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>
44 #include <linux/cpumask.h>
45 #include <linux/smp.h>
47 #include "x86_emulate.h"
48 #include "segment_descriptor.h"
50 MODULE_AUTHOR("Qumranet");
51 MODULE_LICENSE("GPL");
53 static DEFINE_SPINLOCK(kvm_lock);
54 static LIST_HEAD(vm_list);
56 struct kvm_arch_ops *kvm_arch_ops;
58 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
60 static struct kvm_stats_debugfs_item {
63 struct dentry *dentry;
64 } debugfs_entries[] = {
65 { "pf_fixed", STAT_OFFSET(pf_fixed) },
66 { "pf_guest", STAT_OFFSET(pf_guest) },
67 { "tlb_flush", STAT_OFFSET(tlb_flush) },
68 { "invlpg", STAT_OFFSET(invlpg) },
69 { "exits", STAT_OFFSET(exits) },
70 { "io_exits", STAT_OFFSET(io_exits) },
71 { "mmio_exits", STAT_OFFSET(mmio_exits) },
72 { "signal_exits", STAT_OFFSET(signal_exits) },
73 { "irq_window", STAT_OFFSET(irq_window_exits) },
74 { "halt_exits", STAT_OFFSET(halt_exits) },
75 { "request_irq", STAT_OFFSET(request_irq_exits) },
76 { "irq_exits", STAT_OFFSET(irq_exits) },
77 { "light_exits", STAT_OFFSET(light_exits) },
78 { "efer_reload", STAT_OFFSET(efer_reload) },
82 static struct dentry *debugfs_dir;
84 struct vfsmount *kvmfs_mnt;
86 #define MAX_IO_MSRS 256
88 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
89 #define LMSW_GUEST_MASK 0x0eULL
90 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
91 #define CR8_RESEVED_BITS (~0x0fULL)
92 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
95 // LDT or TSS descriptor in the GDT. 16 bytes.
96 struct segment_descriptor_64 {
97 struct segment_descriptor s;
104 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
107 static struct inode *kvmfs_inode(struct file_operations *fops)
110 struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
118 * Mark the inode dirty from the very beginning,
119 * that way it will never be moved to the dirty
120 * list because mark_inode_dirty() will think
121 * that it already _is_ on the dirty list.
123 inode->i_state = I_DIRTY;
124 inode->i_mode = S_IRUSR | S_IWUSR;
125 inode->i_uid = current->fsuid;
126 inode->i_gid = current->fsgid;
127 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
131 return ERR_PTR(error);
134 static struct file *kvmfs_file(struct inode *inode, void *private_data)
136 struct file *file = get_empty_filp();
139 return ERR_PTR(-ENFILE);
141 file->f_path.mnt = mntget(kvmfs_mnt);
142 file->f_path.dentry = d_alloc_anon(inode);
143 if (!file->f_path.dentry)
144 return ERR_PTR(-ENOMEM);
145 file->f_mapping = inode->i_mapping;
148 file->f_flags = O_RDWR;
149 file->f_op = inode->i_fop;
150 file->f_mode = FMODE_READ | FMODE_WRITE;
152 file->private_data = private_data;
156 unsigned long segment_base(u16 selector)
158 struct descriptor_table gdt;
159 struct segment_descriptor *d;
160 unsigned long table_base;
161 typedef unsigned long ul;
167 asm ("sgdt %0" : "=m"(gdt));
168 table_base = gdt.base;
170 if (selector & 4) { /* from ldt */
173 asm ("sldt %0" : "=g"(ldt_selector));
174 table_base = segment_base(ldt_selector);
176 d = (struct segment_descriptor *)(table_base + (selector & ~7));
177 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
180 && (d->type == 2 || d->type == 9 || d->type == 11))
181 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
185 EXPORT_SYMBOL_GPL(segment_base);
187 static inline int valid_vcpu(int n)
189 return likely(n >= 0 && n < KVM_MAX_VCPUS);
192 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
195 unsigned char *host_buf = dest;
196 unsigned long req_size = size;
204 paddr = gva_to_hpa(vcpu, addr);
206 if (is_error_hpa(paddr))
209 guest_buf = (hva_t)kmap_atomic(
210 pfn_to_page(paddr >> PAGE_SHIFT),
212 offset = addr & ~PAGE_MASK;
214 now = min(size, PAGE_SIZE - offset);
215 memcpy(host_buf, (void*)guest_buf, now);
219 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
221 return req_size - size;
223 EXPORT_SYMBOL_GPL(kvm_read_guest);
225 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
228 unsigned char *host_buf = data;
229 unsigned long req_size = size;
238 paddr = gva_to_hpa(vcpu, addr);
240 if (is_error_hpa(paddr))
243 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
244 mark_page_dirty(vcpu->kvm, gfn);
245 guest_buf = (hva_t)kmap_atomic(
246 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
247 offset = addr & ~PAGE_MASK;
249 now = min(size, PAGE_SIZE - offset);
250 memcpy((void*)guest_buf, host_buf, now);
254 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
256 return req_size - size;
258 EXPORT_SYMBOL_GPL(kvm_write_guest);
260 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
262 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
265 vcpu->guest_fpu_loaded = 1;
266 fx_save(vcpu->host_fx_image);
267 fx_restore(vcpu->guest_fx_image);
269 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
271 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
273 if (!vcpu->guest_fpu_loaded)
276 vcpu->guest_fpu_loaded = 0;
277 fx_save(vcpu->guest_fx_image);
278 fx_restore(vcpu->host_fx_image);
280 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
283 * Switches to specified vcpu, until a matching vcpu_put()
285 static void vcpu_load(struct kvm_vcpu *vcpu)
287 mutex_lock(&vcpu->mutex);
288 kvm_arch_ops->vcpu_load(vcpu);
292 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
293 * if the slot is not populated.
295 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
297 struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
299 mutex_lock(&vcpu->mutex);
301 mutex_unlock(&vcpu->mutex);
304 kvm_arch_ops->vcpu_load(vcpu);
308 static void vcpu_put(struct kvm_vcpu *vcpu)
310 kvm_arch_ops->vcpu_put(vcpu);
311 mutex_unlock(&vcpu->mutex);
314 static void ack_flush(void *_completed)
316 atomic_t *completed = _completed;
318 atomic_inc(completed);
321 void kvm_flush_remote_tlbs(struct kvm *kvm)
325 struct kvm_vcpu *vcpu;
328 atomic_set(&completed, 0);
331 for (i = 0; i < kvm->nvcpus; ++i) {
332 vcpu = &kvm->vcpus[i];
333 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
336 if (cpu != -1 && cpu != raw_smp_processor_id())
337 if (!cpu_isset(cpu, cpus)) {
344 * We really want smp_call_function_mask() here. But that's not
345 * available, so ipi all cpus in parallel and wait for them
348 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
349 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
350 while (atomic_read(&completed) != needed) {
356 static struct kvm *kvm_create_vm(void)
358 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
362 return ERR_PTR(-ENOMEM);
364 kvm_io_bus_init(&kvm->pio_bus);
365 spin_lock_init(&kvm->lock);
366 INIT_LIST_HEAD(&kvm->active_mmu_pages);
367 spin_lock(&kvm_lock);
368 list_add(&kvm->vm_list, &vm_list);
369 spin_unlock(&kvm_lock);
370 kvm_io_bus_init(&kvm->mmio_bus);
371 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
372 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
374 mutex_init(&vcpu->mutex);
377 vcpu->mmu.root_hpa = INVALID_PAGE;
382 static int kvm_dev_open(struct inode *inode, struct file *filp)
388 * Free any memory in @free but not in @dont.
