2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
41 #include <asm/processor.h>
44 #include <asm/uaccess.h>
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
53 static cpumask_t cpus_hardware_enabled;
55 struct kvm_arch_ops *kvm_arch_ops;
57 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
59 static struct kvm_stats_debugfs_item {
62 struct dentry *dentry;
63 } debugfs_entries[] = {
64 { "pf_fixed", STAT_OFFSET(pf_fixed) },
65 { "pf_guest", STAT_OFFSET(pf_guest) },
66 { "tlb_flush", STAT_OFFSET(tlb_flush) },
67 { "invlpg", STAT_OFFSET(invlpg) },
68 { "exits", STAT_OFFSET(exits) },
69 { "io_exits", STAT_OFFSET(io_exits) },
70 { "mmio_exits", STAT_OFFSET(mmio_exits) },
71 { "signal_exits", STAT_OFFSET(signal_exits) },
72 { "irq_window", STAT_OFFSET(irq_window_exits) },
73 { "halt_exits", STAT_OFFSET(halt_exits) },
74 { "request_irq", STAT_OFFSET(request_irq_exits) },
75 { "irq_exits", STAT_OFFSET(irq_exits) },
76 { "light_exits", STAT_OFFSET(light_exits) },
77 { "efer_reload", STAT_OFFSET(efer_reload) },
81 static struct dentry *debugfs_dir;
83 #define MAX_IO_MSRS 256
85 #define CR0_RESERVED_BITS \
86 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
87 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
88 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
89 #define CR4_RESERVED_BITS \
90 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
91 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
92 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
93 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
95 #define CR8_RESEVED_BITS (~0x0fULL)
96 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
99 // LDT or TSS descriptor in the GDT. 16 bytes.
100 struct segment_descriptor_64 {
101 struct segment_descriptor s;
108 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
111 unsigned long segment_base(u16 selector)
113 struct descriptor_table gdt;
114 struct segment_descriptor *d;
115 unsigned long table_base;
116 typedef unsigned long ul;
122 asm ("sgdt %0" : "=m"(gdt));
123 table_base = gdt.base;
125 if (selector & 4) { /* from ldt */
128 asm ("sldt %0" : "=g"(ldt_selector));
129 table_base = segment_base(ldt_selector);
131 d = (struct segment_descriptor *)(table_base + (selector & ~7));
132 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
135 && (d->type == 2 || d->type == 9 || d->type == 11))
136 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
140 EXPORT_SYMBOL_GPL(segment_base);
142 static inline int valid_vcpu(int n)
144 return likely(n >= 0 && n < KVM_MAX_VCPUS);
147 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
150 unsigned char *host_buf = dest;
151 unsigned long req_size = size;
159 paddr = gva_to_hpa(vcpu, addr);
161 if (is_error_hpa(paddr))
164 guest_buf = (hva_t)kmap_atomic(
165 pfn_to_page(paddr >> PAGE_SHIFT),
167 offset = addr & ~PAGE_MASK;
169 now = min(size, PAGE_SIZE - offset);
170 memcpy(host_buf, (void*)guest_buf, now);
174 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
176 return req_size - size;
178 EXPORT_SYMBOL_GPL(kvm_read_guest);
180 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
183 unsigned char *host_buf = data;
184 unsigned long req_size = size;
193 paddr = gva_to_hpa(vcpu, addr);
195 if (is_error_hpa(paddr))
198 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
199 mark_page_dirty(vcpu->kvm, gfn);
200 guest_buf = (hva_t)kmap_atomic(
201 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
202 offset = addr & ~PAGE_MASK;
204 now = min(size, PAGE_SIZE - offset);
205 memcpy((void*)guest_buf, host_buf, now);
209 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
211 return req_size - size;
213 EXPORT_SYMBOL_GPL(kvm_write_guest);
215 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
217 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
220 vcpu->guest_fpu_loaded = 1;
221 fx_save(vcpu->host_fx_image);
222 fx_restore(vcpu->guest_fx_image);
224 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
226 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
228 if (!vcpu->guest_fpu_loaded)
231 vcpu->guest_fpu_loaded = 0;
232 fx_save(vcpu->guest_fx_image);
233 fx_restore(vcpu->host_fx_image);
235 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
238 * Switches to specified vcpu, until a matching vcpu_put()
240 static void vcpu_load(struct kvm_vcpu *vcpu)
242 mutex_lock(&vcpu->mutex);
243 kvm_arch_ops->vcpu_load(vcpu);
246 static void vcpu_put(struct kvm_vcpu *vcpu)
248 kvm_arch_ops->vcpu_put(vcpu);
249 mutex_unlock(&vcpu->mutex);
252 static void ack_flush(void *_completed)
254 atomic_t *completed = _completed;
256 atomic_inc(completed);
259 void kvm_flush_remote_tlbs(struct kvm *kvm)
263 struct kvm_vcpu *vcpu;
266 atomic_set(&completed, 0);
269 for (i = 0; i < kvm->nvcpus; ++i) {
270 vcpu = &kvm->vcpus[i];
271 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
274 if (cpu != -1 && cpu != raw_smp_processor_id())
275 if (!cpu_isset(cpu, cpus)) {
282 * We really want smp_call_function_mask() here. But that's not
283 * available, so ipi all cpus in parallel and wait for them
286 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
287 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
288 while (atomic_read(&completed) != needed) {
294 static struct kvm *kvm_create_vm(void)
296 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
300 return ERR_PTR(-ENOMEM);
302 kvm_io_bus_init(&kvm->pio_bus);
303 spin_lock_init(&kvm->lock);
304 INIT_LIST_HEAD(&kvm->active_mmu_pages);
305 kvm_io_bus_init(&kvm->mmio_bus);
306 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
307 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
309 mutex_init(&vcpu->mutex);
312 vcpu->mmu.root_hpa = INVALID_PAGE;
314 spin_lock(&kvm_lock);
315 list_add(&kvm->vm_list, &vm_list);
316 spin_unlock(&kvm_lock);
320 static int kvm_dev_open(struct inode *inode, struct file *filp)
326 * Free any memory in @free but not in @dont.
328 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
329 struct kvm_memory_slot *dont)
333 if (!dont || free->phys_mem != dont->phys_mem)
334 if (free->phys_mem) {
335 for (i = 0; i < free->npages; ++i)
336 if (free->phys_mem[i])
337 __free_page(free->phys_mem[i]);
338 vfree(free->phys_mem);
341 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
342 vfree(free->dirty_bitmap);
344 free->phys_mem = NULL;
346 free->dirty_bitmap = NULL;
349 static void kvm_free_physmem(struct kvm *kvm)
353 for (i = 0; i < kvm->nmemslots; ++i)
354 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
357 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
361 for (i = 0; i < 2; ++i)
362 if (vcpu->pio.guest_pages[i]) {
363 __free_page(vcpu->pio.guest_pages[i]);
364 vcpu->pio.guest_pages[i] = NULL;
368 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
374 kvm_mmu_unload(vcpu);
378 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
384 kvm_mmu_destroy(vcpu);
386 kvm_arch_ops->vcpu_free(vcpu);
387 free_page((unsigned long)vcpu->run);
389 free_page((unsigned long)vcpu->pio_data);
390 vcpu->pio_data = NULL;
391 free_pio_guest_pages(vcpu);
394 static void kvm_free_vcpus(struct kvm *kvm)
399 * Unpin any mmu pages first.
