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 <asm/processor.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <asm/uaccess.h>
31 #include <linux/reboot.h>
33 #include <linux/debugfs.h>
34 #include <linux/highmem.h>
35 #include <linux/file.h>
37 #include <linux/sysdev.h>
38 #include <linux/cpu.h>
39 #include <linux/file.h>
41 #include <linux/mount.h>
43 #include "x86_emulate.h"
44 #include "segment_descriptor.h"
46 MODULE_AUTHOR("Qumranet");
47 MODULE_LICENSE("GPL");
49 static DEFINE_SPINLOCK(kvm_lock);
50 static LIST_HEAD(vm_list);
52 struct kvm_arch_ops *kvm_arch_ops;
53 struct kvm_stat kvm_stat;
54 EXPORT_SYMBOL_GPL(kvm_stat);
56 static struct kvm_stats_debugfs_item {
59 struct dentry *dentry;
60 } debugfs_entries[] = {
61 { "pf_fixed", &kvm_stat.pf_fixed },
62 { "pf_guest", &kvm_stat.pf_guest },
63 { "tlb_flush", &kvm_stat.tlb_flush },
64 { "invlpg", &kvm_stat.invlpg },
65 { "exits", &kvm_stat.exits },
66 { "io_exits", &kvm_stat.io_exits },
67 { "mmio_exits", &kvm_stat.mmio_exits },
68 { "signal_exits", &kvm_stat.signal_exits },
69 { "irq_window", &kvm_stat.irq_window_exits },
70 { "halt_exits", &kvm_stat.halt_exits },
71 { "request_irq", &kvm_stat.request_irq_exits },
72 { "irq_exits", &kvm_stat.irq_exits },
76 static struct dentry *debugfs_dir;
78 #define KVMFS_MAGIC 0x19700426
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);
360 static void kvm_free_vcpus(struct kvm *kvm)
364 for (i = 0; i < KVM_MAX_VCPUS; ++i)
365 kvm_free_vcpu(&kvm->vcpus[i]);
368 static int kvm_dev_release(struct inode *inode, struct file *filp)
373 static void kvm_destroy_vm(struct kvm *kvm)
375 spin_lock(&kvm_lock);
376 list_del(&kvm->vm_list);
377 spin_unlock(&kvm_lock);
379 kvm_free_physmem(kvm);
383 static int kvm_vm_release(struct inode *inode, struct file *filp)
385 struct kvm *kvm = filp->private_data;
391 static void inject_gp(struct kvm_vcpu *vcpu)
393 kvm_arch_ops->inject_gp(vcpu, 0);
397 * Load the pae pdptrs. Return true is they are all valid.
399 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
401 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
402 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
407 struct kvm_memory_slot *memslot;
409 spin_lock(&vcpu->kvm->lock);
410 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
411 /* FIXME: !memslot - emulate? 0xff? */
412 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
415 for (i = 0; i < 4; ++i) {
416 pdpte = pdpt[offset + i];
417 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
423 for (i = 0; i < 4; ++i)
424 vcpu->pdptrs[i] = pdpt[offset + i];
427 kunmap_atomic(pdpt, KM_USER0);
428 spin_unlock(&vcpu->kvm->lock);
433 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
435 if (cr0 & CR0_RESEVED_BITS) {
436 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
442 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
443 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
448 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
449 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
450 "and a clear PE flag\n");
455 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
457 if ((vcpu->shadow_efer & EFER_LME)) {
461 printk(KERN_DEBUG "set_cr0: #GP, start paging "
462 "in long mode while PAE is disabled\n");
466 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
468 printk(KERN_DEBUG "set_cr0: #GP, start paging "
469 "in long mode while CS.L == 1\n");
476 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
477 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
485 kvm_arch_ops->set_cr0(vcpu, cr0);
488 spin_lock(&vcpu->kvm->lock);
489 kvm_mmu_reset_context(vcpu);
490 spin_unlock(&vcpu->kvm->lock);
493 EXPORT_SYMBOL_GPL(set_cr0);
495 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
497 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
498 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
500 EXPORT_SYMBOL_GPL(lmsw);
502 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
504 if (cr4 & CR4_RESEVED_BITS) {
505 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
510 if (is_long_mode(vcpu)) {
511 if (!(cr4 & CR4_PAE_MASK)) {
512 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
517 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
518 && !load_pdptrs(vcpu, vcpu->cr3)) {
519 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
523 if (cr4 & CR4_VMXE_MASK) {
524 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
528 kvm_arch_ops->set_cr4(vcpu, cr4);
529 spin_lock(&vcpu->kvm->lock);
530 kvm_mmu_reset_context(vcpu);
531 spin_unlock(&vcpu->kvm->lock);
533 EXPORT_SYMBOL_GPL(set_cr4);
535 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
537 if (is_long_mode(vcpu)) {
538 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
539 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
544 if (cr3 & CR3_RESEVED_BITS) {
545 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
549 if (is_paging(vcpu) && is_pae(vcpu) &&
550 !load_pdptrs(vcpu, cr3)) {
551 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
559 spin_lock(&vcpu->kvm->lock);
561 * Does the new cr3 value map to physical memory? (Note, we
562 * catch an invalid cr3 even in real-mode, because it would
563 * cause trouble later on when we turn on paging anyway.)
565 * A real CPU would silently accept an invalid cr3 and would
566 * attempt to use it - with largely undefined (and often hard
567 * to debug) behavior on the guest side.