390 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
391 struct kvm_memory_slot *dont)
395 if (!dont || free->phys_mem != dont->phys_mem)
396 if (free->phys_mem) {
397 for (i = 0; i < free->npages; ++i)
398 if (free->phys_mem[i])
399 __free_page(free->phys_mem[i]);
400 vfree(free->phys_mem);
403 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
404 vfree(free->dirty_bitmap);
406 free->phys_mem = NULL;
408 free->dirty_bitmap = NULL;
411 static void kvm_free_physmem(struct kvm *kvm)
415 for (i = 0; i < kvm->nmemslots; ++i)
416 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
419 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
423 for (i = 0; i < 2; ++i)
424 if (vcpu->pio.guest_pages[i]) {
425 __free_page(vcpu->pio.guest_pages[i]);
426 vcpu->pio.guest_pages[i] = NULL;
430 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
436 kvm_mmu_unload(vcpu);
440 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
446 kvm_mmu_destroy(vcpu);
448 kvm_arch_ops->vcpu_free(vcpu);
449 free_page((unsigned long)vcpu->run);
451 free_page((unsigned long)vcpu->pio_data);
452 vcpu->pio_data = NULL;
453 free_pio_guest_pages(vcpu);
456 static void kvm_free_vcpus(struct kvm *kvm)
461 * Unpin any mmu pages first.
463 for (i = 0; i < KVM_MAX_VCPUS; ++i)
464 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
465 for (i = 0; i < KVM_MAX_VCPUS; ++i)
466 kvm_free_vcpu(&kvm->vcpus[i]);
469 static int kvm_dev_release(struct inode *inode, struct file *filp)
474 static void kvm_destroy_vm(struct kvm *kvm)
476 spin_lock(&kvm_lock);
477 list_del(&kvm->vm_list);
478 spin_unlock(&kvm_lock);
479 kvm_io_bus_destroy(&kvm->pio_bus);
480 kvm_io_bus_destroy(&kvm->mmio_bus);
482 kvm_free_physmem(kvm);
486 static int kvm_vm_release(struct inode *inode, struct file *filp)
488 struct kvm *kvm = filp->private_data;
494 static void inject_gp(struct kvm_vcpu *vcpu)
496 kvm_arch_ops->inject_gp(vcpu, 0);
500 * Load the pae pdptrs. Return true is they are all valid.
502 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
504 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
505 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
512 spin_lock(&vcpu->kvm->lock);
513 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
514 /* FIXME: !page - emulate? 0xff? */
515 pdpt = kmap_atomic(page, KM_USER0);
518 for (i = 0; i < 4; ++i) {
519 pdpte = pdpt[offset + i];
520 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
526 for (i = 0; i < 4; ++i)
527 vcpu->pdptrs[i] = pdpt[offset + i];
530 kunmap_atomic(pdpt, KM_USER0);
531 spin_unlock(&vcpu->kvm->lock);
536 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
538 if (cr0 & CR0_RESEVED_BITS) {
539 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
545 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
546 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
551 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
552 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
553 "and a clear PE flag\n");
558 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
560 if ((vcpu->shadow_efer & EFER_LME)) {
564 printk(KERN_DEBUG "set_cr0: #GP, start paging "
565 "in long mode while PAE is disabled\n");
569 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
571 printk(KERN_DEBUG "set_cr0: #GP, start paging "
572 "in long mode while CS.L == 1\n");
579 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
580 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
588 kvm_arch_ops->set_cr0(vcpu, cr0);
591 spin_lock(&vcpu->kvm->lock);
592 kvm_mmu_reset_context(vcpu);
593 spin_unlock(&vcpu->kvm->lock);
596 EXPORT_SYMBOL_GPL(set_cr0);
598 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
600 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
602 EXPORT_SYMBOL_GPL(lmsw);
604 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
606 if (cr4 & CR4_RESEVED_BITS) {
607 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
612 if (is_long_mode(vcpu)) {
613 if (!(cr4 & CR4_PAE_MASK)) {
614 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
619 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
620 && !load_pdptrs(vcpu, vcpu->cr3)) {
621 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
625 if (cr4 & CR4_VMXE_MASK) {
626 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
630 kvm_arch_ops->set_cr4(vcpu, cr4);
631 spin_lock(&vcpu->kvm->lock);
632 kvm_mmu_reset_context(vcpu);
633 spin_unlock(&vcpu->kvm->lock);
635 EXPORT_SYMBOL_GPL(set_cr4);
637 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
639 if (is_long_mode(vcpu)) {
640 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
641 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
646 if (cr3 & CR3_RESEVED_BITS) {
647 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
651 if (is_paging(vcpu) && is_pae(vcpu) &&
652 !load_pdptrs(vcpu, cr3)) {
653 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
661 spin_lock(&vcpu->kvm->lock);
663 * Does the new cr3 value map to physical memory? (Note, we
664 * catch an invalid cr3 even in real-mode, because it would
665 * cause trouble later on when we turn on paging anyway.)
667 * A real CPU would silently accept an invalid cr3 and would
668 * attempt to use it - with largely undefined (and often hard
669 * to debug) behavior on the guest side.
671 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
674 vcpu->mmu.new_cr3(vcpu);
675 spin_unlock(&vcpu->kvm->lock);
677 EXPORT_SYMBOL_GPL(set_cr3);
679 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
681 if ( cr8 & CR8_RESEVED_BITS) {
682 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
688 EXPORT_SYMBOL_GPL(set_cr8);
690 void fx_init(struct kvm_vcpu *vcpu)
692 struct __attribute__ ((__packed__)) fx_image_s {
698 u64 operand;// fpu dp
704 fx_save(vcpu->host_fx_image);
706 fx_save(vcpu->guest_fx_image);
707 fx_restore(vcpu->host_fx_image);
709 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
710 fx_image->mxcsr = 0x1f80;
711 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
712 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
714 EXPORT_SYMBOL_GPL(fx_init);
716 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
718 spin_lock(&vcpu->kvm->lock);
719 kvm_mmu_slot_remove_write_access(vcpu, slot);
720 spin_unlock(&vcpu->kvm->lock);
724 * Allocate some memory and give it an address in the guest physical address
727 * Discontiguous memory is allowed, mostly for framebuffers.
729 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
730 struct kvm_memory_region *mem)
734 unsigned long npages;
736 struct kvm_memory_slot *memslot;
737 struct kvm_memory_slot old, new;
738 int memory_config_version;
741 /* General sanity checks */
742 if (mem->memory_size & (PAGE_SIZE - 1))
744 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
746 if (mem->slot >= KVM_MEMORY_SLOTS)
748 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
751 memslot = &kvm->memslots[mem->slot];
752 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
753 npages = mem->memory_size >> PAGE_SHIFT;
756 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
759 spin_lock(&kvm->lock);
761 memory_config_version = kvm->memory_config_version;
762 new = old = *memslot;
764 new.base_gfn = base_gfn;
766 new.flags = mem->flags;
768 /* Disallow changing a memory slot's size. */
770 if (npages && old.npages && npages != old.npages)
773 /* Check for overlaps */
775 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
776 struct kvm_memory_slot *s = &kvm->memslots[i];
780 if (!((base_gfn + npages <= s->base_gfn) ||
781 (base_gfn >= s->base_gfn + s->npages)))
785 * Do memory allocations outside lock. memory_config_version will
788 spin_unlock(&kvm->lock);
790 /* Deallocate if slot is being removed */
794 /* Free page dirty bitmap if unneeded */
795 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
796 new.dirty_bitmap = NULL;
800 /* Allocate if a slot is being created */
801 if (npages && !new.phys_mem) {
802 new.phys_mem = vmalloc(npages * sizeof(struct page *));
807 memset(new.phys_mem, 0, npages * sizeof(struct page *));
808 for (i = 0; i < npages; ++i) {
809 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
811 if (!new.phys_mem[i])
813 set_page_private(new.phys_mem[i],0);
817 /* Allocate page dirty bitmap if needed */
818 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
819 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
821 new.dirty_bitmap = vmalloc(dirty_bytes);
822 if (!new.dirty_bitmap)
824 memset(new.dirty_bitmap, 0, dirty_bytes);
827 spin_lock(&kvm->lock);
829 if (memory_config_version != kvm->memory_config_version) {
830 spin_unlock(&kvm->lock);
831 kvm_free_physmem_slot(&new, &old);
839 if (mem->slot >= kvm->nmemslots)
840 kvm->nmemslots = mem->slot + 1;
843 ++kvm->memory_config_version;
845 spin_unlock(&kvm->lock);
847 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
848 struct kvm_vcpu *vcpu;
850 vcpu = vcpu_load_slot(kvm, i);
853 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
854 do_remove_write_access(vcpu, mem->slot);
855 kvm_mmu_reset_context(vcpu);
859 kvm_free_physmem_slot(&old, &new);
863 spin_unlock(&kvm->lock);
865 kvm_free_physmem_slot(&new, &old);
871 * Get (and clear) the dirty memory log for a memory slot.