401 for (i = 0; i < KVM_MAX_VCPUS; ++i)
402 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
403 for (i = 0; i < KVM_MAX_VCPUS; ++i)
404 kvm_free_vcpu(&kvm->vcpus[i]);
407 static int kvm_dev_release(struct inode *inode, struct file *filp)
412 static void kvm_destroy_vm(struct kvm *kvm)
414 spin_lock(&kvm_lock);
415 list_del(&kvm->vm_list);
416 spin_unlock(&kvm_lock);
417 kvm_io_bus_destroy(&kvm->pio_bus);
418 kvm_io_bus_destroy(&kvm->mmio_bus);
420 kvm_free_physmem(kvm);
424 static int kvm_vm_release(struct inode *inode, struct file *filp)
426 struct kvm *kvm = filp->private_data;
432 static void inject_gp(struct kvm_vcpu *vcpu)
434 kvm_arch_ops->inject_gp(vcpu, 0);
438 * Load the pae pdptrs. Return true is they are all valid.
440 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
442 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
443 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
450 spin_lock(&vcpu->kvm->lock);
451 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
452 /* FIXME: !page - emulate? 0xff? */
453 pdpt = kmap_atomic(page, KM_USER0);
456 for (i = 0; i < 4; ++i) {
457 pdpte = pdpt[offset + i];
458 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
464 for (i = 0; i < 4; ++i)
465 vcpu->pdptrs[i] = pdpt[offset + i];
468 kunmap_atomic(pdpt, KM_USER0);
469 spin_unlock(&vcpu->kvm->lock);
474 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
476 if (cr0 & CR0_RESERVED_BITS) {
477 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
483 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
484 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
489 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
490 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
491 "and a clear PE flag\n");
496 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
498 if ((vcpu->shadow_efer & EFER_LME)) {
502 printk(KERN_DEBUG "set_cr0: #GP, start paging "
503 "in long mode while PAE is disabled\n");
507 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
509 printk(KERN_DEBUG "set_cr0: #GP, start paging "
510 "in long mode while CS.L == 1\n");
517 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
518 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
526 kvm_arch_ops->set_cr0(vcpu, cr0);
529 spin_lock(&vcpu->kvm->lock);
530 kvm_mmu_reset_context(vcpu);
531 spin_unlock(&vcpu->kvm->lock);
534 EXPORT_SYMBOL_GPL(set_cr0);
536 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
538 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
540 EXPORT_SYMBOL_GPL(lmsw);
542 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
544 if (cr4 & CR4_RESERVED_BITS) {
545 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
550 if (is_long_mode(vcpu)) {
551 if (!(cr4 & X86_CR4_PAE)) {
552 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
557 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
558 && !load_pdptrs(vcpu, vcpu->cr3)) {
559 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
563 if (cr4 & X86_CR4_VMXE) {
564 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
568 kvm_arch_ops->set_cr4(vcpu, cr4);
569 spin_lock(&vcpu->kvm->lock);
570 kvm_mmu_reset_context(vcpu);
571 spin_unlock(&vcpu->kvm->lock);
573 EXPORT_SYMBOL_GPL(set_cr4);
575 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
577 if (is_long_mode(vcpu)) {
578 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
579 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
585 if (cr3 & CR3_PAE_RESERVED_BITS) {
587 "set_cr3: #GP, reserved bits\n");
591 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
592 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
598 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
600 "set_cr3: #GP, reserved bits\n");
608 spin_lock(&vcpu->kvm->lock);
610 * Does the new cr3 value map to physical memory? (Note, we
611 * catch an invalid cr3 even in real-mode, because it would
612 * cause trouble later on when we turn on paging anyway.)
614 * A real CPU would silently accept an invalid cr3 and would
615 * attempt to use it - with largely undefined (and often hard
616 * to debug) behavior on the guest side.
618 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
621 vcpu->mmu.new_cr3(vcpu);
622 spin_unlock(&vcpu->kvm->lock);
624 EXPORT_SYMBOL_GPL(set_cr3);
626 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
628 if ( cr8 & CR8_RESEVED_BITS) {
629 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
635 EXPORT_SYMBOL_GPL(set_cr8);
637 void fx_init(struct kvm_vcpu *vcpu)
639 struct __attribute__ ((__packed__)) fx_image_s {
645 u64 operand;// fpu dp
651 fx_save(vcpu->host_fx_image);
653 fx_save(vcpu->guest_fx_image);
654 fx_restore(vcpu->host_fx_image);
656 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
657 fx_image->mxcsr = 0x1f80;
658 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
659 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
661 EXPORT_SYMBOL_GPL(fx_init);
664 * Allocate some memory and give it an address in the guest physical address
667 * Discontiguous memory is allowed, mostly for framebuffers.
669 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
670 struct kvm_memory_region *mem)
674 unsigned long npages;
676 struct kvm_memory_slot *memslot;
677 struct kvm_memory_slot old, new;
678 int memory_config_version;
681 /* General sanity checks */
682 if (mem->memory_size & (PAGE_SIZE - 1))
684 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
686 if (mem->slot >= KVM_MEMORY_SLOTS)
688 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
691 memslot = &kvm->memslots[mem->slot];
692 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
693 npages = mem->memory_size >> PAGE_SHIFT;
696 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
699 spin_lock(&kvm->lock);
701 memory_config_version = kvm->memory_config_version;
702 new = old = *memslot;
704 new.base_gfn = base_gfn;
706 new.flags = mem->flags;
708 /* Disallow changing a memory slot's size. */
710 if (npages && old.npages && npages != old.npages)
713 /* Check for overlaps */
715 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
716 struct kvm_memory_slot *s = &kvm->memslots[i];
720 if (!((base_gfn + npages <= s->base_gfn) ||
721 (base_gfn >= s->base_gfn + s->npages)))
725 * Do memory allocations outside lock. memory_config_version will
728 spin_unlock(&kvm->lock);
730 /* Deallocate if slot is being removed */
734 /* Free page dirty bitmap if unneeded */
735 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
736 new.dirty_bitmap = NULL;
740 /* Allocate if a slot is being created */
741 if (npages && !new.phys_mem) {
742 new.phys_mem = vmalloc(npages * sizeof(struct page *));
747 memset(new.phys_mem, 0, npages * sizeof(struct page *));
748 for (i = 0; i < npages; ++i) {
749 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
751 if (!new.phys_mem[i])
753 set_page_private(new.phys_mem[i],0);
757 /* Allocate page dirty bitmap if needed */
758 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
759 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
761 new.dirty_bitmap = vmalloc(dirty_bytes);
762 if (!new.dirty_bitmap)
764 memset(new.dirty_bitmap, 0, dirty_bytes);
767 spin_lock(&kvm->lock);
769 if (memory_config_version != kvm->memory_config_version) {
770 spin_unlock(&kvm->lock);
771 kvm_free_physmem_slot(&new, &old);
779 if (mem->slot >= kvm->nmemslots)
780 kvm->nmemslots = mem->slot + 1;
783 ++kvm->memory_config_version;
785 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
786 kvm_flush_remote_tlbs(kvm);
788 spin_unlock(&kvm->lock);
790 kvm_free_physmem_slot(&old, &new);
794 spin_unlock(&kvm->lock);
796 kvm_free_physmem_slot(&new, &old);
802 * Get (and clear) the dirty memory log for a memory slot.
