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>
44 #include "x86_emulate.h"
45 #include "segment_descriptor.h"
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
53 struct kvm_arch_ops *kvm_arch_ops;
54 struct kvm_stat kvm_stat;
55 EXPORT_SYMBOL_GPL(kvm_stat);
57 static struct kvm_stats_debugfs_item {
60 struct dentry *dentry;
61 } debugfs_entries[] = {
62 { "pf_fixed", &kvm_stat.pf_fixed },
63 { "pf_guest", &kvm_stat.pf_guest },
64 { "tlb_flush", &kvm_stat.tlb_flush },
65 { "invlpg", &kvm_stat.invlpg },
66 { "exits", &kvm_stat.exits },
67 { "io_exits", &kvm_stat.io_exits },
68 { "mmio_exits", &kvm_stat.mmio_exits },
69 { "signal_exits", &kvm_stat.signal_exits },
70 { "irq_window", &kvm_stat.irq_window_exits },
71 { "halt_exits", &kvm_stat.halt_exits },
72 { "request_irq", &kvm_stat.request_irq_exits },
73 { "irq_exits", &kvm_stat.irq_exits },
77 static struct dentry *debugfs_dir;
79 struct vfsmount *kvmfs_mnt;
81 #define MAX_IO_MSRS 256
83 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
84 #define LMSW_GUEST_MASK 0x0eULL
85 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
86 #define CR8_RESEVED_BITS (~0x0fULL)
87 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
90 // LDT or TSS descriptor in the GDT. 16 bytes.
91 struct segment_descriptor_64 {
92 struct segment_descriptor s;
99 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
102 static struct inode *kvmfs_inode(struct file_operations *fops)
105 struct inode *inode = new_inode(kvmfs_mnt->mnt_sb);
113 * Mark the inode dirty from the very beginning,
114 * that way it will never be moved to the dirty
115 * list because mark_inode_dirty() will think
116 * that it already _is_ on the dirty list.
118 inode->i_state = I_DIRTY;
119 inode->i_mode = S_IRUSR | S_IWUSR;
120 inode->i_uid = current->fsuid;
121 inode->i_gid = current->fsgid;
122 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
126 return ERR_PTR(error);
129 static struct file *kvmfs_file(struct inode *inode, void *private_data)
131 struct file *file = get_empty_filp();
134 return ERR_PTR(-ENFILE);
136 file->f_path.mnt = mntget(kvmfs_mnt);
137 file->f_path.dentry = d_alloc_anon(inode);
138 if (!file->f_path.dentry)
139 return ERR_PTR(-ENOMEM);
140 file->f_mapping = inode->i_mapping;
143 file->f_flags = O_RDWR;
144 file->f_op = inode->i_fop;
145 file->f_mode = FMODE_READ | FMODE_WRITE;
147 file->private_data = private_data;
151 unsigned long segment_base(u16 selector)
153 struct descriptor_table gdt;
154 struct segment_descriptor *d;
155 unsigned long table_base;
156 typedef unsigned long ul;
162 asm ("sgdt %0" : "=m"(gdt));
163 table_base = gdt.base;
165 if (selector & 4) { /* from ldt */
168 asm ("sldt %0" : "=g"(ldt_selector));
169 table_base = segment_base(ldt_selector);
171 d = (struct segment_descriptor *)(table_base + (selector & ~7));
172 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
175 && (d->type == 2 || d->type == 9 || d->type == 11))
176 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
180 EXPORT_SYMBOL_GPL(segment_base);
182 static inline int valid_vcpu(int n)
184 return likely(n >= 0 && n < KVM_MAX_VCPUS);
187 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
190 unsigned char *host_buf = dest;
191 unsigned long req_size = size;
199 paddr = gva_to_hpa(vcpu, addr);
201 if (is_error_hpa(paddr))
204 guest_buf = (hva_t)kmap_atomic(
205 pfn_to_page(paddr >> PAGE_SHIFT),
207 offset = addr & ~PAGE_MASK;
209 now = min(size, PAGE_SIZE - offset);
210 memcpy(host_buf, (void*)guest_buf, now);
214 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
216 return req_size - size;
218 EXPORT_SYMBOL_GPL(kvm_read_guest);
220 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
223 unsigned char *host_buf = data;
224 unsigned long req_size = size;
233 paddr = gva_to_hpa(vcpu, addr);
235 if (is_error_hpa(paddr))
238 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
239 mark_page_dirty(vcpu->kvm, gfn);
240 guest_buf = (hva_t)kmap_atomic(
241 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
242 offset = addr & ~PAGE_MASK;
244 now = min(size, PAGE_SIZE - offset);
245 memcpy((void*)guest_buf, host_buf, now);
249 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
251 return req_size - size;
253 EXPORT_SYMBOL_GPL(kvm_write_guest);
256 * Switches to specified vcpu, until a matching vcpu_put()
258 static void vcpu_load(struct kvm_vcpu *vcpu)
260 mutex_lock(&vcpu->mutex);
261 kvm_arch_ops->vcpu_load(vcpu);
265 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
266 * if the slot is not populated.
268 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
270 struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
272 mutex_lock(&vcpu->mutex);
274 mutex_unlock(&vcpu->mutex);
277 kvm_arch_ops->vcpu_load(vcpu);
281 static void vcpu_put(struct kvm_vcpu *vcpu)
283 kvm_arch_ops->vcpu_put(vcpu);
284 mutex_unlock(&vcpu->mutex);
287 static struct kvm *kvm_create_vm(void)
289 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
293 return ERR_PTR(-ENOMEM);
295 spin_lock_init(&kvm->lock);
296 INIT_LIST_HEAD(&kvm->active_mmu_pages);
297 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
298 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
300 mutex_init(&vcpu->mutex);
303 vcpu->mmu.root_hpa = INVALID_PAGE;
304 INIT_LIST_HEAD(&vcpu->free_pages);
305 spin_lock(&kvm_lock);
306 list_add(&kvm->vm_list, &vm_list);
307 spin_unlock(&kvm_lock);
312 static int kvm_dev_open(struct inode *inode, struct file *filp)
318 * Free any memory in @free but not in @dont.