873 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
874 struct kvm_dirty_log *log)
876 struct kvm_memory_slot *memslot;
880 unsigned long any = 0;
882 spin_lock(&kvm->lock);
885 * Prevent changes to guest memory configuration even while the lock
889 spin_unlock(&kvm->lock);
891 if (log->slot >= KVM_MEMORY_SLOTS)
894 memslot = &kvm->memslots[log->slot];
896 if (!memslot->dirty_bitmap)
899 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
901 for (i = 0; !any && i < n/sizeof(long); ++i)
902 any = memslot->dirty_bitmap[i];
905 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
910 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
911 struct kvm_vcpu *vcpu;
913 vcpu = vcpu_load_slot(kvm, i);
917 do_remove_write_access(vcpu, log->slot);
918 memset(memslot->dirty_bitmap, 0, n);
921 kvm_arch_ops->tlb_flush(vcpu);
929 spin_lock(&kvm->lock);
931 spin_unlock(&kvm->lock);
936 * Set a new alias region. Aliases map a portion of physical memory into
937 * another portion. This is useful for memory windows, for example the PC
940 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
941 struct kvm_memory_alias *alias)
944 struct kvm_mem_alias *p;
947 /* General sanity checks */
948 if (alias->memory_size & (PAGE_SIZE - 1))
950 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
952 if (alias->slot >= KVM_ALIAS_SLOTS)
954 if (alias->guest_phys_addr + alias->memory_size
955 < alias->guest_phys_addr)
957 if (alias->target_phys_addr + alias->memory_size
958 < alias->target_phys_addr)
961 spin_lock(&kvm->lock);
963 p = &kvm->aliases[alias->slot];
964 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
965 p->npages = alias->memory_size >> PAGE_SHIFT;
966 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
968 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
969 if (kvm->aliases[n - 1].npages)
973 spin_unlock(&kvm->lock);
975 vcpu_load(&kvm->vcpus[0]);
976 spin_lock(&kvm->lock);
977 kvm_mmu_zap_all(&kvm->vcpus[0]);
978 spin_unlock(&kvm->lock);
979 vcpu_put(&kvm->vcpus[0]);
987 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
990 struct kvm_mem_alias *alias;
992 for (i = 0; i < kvm->naliases; ++i) {
993 alias = &kvm->aliases[i];
994 if (gfn >= alias->base_gfn
995 && gfn < alias->base_gfn + alias->npages)
996 return alias->target_gfn + gfn - alias->base_gfn;
1001 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1005 for (i = 0; i < kvm->nmemslots; ++i) {
1006 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1008 if (gfn >= memslot->base_gfn
1009 && gfn < memslot->base_gfn + memslot->npages)
1015 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1017 gfn = unalias_gfn(kvm, gfn);
1018 return __gfn_to_memslot(kvm, gfn);
1021 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1023 struct kvm_memory_slot *slot;
1025 gfn = unalias_gfn(kvm, gfn);
1026 slot = __gfn_to_memslot(kvm, gfn);
1029 return slot->phys_mem[gfn - slot->base_gfn];
1031 EXPORT_SYMBOL_GPL(gfn_to_page);
1033 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1036 struct kvm_memory_slot *memslot;
1037 unsigned long rel_gfn;
1039 for (i = 0; i < kvm->nmemslots; ++i) {
1040 memslot = &kvm->memslots[i];
1042 if (gfn >= memslot->base_gfn
1043 && gfn < memslot->base_gfn + memslot->npages) {
1045 if (!memslot->dirty_bitmap)
1048 rel_gfn = gfn - memslot->base_gfn;
1051 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1052 set_bit(rel_gfn, memslot->dirty_bitmap);
1058 static int emulator_read_std(unsigned long addr,
1061 struct x86_emulate_ctxt *ctxt)
1063 struct kvm_vcpu *vcpu = ctxt->vcpu;
1067 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1068 unsigned offset = addr & (PAGE_SIZE-1);
1069 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1074 if (gpa == UNMAPPED_GVA)
1075 return X86EMUL_PROPAGATE_FAULT;
1076 pfn = gpa >> PAGE_SHIFT;
1077 page = gfn_to_page(vcpu->kvm, pfn);
1079 return X86EMUL_UNHANDLEABLE;
1080 page_virt = kmap_atomic(page, KM_USER0);
1082 memcpy(data, page_virt + offset, tocopy);
1084 kunmap_atomic(page_virt, KM_USER0);
1091 return X86EMUL_CONTINUE;
1094 static int emulator_write_std(unsigned long addr,
1097 struct x86_emulate_ctxt *ctxt)
1099 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1101 return X86EMUL_UNHANDLEABLE;
1104 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1108 * Note that its important to have this wrapper function because
1109 * in the very near future we will be checking for MMIOs against
1110 * the LAPIC as well as the general MMIO bus
1112 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1115 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1118 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1121 static int emulator_read_emulated(unsigned long addr,
1124 struct x86_emulate_ctxt *ctxt)
1126 struct kvm_vcpu *vcpu = ctxt->vcpu;
1127 struct kvm_io_device *mmio_dev;
1130 if (vcpu->mmio_read_completed) {
1131 memcpy(val, vcpu->mmio_data, bytes);
1132 vcpu->mmio_read_completed = 0;
1133 return X86EMUL_CONTINUE;
1134 } else if (emulator_read_std(addr, val, bytes, ctxt)
1135 == X86EMUL_CONTINUE)
1136 return X86EMUL_CONTINUE;
1138 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1139 if (gpa == UNMAPPED_GVA)
1140 return X86EMUL_PROPAGATE_FAULT;
1143 * Is this MMIO handled locally?
1145 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1147 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1148 return X86EMUL_CONTINUE;
1151 vcpu->mmio_needed = 1;
1152 vcpu->mmio_phys_addr = gpa;
1153 vcpu->mmio_size = bytes;
1154 vcpu->mmio_is_write = 0;
1156 return X86EMUL_UNHANDLEABLE;
1159 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1160 const void *val, int bytes)
1164 unsigned offset = offset_in_page(gpa);
1166 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1168 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1171 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1172 virt = kmap_atomic(page, KM_USER0);
1173 kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1174 memcpy(virt + offset_in_page(gpa), val, bytes);
1175 kunmap_atomic(virt, KM_USER0);
1179 static int emulator_write_emulated(unsigned long addr,
1182 struct x86_emulate_ctxt *ctxt)
1184 struct kvm_vcpu *vcpu = ctxt->vcpu;
1185 struct kvm_io_device *mmio_dev;
1186 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1188 if (gpa == UNMAPPED_GVA) {
1189 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1190 return X86EMUL_PROPAGATE_FAULT;
1193 if (emulator_write_phys(vcpu, gpa, val, bytes))
1194 return X86EMUL_CONTINUE;
1197 * Is this MMIO handled locally?