804 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
805 struct kvm_dirty_log *log)
807 struct kvm_memory_slot *memslot;
810 unsigned long any = 0;
812 spin_lock(&kvm->lock);
815 * Prevent changes to guest memory configuration even while the lock
819 spin_unlock(&kvm->lock);
821 if (log->slot >= KVM_MEMORY_SLOTS)
824 memslot = &kvm->memslots[log->slot];
826 if (!memslot->dirty_bitmap)
829 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
831 for (i = 0; !any && i < n/sizeof(long); ++i)
832 any = memslot->dirty_bitmap[i];
835 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
838 spin_lock(&kvm->lock);
839 kvm_mmu_slot_remove_write_access(kvm, log->slot);
840 kvm_flush_remote_tlbs(kvm);
841 memset(memslot->dirty_bitmap, 0, n);
842 spin_unlock(&kvm->lock);
847 spin_lock(&kvm->lock);
849 spin_unlock(&kvm->lock);
854 * Set a new alias region. Aliases map a portion of physical memory into
855 * another portion. This is useful for memory windows, for example the PC
858 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
859 struct kvm_memory_alias *alias)
862 struct kvm_mem_alias *p;
865 /* General sanity checks */
866 if (alias->memory_size & (PAGE_SIZE - 1))
868 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
870 if (alias->slot >= KVM_ALIAS_SLOTS)
872 if (alias->guest_phys_addr + alias->memory_size
873 < alias->guest_phys_addr)
875 if (alias->target_phys_addr + alias->memory_size
876 < alias->target_phys_addr)
879 spin_lock(&kvm->lock);
881 p = &kvm->aliases[alias->slot];
882 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
883 p->npages = alias->memory_size >> PAGE_SHIFT;
884 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
886 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
887 if (kvm->aliases[n - 1].npages)
891 kvm_mmu_zap_all(kvm);
893 spin_unlock(&kvm->lock);
901 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
904 struct kvm_mem_alias *alias;
906 for (i = 0; i < kvm->naliases; ++i) {
907 alias = &kvm->aliases[i];
908 if (gfn >= alias->base_gfn
909 && gfn < alias->base_gfn + alias->npages)
910 return alias->target_gfn + gfn - alias->base_gfn;
915 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
919 for (i = 0; i < kvm->nmemslots; ++i) {
920 struct kvm_memory_slot *memslot = &kvm->memslots[i];
922 if (gfn >= memslot->base_gfn
923 && gfn < memslot->base_gfn + memslot->npages)
929 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
931 gfn = unalias_gfn(kvm, gfn);
932 return __gfn_to_memslot(kvm, gfn);
935 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
937 struct kvm_memory_slot *slot;
939 gfn = unalias_gfn(kvm, gfn);
940 slot = __gfn_to_memslot(kvm, gfn);
943 return slot->phys_mem[gfn - slot->base_gfn];
945 EXPORT_SYMBOL_GPL(gfn_to_page);
947 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
950 struct kvm_memory_slot *memslot;
951 unsigned long rel_gfn;
953 for (i = 0; i < kvm->nmemslots; ++i) {
954 memslot = &kvm->memslots[i];
956 if (gfn >= memslot->base_gfn
957 && gfn < memslot->base_gfn + memslot->npages) {
959 if (!memslot->dirty_bitmap)
962 rel_gfn = gfn - memslot->base_gfn;
965 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
966 set_bit(rel_gfn, memslot->dirty_bitmap);
972 static int emulator_read_std(unsigned long addr,
975 struct x86_emulate_ctxt *ctxt)
977 struct kvm_vcpu *vcpu = ctxt->vcpu;
981 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
982 unsigned offset = addr & (PAGE_SIZE-1);
983 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
988 if (gpa == UNMAPPED_GVA)
989 return X86EMUL_PROPAGATE_FAULT;
990 pfn = gpa >> PAGE_SHIFT;
991 page = gfn_to_page(vcpu->kvm, pfn);
993 return X86EMUL_UNHANDLEABLE;
994 page_virt = kmap_atomic(page, KM_USER0);
996 memcpy(data, page_virt + offset, tocopy);
998 kunmap_atomic(page_virt, KM_USER0);
1005 return X86EMUL_CONTINUE;
1008 static int emulator_write_std(unsigned long addr,
1011 struct x86_emulate_ctxt *ctxt)
1013 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1015 return X86EMUL_UNHANDLEABLE;
1018 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1022 * Note that its important to have this wrapper function because
1023 * in the very near future we will be checking for MMIOs against
1024 * the LAPIC as well as the general MMIO bus
1026 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1029 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1032 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1035 static int emulator_read_emulated(unsigned long addr,
1038 struct x86_emulate_ctxt *ctxt)
1040 struct kvm_vcpu *vcpu = ctxt->vcpu;
1041 struct kvm_io_device *mmio_dev;
1044 if (vcpu->mmio_read_completed) {
1045 memcpy(val, vcpu->mmio_data, bytes);
1046 vcpu->mmio_read_completed = 0;
1047 return X86EMUL_CONTINUE;
1048 } else if (emulator_read_std(addr, val, bytes, ctxt)
1049 == X86EMUL_CONTINUE)
1050 return X86EMUL_CONTINUE;
1052 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1053 if (gpa == UNMAPPED_GVA)
1054 return X86EMUL_PROPAGATE_FAULT;
1057 * Is this MMIO handled locally?
1059 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1061 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1062 return X86EMUL_CONTINUE;
1065 vcpu->mmio_needed = 1;
1066 vcpu->mmio_phys_addr = gpa;
1067 vcpu->mmio_size = bytes;
1068 vcpu->mmio_is_write = 0;
1070 return X86EMUL_UNHANDLEABLE;
1073 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1074 const void *val, int bytes)
1078 unsigned offset = offset_in_page(gpa);
1080 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1082 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1085 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1086 virt = kmap_atomic(page, KM_USER0);
1087 kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1088 memcpy(virt + offset_in_page(gpa), val, bytes);
1089 kunmap_atomic(virt, KM_USER0);
1093 static int emulator_write_emulated_onepage(unsigned long addr,
1096 struct x86_emulate_ctxt *ctxt)
1098 struct kvm_vcpu *vcpu = ctxt->vcpu;
1099 struct kvm_io_device *mmio_dev;
1100 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1102 if (gpa == UNMAPPED_GVA) {
1103 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1104 return X86EMUL_PROPAGATE_FAULT;
1107 if (emulator_write_phys(vcpu, gpa, val, bytes))
1108 return X86EMUL_CONTINUE;
1111 * Is this MMIO handled locally?