320 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
321 struct kvm_memory_slot *dont)
325 if (!dont || free->phys_mem != dont->phys_mem)
326 if (free->phys_mem) {
327 for (i = 0; i < free->npages; ++i)
328 if (free->phys_mem[i])
329 __free_page(free->phys_mem[i]);
330 vfree(free->phys_mem);
333 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
334 vfree(free->dirty_bitmap);
336 free->phys_mem = NULL;
338 free->dirty_bitmap = NULL;
341 static void kvm_free_physmem(struct kvm *kvm)
345 for (i = 0; i < kvm->nmemslots; ++i)
346 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
349 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
355 kvm_mmu_destroy(vcpu);
357 kvm_arch_ops->vcpu_free(vcpu);
358 free_page((unsigned long)vcpu->run);
362 static void kvm_free_vcpus(struct kvm *kvm)
366 for (i = 0; i < KVM_MAX_VCPUS; ++i)
367 kvm_free_vcpu(&kvm->vcpus[i]);
370 static int kvm_dev_release(struct inode *inode, struct file *filp)
375 static void kvm_destroy_vm(struct kvm *kvm)
377 spin_lock(&kvm_lock);
378 list_del(&kvm->vm_list);
379 spin_unlock(&kvm_lock);
381 kvm_free_physmem(kvm);
385 static int kvm_vm_release(struct inode *inode, struct file *filp)
387 struct kvm *kvm = filp->private_data;
393 static void inject_gp(struct kvm_vcpu *vcpu)
395 kvm_arch_ops->inject_gp(vcpu, 0);
399 * Load the pae pdptrs. Return true is they are all valid.
401 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
403 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
404 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
409 struct kvm_memory_slot *memslot;
411 spin_lock(&vcpu->kvm->lock);
412 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
413 /* FIXME: !memslot - emulate? 0xff? */
414 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
417 for (i = 0; i < 4; ++i) {
418 pdpte = pdpt[offset + i];
419 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
425 for (i = 0; i < 4; ++i)
426 vcpu->pdptrs[i] = pdpt[offset + i];
429 kunmap_atomic(pdpt, KM_USER0);
430 spin_unlock(&vcpu->kvm->lock);
435 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
437 if (cr0 & CR0_RESEVED_BITS) {
438 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
444 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
445 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
450 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
451 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
452 "and a clear PE flag\n");
457 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
459 if ((vcpu->shadow_efer & EFER_LME)) {
463 printk(KERN_DEBUG "set_cr0: #GP, start paging "
464 "in long mode while PAE is disabled\n");
468 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
470 printk(KERN_DEBUG "set_cr0: #GP, start paging "
471 "in long mode while CS.L == 1\n");
478 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
479 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
487 kvm_arch_ops->set_cr0(vcpu, cr0);
490 spin_lock(&vcpu->kvm->lock);
491 kvm_mmu_reset_context(vcpu);
492 spin_unlock(&vcpu->kvm->lock);
495 EXPORT_SYMBOL_GPL(set_cr0);
497 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
499 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
500 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
502 EXPORT_SYMBOL_GPL(lmsw);
504 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
506 if (cr4 & CR4_RESEVED_BITS) {
507 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
512 if (is_long_mode(vcpu)) {
513 if (!(cr4 & CR4_PAE_MASK)) {
514 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
519 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
520 && !load_pdptrs(vcpu, vcpu->cr3)) {
521 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
525 if (cr4 & CR4_VMXE_MASK) {
526 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
530 kvm_arch_ops->set_cr4(vcpu, cr4);
531 spin_lock(&vcpu->kvm->lock);
532 kvm_mmu_reset_context(vcpu);
533 spin_unlock(&vcpu->kvm->lock);
535 EXPORT_SYMBOL_GPL(set_cr4);
537 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
539 if (is_long_mode(vcpu)) {
540 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
541 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
546 if (cr3 & CR3_RESEVED_BITS) {
547 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
551 if (is_paging(vcpu) && is_pae(vcpu) &&
552 !load_pdptrs(vcpu, cr3)) {
553 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
561 spin_lock(&vcpu->kvm->lock);
563 * Does the new cr3 value map to physical memory? (Note, we
564 * catch an invalid cr3 even in real-mode, because it would
565 * cause trouble later on when we turn on paging anyway.)
567 * A real CPU would silently accept an invalid cr3 and would
568 * attempt to use it - with largely undefined (and often hard
569 * to debug) behavior on the guest side.
571 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
574 vcpu->mmu.new_cr3(vcpu);
575 spin_unlock(&vcpu->kvm->lock);
577 EXPORT_SYMBOL_GPL(set_cr3);
579 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
581 if ( cr8 & CR8_RESEVED_BITS) {
582 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
588 EXPORT_SYMBOL_GPL(set_cr8);
590 void fx_init(struct kvm_vcpu *vcpu)
592 struct __attribute__ ((__packed__)) fx_image_s {
598 u64 operand;// fpu dp
604 fx_save(vcpu->host_fx_image);
606 fx_save(vcpu->guest_fx_image);
607 fx_restore(vcpu->host_fx_image);
609 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
610 fx_image->mxcsr = 0x1f80;
611 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
612 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
614 EXPORT_SYMBOL_GPL(fx_init);
616 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
618 spin_lock(&vcpu->kvm->lock);
619 kvm_mmu_slot_remove_write_access(vcpu, slot);
620 spin_unlock(&vcpu->kvm->lock);
624 * Allocate some memory and give it an address in the guest physical address
627 * Discontiguous memory is allowed, mostly for framebuffers.