1199 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1201 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1202 return X86EMUL_CONTINUE;
1205 vcpu->mmio_needed = 1;
1206 vcpu->mmio_phys_addr = gpa;
1207 vcpu->mmio_size = bytes;
1208 vcpu->mmio_is_write = 1;
1209 memcpy(vcpu->mmio_data, val, bytes);
1211 return X86EMUL_CONTINUE;
1214 static int emulator_cmpxchg_emulated(unsigned long addr,
1218 struct x86_emulate_ctxt *ctxt)
1220 static int reported;
1224 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1226 return emulator_write_emulated(addr, new, bytes, ctxt);
1229 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1231 return kvm_arch_ops->get_segment_base(vcpu, seg);
1234 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1236 return X86EMUL_CONTINUE;
1239 int emulate_clts(struct kvm_vcpu *vcpu)
1243 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1244 kvm_arch_ops->set_cr0(vcpu, cr0);
1245 return X86EMUL_CONTINUE;
1248 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1250 struct kvm_vcpu *vcpu = ctxt->vcpu;
1254 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1255 return X86EMUL_CONTINUE;
1257 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1259 return X86EMUL_UNHANDLEABLE;
1263 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1265 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1268 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1270 /* FIXME: better handling */
1271 return X86EMUL_UNHANDLEABLE;
1273 return X86EMUL_CONTINUE;
1276 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1278 static int reported;
1280 unsigned long rip = ctxt->vcpu->rip;
1281 unsigned long rip_linear;
1283 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1288 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1290 printk(KERN_ERR "emulation failed but !mmio_needed?"
1291 " rip %lx %02x %02x %02x %02x\n",
1292 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1296 struct x86_emulate_ops emulate_ops = {
1297 .read_std = emulator_read_std,
1298 .write_std = emulator_write_std,
1299 .read_emulated = emulator_read_emulated,
1300 .write_emulated = emulator_write_emulated,
1301 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1304 int emulate_instruction(struct kvm_vcpu *vcpu,
1305 struct kvm_run *run,
1309 struct x86_emulate_ctxt emulate_ctxt;
1313 vcpu->mmio_fault_cr2 = cr2;
1314 kvm_arch_ops->cache_regs(vcpu);
1316 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1318 emulate_ctxt.vcpu = vcpu;
1319 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1320 emulate_ctxt.cr2 = cr2;
1321 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1322 ? X86EMUL_MODE_REAL : cs_l
1323 ? X86EMUL_MODE_PROT64 : cs_db
1324 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1326 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1327 emulate_ctxt.cs_base = 0;
1328 emulate_ctxt.ds_base = 0;
1329 emulate_ctxt.es_base = 0;
1330 emulate_ctxt.ss_base = 0;
1332 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1333 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1334 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1335 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1338 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1339 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1341 vcpu->mmio_is_write = 0;
1342 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1344 if ((r || vcpu->mmio_is_write) && run) {
1345 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1346 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1347 run->mmio.len = vcpu->mmio_size;
1348 run->mmio.is_write = vcpu->mmio_is_write;
1352 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1353 return EMULATE_DONE;
1354 if (!vcpu->mmio_needed) {
1355 report_emulation_failure(&emulate_ctxt);
1356 return EMULATE_FAIL;
1358 return EMULATE_DO_MMIO;
1361 kvm_arch_ops->decache_regs(vcpu);
1362 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1364 if (vcpu->mmio_is_write) {
1365 vcpu->mmio_needed = 0;
1366 return EMULATE_DO_MMIO;
1369 return EMULATE_DONE;
1371 EXPORT_SYMBOL_GPL(emulate_instruction);
1373 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1375 if (vcpu->irq_summary)
1378 vcpu->run->exit_reason = KVM_EXIT_HLT;
1379 ++vcpu->stat.halt_exits;
1382 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1384 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1386 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1388 kvm_arch_ops->cache_regs(vcpu);
1390 #ifdef CONFIG_X86_64
1391 if (is_long_mode(vcpu)) {
1392 nr = vcpu->regs[VCPU_REGS_RAX];
1393 a0 = vcpu->regs[VCPU_REGS_RDI];
1394 a1 = vcpu->regs[VCPU_REGS_RSI];
1395 a2 = vcpu->regs[VCPU_REGS_RDX];
1396 a3 = vcpu->regs[VCPU_REGS_RCX];
1397 a4 = vcpu->regs[VCPU_REGS_R8];
1398 a5 = vcpu->regs[VCPU_REGS_R9];
1402 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1403 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1404 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1405 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1406 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1407 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1408 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1412 run->hypercall.args[0] = a0;
1413 run->hypercall.args[1] = a1;
1414 run->hypercall.args[2] = a2;
1415 run->hypercall.args[3] = a3;
1416 run->hypercall.args[4] = a4;
1417 run->hypercall.args[5] = a5;
1418 run->hypercall.ret = ret;
1419 run->hypercall.longmode = is_long_mode(vcpu);
1420 kvm_arch_ops->decache_regs(vcpu);
1423 vcpu->regs[VCPU_REGS_RAX] = ret;
1424 kvm_arch_ops->decache_regs(vcpu);
1427 EXPORT_SYMBOL_GPL(kvm_hypercall);
1429 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1431 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1434 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1436 struct descriptor_table dt = { limit, base };
1438 kvm_arch_ops->set_gdt(vcpu, &dt);
1441 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1443 struct descriptor_table dt = { limit, base };
1445 kvm_arch_ops->set_idt(vcpu, &dt);
1448 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1449 unsigned long *rflags)
1452 *rflags = kvm_arch_ops->get_rflags(vcpu);
1455 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1457 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1468 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1473 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1474 unsigned long *rflags)
1478 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1479 *rflags = kvm_arch_ops->get_rflags(vcpu);
1488 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1491 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1496 * Register the para guest with the host:
1498 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1500 struct kvm_vcpu_para_state *para_state;
1501 hpa_t para_state_hpa, hypercall_hpa;
1502 struct page *para_state_page;
1503 unsigned char *hypercall;
1504 gpa_t hypercall_gpa;
1506 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1507 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1510 * Needs to be page aligned:
1512 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1515 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1516 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1517 if (is_error_hpa(para_state_hpa))
1520 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1521 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1522 para_state = kmap_atomic(para_state_page, KM_USER0);
1524 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1525 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1527 para_state->host_version = KVM_PARA_API_VERSION;
1529 * We cannot support guests that try to register themselves
1530 * with a newer API version than the host supports:
1532 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1533 para_state->ret = -KVM_EINVAL;
1534 goto err_kunmap_skip;
1537 hypercall_gpa = para_state->hypercall_gpa;
1538 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1539 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1540 if (is_error_hpa(hypercall_hpa)) {
1541 para_state->ret = -KVM_EINVAL;
1542 goto err_kunmap_skip;
1545 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1546 vcpu->para_state_page = para_state_page;
1547 vcpu->para_state_gpa = para_state_gpa;
1548 vcpu->hypercall_gpa = hypercall_gpa;
1550 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1551 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1552 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1553 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1554 kunmap_atomic(hypercall, KM_USER1);
1556 para_state->ret = 0;
1558 kunmap_atomic(para_state, KM_USER0);
1564 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1569 case 0xc0010010: /* SYSCFG */
1570 case 0xc0010015: /* HWCR */
1571 case MSR_IA32_PLATFORM_ID:
1572 case MSR_IA32_P5_MC_ADDR:
1573 case MSR_IA32_P5_MC_TYPE:
1574 case MSR_IA32_MC0_CTL:
1575 case MSR_IA32_MCG_STATUS:
1576 case MSR_IA32_MCG_CAP:
1577 case MSR_IA32_MC0_MISC:
1578 case MSR_IA32_MC0_MISC+4:
1579 case MSR_IA32_MC0_MISC+8:
1580 case MSR_IA32_MC0_MISC+12:
1581 case MSR_IA32_MC0_MISC+16:
1582 case MSR_IA32_UCODE_REV:
1583 case MSR_IA32_PERF_STATUS:
1584 case MSR_IA32_EBL_CR_POWERON:
1585 /* MTRR registers */
1587 case 0x200 ... 0x2ff:
1590 case 0xcd: /* fsb frequency */
1593 case MSR_IA32_APICBASE:
1594 data = vcpu->apic_base;
1596 case MSR_IA32_MISC_ENABLE:
1597 data = vcpu->ia32_misc_enable_msr;
1599 #ifdef CONFIG_X86_64
1601 data = vcpu->shadow_efer;
1605 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1611 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1614 * Reads an msr value (of 'msr_index') into 'pdata'.