1113 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1115 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1116 return X86EMUL_CONTINUE;
1119 vcpu->mmio_needed = 1;
1120 vcpu->mmio_phys_addr = gpa;
1121 vcpu->mmio_size = bytes;
1122 vcpu->mmio_is_write = 1;
1123 memcpy(vcpu->mmio_data, val, bytes);
1125 return X86EMUL_CONTINUE;
1128 static int emulator_write_emulated(unsigned long addr,
1131 struct x86_emulate_ctxt *ctxt)
1133 /* Crossing a page boundary? */
1134 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1137 now = -addr & ~PAGE_MASK;
1138 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1139 if (rc != X86EMUL_CONTINUE)
1145 return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1148 static int emulator_cmpxchg_emulated(unsigned long addr,
1152 struct x86_emulate_ctxt *ctxt)
1154 static int reported;
1158 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1160 return emulator_write_emulated(addr, new, bytes, ctxt);
1163 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1165 return kvm_arch_ops->get_segment_base(vcpu, seg);
1168 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1170 return X86EMUL_CONTINUE;
1173 int emulate_clts(struct kvm_vcpu *vcpu)
1177 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1178 kvm_arch_ops->set_cr0(vcpu, cr0);
1179 return X86EMUL_CONTINUE;
1182 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1184 struct kvm_vcpu *vcpu = ctxt->vcpu;
1188 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1189 return X86EMUL_CONTINUE;
1191 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1193 return X86EMUL_UNHANDLEABLE;
1197 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1199 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1202 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1204 /* FIXME: better handling */
1205 return X86EMUL_UNHANDLEABLE;
1207 return X86EMUL_CONTINUE;
1210 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1212 static int reported;
1214 unsigned long rip = ctxt->vcpu->rip;
1215 unsigned long rip_linear;
1217 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1222 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1224 printk(KERN_ERR "emulation failed but !mmio_needed?"
1225 " rip %lx %02x %02x %02x %02x\n",
1226 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1230 struct x86_emulate_ops emulate_ops = {
1231 .read_std = emulator_read_std,
1232 .write_std = emulator_write_std,
1233 .read_emulated = emulator_read_emulated,
1234 .write_emulated = emulator_write_emulated,
1235 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1238 int emulate_instruction(struct kvm_vcpu *vcpu,
1239 struct kvm_run *run,
1243 struct x86_emulate_ctxt emulate_ctxt;
1247 vcpu->mmio_fault_cr2 = cr2;
1248 kvm_arch_ops->cache_regs(vcpu);
1250 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1252 emulate_ctxt.vcpu = vcpu;
1253 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1254 emulate_ctxt.cr2 = cr2;
1255 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1256 ? X86EMUL_MODE_REAL : cs_l
1257 ? X86EMUL_MODE_PROT64 : cs_db
1258 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1260 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1261 emulate_ctxt.cs_base = 0;
1262 emulate_ctxt.ds_base = 0;
1263 emulate_ctxt.es_base = 0;
1264 emulate_ctxt.ss_base = 0;
1266 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1267 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1268 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1269 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1272 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1273 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1275 vcpu->mmio_is_write = 0;
1276 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1278 if ((r || vcpu->mmio_is_write) && run) {
1279 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1280 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1281 run->mmio.len = vcpu->mmio_size;
1282 run->mmio.is_write = vcpu->mmio_is_write;
1286 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1287 return EMULATE_DONE;
1288 if (!vcpu->mmio_needed) {
1289 report_emulation_failure(&emulate_ctxt);
1290 return EMULATE_FAIL;
1292 return EMULATE_DO_MMIO;
1295 kvm_arch_ops->decache_regs(vcpu);
1296 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1298 if (vcpu->mmio_is_write) {
1299 vcpu->mmio_needed = 0;
1300 return EMULATE_DO_MMIO;
1303 return EMULATE_DONE;
1305 EXPORT_SYMBOL_GPL(emulate_instruction);
1307 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1309 if (vcpu->irq_summary)
1312 vcpu->run->exit_reason = KVM_EXIT_HLT;
1313 ++vcpu->stat.halt_exits;
1316 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1318 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1320 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1322 kvm_arch_ops->cache_regs(vcpu);
1324 #ifdef CONFIG_X86_64
1325 if (is_long_mode(vcpu)) {
1326 nr = vcpu->regs[VCPU_REGS_RAX];
1327 a0 = vcpu->regs[VCPU_REGS_RDI];
1328 a1 = vcpu->regs[VCPU_REGS_RSI];
1329 a2 = vcpu->regs[VCPU_REGS_RDX];
1330 a3 = vcpu->regs[VCPU_REGS_RCX];
1331 a4 = vcpu->regs[VCPU_REGS_R8];
1332 a5 = vcpu->regs[VCPU_REGS_R9];
1336 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1337 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1338 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1339 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1340 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1341 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1342 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1346 run->hypercall.args[0] = a0;
1347 run->hypercall.args[1] = a1;
1348 run->hypercall.args[2] = a2;
1349 run->hypercall.args[3] = a3;
1350 run->hypercall.args[4] = a4;
1351 run->hypercall.args[5] = a5;
1352 run->hypercall.ret = ret;
1353 run->hypercall.longmode = is_long_mode(vcpu);
1354 kvm_arch_ops->decache_regs(vcpu);
1357 vcpu->regs[VCPU_REGS_RAX] = ret;
1358 kvm_arch_ops->decache_regs(vcpu);
1361 EXPORT_SYMBOL_GPL(kvm_hypercall);
1363 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1365 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1368 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1370 struct descriptor_table dt = { limit, base };
1372 kvm_arch_ops->set_gdt(vcpu, &dt);
1375 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1377 struct descriptor_table dt = { limit, base };
1379 kvm_arch_ops->set_idt(vcpu, &dt);
1382 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1383 unsigned long *rflags)
1386 *rflags = kvm_arch_ops->get_rflags(vcpu);
1389 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1391 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1402 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1407 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1408 unsigned long *rflags)
1412 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1413 *rflags = kvm_arch_ops->get_rflags(vcpu);
1422 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1425 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1430 * Register the para guest with the host:
1432 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1434 struct kvm_vcpu_para_state *para_state;
1435 hpa_t para_state_hpa, hypercall_hpa;
1436 struct page *para_state_page;
1437 unsigned char *hypercall;
1438 gpa_t hypercall_gpa;
1440 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1441 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1444 * Needs to be page aligned:
1446 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1449 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1450 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1451 if (is_error_hpa(para_state_hpa))
1454 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1455 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1456 para_state = kmap_atomic(para_state_page, KM_USER0);
1458 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1459 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1461 para_state->host_version = KVM_PARA_API_VERSION;
1463 * We cannot support guests that try to register themselves
1464 * with a newer API version than the host supports:
1466 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1467 para_state->ret = -KVM_EINVAL;
1468 goto err_kunmap_skip;
1471 hypercall_gpa = para_state->hypercall_gpa;
1472 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1473 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1474 if (is_error_hpa(hypercall_hpa)) {
1475 para_state->ret = -KVM_EINVAL;
1476 goto err_kunmap_skip;
1479 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1480 vcpu->para_state_page = para_state_page;
1481 vcpu->para_state_gpa = para_state_gpa;
1482 vcpu->hypercall_gpa = hypercall_gpa;
1484 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1485 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1486 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1487 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1488 kunmap_atomic(hypercall, KM_USER1);
1490 para_state->ret = 0;
1492 kunmap_atomic(para_state, KM_USER0);
1498 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1503 case 0xc0010010: /* SYSCFG */
1504 case 0xc0010015: /* HWCR */
1505 case MSR_IA32_PLATFORM_ID:
1506 case MSR_IA32_P5_MC_ADDR:
1507 case MSR_IA32_P5_MC_TYPE:
1508 case MSR_IA32_MC0_CTL:
1509 case MSR_IA32_MCG_STATUS:
1510 case MSR_IA32_MCG_CAP:
1511 case MSR_IA32_MC0_MISC:
1512 case MSR_IA32_MC0_MISC+4:
1513 case MSR_IA32_MC0_MISC+8:
1514 case MSR_IA32_MC0_MISC+12:
1515 case MSR_IA32_MC0_MISC+16:
1516 case MSR_IA32_UCODE_REV:
1517 case MSR_IA32_PERF_STATUS:
1518 case MSR_IA32_EBL_CR_POWERON:
1519 /* MTRR registers */
1521 case 0x200 ... 0x2ff:
1524 case 0xcd: /* fsb frequency */
1527 case MSR_IA32_APICBASE:
1528 data = vcpu->apic_base;
1530 case MSR_IA32_MISC_ENABLE:
1531 data = vcpu->ia32_misc_enable_msr;
1533 #ifdef CONFIG_X86_64
1535 data = vcpu->shadow_efer;
1539 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1545 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1548 * Reads an msr value (of 'msr_index') into 'pdata'.