569 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
572 vcpu->mmu.new_cr3(vcpu);
573 spin_unlock(&vcpu->kvm->lock);
575 EXPORT_SYMBOL_GPL(set_cr3);
577 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
579 if ( cr8 & CR8_RESEVED_BITS) {
580 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
586 EXPORT_SYMBOL_GPL(set_cr8);
588 void fx_init(struct kvm_vcpu *vcpu)
590 struct __attribute__ ((__packed__)) fx_image_s {
596 u64 operand;// fpu dp
602 fx_save(vcpu->host_fx_image);
604 fx_save(vcpu->guest_fx_image);
605 fx_restore(vcpu->host_fx_image);
607 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
608 fx_image->mxcsr = 0x1f80;
609 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
610 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
612 EXPORT_SYMBOL_GPL(fx_init);
615 * Allocate some memory and give it an address in the guest physical address
618 * Discontiguous memory is allowed, mostly for framebuffers.
620 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
621 struct kvm_memory_region *mem)
625 unsigned long npages;
627 struct kvm_memory_slot *memslot;
628 struct kvm_memory_slot old, new;
629 int memory_config_version;
632 /* General sanity checks */
633 if (mem->memory_size & (PAGE_SIZE - 1))
635 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
637 if (mem->slot >= KVM_MEMORY_SLOTS)
639 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
642 memslot = &kvm->memslots[mem->slot];
643 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
644 npages = mem->memory_size >> PAGE_SHIFT;
647 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
650 spin_lock(&kvm->lock);
652 memory_config_version = kvm->memory_config_version;
653 new = old = *memslot;
655 new.base_gfn = base_gfn;
657 new.flags = mem->flags;
659 /* Disallow changing a memory slot's size. */
661 if (npages && old.npages && npages != old.npages)
664 /* Check for overlaps */
666 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
667 struct kvm_memory_slot *s = &kvm->memslots[i];
671 if (!((base_gfn + npages <= s->base_gfn) ||
672 (base_gfn >= s->base_gfn + s->npages)))
676 * Do memory allocations outside lock. memory_config_version will
679 spin_unlock(&kvm->lock);
681 /* Deallocate if slot is being removed */
685 /* Free page dirty bitmap if unneeded */
686 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
687 new.dirty_bitmap = NULL;
691 /* Allocate if a slot is being created */
692 if (npages && !new.phys_mem) {
693 new.phys_mem = vmalloc(npages * sizeof(struct page *));
698 memset(new.phys_mem, 0, npages * sizeof(struct page *));
699 for (i = 0; i < npages; ++i) {
700 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
702 if (!new.phys_mem[i])
704 set_page_private(new.phys_mem[i],0);
708 /* Allocate page dirty bitmap if needed */
709 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
710 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
712 new.dirty_bitmap = vmalloc(dirty_bytes);
713 if (!new.dirty_bitmap)
715 memset(new.dirty_bitmap, 0, dirty_bytes);
718 spin_lock(&kvm->lock);
720 if (memory_config_version != kvm->memory_config_version) {
721 spin_unlock(&kvm->lock);
722 kvm_free_physmem_slot(&new, &old);
730 if (mem->slot >= kvm->nmemslots)
731 kvm->nmemslots = mem->slot + 1;
734 ++kvm->memory_config_version;
736 spin_unlock(&kvm->lock);
738 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
739 struct kvm_vcpu *vcpu;
741 vcpu = vcpu_load_slot(kvm, i);
744 kvm_mmu_reset_context(vcpu);
748 kvm_free_physmem_slot(&old, &new);
752 spin_unlock(&kvm->lock);
754 kvm_free_physmem_slot(&new, &old);
759 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
761 spin_lock(&vcpu->kvm->lock);
762 kvm_mmu_slot_remove_write_access(vcpu, slot);
763 spin_unlock(&vcpu->kvm->lock);
767 * Get (and clear) the dirty memory log for a memory slot.