1615 * Returns 0 on success, non-0 otherwise.
1616 * Assumes vcpu_load() was already called.
1618 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1620 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1623 #ifdef CONFIG_X86_64
1625 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1627 if (efer & EFER_RESERVED_BITS) {
1628 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1635 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1636 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1641 kvm_arch_ops->set_efer(vcpu, efer);
1644 efer |= vcpu->shadow_efer & EFER_LMA;
1646 vcpu->shadow_efer = efer;
1651 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1654 #ifdef CONFIG_X86_64
1656 set_efer(vcpu, data);
1659 case MSR_IA32_MC0_STATUS:
1660 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1661 __FUNCTION__, data);
1663 case MSR_IA32_MCG_STATUS:
1664 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1665 __FUNCTION__, data);
1667 case MSR_IA32_UCODE_REV:
1668 case MSR_IA32_UCODE_WRITE:
1669 case 0x200 ... 0x2ff: /* MTRRs */
1671 case MSR_IA32_APICBASE:
1672 vcpu->apic_base = data;
1674 case MSR_IA32_MISC_ENABLE:
1675 vcpu->ia32_misc_enable_msr = data;
1678 * This is the 'probe whether the host is KVM' logic:
1680 case MSR_KVM_API_MAGIC:
1681 return vcpu_register_para(vcpu, data);
1684 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1689 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1692 * Writes msr value into into the appropriate "register".
1693 * Returns 0 on success, non-0 otherwise.
1694 * Assumes vcpu_load() was already called.
1696 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1698 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1701 void kvm_resched(struct kvm_vcpu *vcpu)
1703 if (!need_resched())
1709 EXPORT_SYMBOL_GPL(kvm_resched);
1711 void load_msrs(struct vmx_msr_entry *e, int n)
1715 for (i = 0; i < n; ++i)
1716 wrmsrl(e[i].index, e[i].data);
1718 EXPORT_SYMBOL_GPL(load_msrs);
1720 void save_msrs(struct vmx_msr_entry *e, int n)
1724 for (i = 0; i < n; ++i)
1725 rdmsrl(e[i].index, e[i].data);
1727 EXPORT_SYMBOL_GPL(save_msrs);
1729 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1733 struct kvm_cpuid_entry *e, *best;
1735 kvm_arch_ops->cache_regs(vcpu);
1736 function = vcpu->regs[VCPU_REGS_RAX];
1737 vcpu->regs[VCPU_REGS_RAX] = 0;
1738 vcpu->regs[VCPU_REGS_RBX] = 0;
1739 vcpu->regs[VCPU_REGS_RCX] = 0;
1740 vcpu->regs[VCPU_REGS_RDX] = 0;
1742 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1743 e = &vcpu->cpuid_entries[i];
1744 if (e->function == function) {
1749 * Both basic or both extended?
1751 if (((e->function ^ function) & 0x80000000) == 0)
1752 if (!best || e->function > best->function)
1756 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1757 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1758 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1759 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1761 kvm_arch_ops->decache_regs(vcpu);
1762 kvm_arch_ops->skip_emulated_instruction(vcpu);
1764 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1766 static int pio_copy_data(struct kvm_vcpu *vcpu)
1768 void *p = vcpu->pio_data;
1771 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1773 kvm_arch_ops->vcpu_put(vcpu);
1774 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1777 kvm_arch_ops->vcpu_load(vcpu);
1778 free_pio_guest_pages(vcpu);
1781 q += vcpu->pio.guest_page_offset;
1782 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1784 memcpy(q, p, bytes);
1786 memcpy(p, q, bytes);
1787 q -= vcpu->pio.guest_page_offset;
1789 kvm_arch_ops->vcpu_load(vcpu);
1790 free_pio_guest_pages(vcpu);
1794 static int complete_pio(struct kvm_vcpu *vcpu)
1796 struct kvm_pio_request *io = &vcpu->pio;
1800 kvm_arch_ops->cache_regs(vcpu);
1804 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1808 r = pio_copy_data(vcpu);
1810 kvm_arch_ops->cache_regs(vcpu);
1817 delta *= io->cur_count;
1819 * The size of the register should really depend on
1820 * current address size.
1822 vcpu->regs[VCPU_REGS_RCX] -= delta;
1828 vcpu->regs[VCPU_REGS_RDI] += delta;
1830 vcpu->regs[VCPU_REGS_RSI] += delta;
1833 kvm_arch_ops->decache_regs(vcpu);
1835 io->count -= io->cur_count;
1839 kvm_arch_ops->skip_emulated_instruction(vcpu);
1843 void kernel_pio(struct kvm_io_device *pio_dev, struct kvm_vcpu *vcpu)
1845 /* TODO: String I/O for in kernel device */
1848 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1852 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1857 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1858 int size, unsigned long count, int string, int down,
1859 gva_t address, int rep, unsigned port)
1861 unsigned now, in_page;
1865 struct kvm_io_device *pio_dev;
1867 vcpu->run->exit_reason = KVM_EXIT_IO;
1868 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1869 vcpu->run->io.size = size;
1870 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1871 vcpu->run->io.count = count;
1872 vcpu->run->io.port = port;
1873 vcpu->pio.count = count;
1874 vcpu->pio.cur_count = count;
1875 vcpu->pio.size = size;
1877 vcpu->pio.port = port;
1878 vcpu->pio.string = string;
1879 vcpu->pio.down = down;
1880 vcpu->pio.guest_page_offset = offset_in_page(address);
1881 vcpu->pio.rep = rep;
1883 pio_dev = vcpu_find_pio_dev(vcpu, port);
1885 kvm_arch_ops->cache_regs(vcpu);
1886 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1887 kvm_arch_ops->decache_regs(vcpu);
1889 kernel_pio(pio_dev, vcpu);
1895 /* TODO: String I/O for in kernel device */
1897 printk(KERN_ERR "kvm_setup_pio: no string io support\n");
1900 kvm_arch_ops->skip_emulated_instruction(vcpu);
1904 now = min(count, PAGE_SIZE / size);
1907 in_page = PAGE_SIZE - offset_in_page(address);
1909 in_page = offset_in_page(address) + size;
1910 now = min(count, (unsigned long)in_page / size);
1913 * String I/O straddles page boundary. Pin two guest pages
1914 * so that we satisfy atomicity constraints. Do just one
1915 * transaction to avoid complexity.
1922 * String I/O in reverse. Yuck. Kill the guest, fix later.
