102 r = kvm_mmu_module_init();
3106 r = register_filesystem(&kvm_fs_type);
3110 kvmfs_mnt = kern_mount(&kvm_fs_type);
3111 r = PTR_ERR(kvmfs_mnt);
3112 if (IS_ERR(kvmfs_mnt))
3116 kvm_init_msr_list();
3118 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3123 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3124 memset(__va(bad_page_address), 0, PAGE_SIZE);
3132 unregister_filesystem(&kvm_fs_type);
3134 kvm_mmu_module_exit();
3139 static __exit void kvm_exit(void)
3142 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3144 unregister_filesystem(&kvm_fs_type);
3145 kvm_mmu_module_exit();
3148 module_init(kvm_init)
3149 module_exit(kvm_exit)
3151 EXPORT_SYMBOL_GPL(kvm_init_arch);
3152 EXPORT_SYMBOL_GPL(kvm_exit_arch);