769 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
770 struct kvm_dirty_log *log)
772 struct kvm_memory_slot *memslot;
776 unsigned long any = 0;
778 spin_lock(&kvm->lock);
781 * Prevent changes to guest memory configuration even while the lock
785 spin_unlock(&kvm->lock);
787 if (log->slot >= KVM_MEMORY_SLOTS)
790 memslot = &kvm->memslots[log->slot];
792 if (!memslot->dirty_bitmap)
795 n = ALIGN(memslot->npages, 8) / 8;
797 for (i = 0; !any && i < n; ++i)
798 any = memslot->dirty_bitmap[i];
801 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
806 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
807 struct kvm_vcpu *vcpu;
809 vcpu = vcpu_load_slot(kvm, i);
813 do_remove_write_access(vcpu, log->slot);
814 memset(memslot->dirty_bitmap, 0, n);
817 kvm_arch_ops->tlb_flush(vcpu);
825 spin_lock(&kvm->lock);
827 spin_unlock(&kvm->lock);
831 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
835 for (i = 0; i < kvm->nmemslots; ++i) {
836 struct kvm_memory_slot *memslot = &kvm->memslots[i];
838 if (gfn >= memslot->base_gfn
839 && gfn < memslot->base_gfn + memslot->npages)
844 EXPORT_SYMBOL_GPL(gfn_to_memslot);
846 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
849 struct kvm_memory_slot *memslot = NULL;
850 unsigned long rel_gfn;
852 for (i = 0; i < kvm->nmemslots; ++i) {
853 memslot = &kvm->memslots[i];
855 if (gfn >= memslot->base_gfn
856 && gfn < memslot->base_gfn + memslot->npages) {
858 if (!memslot || !memslot->dirty_bitmap)
861 rel_gfn = gfn - memslot->base_gfn;
864 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
865 set_bit(rel_gfn, memslot->dirty_bitmap);
871 static int emulator_read_std(unsigned long addr,
874 struct x86_emulate_ctxt *ctxt)
876 struct kvm_vcpu *vcpu = ctxt->vcpu;
880 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
881 unsigned offset = addr & (PAGE_SIZE-1);
882 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
884 struct kvm_memory_slot *memslot;
887 if (gpa == UNMAPPED_GVA)
888 return X86EMUL_PROPAGATE_FAULT;
889 pfn = gpa >> PAGE_SHIFT;
890 memslot = gfn_to_memslot(vcpu->kvm, pfn);
892 return X86EMUL_UNHANDLEABLE;
893 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
895 memcpy(data, page + offset, tocopy);
897 kunmap_atomic(page, KM_USER0);
904 return X86EMUL_CONTINUE;
907 static int emulator_write_std(unsigned long addr,
910 struct x86_emulate_ctxt *ctxt)
912 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
914 return X86EMUL_UNHANDLEABLE;
917 static int emulator_read_emulated(unsigned long addr,
920 struct x86_emulate_ctxt *ctxt)
922 struct kvm_vcpu *vcpu = ctxt->vcpu;
924 if (vcpu->mmio_read_completed) {
925 memcpy(val, vcpu->mmio_data, bytes);
926 vcpu->mmio_read_completed = 0;
927 return X86EMUL_CONTINUE;
928 } else if (emulator_read_std(addr, val, bytes, ctxt)
930 return X86EMUL_CONTINUE;
932 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
934 if (gpa == UNMAPPED_GVA)
935 return X86EMUL_PROPAGATE_FAULT;
936 vcpu->mmio_needed = 1;
937 vcpu->mmio_phys_addr = gpa;
938 vcpu->mmio_size = bytes;
939 vcpu->mmio_is_write = 0;
941 return X86EMUL_UNHANDLEABLE;
945 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
946 unsigned long val, int bytes)
948 struct kvm_memory_slot *m;
952 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
954 m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
957 page = gfn_to_page(m, gpa >> PAGE_SHIFT);
958 kvm_mmu_pre_write(vcpu, gpa, bytes);
959 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
960 virt = kmap_atomic(page, KM_USER0);
961 memcpy(virt + offset_in_page(gpa), &val, bytes);
962 kunmap_atomic(virt, KM_USER0);
963 kvm_mmu_post_write(vcpu, gpa, bytes);
967 static int emulator_write_emulated(unsigned long addr,
970 struct x86_emulate_ctxt *ctxt)
972 struct kvm_vcpu *vcpu = ctxt->vcpu;
973 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
975 if (gpa == UNMAPPED_GVA)
976 return X86EMUL_PROPAGATE_FAULT;
978 if (emulator_write_phys(vcpu, gpa, val, bytes))
979 return X86EMUL_CONTINUE;
981 vcpu->mmio_needed = 1;
982 vcpu->mmio_phys_addr = gpa;
983 vcpu->mmio_size = bytes;
984 vcpu->mmio_is_write = 1;
985 memcpy(vcpu->mmio_data, &val, bytes);
987 return X86EMUL_CONTINUE;
990 static int emulator_cmpxchg_emulated(unsigned long addr,
994 struct x86_emulate_ctxt *ctxt)
1000 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1002 return emulator_write_emulated(addr, new, bytes, ctxt);
1005 #ifdef CONFIG_X86_32
1007 static int emulator_cmpxchg8b_emulated(unsigned long addr,
1008 unsigned long old_lo,
1009 unsigned long old_hi,
1010 unsigned long new_lo,
1011 unsigned long new_hi,
1012 struct x86_emulate_ctxt *ctxt)
1014 static int reported;
1019 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
1021 r = emulator_write_emulated(addr, new_lo, 4, ctxt);
1022 if (r != X86EMUL_CONTINUE)
1024 return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
1029 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1031 return kvm_arch_ops->get_segment_base(vcpu, seg);
1034 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1036 return X86EMUL_CONTINUE;
1039 int emulate_clts(struct kvm_vcpu *vcpu)
1043 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1044 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1045 kvm_arch_ops->set_cr0(vcpu, cr0);
1046 return X86EMUL_CONTINUE;
1049 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1051 struct kvm_vcpu *vcpu = ctxt->vcpu;
1055 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1056 return X86EMUL_CONTINUE;
1058 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1060 return X86EMUL_UNHANDLEABLE;
1064 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1066 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1069 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1071 /* FIXME: better handling */
1072 return X86EMUL_UNHANDLEABLE;
1074 return X86EMUL_CONTINUE;
1077 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1079 static int reported;
1081 unsigned long rip = ctxt->vcpu->rip;
1082 unsigned long rip_linear;
1084 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1089 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1091 printk(KERN_ERR "emulation failed but !mmio_needed?"