1924 printk(KERN_ERR "kvm: guest string pio down\n");
1928 vcpu->run->io.count = now;
1929 vcpu->pio.cur_count = now;
1931 for (i = 0; i < nr_pages; ++i) {
1932 spin_lock(&vcpu->kvm->lock);
1933 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1936 vcpu->pio.guest_pages[i] = page;
1937 spin_unlock(&vcpu->kvm->lock);
1940 free_pio_guest_pages(vcpu);
1946 return pio_copy_data(vcpu);
1949 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1951 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1958 if (vcpu->sigset_active)
1959 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1961 /* re-sync apic's tpr */
1962 vcpu->cr8 = kvm_run->cr8;
1964 if (vcpu->pio.cur_count) {
1965 r = complete_pio(vcpu);
1970 if (vcpu->mmio_needed) {
1971 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1972 vcpu->mmio_read_completed = 1;
1973 vcpu->mmio_needed = 0;
1974 r = emulate_instruction(vcpu, kvm_run,
1975 vcpu->mmio_fault_cr2, 0);
1976 if (r == EMULATE_DO_MMIO) {
1978 * Read-modify-write. Back to userspace.
1980 kvm_run->exit_reason = KVM_EXIT_MMIO;
1986 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1987 kvm_arch_ops->cache_regs(vcpu);
1988 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1989 kvm_arch_ops->decache_regs(vcpu);
1992 r = kvm_arch_ops->run(vcpu, kvm_run);
1995 if (vcpu->sigset_active)
1996 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2002 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2003 struct kvm_regs *regs)
2007 kvm_arch_ops->cache_regs(vcpu);
2009 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2010 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2011 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2012 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2013 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2014 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2015 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2016 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2017 #ifdef CONFIG_X86_64
2018 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2019 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2020 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2021 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2022 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2023 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2024 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2025 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2028 regs->rip = vcpu->rip;
2029 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2032 * Don't leak debug flags in case they were set for guest debugging
2034 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2035 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2042 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2043 struct kvm_regs *regs)
2047 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2048 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2049 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2050 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2051 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2052 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2053 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2054 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2055 #ifdef CONFIG_X86_64
2056 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2057 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2058 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2059 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2060 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2061 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2062 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2063 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2066 vcpu->rip = regs->rip;
2067 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2069 kvm_arch_ops->decache_regs(vcpu);
2076 static void get_segment(struct kvm_vcpu *vcpu,
2077 struct kvm_segment *var, int seg)
2079 return kvm_arch_ops->get_segment(vcpu, var, seg);
2082 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2083 struct kvm_sregs *sregs)
2085 struct descriptor_table dt;
2089 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2090 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2091 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2092 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2093 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2094 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2096 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2097 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2099 kvm_arch_ops->get_idt(vcpu, &dt);
2100 sregs->idt.limit = dt.limit;
2101 sregs->idt.base = dt.base;
2102 kvm_arch_ops->get_gdt(vcpu, &dt);
2103 sregs->gdt.limit = dt.limit;
2104 sregs->gdt.base = dt.base;
2106 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2107 sregs->cr0 = vcpu->cr0;
2108 sregs->cr2 = vcpu->cr2;
2109 sregs->cr3 = vcpu->cr3;
2110 sregs->cr4 = vcpu->cr4;
2111 sregs->cr8 = vcpu->cr8;
2112 sregs->efer = vcpu->shadow_efer;
2113 sregs->apic_base = vcpu->apic_base;
2115 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2116 sizeof sregs->interrupt_bitmap);
2123 static void set_segment(struct kvm_vcpu *vcpu,
2124 struct kvm_segment *var, int seg)
2126 return kvm_arch_ops->set_segment(vcpu, var, seg);
2129 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2130 struct kvm_sregs *sregs)
2132 int mmu_reset_needed = 0;
2134 struct descriptor_table dt;
2138 dt.limit = sregs->idt.limit;
2139 dt.base = sregs->idt.base;
2140 kvm_arch_ops->set_idt(vcpu, &dt);
2141 dt.limit = sregs->gdt.limit;
2142 dt.base = sregs->gdt.base;
2143 kvm_arch_ops->set_gdt(vcpu, &dt);
2145 vcpu->cr2 = sregs->cr2;
2146 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2147 vcpu->cr3 = sregs->cr3;
2149 vcpu->cr8 = sregs->cr8;
2151 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2152 #ifdef CONFIG_X86_64
2153 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2155 vcpu->apic_base = sregs->apic_base;
2157 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2159 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2160 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2162 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2163 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2164 if (!is_long_mode(vcpu) && is_pae(vcpu))
2165 load_pdptrs(vcpu, vcpu->cr3);
2167 if (mmu_reset_needed)
2168 kvm_mmu_reset_context(vcpu);
2170 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2171 sizeof vcpu->irq_pending);
2172 vcpu->irq_summary = 0;
2173 for (i = 0; i < NR_IRQ_WORDS; ++i)
2174 if (vcpu->irq_pending[i])
2175 __set_bit(i, &vcpu->irq_summary);
2177 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2178 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2179 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2180 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2181 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2182 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2184 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2185 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2193 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2194 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2196 * This list is modified at module load time to reflect the
2197 * capabilities of the host cpu.
2199 static u32 msrs_to_save[] = {
2200 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2202 #ifdef CONFIG_X86_64
2203 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2205 MSR_IA32_TIME_STAMP_COUNTER,
2208 static unsigned num_msrs_to_save;
2210 static u32 emulated_msrs[] = {
2211 MSR_IA32_MISC_ENABLE,
2214 static __init void kvm_init_msr_list(void)
2219 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2220 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2223 msrs_to_save[j] = msrs_to_save[i];
2226 num_msrs_to_save = j;
2230 * Adapt set_msr() to msr_io()'s calling convention
2232 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2234 return set_msr(vcpu, index, *data);
2238 * Read or write a bunch of msrs. All parameters are kernel addresses.
2240 * @return number of msrs set successfully.
2242 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2243 struct kvm_msr_entry *entries,
2244 int (*do_msr)(struct kvm_vcpu *vcpu,
2245 unsigned index, u64 *data))
2251 for (i = 0; i < msrs->nmsrs; ++i)
2252 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2261 * Read or write a bunch of msrs. Parameters are user addresses.
2263 * @return number of msrs set successfully.
2265 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2266 int (*do_msr)(struct kvm_vcpu *vcpu,
2267 unsigned index, u64 *data),
2270 struct kvm_msrs msrs;
2271 struct kvm_msr_entry *entries;
2276 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2280 if (msrs.nmsrs >= MAX_IO_MSRS)
2284 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2285 entries = vmalloc(size);
2290 if (copy_from_user(entries, user_msrs->entries, size))
2293 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2298 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2310 * Translate a guest virtual address to a guest physical address.
2312 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2313 struct kvm_translation *tr)
2315 unsigned long vaddr = tr->linear_address;
2319 spin_lock(&vcpu->kvm->lock);
2320 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2321 tr->physical_address = gpa;
2322 tr->valid = gpa != UNMAPPED_GVA;
2325 spin_unlock(&vcpu->kvm->lock);
2331 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2332 struct kvm_interrupt *irq)
2334 if (irq->irq < 0 || irq->irq >= 256)
2338 set_bit(irq->irq, vcpu->irq_pending);
2339 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2346 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2347 struct kvm_debug_guest *dbg)
2353 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2360 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2361 unsigned long address,
2364 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2365 unsigned long pgoff;
2368 *type = VM_FAULT_MINOR;
2369 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2371 page = virt_to_page(vcpu->run);
2372 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2373 page = virt_to_page(vcpu->pio_data);
2375 return NOPAGE_SIGBUS;
2380 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2381 .nopage = kvm_vcpu_nopage,
2384 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2386 vma->vm_ops = &kvm_vcpu_vm_ops;
2390 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2392 struct kvm_vcpu *vcpu = filp->private_data;
2394 fput(vcpu->kvm->filp);
2398 static struct file_operations kvm_vcpu_fops = {
2399 .release = kvm_vcpu_release,
2400 .unlocked_ioctl = kvm_vcpu_ioctl,
2401 .compat_ioctl = kvm_vcpu_ioctl,
2402 .mmap = kvm_vcpu_mmap,
2406 * Allocates an inode for the vcpu.