1549 * Returns 0 on success, non-0 otherwise.
1550 * Assumes vcpu_load() was already called.
1552 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1554 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1557 #ifdef CONFIG_X86_64
1559 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1561 if (efer & EFER_RESERVED_BITS) {
1562 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1569 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1570 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1575 kvm_arch_ops->set_efer(vcpu, efer);
1578 efer |= vcpu->shadow_efer & EFER_LMA;
1580 vcpu->shadow_efer = efer;
1585 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1588 #ifdef CONFIG_X86_64
1590 set_efer(vcpu, data);
1593 case MSR_IA32_MC0_STATUS:
1594 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1595 __FUNCTION__, data);
1597 case MSR_IA32_MCG_STATUS:
1598 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1599 __FUNCTION__, data);
1601 case MSR_IA32_UCODE_REV:
1602 case MSR_IA32_UCODE_WRITE:
1603 case 0x200 ... 0x2ff: /* MTRRs */
1605 case MSR_IA32_APICBASE:
1606 vcpu->apic_base = data;
1608 case MSR_IA32_MISC_ENABLE:
1609 vcpu->ia32_misc_enable_msr = data;
1612 * This is the 'probe whether the host is KVM' logic:
1614 case MSR_KVM_API_MAGIC:
1615 return vcpu_register_para(vcpu, data);
1618 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1623 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1626 * Writes msr value into into the appropriate "register".
1627 * Returns 0 on success, non-0 otherwise.
1628 * Assumes vcpu_load() was already called.
1630 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1632 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1635 void kvm_resched(struct kvm_vcpu *vcpu)
1637 if (!need_resched())
1643 EXPORT_SYMBOL_GPL(kvm_resched);
1645 void load_msrs(struct vmx_msr_entry *e, int n)
1649 for (i = 0; i < n; ++i)
1650 wrmsrl(e[i].index, e[i].data);
1652 EXPORT_SYMBOL_GPL(load_msrs);
1654 void save_msrs(struct vmx_msr_entry *e, int n)
1658 for (i = 0; i < n; ++i)
1659 rdmsrl(e[i].index, e[i].data);
1661 EXPORT_SYMBOL_GPL(save_msrs);
1663 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1667 struct kvm_cpuid_entry *e, *best;
1669 kvm_arch_ops->cache_regs(vcpu);
1670 function = vcpu->regs[VCPU_REGS_RAX];
1671 vcpu->regs[VCPU_REGS_RAX] = 0;
1672 vcpu->regs[VCPU_REGS_RBX] = 0;
1673 vcpu->regs[VCPU_REGS_RCX] = 0;
1674 vcpu->regs[VCPU_REGS_RDX] = 0;
1676 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1677 e = &vcpu->cpuid_entries[i];
1678 if (e->function == function) {
1683 * Both basic or both extended?
1685 if (((e->function ^ function) & 0x80000000) == 0)
1686 if (!best || e->function > best->function)
1690 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1691 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1692 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1693 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1695 kvm_arch_ops->decache_regs(vcpu);
1696 kvm_arch_ops->skip_emulated_instruction(vcpu);
1698 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1700 static int pio_copy_data(struct kvm_vcpu *vcpu)
1702 void *p = vcpu->pio_data;
1705 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1707 kvm_arch_ops->vcpu_put(vcpu);
1708 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1711 kvm_arch_ops->vcpu_load(vcpu);
1712 free_pio_guest_pages(vcpu);
1715 q += vcpu->pio.guest_page_offset;
1716 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1718 memcpy(q, p, bytes);
1720 memcpy(p, q, bytes);
1721 q -= vcpu->pio.guest_page_offset;
1723 kvm_arch_ops->vcpu_load(vcpu);
1724 free_pio_guest_pages(vcpu);
1728 static int complete_pio(struct kvm_vcpu *vcpu)
1730 struct kvm_pio_request *io = &vcpu->pio;
1734 kvm_arch_ops->cache_regs(vcpu);
1738 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1742 r = pio_copy_data(vcpu);
1744 kvm_arch_ops->cache_regs(vcpu);
1751 delta *= io->cur_count;
1753 * The size of the register should really depend on
1754 * current address size.
1756 vcpu->regs[VCPU_REGS_RCX] -= delta;
1762 vcpu->regs[VCPU_REGS_RDI] += delta;
1764 vcpu->regs[VCPU_REGS_RSI] += delta;
1767 kvm_arch_ops->decache_regs(vcpu);
1769 io->count -= io->cur_count;
1773 kvm_arch_ops->skip_emulated_instruction(vcpu);
1777 static void kernel_pio(struct kvm_io_device *pio_dev,
1778 struct kvm_vcpu *vcpu,
1781 /* TODO: String I/O for in kernel device */
1784 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1788 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1793 static void pio_string_write(struct kvm_io_device *pio_dev,
1794 struct kvm_vcpu *vcpu)
1796 struct kvm_pio_request *io = &vcpu->pio;
1797 void *pd = vcpu->pio_data;
1800 for (i = 0; i < io->cur_count; i++) {
1801 kvm_iodevice_write(pio_dev, io->port,
1808 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1809 int size, unsigned long count, int string, int down,
1810 gva_t address, int rep, unsigned port)
1812 unsigned now, in_page;
1816 struct kvm_io_device *pio_dev;
1818 vcpu->run->exit_reason = KVM_EXIT_IO;
1819 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1820 vcpu->run->io.size = size;
1821 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1822 vcpu->run->io.count = count;
1823 vcpu->run->io.port = port;
1824 vcpu->pio.count = count;
1825 vcpu->pio.cur_count = count;
1826 vcpu->pio.size = size;
1828 vcpu->pio.port = port;
1829 vcpu->pio.string = string;
1830 vcpu->pio.down = down;
1831 vcpu->pio.guest_page_offset = offset_in_page(address);
1832 vcpu->pio.rep = rep;
1834 pio_dev = vcpu_find_pio_dev(vcpu, port);
1836 kvm_arch_ops->cache_regs(vcpu);
1837 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1838 kvm_arch_ops->decache_regs(vcpu);
1840 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1848 kvm_arch_ops->skip_emulated_instruction(vcpu);
1852 now = min(count, PAGE_SIZE / size);
1855 in_page = PAGE_SIZE - offset_in_page(address);
1857 in_page = offset_in_page(address) + size;
1858 now = min(count, (unsigned long)in_page / size);
1861 * String I/O straddles page boundary. Pin two guest pages
1862 * so that we satisfy atomicity constraints. Do just one
1863 * transaction to avoid complexity.