1092 " rip %lx %02x %02x %02x %02x\n",
1093 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1097 struct x86_emulate_ops emulate_ops = {
1098 .read_std = emulator_read_std,
1099 .write_std = emulator_write_std,
1100 .read_emulated = emulator_read_emulated,
1101 .write_emulated = emulator_write_emulated,
1102 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1103 #ifdef CONFIG_X86_32
1104 .cmpxchg8b_emulated = emulator_cmpxchg8b_emulated,
1108 int emulate_instruction(struct kvm_vcpu *vcpu,
1109 struct kvm_run *run,
1113 struct x86_emulate_ctxt emulate_ctxt;
1117 kvm_arch_ops->cache_regs(vcpu);
1119 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1121 emulate_ctxt.vcpu = vcpu;
1122 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1123 emulate_ctxt.cr2 = cr2;
1124 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1125 ? X86EMUL_MODE_REAL : cs_l
1126 ? X86EMUL_MODE_PROT64 : cs_db
1127 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1129 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1130 emulate_ctxt.cs_base = 0;
1131 emulate_ctxt.ds_base = 0;
1132 emulate_ctxt.es_base = 0;
1133 emulate_ctxt.ss_base = 0;
1135 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1136 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1137 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1138 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1141 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1142 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1144 vcpu->mmio_is_write = 0;
1145 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1147 if ((r || vcpu->mmio_is_write) && run) {
1148 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1149 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1150 run->mmio.len = vcpu->mmio_size;
1151 run->mmio.is_write = vcpu->mmio_is_write;
1155 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1156 return EMULATE_DONE;
1157 if (!vcpu->mmio_needed) {
1158 report_emulation_failure(&emulate_ctxt);
1159 return EMULATE_FAIL;
1161 return EMULATE_DO_MMIO;
1164 kvm_arch_ops->decache_regs(vcpu);
1165 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1167 if (vcpu->mmio_is_write)
1168 return EMULATE_DO_MMIO;
1170 return EMULATE_DONE;
1172 EXPORT_SYMBOL_GPL(emulate_instruction);
1174 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1176 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1178 kvm_arch_ops->decache_regs(vcpu);
1180 #ifdef CONFIG_X86_64
1181 if (is_long_mode(vcpu)) {
1182 nr = vcpu->regs[VCPU_REGS_RAX];
1183 a0 = vcpu->regs[VCPU_REGS_RDI];
1184 a1 = vcpu->regs[VCPU_REGS_RSI];
1185 a2 = vcpu->regs[VCPU_REGS_RDX];
1186 a3 = vcpu->regs[VCPU_REGS_RCX];
1187 a4 = vcpu->regs[VCPU_REGS_R8];
1188 a5 = vcpu->regs[VCPU_REGS_R9];
1192 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1193 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1194 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1195 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1196 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1197 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1198 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1204 vcpu->regs[VCPU_REGS_RAX] = ret;
1205 kvm_arch_ops->cache_regs(vcpu);
1208 EXPORT_SYMBOL_GPL(kvm_hypercall);
1210 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1212 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1215 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1217 struct descriptor_table dt = { limit, base };
1219 kvm_arch_ops->set_gdt(vcpu, &dt);
1222 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1224 struct descriptor_table dt = { limit, base };
1226 kvm_arch_ops->set_idt(vcpu, &dt);
1229 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1230 unsigned long *rflags)
1233 *rflags = kvm_arch_ops->get_rflags(vcpu);
1236 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1238 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1249 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1254 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1255 unsigned long *rflags)
1259 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1260 *rflags = kvm_arch_ops->get_rflags(vcpu);
1269 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1272 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1277 * Register the para guest with the host:
1279 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1281 struct kvm_vcpu_para_state *para_state;
1282 hpa_t para_state_hpa, hypercall_hpa;
1283 struct page *para_state_page;
1284 unsigned char *hypercall;
1285 gpa_t hypercall_gpa;
1287 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1288 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1291 * Needs to be page aligned:
1293 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1296 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1297 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1298 if (is_error_hpa(para_state_hpa))
1301 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1302 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1303 para_state = kmap_atomic(para_state_page, KM_USER0);
1305 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1306 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1308 para_state->host_version = KVM_PARA_API_VERSION;
1310 * We cannot support guests that try to register themselves
1311 * with a newer API version than the host supports:
1313 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1314 para_state->ret = -KVM_EINVAL;
1315 goto err_kunmap_skip;
1318 hypercall_gpa = para_state->hypercall_gpa;
1319 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1320 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1321 if (is_error_hpa(hypercall_hpa)) {
1322 para_state->ret = -KVM_EINVAL;
1323 goto err_kunmap_skip;
1326 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1327 vcpu->para_state_page = para_state_page;
1328 vcpu->para_state_gpa = para_state_gpa;
1329 vcpu->hypercall_gpa = hypercall_gpa;
1331 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1332 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1333 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1334 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1335 kunmap_atomic(hypercall, KM_USER1);
1337 para_state->ret = 0;
1339 kunmap_atomic(para_state, KM_USER0);
1345 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1350 case 0xc0010010: /* SYSCFG */
1351 case 0xc0010015: /* HWCR */
1352 case MSR_IA32_PLATFORM_ID:
1353 case MSR_IA32_P5_MC_ADDR:
1354 case MSR_IA32_P5_MC_TYPE:
1355 case MSR_IA32_MC0_CTL:
1356 case MSR_IA32_MCG_STATUS:
1357 case MSR_IA32_MCG_CAP:
1358 case MSR_IA32_MC0_MISC:
1359 case MSR_IA32_MC0_MISC+4:
1360 case MSR_IA32_MC0_MISC+8:
1361 case MSR_IA32_MC0_MISC+12:
1362 case MSR_IA32_MC0_MISC+16:
1363 case MSR_IA32_UCODE_REV:
1364 case MSR_IA32_PERF_STATUS:
1365 /* MTRR registers */
1367 case 0x200 ... 0x2ff:
1370 case 0xcd: /* fsb frequency */
1373 case MSR_IA32_APICBASE:
1374 data = vcpu->apic_base;
1376 case MSR_IA32_MISC_ENABLE:
1377 data = vcpu->ia32_misc_enable_msr;
1379 #ifdef CONFIG_X86_64
1381 data = vcpu->shadow_efer;
1385 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1391 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1394 * Reads an msr value (of 'msr_index') into 'pdata'.