2408 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2411 struct inode *inode;
2414 atomic_inc(&vcpu->kvm->filp->f_count);
2415 inode = kvmfs_inode(&kvm_vcpu_fops);
2416 if (IS_ERR(inode)) {
2421 file = kvmfs_file(inode, vcpu);
2427 r = get_unused_fd();
2431 fd_install(fd, file);
2440 fput(vcpu->kvm->filp);
2445 * Creates some virtual cpus. Good luck creating more than one.
2447 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2450 struct kvm_vcpu *vcpu;
2457 vcpu = &kvm->vcpus[n];
2459 mutex_lock(&vcpu->mutex);
2462 mutex_unlock(&vcpu->mutex);
2466 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2470 vcpu->run = page_address(page);
2472 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2476 vcpu->pio_data = page_address(page);
2478 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2480 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2483 r = kvm_arch_ops->vcpu_create(vcpu);
2485 goto out_free_vcpus;
2487 r = kvm_mmu_create(vcpu);
2489 goto out_free_vcpus;
2491 kvm_arch_ops->vcpu_load(vcpu);
2492 r = kvm_mmu_setup(vcpu);
2494 r = kvm_arch_ops->vcpu_setup(vcpu);
2498 goto out_free_vcpus;
2500 r = create_vcpu_fd(vcpu);
2502 goto out_free_vcpus;
2504 spin_lock(&kvm_lock);
2505 if (n >= kvm->nvcpus)
2506 kvm->nvcpus = n + 1;
2507 spin_unlock(&kvm_lock);
2512 kvm_free_vcpu(vcpu);
2514 free_page((unsigned long)vcpu->run);
2517 mutex_unlock(&vcpu->mutex);
2522 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2526 struct kvm_cpuid_entry *e, *entry;
2528 rdmsrl(MSR_EFER, efer);
2530 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2531 e = &vcpu->cpuid_entries[i];
2532 if (e->function == 0x80000001) {
2537 if (entry && (entry->edx & EFER_NX) && !(efer & EFER_NX)) {
2538 entry->edx &= ~(1 << 20);
2539 printk(KERN_INFO ": guest NX capability removed\n");
2543 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2544 struct kvm_cpuid *cpuid,
2545 struct kvm_cpuid_entry __user *entries)
2550 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2553 if (copy_from_user(&vcpu->cpuid_entries, entries,
2554 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2556 vcpu->cpuid_nent = cpuid->nent;
2557 cpuid_fix_nx_cap(vcpu);
2564 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2567 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2568 vcpu->sigset_active = 1;
2569 vcpu->sigset = *sigset;
2571 vcpu->sigset_active = 0;
2576 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2577 * we have asm/x86/processor.h
2588 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2589 #ifdef CONFIG_X86_64
2590 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2592 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2596 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2598 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2602 memcpy(fpu->fpr, fxsave->st_space, 128);
2603 fpu->fcw = fxsave->cwd;
2604 fpu->fsw = fxsave->swd;
2605 fpu->ftwx = fxsave->twd;
2606 fpu->last_opcode = fxsave->fop;
2607 fpu->last_ip = fxsave->rip;
2608 fpu->last_dp = fxsave->rdp;
2609 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2616 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2618 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2622 memcpy(fxsave->st_space, fpu->fpr, 128);
2623 fxsave->cwd = fpu->fcw;
2624 fxsave->swd = fpu->fsw;
2625 fxsave->twd = fpu->ftwx;
2626 fxsave->fop = fpu->last_opcode;
2627 fxsave->rip = fpu->last_ip;
2628 fxsave->rdp = fpu->last_dp;
2629 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2636 static long kvm_vcpu_ioctl(struct file *filp,
2637 unsigned int ioctl, unsigned long arg)
2639 struct kvm_vcpu *vcpu = filp->private_data;
2640 void __user *argp = (void __user *)arg;
2648 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2650 case KVM_GET_REGS: {
2651 struct kvm_regs kvm_regs;
2653 memset(&kvm_regs, 0, sizeof kvm_regs);
2654 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2658 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2663 case KVM_SET_REGS: {
2664 struct kvm_regs kvm_regs;
2667 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2669 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2675 case KVM_GET_SREGS: {
2676 struct kvm_sregs kvm_sregs;
2678 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2679 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2683 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2688 case KVM_SET_SREGS: {
2689 struct kvm_sregs kvm_sregs;
2692 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2694 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2700 case KVM_TRANSLATE: {
2701 struct kvm_translation tr;
2704 if (copy_from_user(&tr, argp, sizeof tr))
2706 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2710 if (copy_to_user(argp, &tr, sizeof tr))
2715 case KVM_INTERRUPT: {
2716 struct kvm_interrupt irq;
2719 if (copy_from_user(&irq, argp, sizeof irq))
2721 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2727 case KVM_DEBUG_GUEST: {
2728 struct kvm_debug_guest dbg;
2731 if (copy_from_user(&dbg, argp, sizeof dbg))
2733 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2740 r = msr_io(vcpu, argp, get_msr, 1);
2743 r = msr_io(vcpu, argp, do_set_msr, 0);
2745 case KVM_SET_CPUID: {
2746 struct kvm_cpuid __user *cpuid_arg = argp;
2747 struct kvm_cpuid cpuid;
2750 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2752 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2757 case KVM_SET_SIGNAL_MASK: {
2758 struct kvm_signal_mask __user *sigmask_arg = argp;
2759 struct kvm_signal_mask kvm_sigmask;
2760 sigset_t sigset, *p;
2765 if (copy_from_user(&kvm_sigmask, argp,
2766 sizeof kvm_sigmask))
2769 if (kvm_sigmask.len != sizeof sigset)
2772 if (copy_from_user(&sigset, sigmask_arg->sigset,
2777 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2783 memset(&fpu, 0, sizeof fpu);
2784 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2788 if (copy_to_user(argp, &fpu, sizeof fpu))
2797 if (copy_from_user(&fpu, argp, sizeof fpu))
2799 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2812 static long kvm_vm_ioctl(struct file *filp,
2813 unsigned int ioctl, unsigned long arg)
2815 struct kvm *kvm = filp->private_data;
2816 void __user *argp = (void __user *)arg;
2820 case KVM_CREATE_VCPU:
2821 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2825 case KVM_SET_MEMORY_REGION: {
2826 struct kvm_memory_region kvm_mem;
2829 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2831 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2836 case KVM_GET_DIRTY_LOG: {
2837 struct kvm_dirty_log log;
2840 if (copy_from_user(&log, argp, sizeof log))
2842 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2847 case KVM_SET_MEMORY_ALIAS: {
2848 struct kvm_memory_alias alias;
2851 if (copy_from_user(&alias, argp, sizeof alias))
2853 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2865 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2866 unsigned long address,
2869 struct kvm *kvm = vma->vm_file->private_data;
2870 unsigned long pgoff;
2873 *type = VM_FAULT_MINOR;
2874 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2875 page = gfn_to_page(kvm, pgoff);
2877 return NOPAGE_SIGBUS;
2882 static struct vm_operations_struct kvm_vm_vm_ops = {
2883 .nopage = kvm_vm_nopage,
2886 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2888 vma->vm_ops = &kvm_vm_vm_ops;
2892 static struct file_operations kvm_vm_fops = {
2893 .release = kvm_vm_release,
2894 .unlocked_ioctl = kvm_vm_ioctl,
2895 .compat_ioctl = kvm_vm_ioctl,
2896 .mmap = kvm_vm_mmap,
2899 static int kvm_dev_ioctl_create_vm(void)
2902 struct inode *inode;
2906 inode = kvmfs_inode(&kvm_vm_fops);
2907 if (IS_ERR(inode)) {
2912 kvm = kvm_create_vm();
2918 file = kvmfs_file(inode, kvm);
2925 r = get_unused_fd();
2929 fd_install(fd, file);
2936 kvm_destroy_vm(kvm);
2943 static long kvm_dev_ioctl(struct file *filp,
2944 unsigned int ioctl, unsigned long arg)
2946 void __user *argp = (void __user *)arg;
2950 case KVM_GET_API_VERSION:
2954 r = KVM_API_VERSION;
2960 r = kvm_dev_ioctl_create_vm();
2962 case KVM_GET_MSR_INDEX_LIST: {
2963 struct kvm_msr_list __user *user_msr_list = argp;
2964 struct kvm_msr_list msr_list;
2968 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2971 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2972 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2975 if (n < num_msrs_to_save)
2978 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2979 num_msrs_to_save * sizeof(u32)))
2981 if (copy_to_user(user_msr_list->indices
2982 + num_msrs_to_save * sizeof(u32),
2984 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2989 case KVM_CHECK_EXTENSION:
2991 * No extensions defined at present.