1870 * String I/O in reverse. Yuck. Kill the guest, fix later.
1872 printk(KERN_ERR "kvm: guest string pio down\n");
1876 vcpu->run->io.count = now;
1877 vcpu->pio.cur_count = now;
1879 for (i = 0; i < nr_pages; ++i) {
1880 spin_lock(&vcpu->kvm->lock);
1881 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1884 vcpu->pio.guest_pages[i] = page;
1885 spin_unlock(&vcpu->kvm->lock);
1888 free_pio_guest_pages(vcpu);
1893 if (!vcpu->pio.in) {
1894 /* string PIO write */
1895 ret = pio_copy_data(vcpu);
1896 if (ret >= 0 && pio_dev) {
1897 pio_string_write(pio_dev, vcpu);
1899 if (vcpu->pio.count == 0)
1903 printk(KERN_ERR "no string pio read support yet, "
1904 "port %x size %d count %ld\n",
1909 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1911 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1918 if (vcpu->sigset_active)
1919 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1921 /* re-sync apic's tpr */
1922 vcpu->cr8 = kvm_run->cr8;
1924 if (vcpu->pio.cur_count) {
1925 r = complete_pio(vcpu);
1930 if (vcpu->mmio_needed) {
1931 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1932 vcpu->mmio_read_completed = 1;
1933 vcpu->mmio_needed = 0;
1934 r = emulate_instruction(vcpu, kvm_run,
1935 vcpu->mmio_fault_cr2, 0);
1936 if (r == EMULATE_DO_MMIO) {
1938 * Read-modify-write. Back to userspace.
1940 kvm_run->exit_reason = KVM_EXIT_MMIO;
1946 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1947 kvm_arch_ops->cache_regs(vcpu);
1948 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1949 kvm_arch_ops->decache_regs(vcpu);
1952 r = kvm_arch_ops->run(vcpu, kvm_run);
1955 if (vcpu->sigset_active)
1956 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1962 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1963 struct kvm_regs *regs)
1967 kvm_arch_ops->cache_regs(vcpu);
1969 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1970 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1971 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1972 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1973 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1974 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1975 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1976 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1977 #ifdef CONFIG_X86_64
1978 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1979 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1980 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1981 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1982 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1983 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1984 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1985 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1988 regs->rip = vcpu->rip;
1989 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1992 * Don't leak debug flags in case they were set for guest debugging
1994 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1995 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2002 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2003 struct kvm_regs *regs)
2007 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2008 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2009 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2010 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2011 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2012 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2013 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2014 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2015 #ifdef CONFIG_X86_64
2016 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2017 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2018 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2019 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2020 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2021 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2022 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2023 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2026 vcpu->rip = regs->rip;
2027 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2029 kvm_arch_ops->decache_regs(vcpu);
2036 static void get_segment(struct kvm_vcpu *vcpu,
2037 struct kvm_segment *var, int seg)
2039 return kvm_arch_ops->get_segment(vcpu, var, seg);
2042 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2043 struct kvm_sregs *sregs)
2045 struct descriptor_table dt;
2049 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2050 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2051 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2052 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2053 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2054 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2056 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2057 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2059 kvm_arch_ops->get_idt(vcpu, &dt);
2060 sregs->idt.limit = dt.limit;
2061 sregs->idt.base = dt.base;
2062 kvm_arch_ops->get_gdt(vcpu, &dt);
2063 sregs->gdt.limit = dt.limit;
2064 sregs->gdt.base = dt.base;
2066 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2067 sregs->cr0 = vcpu->cr0;
2068 sregs->cr2 = vcpu->cr2;
2069 sregs->cr3 = vcpu->cr3;
2070 sregs->cr4 = vcpu->cr4;
2071 sregs->cr8 = vcpu->cr8;
2072 sregs->efer = vcpu->shadow_efer;
2073 sregs->apic_base = vcpu->apic_base;
2075 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2076 sizeof sregs->interrupt_bitmap);
2083 static void set_segment(struct kvm_vcpu *vcpu,
2084 struct kvm_segment *var, int seg)
2086 return kvm_arch_ops->set_segment(vcpu, var, seg);
2089 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2090 struct kvm_sregs *sregs)
2092 int mmu_reset_needed = 0;
2094 struct descriptor_table dt;
2098 dt.limit = sregs->idt.limit;
2099 dt.base = sregs->idt.base;
2100 kvm_arch_ops->set_idt(vcpu, &dt);
2101 dt.limit = sregs->gdt.limit;
2102 dt.base = sregs->gdt.base;
2103 kvm_arch_ops->set_gdt(vcpu, &dt);
2105 vcpu->cr2 = sregs->cr2;
2106 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2107 vcpu->cr3 = sregs->cr3;
2109 vcpu->cr8 = sregs->cr8;
2111 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2112 #ifdef CONFIG_X86_64
2113 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2115 vcpu->apic_base = sregs->apic_base;
2117 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2119 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2120 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2122 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2123 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2124 if (!is_long_mode(vcpu) && is_pae(vcpu))
2125 load_pdptrs(vcpu, vcpu->cr3);
2127 if (mmu_reset_needed)
2128 kvm_mmu_reset_context(vcpu);
2130 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2131 sizeof vcpu->irq_pending);
2132 vcpu->irq_summary = 0;
2133 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2134 if (vcpu->irq_pending[i])
2135 __set_bit(i, &vcpu->irq_summary);
2137 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2138 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2139 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2140 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2141 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2142 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2144 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2145 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2153 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2154 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2156 * This list is modified at module load time to reflect the
2157 * capabilities of the host cpu.
2159 static u32 msrs_to_save[] = {
2160 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2162 #ifdef CONFIG_X86_64
2163 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2165 MSR_IA32_TIME_STAMP_COUNTER,
2168 static unsigned num_msrs_to_save;
2170 static u32 emulated_msrs[] = {
2171 MSR_IA32_MISC_ENABLE,
2174 static __init void kvm_init_msr_list(void)
2179 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2180 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2183 msrs_to_save[j] = msrs_to_save[i];
2186 num_msrs_to_save = j;
2190 * Adapt set_msr() to msr_io()'s calling convention
2192 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2194 return kvm_set_msr(vcpu, index, *data);
2198 * Read or write a bunch of msrs. All parameters are kernel addresses.
2200 * @return number of msrs set successfully.
2202 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2203 struct kvm_msr_entry *entries,
2204 int (*do_msr)(struct kvm_vcpu *vcpu,
2205 unsigned index, u64 *data))
2211 for (i = 0; i < msrs->nmsrs; ++i)
2212 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2221 * Read or write a bunch of msrs. Parameters are user addresses.
2223 * @return number of msrs set successfully.
2225 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2226 int (*do_msr)(struct kvm_vcpu *vcpu,
2227 unsigned index, u64 *data),
2230 struct kvm_msrs msrs;
2231 struct kvm_msr_entry *entries;
2236 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2240 if (msrs.nmsrs >= MAX_IO_MSRS)
2244 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2245 entries = vmalloc(size);
2250 if (copy_from_user(entries, user_msrs->entries, size))
2253 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2258 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2270 * Translate a guest virtual address to a guest physical address.