1395 * Returns 0 on success, non-0 otherwise.
1396 * Assumes vcpu_load() was already called.
1398 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1400 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1403 #ifdef CONFIG_X86_64
1405 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1407 if (efer & EFER_RESERVED_BITS) {
1408 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1415 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1416 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1421 kvm_arch_ops->set_efer(vcpu, efer);
1424 efer |= vcpu->shadow_efer & EFER_LMA;
1426 vcpu->shadow_efer = efer;
1431 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1434 #ifdef CONFIG_X86_64
1436 set_efer(vcpu, data);
1439 case MSR_IA32_MC0_STATUS:
1440 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1441 __FUNCTION__, data);
1443 case MSR_IA32_UCODE_REV:
1444 case MSR_IA32_UCODE_WRITE:
1445 case 0x200 ... 0x2ff: /* MTRRs */
1447 case MSR_IA32_APICBASE:
1448 vcpu->apic_base = data;
1450 case MSR_IA32_MISC_ENABLE:
1451 vcpu->ia32_misc_enable_msr = data;
1454 * This is the 'probe whether the host is KVM' logic:
1456 case MSR_KVM_API_MAGIC:
1457 return vcpu_register_para(vcpu, data);
1460 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1465 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1468 * Writes msr value into into the appropriate "register".
1469 * Returns 0 on success, non-0 otherwise.
1470 * Assumes vcpu_load() was already called.
1472 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1474 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1477 void kvm_resched(struct kvm_vcpu *vcpu)
1483 EXPORT_SYMBOL_GPL(kvm_resched);
1485 void load_msrs(struct vmx_msr_entry *e, int n)
1489 for (i = 0; i < n; ++i)
1490 wrmsrl(e[i].index, e[i].data);
1492 EXPORT_SYMBOL_GPL(load_msrs);
1494 void save_msrs(struct vmx_msr_entry *e, int n)
1498 for (i = 0; i < n; ++i)
1499 rdmsrl(e[i].index, e[i].data);
1501 EXPORT_SYMBOL_GPL(save_msrs);
1503 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1509 /* re-sync apic's tpr */
1510 vcpu->cr8 = kvm_run->cr8;
1512 if (kvm_run->emulated) {
1513 kvm_arch_ops->skip_emulated_instruction(vcpu);
1514 kvm_run->emulated = 0;
1517 if (kvm_run->mmio_completed) {
1518 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1519 vcpu->mmio_read_completed = 1;
1522 vcpu->mmio_needed = 0;
1524 r = kvm_arch_ops->run(vcpu, kvm_run);
1530 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1531 struct kvm_regs *regs)
1535 kvm_arch_ops->cache_regs(vcpu);
1537 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1538 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1539 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1540 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1541 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1542 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1543 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1544 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1545 #ifdef CONFIG_X86_64
1546 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1547 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1548 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1549 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1550 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1551 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1552 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1553 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1556 regs->rip = vcpu->rip;
1557 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1560 * Don't leak debug flags in case they were set for guest debugging
1562 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1563 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1570 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1571 struct kvm_regs *regs)
1575 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1576 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1577 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1578 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1579 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1580 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1581 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1582 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1583 #ifdef CONFIG_X86_64
1584 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1585 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1586 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1587 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1588 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1589 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1590 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1591 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1594 vcpu->rip = regs->rip;
1595 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1597 kvm_arch_ops->decache_regs(vcpu);
1604 static void get_segment(struct kvm_vcpu *vcpu,
1605 struct kvm_segment *var, int seg)
1607 return kvm_arch_ops->get_segment(vcpu, var, seg);
1610 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1611 struct kvm_sregs *sregs)
1613 struct descriptor_table dt;
1617 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1618 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1619 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1620 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1621 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1622 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1624 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1625 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1627 kvm_arch_ops->get_idt(vcpu, &dt);
1628 sregs->idt.limit = dt.limit;
1629 sregs->idt.base = dt.base;
1630 kvm_arch_ops->get_gdt(vcpu, &dt);
1631 sregs->gdt.limit = dt.limit;