629 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
630 struct kvm_memory_region *mem)
634 unsigned long npages;
636 struct kvm_memory_slot *memslot;
637 struct kvm_memory_slot old, new;
638 int memory_config_version;
641 /* General sanity checks */
642 if (mem->memory_size & (PAGE_SIZE - 1))
644 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
646 if (mem->slot >= KVM_MEMORY_SLOTS)
648 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
651 memslot = &kvm->memslots[mem->slot];
652 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
653 npages = mem->memory_size >> PAGE_SHIFT;
656 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
659 spin_lock(&kvm->lock);
661 memory_config_version = kvm->memory_config_version;
662 new = old = *memslot;
664 new.base_gfn = base_gfn;
666 new.flags = mem->flags;
668 /* Disallow changing a memory slot's size. */
670 if (npages && old.npages && npages != old.npages)
673 /* Check for overlaps */
675 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
676 struct kvm_memory_slot *s = &kvm->memslots[i];
680 if (!((base_gfn + npages <= s->base_gfn) ||
681 (base_gfn >= s->base_gfn + s->npages)))
685 * Do memory allocations outside lock. memory_config_version will
688 spin_unlock(&kvm->lock);
690 /* Deallocate if slot is being removed */
694 /* Free page dirty bitmap if unneeded */
695 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
696 new.dirty_bitmap = NULL;
700 /* Allocate if a slot is being created */
701 if (npages && !new.phys_mem) {
702 new.phys_mem = vmalloc(npages * sizeof(struct page *));
707 memset(new.phys_mem, 0, npages * sizeof(struct page *));
708 for (i = 0; i < npages; ++i) {
709 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
711 if (!new.phys_mem[i])
713 set_page_private(new.phys_mem[i],0);
717 /* Allocate page dirty bitmap if needed */
718 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
719 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
721 new.dirty_bitmap = vmalloc(dirty_bytes);
722 if (!new.dirty_bitmap)
724 memset(new.dirty_bitmap, 0, dirty_bytes);
727 spin_lock(&kvm->lock);
729 if (memory_config_version != kvm->memory_config_version) {
730 spin_unlock(&kvm->lock);
731 kvm_free_physmem_slot(&new, &old);
739 if (mem->slot >= kvm->nmemslots)
740 kvm->nmemslots = mem->slot + 1;
743 ++kvm->memory_config_version;
745 spin_unlock(&kvm->lock);
747 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
748 struct kvm_vcpu *vcpu;
750 vcpu = vcpu_load_slot(kvm, i);
753 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
754 do_remove_write_access(vcpu, mem->slot);
755 kvm_mmu_reset_context(vcpu);
759 kvm_free_physmem_slot(&old, &new);
763 spin_unlock(&kvm->lock);
765 kvm_free_physmem_slot(&new, &old);
771 * Get (and clear) the dirty memory log for a memory slot.
773 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
774 struct kvm_dirty_log *log)
776 struct kvm_memory_slot *memslot;
780 unsigned long any = 0;
782 spin_lock(&kvm->lock);
785 * Prevent changes to guest memory configuration even while the lock
789 spin_unlock(&kvm->lock);
791 if (log->slot >= KVM_MEMORY_SLOTS)
794 memslot = &kvm->memslots[log->slot];
796 if (!memslot->dirty_bitmap)
799 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
801 for (i = 0; !any && i < n/sizeof(long); ++i)
802 any = memslot->dirty_bitmap[i];
805 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
810 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
811 struct kvm_vcpu *vcpu;
813 vcpu = vcpu_load_slot(kvm, i);
817 do_remove_write_access(vcpu, log->slot);
818 memset(memslot->dirty_bitmap, 0, n);
821 kvm_arch_ops->tlb_flush(vcpu);
829 spin_lock(&kvm->lock);
831 spin_unlock(&kvm->lock);
835 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
839 for (i = 0; i < kvm->nmemslots; ++i) {
840 struct kvm_memory_slot *memslot = &kvm->memslots[i];
842 if (gfn >= memslot->base_gfn
843 && gfn < memslot->base_gfn + memslot->npages)
848 EXPORT_SYMBOL_GPL(gfn_to_memslot);
850 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
853 struct kvm_memory_slot *memslot = NULL;
854 unsigned long rel_gfn;
856 for (i = 0; i < kvm->nmemslots; ++i) {
857 memslot = &kvm->memslots[i];
859 if (gfn >= memslot->base_gfn
860 && gfn < memslot->base_gfn + memslot->npages) {
862 if (!memslot || !memslot->dirty_bitmap)
865 rel_gfn = gfn - memslot->base_gfn;
868 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
869 set_bit(rel_gfn, memslot->dirty_bitmap);
875 static int emulator_read_std(unsigned long addr,
878 struct x86_emulate_ctxt *ctxt)
880 struct kvm_vcpu *vcpu = ctxt->vcpu;
884 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
885 unsigned offset = addr & (PAGE_SIZE-1);
886 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
888 struct kvm_memory_slot *memslot;
891 if (gpa == UNMAPPED_GVA)
892 return X86EMUL_PROPAGATE_FAULT;
893 pfn = gpa >> PAGE_SHIFT;
894 memslot = gfn_to_memslot(vcpu->kvm, pfn);
896 return X86EMUL_UNHANDLEABLE;
897 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
899 memcpy(data, page + offset, tocopy);
901 kunmap_atomic(page, KM_USER0);
908 return X86EMUL_CONTINUE;
911 static int emulator_write_std(unsigned long addr,
914 struct x86_emulate_ctxt *ctxt)
916 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
918 return X86EMUL_UNHANDLEABLE;
921 static int emulator_read_emulated(unsigned long addr,
924 struct x86_emulate_ctxt *ctxt)
926 struct kvm_vcpu *vcpu = ctxt->vcpu;
928 if (vcpu->mmio_read_completed) {
929 memcpy(val, vcpu->mmio_data, bytes);
930 vcpu->mmio_read_completed = 0;
931 return X86EMUL_CONTINUE;
932 } else if (emulator_read_std(addr, val, bytes, ctxt)
934 return X86EMUL_CONTINUE;
936 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
938 if (gpa == UNMAPPED_GVA)
939 return X86EMUL_PROPAGATE_FAULT;
940 vcpu->mmio_needed = 1;
941 vcpu->mmio_phys_addr = gpa;
942 vcpu->mmio_size = bytes;
943 vcpu->mmio_is_write = 0;
945 return X86EMUL_UNHANDLEABLE;