2272 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2273 struct kvm_translation *tr)
2275 unsigned long vaddr = tr->linear_address;
2279 spin_lock(&vcpu->kvm->lock);
2280 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2281 tr->physical_address = gpa;
2282 tr->valid = gpa != UNMAPPED_GVA;
2285 spin_unlock(&vcpu->kvm->lock);
2291 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2292 struct kvm_interrupt *irq)
2294 if (irq->irq < 0 || irq->irq >= 256)
2298 set_bit(irq->irq, vcpu->irq_pending);
2299 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2306 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2307 struct kvm_debug_guest *dbg)
2313 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2320 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2321 unsigned long address,
2324 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2325 unsigned long pgoff;
2328 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2330 page = virt_to_page(vcpu->run);
2331 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2332 page = virt_to_page(vcpu->pio_data);
2334 return NOPAGE_SIGBUS;
2337 *type = VM_FAULT_MINOR;
2342 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2343 .nopage = kvm_vcpu_nopage,
2346 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2348 vma->vm_ops = &kvm_vcpu_vm_ops;
2352 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2354 struct kvm_vcpu *vcpu = filp->private_data;
2356 fput(vcpu->kvm->filp);
2360 static struct file_operations kvm_vcpu_fops = {
2361 .release = kvm_vcpu_release,
2362 .unlocked_ioctl = kvm_vcpu_ioctl,
2363 .compat_ioctl = kvm_vcpu_ioctl,
2364 .mmap = kvm_vcpu_mmap,
2368 * Allocates an inode for the vcpu.
2370 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2373 struct inode *inode;
2376 r = anon_inode_getfd(&fd, &inode, &file,
2377 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2380 atomic_inc(&vcpu->kvm->filp->f_count);
2385 * Creates some virtual cpus. Good luck creating more than one.
2387 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2390 struct kvm_vcpu *vcpu;
2397 vcpu = &kvm->vcpus[n];
2400 mutex_lock(&vcpu->mutex);
2403 mutex_unlock(&vcpu->mutex);
2407 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2411 vcpu->run = page_address(page);
2413 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2417 vcpu->pio_data = page_address(page);
2419 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2421 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2424 r = kvm_arch_ops->vcpu_create(vcpu);
2426 goto out_free_vcpus;
2428 r = kvm_mmu_create(vcpu);
2430 goto out_free_vcpus;
2432 kvm_arch_ops->vcpu_load(vcpu);
2433 r = kvm_mmu_setup(vcpu);
2435 r = kvm_arch_ops->vcpu_setup(vcpu);
2439 goto out_free_vcpus;
2441 r = create_vcpu_fd(vcpu);
2443 goto out_free_vcpus;
2445 spin_lock(&kvm_lock);
2446 if (n >= kvm->nvcpus)
2447 kvm->nvcpus = n + 1;
2448 spin_unlock(&kvm_lock);
2453 kvm_free_vcpu(vcpu);
2455 free_page((unsigned long)vcpu->run);
2458 mutex_unlock(&vcpu->mutex);
2463 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2467 struct kvm_cpuid_entry *e, *entry;
2469 rdmsrl(MSR_EFER, efer);
2471 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2472 e = &vcpu->cpuid_entries[i];
2473 if (e->function == 0x80000001) {
2478 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2479 entry->edx &= ~(1 << 20);
2480 printk(KERN_INFO "kvm: guest NX capability removed\n");
2484 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2485 struct kvm_cpuid *cpuid,
2486 struct kvm_cpuid_entry __user *entries)
2491 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2494 if (copy_from_user(&vcpu->cpuid_entries, entries,
2495 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2497 vcpu->cpuid_nent = cpuid->nent;
2498 cpuid_fix_nx_cap(vcpu);
2505 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2508 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2509 vcpu->sigset_active = 1;
2510 vcpu->sigset = *sigset;
2512 vcpu->sigset_active = 0;
2517 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2518 * we have asm/x86/processor.h
2529 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2530 #ifdef CONFIG_X86_64
2531 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2533 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2537 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2539 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2543 memcpy(fpu->fpr, fxsave->st_space, 128);
2544 fpu->fcw = fxsave->cwd;
2545 fpu->fsw = fxsave->swd;
2546 fpu->ftwx = fxsave->twd;
2547 fpu->last_opcode = fxsave->fop;
2548 fpu->last_ip = fxsave->rip;
2549 fpu->last_dp = fxsave->rdp;
2550 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2557 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2559 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2563 memcpy(fxsave->st_space, fpu->fpr, 128);
2564 fxsave->cwd = fpu->fcw;
2565 fxsave->swd = fpu->fsw;
2566 fxsave->twd = fpu->ftwx;
2567 fxsave->fop = fpu->last_opcode;
2568 fxsave->rip = fpu->last_ip;
2569 fxsave->rdp = fpu->last_dp;
2570 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2577 static long kvm_vcpu_ioctl(struct file *filp,
2578 unsigned int ioctl, unsigned long arg)
2580 struct kvm_vcpu *vcpu = filp->private_data;
2581 void __user *argp = (void __user *)arg;
2589 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2591 case KVM_GET_REGS: {
2592 struct kvm_regs kvm_regs;
2594 memset(&kvm_regs, 0, sizeof kvm_regs);
2595 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2599 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2604 case KVM_SET_REGS: {
2605 struct kvm_regs kvm_regs;
2608 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2610 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2616 case KVM_GET_SREGS: {
2617 struct kvm_sregs kvm_sregs;
2619 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2620 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2624 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2629 case KVM_SET_SREGS: {
2630 struct kvm_sregs kvm_sregs;
2633 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2635 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2641 case KVM_TRANSLATE: {
2642 struct kvm_translation tr;
2645 if (copy_from_user(&tr, argp, sizeof tr))
2647 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2651 if (copy_to_user(argp, &tr, sizeof tr))
2656 case KVM_INTERRUPT: {
2657 struct kvm_interrupt irq;
2660 if (copy_from_user(&irq, argp, sizeof irq))
2662 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2668 case KVM_DEBUG_GUEST: {
2669 struct kvm_debug_guest dbg;
2672 if (copy_from_user(&dbg, argp, sizeof dbg))
2674 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2681 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2684 r = msr_io(vcpu, argp, do_set_msr, 0);
2686 case KVM_SET_CPUID: {
2687 struct kvm_cpuid __user *cpuid_arg = argp;
2688 struct kvm_cpuid cpuid;
2691 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2693 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2698 case KVM_SET_SIGNAL_MASK: {
2699 struct kvm_signal_mask __user *sigmask_arg = argp;
2700 struct kvm_signal_mask kvm_sigmask;
2701 sigset_t sigset, *p;
2706 if (copy_from_user(&kvm_sigmask, argp,
2707 sizeof kvm_sigmask))
2710 if (kvm_sigmask.len != sizeof sigset)
2713 if (copy_from_user(&sigset, sigmask_arg->sigset,