1632 sregs->gdt.base = dt.base;
1634 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1635 sregs->cr0 = vcpu->cr0;
1636 sregs->cr2 = vcpu->cr2;
1637 sregs->cr3 = vcpu->cr3;
1638 sregs->cr4 = vcpu->cr4;
1639 sregs->cr8 = vcpu->cr8;
1640 sregs->efer = vcpu->shadow_efer;
1641 sregs->apic_base = vcpu->apic_base;
1643 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1644 sizeof sregs->interrupt_bitmap);
1651 static void set_segment(struct kvm_vcpu *vcpu,
1652 struct kvm_segment *var, int seg)
1654 return kvm_arch_ops->set_segment(vcpu, var, seg);
1657 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1658 struct kvm_sregs *sregs)
1660 int mmu_reset_needed = 0;
1662 struct descriptor_table dt;
1666 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1667 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1668 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1669 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1670 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1671 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1673 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1674 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1676 dt.limit = sregs->idt.limit;
1677 dt.base = sregs->idt.base;
1678 kvm_arch_ops->set_idt(vcpu, &dt);
1679 dt.limit = sregs->gdt.limit;
1680 dt.base = sregs->gdt.base;
1681 kvm_arch_ops->set_gdt(vcpu, &dt);
1683 vcpu->cr2 = sregs->cr2;
1684 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1685 vcpu->cr3 = sregs->cr3;
1687 vcpu->cr8 = sregs->cr8;
1689 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1690 #ifdef CONFIG_X86_64
1691 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1693 vcpu->apic_base = sregs->apic_base;
1695 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1697 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1698 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1700 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1701 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1702 if (!is_long_mode(vcpu) && is_pae(vcpu))
1703 load_pdptrs(vcpu, vcpu->cr3);
1705 if (mmu_reset_needed)
1706 kvm_mmu_reset_context(vcpu);
1708 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1709 sizeof vcpu->irq_pending);
1710 vcpu->irq_summary = 0;
1711 for (i = 0; i < NR_IRQ_WORDS; ++i)
1712 if (vcpu->irq_pending[i])
1713 __set_bit(i, &vcpu->irq_summary);
1721 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1722 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1724 * This list is modified at module load time to reflect the
1725 * capabilities of the host cpu.
1727 static u32 msrs_to_save[] = {
1728 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1730 #ifdef CONFIG_X86_64
1731 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1733 MSR_IA32_TIME_STAMP_COUNTER,
1736 static unsigned num_msrs_to_save;
1738 static u32 emulated_msrs[] = {
1739 MSR_IA32_MISC_ENABLE,
1742 static __init void kvm_init_msr_list(void)
1747 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1748 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1751 msrs_to_save[j] = msrs_to_save[i];
1754 num_msrs_to_save = j;
1758 * Adapt set_msr() to msr_io()'s calling convention
1760 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1762 return set_msr(vcpu, index, *data);
1766 * Read or write a bunch of msrs. All parameters are kernel addresses.
1768 * @return number of msrs set successfully.
1770 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1771 struct kvm_msr_entry *entries,
1772 int (*do_msr)(struct kvm_vcpu *vcpu,
1773 unsigned index, u64 *data))
1779 for (i = 0; i < msrs->nmsrs; ++i)
1780 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1789 * Read or write a bunch of msrs. Parameters are user addresses.
1791 * @return number of msrs set successfully.
1793 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1794 int (*do_msr)(struct kvm_vcpu *vcpu,
1795 unsigned index, u64 *data),
1798 struct kvm_msrs msrs;
1799 struct kvm_msr_entry *entries;
1804 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1808 if (msrs.nmsrs >= MAX_IO_MSRS)
1812 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1813 entries = vmalloc(size);
1818 if (copy_from_user(entries, user_msrs->entries, size))
1821 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1826 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1838 * Translate a guest virtual address to a guest physical address.
1840 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1841 struct kvm_translation *tr)
1843 unsigned long vaddr = tr->linear_address;
1847 spin_lock(&vcpu->kvm->lock);
1848 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1849 tr->physical_address = gpa;
1850 tr->valid = gpa != UNMAPPED_GVA;
1853 spin_unlock(&vcpu->kvm->lock);
1859 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1860 struct kvm_interrupt *irq)
1862 if (irq->irq < 0 || irq->irq >= 256)
1866 set_bit(irq->irq, vcpu->irq_pending);
1867 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1874 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1875 struct kvm_debug_guest *dbg)
1881 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1888 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1890 struct kvm_vcpu *vcpu = filp->private_data;
1892 fput(vcpu->kvm->filp);
1896 static struct file_operations kvm_vcpu_fops = {
1897 .release = kvm_vcpu_release,
1898 .unlocked_ioctl = kvm_vcpu_ioctl,
1899 .compat_ioctl = kvm_vcpu_ioctl,
1903 * Allocates an inode for the vcpu.
1905 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1908 struct inode *inode;
1911 atomic_inc(&vcpu->kvm->filp->f_count);
1912 inode = kvmfs_inode(&kvm_vcpu_fops);
1913 if (IS_ERR(inode)) {
1918 file = kvmfs_file(inode, vcpu);
1924 r = get_unused_fd();
1928 fd_install(fd, file);
1937 fput(vcpu->kvm->filp);
1942 * Creates some virtual cpus. Good luck creating more than one.