949 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
950 unsigned long val, int bytes)
952 struct kvm_memory_slot *m;
956 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
958 m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
961 page = gfn_to_page(m, gpa >> PAGE_SHIFT);
962 kvm_mmu_pre_write(vcpu, gpa, bytes);
963 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
964 virt = kmap_atomic(page, KM_USER0);
965 memcpy(virt + offset_in_page(gpa), &val, bytes);
966 kunmap_atomic(virt, KM_USER0);
967 kvm_mmu_post_write(vcpu, gpa, bytes);
971 static int emulator_write_emulated(unsigned long addr,
974 struct x86_emulate_ctxt *ctxt)
976 struct kvm_vcpu *vcpu = ctxt->vcpu;
977 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
979 if (gpa == UNMAPPED_GVA)
980 return X86EMUL_PROPAGATE_FAULT;
982 if (emulator_write_phys(vcpu, gpa, val, bytes))
983 return X86EMUL_CONTINUE;
985 vcpu->mmio_needed = 1;
986 vcpu->mmio_phys_addr = gpa;
987 vcpu->mmio_size = bytes;
988 vcpu->mmio_is_write = 1;
989 memcpy(vcpu->mmio_data, &val, bytes);
991 return X86EMUL_CONTINUE;
994 static int emulator_cmpxchg_emulated(unsigned long addr,
998 struct x86_emulate_ctxt *ctxt)
1000 static int reported;
1004 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1006 return emulator_write_emulated(addr, new, bytes, ctxt);
1009 #ifdef CONFIG_X86_32
1011 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1012 unsigned long old_lo,
1013 unsigned long old_hi,
1014 unsigned long new_lo,
1015 unsigned long new_hi,
1016 struct x86_emulate_ctxt *ctxt)
1018 static int reported;
1023 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1025 r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1026 if (r != X86EMUL_CONTINUE)
1028 return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1033 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1035 return kvm_arch_ops->get_segment_base(vcpu, seg);
1038 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1040 return X86EMUL_CONTINUE;
1043 int emulate_clts(struct kvm_vcpu *vcpu)
1047 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1048 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1049 kvm_arch_ops->set_cr0(vcpu, cr0);
1050 return X86EMUL_CONTINUE;
1053 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1055 struct kvm_vcpu *vcpu = ctxt->vcpu;
1059 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1060 return X86EMUL_CONTINUE;
1062 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1064 return X86EMUL_UNHANDLEABLE;
1068 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1070 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1073 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1075 /* FIXME: better handling */
1076 return X86EMUL_UNHANDLEABLE;
1078 return X86EMUL_CONTINUE;
1081 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1083 static int reported;
1085 unsigned long rip = ctxt->vcpu->rip;
1086 unsigned long rip_linear;
1088 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1093 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1095 printk(KERN_ERR "emulation failed but !mmio_needed?"
1096 " rip %lx %02x %02x %02x %02x\n",
1097 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1101 struct x86_emulate_ops emulate_ops = {
1102 .read_std = emulator_read_std,
1103 .write_std = emulator_write_std,
1104 .read_emulated = emulator_read_emulated,
1105 .write_emulated = emulator_write_emulated,
1106 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1107 #ifdef CONFIG_X86_32
1108 .cmpxchg8b_emulated = emulator_cmpxchg8b_emulated,
1112 int emulate_instruction(struct kvm_vcpu *vcpu,
1113 struct kvm_run *run,
1117 struct x86_emulate_ctxt emulate_ctxt;
1121 kvm_arch_ops->cache_regs(vcpu);
1123 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1125 emulate_ctxt.vcpu = vcpu;
1126 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1127 emulate_ctxt.cr2 = cr2;
1128 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1129 ? X86EMUL_MODE_REAL : cs_l
1130 ? X86EMUL_MODE_PROT64 : cs_db
1131 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1133 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1134 emulate_ctxt.cs_base = 0;
1135 emulate_ctxt.ds_base = 0;
1136 emulate_ctxt.es_base = 0;
1137 emulate_ctxt.ss_base = 0;
1139 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1140 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1141 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1142 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1145 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1146 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1148 vcpu->mmio_is_write = 0;
1149 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1151 if ((r || vcpu->mmio_is_write) && run) {
1152 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1153 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1154 run->mmio.len = vcpu->mmio_size;
1155 run->mmio.is_write = vcpu->mmio_is_write;
1159 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1160 return EMULATE_DONE;
1161 if (!vcpu->mmio_needed) {
1162 report_emulation_failure(&emulate_ctxt);
1163 return EMULATE_FAIL;
1165 return EMULATE_DO_MMIO;
1168 kvm_arch_ops->decache_regs(vcpu);
1169 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1171 if (vcpu->mmio_is_write)
1172 return EMULATE_DO_MMIO;
1174 return EMULATE_DONE;
1176 EXPORT_SYMBOL_GPL(emulate_instruction);
1178 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1180 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1182 kvm_arch_ops->cache_regs(vcpu);
1184 #ifdef CONFIG_X86_64
1185 if (is_long_mode(vcpu)) {
1186 nr = vcpu->regs[VCPU_REGS_RAX];
1187 a0 = vcpu->regs[VCPU_REGS_RDI];
1188 a1 = vcpu->regs[VCPU_REGS_RSI];
1189 a2 = vcpu->regs[VCPU_REGS_RDX];
1190 a3 = vcpu->regs[VCPU_REGS_RCX];
1191 a4 = vcpu->regs[VCPU_REGS_R8];
1192 a5 = vcpu->regs[VCPU_REGS_R9];