2995 case KVM_GET_VCPU_MMAP_SIZE:
3008 static struct file_operations kvm_chardev_ops = {
3009 .open = kvm_dev_open,
3010 .release = kvm_dev_release,
3011 .unlocked_ioctl = kvm_dev_ioctl,
3012 .compat_ioctl = kvm_dev_ioctl,
3015 static struct miscdevice kvm_dev = {
3021 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3024 if (val == SYS_RESTART) {
3026 * Some (well, at least mine) BIOSes hang on reboot if
3029 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3030 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3035 static struct notifier_block kvm_reboot_notifier = {
3036 .notifier_call = kvm_reboot,
3041 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3044 static void decache_vcpus_on_cpu(int cpu)
3047 struct kvm_vcpu *vcpu;
3050 spin_lock(&kvm_lock);
3051 list_for_each_entry(vm, &vm_list, vm_list)
3052 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3053 vcpu = &vm->vcpus[i];
3055 * If the vcpu is locked, then it is running on some
3056 * other cpu and therefore it is not cached on the
3059 * If it's not locked, check the last cpu it executed
3062 if (mutex_trylock(&vcpu->mutex)) {
3063 if (vcpu->cpu == cpu) {
3064 kvm_arch_ops->vcpu_decache(vcpu);
3067 mutex_unlock(&vcpu->mutex);
3070 spin_unlock(&kvm_lock);
3073 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3079 case CPU_DOWN_PREPARE:
3080 case CPU_DOWN_PREPARE_FROZEN:
3081 case CPU_UP_CANCELED:
3082 case CPU_UP_CANCELED_FROZEN:
3083 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3085 decache_vcpus_on_cpu(cpu);
3086 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
3090 case CPU_ONLINE_FROZEN:
3091 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3093 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
3100 void kvm_io_bus_init(struct kvm_io_bus *bus)
3102 memset(bus, 0, sizeof(*bus));
3105 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3109 for (i = 0; i < bus->dev_count; i++) {
3110 struct kvm_io_device *pos = bus->devs[i];
3112 kvm_iodevice_destructor(pos);
3116 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3120 for (i = 0; i < bus->dev_count; i++) {
3121 struct kvm_io_device *pos = bus->devs[i];
3123 if (pos->in_range(pos, addr))
3130 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3132 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3134 bus->devs[bus->dev_count++] = dev;
3137 static struct notifier_block kvm_cpu_notifier = {
3138 .notifier_call = kvm_cpu_hotplug,
3139 .priority = 20, /* must be > scheduler priority */
3142 static u64 stat_get(void *_offset)
3144 unsigned offset = (long)_offset;
3147 struct kvm_vcpu *vcpu;
3150 spin_lock(&kvm_lock);
3151 list_for_each_entry(kvm, &vm_list, vm_list)
3152 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3153 vcpu = &kvm->vcpus[i];
3154 total += *(u32 *)((void *)vcpu + offset);
3156 spin_unlock(&kvm_lock);
3160 static void stat_set(void *offset, u64 val)
3164 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3166 static __init void kvm_init_debug(void)
3168 struct kvm_stats_debugfs_item *p;
3170 debugfs_dir = debugfs_create_dir("kvm", NULL);
3171 for (p = debugfs_entries; p->name; ++p)
3172 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3173 (void *)(long)p->offset,
3177 static void kvm_exit_debug(void)
3179 struct kvm_stats_debugfs_item *p;
3181 for (p = debugfs_entries; p->name; ++p)
3182 debugfs_remove(p->dentry);
3183 debugfs_remove(debugfs_dir);
3186 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3188 decache_vcpus_on_cpu(raw_smp_processor_id());
3189 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3193 static int kvm_resume(struct sys_device *dev)
3195 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3199 static struct sysdev_class kvm_sysdev_class = {
3200 set_kset_name("kvm"),
3201 .suspend = kvm_suspend,
3202 .resume = kvm_resume,
3205 static struct sys_device kvm_sysdev = {
3207 .cls = &kvm_sysdev_class,
3210 hpa_t bad_page_address;
3212 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
3213 const char *dev_name, void *data, struct vfsmount *mnt)
3215 return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
3218 static struct file_system_type kvm_fs_type = {
3220 .get_sb = kvmfs_get_sb,
3221 .kill_sb = kill_anon_super,
3224 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3229 printk(KERN_ERR "kvm: already loaded the other module\n");
3233 if (!ops->cpu_has_kvm_support()) {
3234 printk(KERN_ERR "kvm: no hardware support\n");
3237 if (ops->disabled_by_bios()) {
3238 printk(KERN_ERR "kvm: disabled by bios\n");
3244 r = kvm_arch_ops->hardware_setup();
3248 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
3249 r = register_cpu_notifier(&kvm_cpu_notifier);
3252 register_reboot_notifier(&kvm_reboot_notifier);
3254 r = sysdev_class_register(&kvm_sysdev_class);
3258 r = sysdev_register(&kvm_sysdev);
3262 kvm_chardev_ops.owner = module;
3264 r = misc_register(&kvm_dev);
3266 printk (KERN_ERR "kvm: misc device register failed\n");
3273 sysdev_unregister(&kvm_sysdev);
3275 sysdev_class_unregister(&kvm_sysdev_class);
3277 unregister_reboot_notifier(&kvm_reboot_notifier);
3278 unregister_cpu_notifier(&kvm_cpu_notifier);
3280 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3281 kvm_arch_ops->hardware_unsetup();
3283 kvm_arch_ops = NULL;
3287 void kvm_exit_arch(void)
3289 misc_deregister(&kvm_dev);
3290 sysdev_unregister(&kvm_sysdev);
3291 sysdev_class_unregister(&kvm_sysdev_class);
3292 unregister_reboot_notifier(&kvm_reboot_notifier);
3293 unregister_cpu_notifier(&kvm_cpu_notifier);
3294 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
3295 kvm_arch_ops->hardware_unsetup();
3296 kvm_arch_ops = NULL;
3299 static __init int kvm_init(void)
3301 static struct page *bad_page;
3304 r = kvm_mmu_module_init();
3308 r = register_filesystem(&kvm_fs_type);
3312 kvmfs_mnt = kern_mount(&kvm_fs_type);
3313 r = PTR_ERR(kvmfs_mnt);
3314 if (IS_ERR(kvmfs_mnt))
3318 kvm_init_msr_list();
3320 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3325 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3326 memset(__va(bad_page_address), 0, PAGE_SIZE);
3334 unregister_filesystem(&kvm_fs_type);
3336 kvm_mmu_module_exit();
3341 static __exit void kvm_exit(void)
3344 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3346 unregister_filesystem(&kvm_fs_type);
3347 kvm_mmu_module_exit();
3350 module_init(kvm_init)
3351 module_exit(kvm_exit)
3353 EXPORT_SYMBOL_GPL(kvm_init_arch);
3354 EXPORT_SYMBOL_GPL(kvm_exit_arch);