1944 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1947 struct kvm_vcpu *vcpu;
1953 vcpu = &kvm->vcpus[n];
1955 mutex_lock(&vcpu->mutex);
1958 mutex_unlock(&vcpu->mutex);
1962 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
1964 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
1966 r = kvm_arch_ops->vcpu_create(vcpu);
1968 goto out_free_vcpus;
1970 r = kvm_mmu_create(vcpu);
1972 goto out_free_vcpus;
1974 kvm_arch_ops->vcpu_load(vcpu);
1975 r = kvm_mmu_setup(vcpu);
1977 r = kvm_arch_ops->vcpu_setup(vcpu);
1981 goto out_free_vcpus;
1983 r = create_vcpu_fd(vcpu);
1985 goto out_free_vcpus;
1990 kvm_free_vcpu(vcpu);
1991 mutex_unlock(&vcpu->mutex);
1996 static long kvm_vcpu_ioctl(struct file *filp,
1997 unsigned int ioctl, unsigned long arg)
1999 struct kvm_vcpu *vcpu = filp->private_data;
2000 void __user *argp = (void __user *)arg;
2005 struct kvm_run kvm_run;
2008 if (copy_from_user(&kvm_run, argp, sizeof kvm_run))
2010 r = kvm_vcpu_ioctl_run(vcpu, &kvm_run);
2011 if (r < 0 && r != -EINTR)
2013 if (copy_to_user(argp, &kvm_run, sizeof kvm_run)) {
2019 case KVM_GET_REGS: {
2020 struct kvm_regs kvm_regs;
2022 memset(&kvm_regs, 0, sizeof kvm_regs);
2023 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2027 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2032 case KVM_SET_REGS: {
2033 struct kvm_regs kvm_regs;
2036 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2038 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2044 case KVM_GET_SREGS: {
2045 struct kvm_sregs kvm_sregs;
2047 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2048 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2052 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2057 case KVM_SET_SREGS: {
2058 struct kvm_sregs kvm_sregs;
2061 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2063 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2069 case KVM_TRANSLATE: {
2070 struct kvm_translation tr;
2073 if (copy_from_user(&tr, argp, sizeof tr))
2075 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2079 if (copy_to_user(argp, &tr, sizeof tr))
2084 case KVM_INTERRUPT: {
2085 struct kvm_interrupt irq;
2088 if (copy_from_user(&irq, argp, sizeof irq))
2090 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2096 case KVM_DEBUG_GUEST: {
2097 struct kvm_debug_guest dbg;
2100 if (copy_from_user(&dbg, argp, sizeof dbg))
2102 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2109 r = msr_io(vcpu, argp, get_msr, 1);
2112 r = msr_io(vcpu, argp, do_set_msr, 0);
2121 static long kvm_vm_ioctl(struct file *filp,
2122 unsigned int ioctl, unsigned long arg)
2124 struct kvm *kvm = filp->private_data;
2125 void __user *argp = (void __user *)arg;
2129 case KVM_CREATE_VCPU:
2130 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2134 case KVM_SET_MEMORY_REGION: {
2135 struct kvm_memory_region kvm_mem;
2138 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2140 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2145 case KVM_GET_DIRTY_LOG: {
2146 struct kvm_dirty_log log;
2149 if (copy_from_user(&log, argp, sizeof log))
2151 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2163 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2164 unsigned long address,
2167 struct kvm *kvm = vma->vm_file->private_data;
2168 unsigned long pgoff;
2169 struct kvm_memory_slot *slot;
2172 *type = VM_FAULT_MINOR;
2173 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2174 slot = gfn_to_memslot(kvm, pgoff);
2176 return NOPAGE_SIGBUS;
2177 page = gfn_to_page(slot, pgoff);
2179 return NOPAGE_SIGBUS;
2184 static struct vm_operations_struct kvm_vm_vm_ops = {
2185 .nopage = kvm_vm_nopage,
2188 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2190 vma->vm_ops = &kvm_vm_vm_ops;
2194 static struct file_operations kvm_vm_fops = {
2195 .release = kvm_vm_release,
2196 .unlocked_ioctl = kvm_vm_ioctl,
2197 .compat_ioctl = kvm_vm_ioctl,
2198 .mmap = kvm_vm_mmap,
2201 static int kvm_dev_ioctl_create_vm(void)
2204 struct inode *inode;
2208 inode = kvmfs_inode(&kvm_vm_fops);
2209 if (IS_ERR(inode)) {
2214 kvm = kvm_create_vm();
2220 file = kvmfs_file(inode, kvm);
2227 r = get_unused_fd();
2231 fd_install(fd, file);
2238 kvm_destroy_vm(kvm);
2245 static long kvm_dev_ioctl(struct file *filp,
2246 unsigned int ioctl, unsigned long arg)
2248 void __user *argp = (void __user *)arg;
2252 case KVM_GET_API_VERSION:
2253 r = KVM_API_VERSION;
2256 r = kvm_dev_ioctl_create_vm();
2258 case KVM_GET_MSR_INDEX_LIST: {
2259 struct kvm_msr_list __user *user_msr_list = argp;
2260 struct kvm_msr_list msr_list;
2264 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2267 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2268 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2271 if (n < num_msrs_to_save)
2274 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2275 num_msrs_to_save * sizeof(u32)))
2277 if (copy_to_user(user_msr_list->indices
2278 + num_msrs_to_save * sizeof(u32),
2280 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2292 static struct file_operations kvm_chardev_ops = {