1196 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1197 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1198 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1199 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1200 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1201 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1202 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1208 vcpu->regs[VCPU_REGS_RAX] = ret;
1209 kvm_arch_ops->decache_regs(vcpu);
1212 EXPORT_SYMBOL_GPL(kvm_hypercall);
1214 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1216 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1219 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1221 struct descriptor_table dt = { limit, base };
1223 kvm_arch_ops->set_gdt(vcpu, &dt);
1226 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1228 struct descriptor_table dt = { limit, base };
1230 kvm_arch_ops->set_idt(vcpu, &dt);
1233 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1234 unsigned long *rflags)
1237 *rflags = kvm_arch_ops->get_rflags(vcpu);
1240 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1242 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1253 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1258 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1259 unsigned long *rflags)
1263 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1264 *rflags = kvm_arch_ops->get_rflags(vcpu);
1273 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1276 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1281 * Register the para guest with the host:
1283 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1285 struct kvm_vcpu_para_state *para_state;
1286 hpa_t para_state_hpa, hypercall_hpa;
1287 struct page *para_state_page;
1288 unsigned char *hypercall;
1289 gpa_t hypercall_gpa;
1291 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1292 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1295 * Needs to be page aligned:
1297 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1300 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1301 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1302 if (is_error_hpa(para_state_hpa))
1305 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1306 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1307 para_state = kmap_atomic(para_state_page, KM_USER0);
1309 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1310 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1312 para_state->host_version = KVM_PARA_API_VERSION;
1314 * We cannot support guests that try to register themselves
1315 * with a newer API version than the host supports:
1317 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1318 para_state->ret = -KVM_EINVAL;
1319 goto err_kunmap_skip;
1322 hypercall_gpa = para_state->hypercall_gpa;
1323 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1324 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1325 if (is_error_hpa(hypercall_hpa)) {
1326 para_state->ret = -KVM_EINVAL;
1327 goto err_kunmap_skip;
1330 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1331 vcpu->para_state_page = para_state_page;
1332 vcpu->para_state_gpa = para_state_gpa;
1333 vcpu->hypercall_gpa = hypercall_gpa;
1335 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1336 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1337 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1338 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1339 kunmap_atomic(hypercall, KM_USER1);
1341 para_state->ret = 0;
1343 kunmap_atomic(para_state, KM_USER0);
1349 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1354 case 0xc0010010: /* SYSCFG */
1355 case 0xc0010015: /* HWCR */
1356 case MSR_IA32_PLATFORM_ID:
1357 case MSR_IA32_P5_MC_ADDR:
1358 case MSR_IA32_P5_MC_TYPE:
1359 case MSR_IA32_MC0_CTL:
1360 case MSR_IA32_MCG_STATUS:
1361 case MSR_IA32_MCG_CAP:
1362 case MSR_IA32_MC0_MISC:
1363 case MSR_IA32_MC0_MISC+4:
1364 case MSR_IA32_MC0_MISC+8:
1365 case MSR_IA32_MC0_MISC+12:
1366 case MSR_IA32_MC0_MISC+16:
1367 case MSR_IA32_UCODE_REV:
1368 case MSR_IA32_PERF_STATUS:
1369 /* MTRR registers */
1371 case 0x200 ... 0x2ff:
1374 case 0xcd: /* fsb frequency */
1377 case MSR_IA32_APICBASE:
1378 data = vcpu->apic_base;
1380 case MSR_IA32_MISC_ENABLE:
1381 data = vcpu->ia32_misc_enable_msr;
1383 #ifdef CONFIG_X86_64
1385 data = vcpu->shadow_efer;
1389 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1395 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1398 * Reads an msr value (of 'msr_index') into 'pdata'.
1399 * Returns 0 on success, non-0 otherwise.
1400 * Assumes vcpu_load() was already called.
1402 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1404 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1407 #ifdef CONFIG_X86_64
1409 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1411 if (efer & EFER_RESERVED_BITS) {
1412 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1419 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1420 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1425 kvm_arch_ops->set_efer(vcpu, efer);
1428 efer |= vcpu->shadow_efer & EFER_LMA;
1430 vcpu->shadow_efer = efer;
1435 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1438 #ifdef CONFIG_X86_64
1440 set_efer(vcpu, data);
1443 case MSR_IA32_MC0_STATUS:
1444 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1445 __FUNCTION__, data);
1447 case MSR_IA32_UCODE_REV:
1448 case MSR_IA32_UCODE_WRITE:
1449 case 0x200 ... 0x2ff: /* MTRRs */
1451 case MSR_IA32_APICBASE:
1452 vcpu->apic_base = data;
1454 case MSR_IA32_MISC_ENABLE:
1455 vcpu->ia32_misc_enable_msr = data;
1458 * This is the 'probe whether the host is KVM' logic:
1460 case MSR_KVM_API_MAGIC:
1461 return vcpu_register_para(vcpu, data);
1464 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1469 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1472 * Writes msr value into into the appropriate "register".
1473 * Returns 0 on success, non-0 otherwise.
1474 * Assumes vcpu_load() was already called.