2293 .open = kvm_dev_open,
2294 .release = kvm_dev_release,
2295 .unlocked_ioctl = kvm_dev_ioctl,
2296 .compat_ioctl = kvm_dev_ioctl,
2299 static struct miscdevice kvm_dev = {
2305 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2308 if (val == SYS_RESTART) {
2310 * Some (well, at least mine) BIOSes hang on reboot if
2313 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2314 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2319 static struct notifier_block kvm_reboot_notifier = {
2320 .notifier_call = kvm_reboot,
2325 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2328 static void decache_vcpus_on_cpu(int cpu)
2331 struct kvm_vcpu *vcpu;
2334 spin_lock(&kvm_lock);
2335 list_for_each_entry(vm, &vm_list, vm_list)
2336 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2337 vcpu = &vm->vcpus[i];
2339 * If the vcpu is locked, then it is running on some
2340 * other cpu and therefore it is not cached on the
2343 * If it's not locked, check the last cpu it executed
2346 if (mutex_trylock(&vcpu->mutex)) {
2347 if (vcpu->cpu == cpu) {
2348 kvm_arch_ops->vcpu_decache(vcpu);
2351 mutex_unlock(&vcpu->mutex);
2354 spin_unlock(&kvm_lock);
2357 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2363 case CPU_DOWN_PREPARE:
2364 case CPU_UP_CANCELED:
2365 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2367 decache_vcpus_on_cpu(cpu);
2368 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2372 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2374 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2381 static struct notifier_block kvm_cpu_notifier = {
2382 .notifier_call = kvm_cpu_hotplug,
2383 .priority = 20, /* must be > scheduler priority */
2386 static __init void kvm_init_debug(void)
2388 struct kvm_stats_debugfs_item *p;
2390 debugfs_dir = debugfs_create_dir("kvm", NULL);
2391 for (p = debugfs_entries; p->name; ++p)
2392 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2396 static void kvm_exit_debug(void)
2398 struct kvm_stats_debugfs_item *p;
2400 for (p = debugfs_entries; p->name; ++p)
2401 debugfs_remove(p->dentry);
2402 debugfs_remove(debugfs_dir);
2405 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2407 decache_vcpus_on_cpu(raw_smp_processor_id());
2408 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2412 static int kvm_resume(struct sys_device *dev)
2414 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2418 static struct sysdev_class kvm_sysdev_class = {
2419 set_kset_name("kvm"),
2420 .suspend = kvm_suspend,
2421 .resume = kvm_resume,
2424 static struct sys_device kvm_sysdev = {
2426 .cls = &kvm_sysdev_class,
2429 hpa_t bad_page_address;
2431 static int kvmfs_get_sb(struct file_system_type *fs_type, int flags,
2432 const char *dev_name, void *data, struct vfsmount *mnt)
2434 return get_sb_pseudo(fs_type, "kvm:", NULL, KVMFS_MAGIC, mnt);
2437 static struct file_system_type kvm_fs_type = {
2439 .get_sb = kvmfs_get_sb,
2440 .kill_sb = kill_anon_super,
2443 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2448 printk(KERN_ERR "kvm: already loaded the other module\n");
2452 if (!ops->cpu_has_kvm_support()) {
2453 printk(KERN_ERR "kvm: no hardware support\n");
2456 if (ops->disabled_by_bios()) {
2457 printk(KERN_ERR "kvm: disabled by bios\n");
2463 r = kvm_arch_ops->hardware_setup();
2467 on_each_cpu(kvm_arch_ops->hardware_enable, NULL, 0, 1);
2468 r = register_cpu_notifier(&kvm_cpu_notifier);
2471 register_reboot_notifier(&kvm_reboot_notifier);
2473 r = sysdev_class_register(&kvm_sysdev_class);
2477 r = sysdev_register(&kvm_sysdev);
2481 kvm_chardev_ops.owner = module;
2483 r = misc_register(&kvm_dev);
2485 printk (KERN_ERR "kvm: misc device register failed\n");
2492 sysdev_unregister(&kvm_sysdev);
2494 sysdev_class_unregister(&kvm_sysdev_class);
2496 unregister_reboot_notifier(&kvm_reboot_notifier);
2497 unregister_cpu_notifier(&kvm_cpu_notifier);
2499 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2500 kvm_arch_ops->hardware_unsetup();
2504 void kvm_exit_arch(void)
2506 misc_deregister(&kvm_dev);
2507 sysdev_unregister(&kvm_sysdev);
2508 sysdev_class_unregister(&kvm_sysdev_class);
2509 unregister_reboot_notifier(&kvm_reboot_notifier);
2510 unregister_cpu_notifier(&kvm_cpu_notifier);
2511 on_each_cpu(kvm_arch_ops->hardware_disable, NULL, 0, 1);
2512 kvm_arch_ops->hardware_unsetup();
2513 kvm_arch_ops = NULL;
2516 static __init int kvm_init(void)
2518 static struct page *bad_page;
2521 r = register_filesystem(&kvm_fs_type);
2525 kvmfs_mnt = kern_mount(&kvm_fs_type);
2526 r = PTR_ERR(kvmfs_mnt);
2527 if (IS_ERR(kvmfs_mnt))
2531 kvm_init_msr_list();
2533 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2538 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2539 memset(__va(bad_page_address), 0, PAGE_SIZE);
2547 unregister_filesystem(&kvm_fs_type);
2552 static __exit void kvm_exit(void)
2555 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2557 unregister_filesystem(&kvm_fs_type);
2560 module_init(kvm_init)
2561 module_exit(kvm_exit)
2563 EXPORT_SYMBOL_GPL(kvm_init_arch);
2564 EXPORT_SYMBOL_GPL(kvm_exit_arch);