1476 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1478 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1481 void kvm_resched(struct kvm_vcpu *vcpu)
1487 EXPORT_SYMBOL_GPL(kvm_resched);
1489 void load_msrs(struct vmx_msr_entry *e, int n)
1493 for (i = 0; i < n; ++i)
1494 wrmsrl(e[i].index, e[i].data);
1496 EXPORT_SYMBOL_GPL(load_msrs);
1498 void save_msrs(struct vmx_msr_entry *e, int n)
1502 for (i = 0; i < n; ++i)
1503 rdmsrl(e[i].index, e[i].data);
1505 EXPORT_SYMBOL_GPL(save_msrs);
1507 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1513 /* re-sync apic's tpr */
1514 vcpu->cr8 = kvm_run->cr8;
1516 if (kvm_run->emulated) {
1517 kvm_arch_ops->skip_emulated_instruction(vcpu);
1518 kvm_run->emulated = 0;
1521 if (kvm_run->mmio_completed) {
1522 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1523 vcpu->mmio_read_completed = 1;
1526 vcpu->mmio_needed = 0;
1528 r = kvm_arch_ops->run(vcpu, kvm_run);
1534 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1535 struct kvm_regs *regs)
1539 kvm_arch_ops->cache_regs(vcpu);
1541 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1542 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1543 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1544 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1545 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1546 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1547 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1548 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1549 #ifdef CONFIG_X86_64
1550 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1551 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1552 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1553 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1554 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1555 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1556 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1557 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1560 regs->rip = vcpu->rip;
1561 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1564 * Don't leak debug flags in case they were set for guest debugging
1566 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1567 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1574 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1575 struct kvm_regs *regs)
1579 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1580 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1581 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1582 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1583 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1584 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1585 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1586 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1587 #ifdef CONFIG_X86_64
1588 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1589 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1590 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1591 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1592 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1593 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1594 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1595 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1598 vcpu->rip = regs->rip;
1599 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1601 kvm_arch_ops->decache_regs(vcpu);
1608 static void get_segment(struct kvm_vcpu *vcpu,
1609 struct kvm_segment *var, int seg)
1611 return kvm_arch_ops->get_segment(vcpu, var, seg);
1614 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1615 struct kvm_sregs *sregs)
1617 struct descriptor_table dt;
1621 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1622 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1623 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1624 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1625 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1626 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1628 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1629 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1631 kvm_arch_ops->get_idt(vcpu, &dt);
1632 sregs->idt.limit = dt.limit;
1633 sregs->idt.base = dt.base;
1634 kvm_arch_ops->get_gdt(vcpu, &dt);
1635 sregs->gdt.limit = dt.limit;
1636 sregs->gdt.base = dt.base;
1638 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1639 sregs->cr0 = vcpu->cr0;
1640 sregs->cr2 = vcpu->cr2;
1641 sregs->cr3 = vcpu->cr3;
1642 sregs->cr4 = vcpu->cr4;
1643 sregs->cr8 = vcpu->cr8;
1644 sregs->efer = vcpu->shadow_efer;
1645 sregs->apic_base = vcpu->apic_base;
1647 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1648 sizeof sregs->interrupt_bitmap);
1655 static void set_segment(struct kvm_vcpu *vcpu,
1656 struct kvm_segment *var, int seg)
1658 return kvm_arch_ops->set_segment(vcpu, var, seg);
1661 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1662 struct kvm_sregs *sregs)
1664 int mmu_reset_needed = 0;
1666 struct descriptor_table dt;
1670 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1671 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1672 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1673 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1674 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1675 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1677 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1678 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1680 dt.limit = sregs->idt.limit;
1681 dt.base = sregs->idt.base;
1682 kvm_arch_ops->set_idt(vcpu, &dt);
1683 dt.limit = sregs->gdt.limit;
1684 dt.base = sregs->gdt.base;
1685 kvm_arch_ops->set_gdt(vcpu, &dt);
1687 vcpu->cr2 = sregs->cr2;
1688 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1689 vcpu->cr3 = sregs->cr3;
1691 vcpu->cr8 = sregs->cr8;
1693 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1694 #ifdef CONFIG_X86_64
1695 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1697 vcpu->apic_base = sregs->apic_base;
1699 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1701 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1702 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1704 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1705 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1706 if (!is_long_mode(vcpu) && is_pae(vcpu))
1707 load_pdptrs(vcpu, vcpu->cr3);
1709 if (mmu_reset_needed)
1710 kvm_mmu_reset_context(vcpu);
1712 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1713 sizeof vcpu->irq_pending);
1714 vcpu->irq_summary = 0;
1715 for (i = 0; i < NR_IRQ_WORDS; ++i)
1716 if (vcpu->irq_pending[i])
1717 __set_bit(i, &vcpu->irq_summary);
1725 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1726 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1728 * This list is modified at module load time to reflect the
1729 * capabilities of the host cpu.
1731 static u32 msrs_to_save[] = {
1732 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1734 #ifdef CONFIG_X86_64
1735 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1737 MSR_IA32_TIME_STAMP_COUNTER,
1740 static unsigned num_msrs_to_save;
1742 static u32 emulated_msrs[] = {
1743 MSR_IA32_MISC_ENABLE,
1746 static __init void kvm_init_msr_list(void)
1751 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1752 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1755 msrs_to_save[j] = msrs_to_save[i];
1758 num_msrs_to_save = j;
1762 * Adapt set_msr() to msr_io()'s calling convention
1764 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1766 return set_msr(vcpu, index, *data);
1770 * Read or write a bunch of msrs. All parameters are kernel addresses.
1772 * @return number of msrs set successfully.
1774 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1775 struct kvm_msr_entry *entries,
1776 int (*do_msr)(struct kvm_vcpu *vcpu,
1777 unsigned index, u64 *data))
1783 for (i = 0; i < msrs->nmsrs; ++i)
1784 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1793 * Read or write a bunch of msrs. Parameters are user addresses.
1795 * @return number of msrs set successfully.
1797 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1798 int (*do_msr)(struct kvm_vcpu *vcpu,
1799 unsigned index, u64 *data),
1802 struct kvm_msrs msrs;
1803 struct kvm_msr_entry *entries;
1808 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1812 if (msrs.nmsrs >= MAX_IO_MSRS)
1816 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1817 entries = vmalloc(size);
1822 if (copy_from_user(entries, user_msrs->entries, size))
1825 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1830 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1842 * Translate a guest virtual address to a guest physical address.
1844 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1845 struct kvm_translation *tr)
1847 unsigned long vaddr = tr->linear_address;
1851 spin_lock(&vcpu->kvm->lock);
1852 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1853 tr->physical_address = gpa;
1854 tr->valid = gpa != UNMAPPED_GVA;
1857 spin_unlock(&vcpu->kvm->lock);
1863 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1864 struct kvm_interrupt *irq)
1866 if (irq->irq < 0 || irq->irq >= 256)
1870 set_bit(irq->irq, vcpu->irq_pending);
1871 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1878 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1879 struct kvm_debug_guest *dbg)
1885 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1892 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1893 unsigned long address,
1896 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1897 unsigned long pgoff;
1900 *type = VM_FAULT_MINOR;
1901 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1903 return NOPAGE_SIGBUS;
1904 page = virt_to_page(vcpu->run);
1909 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1910 .nopage = kvm_vcpu_nopage,
1913 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1915 vma->vm_ops = &kvm_vcpu_vm_ops;
1919 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1921 struct kvm_vcpu *vcpu = filp->private_data;
1923 fput(vcpu->kvm->filp);
1927 static struct file_operations kvm_vcpu_fops = {
1928 .release = kvm_vcpu_release,
1929 .unlocked_ioctl = kvm_vcpu_ioctl,
1930 .compat_ioctl = kvm_vcpu_ioctl,
1931 .mmap = kvm_vcpu_mmap,
1935 * Allocates an inode for the vcpu.
1937 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1940 struct inode *inode;
1943 atomic_inc(&vcpu->kvm->filp->f_count);
1944 inode = kvmfs_inode(&kvm_vcpu_fops);
1945 if (IS_ERR(inode)) {
1950 file = kvmfs_file(inode, vcpu);
1956 r = get_unused_fd();
1960 fd_install(fd, file);
1969 fput(vcpu->kvm->filp);
1974 * Creates some virtual cpus. Good luck creating more than one.
1976 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1979 struct kvm_vcpu *vcpu;
1986 vcpu = &kvm->vcpus[n];
1988 mutex_lock(&vcpu->mutex);
1991 mutex_unlock(&vcpu->mutex);
1995 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1999 vcpu->run = page_address(page);
2001 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2003 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2005 r = kvm_arch_ops->vcpu_create(vcpu);
2007 goto out_free_vcpus;
2009 r = kvm_mmu_create(vcpu);
2011 goto out_free_vcpus;
2013 kvm_arch_ops->vcpu_load(vcpu);
2014 r = kvm_mmu_setup(vcpu);
2016 r = kvm_arch_ops->vcpu_setup(vcpu);
2020 goto out_free_vcpus;
2022 r = create_vcpu_fd(vcpu);
2024 goto out_free_vcpus;
2029 kvm_free_vcpu(vcpu);
2031 mutex_unlock(&vcpu->mutex);
2036 static long kvm_vcpu_ioctl(struct file *filp,
2037 unsigned int ioctl, unsigned long arg)
2039 struct kvm_vcpu *vcpu = filp->private_data;
2040 void __user *argp = (void __user *)arg;
2045 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2047 case KVM_GET_REGS: {
2048 struct kvm_regs kvm_regs;
2050 memset(&kvm_regs, 0, sizeof kvm_regs);
2051 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2055 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2060 case KVM_SET_REGS: {
2061 struct kvm_regs kvm_regs;
2064 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2066 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2072 case KVM_GET_SREGS: {
2073 struct kvm_sregs kvm_sregs;
2075 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2076 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2080 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2085 case KVM_SET_SREGS: {
2086 struct kvm_sregs kvm_sregs;
2089 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2091 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2097 case KVM_TRANSLATE: {
2098 struct kvm_translation tr;
2101 if (copy_from_user(&tr, argp, sizeof tr))
2103 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2107 if (copy_to_user(argp, &tr, sizeof tr))
2112 case KVM_INTERRUPT: {
2113 struct kvm_interrupt irq;
2116 if (copy_from_user(&irq, argp, sizeof irq))
2118 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2124 case KVM_DEBUG_GUEST: {
2125 struct kvm_debug_guest dbg;
2128 if (copy_from_user(&dbg, argp, sizeof dbg))
2130 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2137 r = msr_io(vcpu, argp, get_msr, 1);
2140 r = msr_io(vcpu, argp, do_set_msr, 0);
2149 static long kvm_vm_ioctl(struct file *filp,
2150 unsigned int ioctl, unsigned long arg)
2152 struct kvm *kvm = filp->private_data;
2153 void __user *argp = (void __user *)arg;
2157 case KVM_CREATE_VCPU:
2158 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2162 case KVM_SET_MEMORY_REGION: {
2163 struct kvm_memory_region kvm_mem;
2166 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2168 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2173 case KVM_GET_DIRTY_LOG: {
2174 struct kvm_dirty_log log;
2177 if (copy_from_user(&log, argp, sizeof log))
2179 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2191 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2192 unsigned long address,
2195 struct kvm *kvm = vma->vm_file->private_data;
2196 unsigned long pgoff;
2197 struct kvm_memory_slot *slot;
2200 *type = VM_FAULT_MINOR;
2201 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2202 slot = gfn_to_memslot(kvm, pgoff);
2204 return NOPAGE_SIGBUS;
2205 page = gfn_to_page(slot, pgoff);
2207 return NOPAGE_SIGBUS;
2212 static struct vm_operations_struct kvm_vm_vm_ops = {
2213 .nopage = kvm_vm_nopage,
2216 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2218 vma->vm_ops = &kvm_vm_vm_ops;
2222 static struct file_operations kvm_vm_fops = {
2223 .release = kvm_vm_release,
2224 .unlocked_ioctl = kvm_vm_ioctl,
2225 .compat_ioctl = kvm_vm_ioctl,
2226 .mmap = kvm_vm_mmap,
2229 static int kvm_dev_ioctl_create_vm(void)
2232 struct inode *inode;
2236 inode = kvmfs_inode(&kvm_vm_fops);
2237 if (IS_ERR(inode)) {
2242 kvm = kvm_create_vm();
2248 file = kvmfs_file(inode, kvm);
2255 r = get_unused_fd();
2259 fd_install(fd, file);
2266 kvm_destroy_vm(kvm);
2273 static long kvm_dev_ioctl(struct file *filp,
2274 unsigned int ioctl, unsigned long arg)
2276 void __user *argp = (void __user *)arg;
2280 case KVM_GET_API_VERSION:
2281 r = KVM_API_VERSION;
2284 r = kvm_dev_ioctl_create_vm();
2286 case KVM_GET_MSR_INDEX_LIST: {
2287 struct kvm_msr_list __user *user_msr_list = argp;
2288 struct kvm_msr_list msr_list;
2292 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2295 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2296 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2299 if (n < num_msrs_to_save)
2302 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2303 num_msrs_to_save * sizeof(u32)))
2305 if (copy_to_user(user_msr_list->indices
2306 + num_msrs_to_save * sizeof(u32),
2308 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2320 static struct file_operations kvm_chardev_ops = {
2321 .open = kvm_dev_open,
2322 .release = kvm_dev_release,
2323 .unlocked_ioctl = kvm_dev_ioctl,
2324 .compat_ioctl = kvm_dev_ioctl,
2327 static struct miscdevice kvm_dev = {
2333 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2336 if (val == SYS_RESTART) {
2338 * Some (well, at least mine) BIOSes hang on reboot if
2341 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2342 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2347 static struct notifier_block kvm_reboot_notifier = {
2348 .notifier_call = kvm_reboot,
2353 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2356 static void decache_vcpus_on_cpu(int cpu)
2359 struct kvm_vcpu *vcpu;
2362 spin_lock(&kvm_lock);
2363 list_for_each_entry(vm, &vm_list, vm_list)
2364 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2365 vcpu = &vm->vcpus[i];
2367 * If the vcpu is locked, then it is running on some
2368 * other cpu and therefore it is not cached on the
2371 * If it's not locked, check the last cpu it executed
2374 if (mutex_trylock(&vcpu->mutex)) {
2375 if (vcpu->cpu == cpu) {
2376 kvm_arch_ops->vcpu_decache(vcpu);
2379 mutex_unlock(&vcpu->mutex);
2382 spin_unlock(&kvm_lock);
2385 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2391 case CPU_DOWN_PREPARE:
2392 case CPU_UP_CANCELED:
2393 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2395 decache_vcpus_on_cpu(cpu);
2396 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2400 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2402 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2409 static struct notifier_block kvm_cpu_notifier = {
2410 .notifier_call = kvm_cpu_hotplug,
2411 .priority = 20, /* must be > scheduler priority */
2414 static __init void kvm_init_debug(void)
2416 struct kvm_stats_debugfs_item *p;
2418 debugfs_dir = debugfs_create_dir("kvm", NULL);
2419 for (p = debugfs_entries; p->name; ++p)
2420 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2424 static void kvm_exit_debug(void)
2426 struct kvm_stats_debugfs_item *p;
2428 for (p = debugfs_entries; p->name; ++p)
2429 debugfs_remove(p->dentry);
2430 debugfs_remove(debugfs_dir);
2433 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2435 decache_vcpus_on_cpu(raw_smp_processor_id());
2436 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2440 static int kvm_resume(struct sys_device *dev)
2442 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2446 static struct sysdev_class kvm_sysdev_class = {
2447 set_kset_name("kvm"),
2448 .suspend = kvm_suspend,
2449 .resume = kvm_resume,
2452 static struct sys_device kvm_sysdev = {
2454 .cls = &kvm_sysdev_class,
2457 hpa_t bad_page_address;
2459 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2460 const char *dev_name, void *data, struct vfsmount *mnt)
2462 return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_SUPER_MAGIC, mnt);
2465 static struct file_system_type kvm_fs_type = {
2467 .get_sb = kvmfs_get_sb,
2468 .kill_sb = kill_anon_super,
2471 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2476 printk(KERN_ERR "kvm: already loaded the other module\n");
2480 if (!ops->cpu_has_kvm_support()) {
2481 printk(KERN_ERR "kvm: no hardware support\n");
2484 if (ops->disabled_by_bios()) {
2485 printk(KERN_ERR "kvm: disabled by bios\n");
2491 r = kvm_arch_ops->hardware_setup();
2495 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2496 r = register_cpu_notifier(&kvm_cpu_notifier);
2499 register_reboot_notifier(&kvm_reboot_notifier);
2501 r = sysdev_class_register(&kvm_sysdev_class);
2505 r = sysdev_register(&kvm_sysdev);
2509 kvm_chardev_ops.owner = module;
2511 r = misc_register(&kvm_dev);
2513 printk (KERN_ERR "kvm: misc device register failed\n");
2520 sysdev_unregister(&kvm_sysdev);
2522 sysdev_class_unregister(&kvm_sysdev_class);
2524 unregister_reboot_notifier(&kvm_reboot_notifier);
2525 unregister_cpu_notifier(&kvm_cpu_notifier);
2527 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2528 kvm_arch_ops->hardware_unsetup();
2530 kvm_arch_ops = NULL;
2534 void kvm_exit_arch(void)
2536 misc_deregister(&kvm_dev);
2537 sysdev_unregister(&kvm_sysdev);
2538 sysdev_class_unregister(&kvm_sysdev_class);
2539 unregister_reboot_notifier(&kvm_reboot_notifier);
2540 unregister_cpu_notifier(&kvm_cpu_notifier);
2541 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2542 kvm_arch_ops->hardware_unsetup();
2543 kvm_arch_ops = NULL;
2546 static __init int kvm_init(void)
2548 static struct page *bad_page;
2551 r = register_filesystem(&kvm_fs_type);
2555 kvmfs_mnt = kern_mount(&kvm_fs_type);
2556 r = PTR_ERR(kvmfs_mnt);
2557 if (IS_ERR(kvmfs_mnt))
2561 kvm_init_msr_list();
2563 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2568 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2569 memset(__va(bad_page_address), 0, PAGE_SIZE);
2577 unregister_filesystem(&kvm_fs_type);
2582 static __exit void kvm_exit(void)
2585 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2587 unregister_filesystem(&kvm_fs_type);
2590 module_init(kvm_init)
2591 module_exit(kvm_exit)
2593 EXPORT_SYMBOL_GPL(kvm_init_arch);
2594 EXPORT_SYMBOL_GPL(kvm_exit_arch);