2718 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2724 memset(&fpu, 0, sizeof fpu);
2725 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2729 if (copy_to_user(argp, &fpu, sizeof fpu))
2738 if (copy_from_user(&fpu, argp, sizeof fpu))
2740 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2753 static long kvm_vm_ioctl(struct file *filp,
2754 unsigned int ioctl, unsigned long arg)
2756 struct kvm *kvm = filp->private_data;
2757 void __user *argp = (void __user *)arg;
2761 case KVM_CREATE_VCPU:
2762 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2766 case KVM_SET_MEMORY_REGION: {
2767 struct kvm_memory_region kvm_mem;
2770 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2772 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2777 case KVM_GET_DIRTY_LOG: {
2778 struct kvm_dirty_log log;
2781 if (copy_from_user(&log, argp, sizeof log))
2783 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2788 case KVM_SET_MEMORY_ALIAS: {
2789 struct kvm_memory_alias alias;
2792 if (copy_from_user(&alias, argp, sizeof alias))
2794 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2806 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2807 unsigned long address,
2810 struct kvm *kvm = vma->vm_file->private_data;
2811 unsigned long pgoff;
2814 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2815 page = gfn_to_page(kvm, pgoff);
2817 return NOPAGE_SIGBUS;
2820 *type = VM_FAULT_MINOR;
2825 static struct vm_operations_struct kvm_vm_vm_ops = {
2826 .nopage = kvm_vm_nopage,
2829 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2831 vma->vm_ops = &kvm_vm_vm_ops;
2835 static struct file_operations kvm_vm_fops = {
2836 .release = kvm_vm_release,
2837 .unlocked_ioctl = kvm_vm_ioctl,
2838 .compat_ioctl = kvm_vm_ioctl,
2839 .mmap = kvm_vm_mmap,
2842 static int kvm_dev_ioctl_create_vm(void)
2845 struct inode *inode;
2849 kvm = kvm_create_vm();
2851 return PTR_ERR(kvm);
2852 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2854 kvm_destroy_vm(kvm);
2863 static long kvm_dev_ioctl(struct file *filp,
2864 unsigned int ioctl, unsigned long arg)
2866 void __user *argp = (void __user *)arg;
2870 case KVM_GET_API_VERSION:
2874 r = KVM_API_VERSION;
2880 r = kvm_dev_ioctl_create_vm();
2882 case KVM_GET_MSR_INDEX_LIST: {
2883 struct kvm_msr_list __user *user_msr_list = argp;
2884 struct kvm_msr_list msr_list;
2888 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2891 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2892 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2895 if (n < num_msrs_to_save)
2898 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2899 num_msrs_to_save * sizeof(u32)))
2901 if (copy_to_user(user_msr_list->indices
2902 + num_msrs_to_save * sizeof(u32),
2904 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2909 case KVM_CHECK_EXTENSION:
2911 * No extensions defined at present.
2915 case KVM_GET_VCPU_MMAP_SIZE:
2928 static struct file_operations kvm_chardev_ops = {
2929 .open = kvm_dev_open,
2930 .release = kvm_dev_release,
2931 .unlocked_ioctl = kvm_dev_ioctl,
2932 .compat_ioctl = kvm_dev_ioctl,
2935 static struct miscdevice kvm_dev = {
2942 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2945 static void decache_vcpus_on_cpu(int cpu)
2948 struct kvm_vcpu *vcpu;
2951 spin_lock(&kvm_lock);
2952 list_for_each_entry(vm, &vm_list, vm_list)
2953 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2954 vcpu = &vm->vcpus[i];
2956 * If the vcpu is locked, then it is running on some
2957 * other cpu and therefore it is not cached on the
2960 * If it's not locked, check the last cpu it executed
2963 if (mutex_trylock(&vcpu->mutex)) {
2964 if (vcpu->cpu == cpu) {
2965 kvm_arch_ops->vcpu_decache(vcpu);
2968 mutex_unlock(&vcpu->mutex);
2971 spin_unlock(&kvm_lock);
2974 static void hardware_enable(void *junk)
2976 int cpu = raw_smp_processor_id();
2978 if (cpu_isset(cpu, cpus_hardware_enabled))
2980 cpu_set(cpu, cpus_hardware_enabled);
2981 kvm_arch_ops->hardware_enable(NULL);
2984 static void hardware_disable(void *junk)
2986 int cpu = raw_smp_processor_id();
2988 if (!cpu_isset(cpu, cpus_hardware_enabled))
2990 cpu_clear(cpu, cpus_hardware_enabled);
2991 decache_vcpus_on_cpu(cpu);
2992 kvm_arch_ops->hardware_disable(NULL);
2995 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3002 case CPU_DYING_FROZEN:
3003 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3005 hardware_disable(NULL);
3007 case CPU_UP_CANCELED:
3008 case CPU_UP_CANCELED_FROZEN:
3009 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3011 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3014 case CPU_ONLINE_FROZEN:
3015 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3017 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3023 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3026 if (val == SYS_RESTART) {
3028 * Some (well, at least mine) BIOSes hang on reboot if
3031 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3032 on_each_cpu(hardware_disable, NULL, 0, 1);
3037 static struct notifier_block kvm_reboot_notifier = {
3038 .notifier_call = kvm_reboot,
3042 void kvm_io_bus_init(struct kvm_io_bus *bus)
3044 memset(bus, 0, sizeof(*bus));
3047 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3051 for (i = 0; i < bus->dev_count; i++) {
3052 struct kvm_io_device *pos = bus->devs[i];
3054 kvm_iodevice_destructor(pos);
3058 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3062 for (i = 0; i < bus->dev_count; i++) {
3063 struct kvm_io_device *pos = bus->devs[i];
3065 if (pos->in_range(pos, addr))
3072 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3074 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3076 bus->devs[bus->dev_count++] = dev;
3079 static struct notifier_block kvm_cpu_notifier = {
3080 .notifier_call = kvm_cpu_hotplug,
3081 .priority = 20, /* must be > scheduler priority */
3084 static u64 stat_get(void *_offset)
3086 unsigned offset = (long)_offset;
3089 struct kvm_vcpu *vcpu;
3092 spin_lock(&kvm_lock);
3093 list_for_each_entry(kvm, &vm_list, vm_list)
3094 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3095 vcpu = &kvm->vcpus[i];
3096 total += *(u32 *)((void *)vcpu + offset);
3098 spin_unlock(&kvm_lock);
3102 static void stat_set(void *offset, u64 val)
3106 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3108 static __init void kvm_init_debug(void)
3110 struct kvm_stats_debugfs_item *p;
3112 debugfs_dir = debugfs_create_dir("kvm", NULL);
3113 for (p = debugfs_entries; p->name; ++p)
3114 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3115 (void *)(long)p->offset,
3119 static void kvm_exit_debug(void)
3121 struct kvm_stats_debugfs_item *p;
3123 for (p = debugfs_entries; p->name; ++p)
3124 debugfs_remove(p->dentry);
3125 debugfs_remove(debugfs_dir);
3128 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3130 hardware_disable(NULL);
3134 static int kvm_resume(struct sys_device *dev)
3136 hardware_enable(NULL);
3140 static struct sysdev_class kvm_sysdev_class = {
3141 set_kset_name("kvm"),
3142 .suspend = kvm_suspend,
3143 .resume = kvm_resume,
3146 static struct sys_device kvm_sysdev = {
3148 .cls = &kvm_sysdev_class,
3151 hpa_t bad_page_address;
3153 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3158 printk(KERN_ERR "kvm: already loaded the other module\n");
3162 if (!ops->cpu_has_kvm_support()) {
3163 printk(KERN_ERR "kvm: no hardware support\n");
3166 if (ops->disabled_by_bios()) {
3167 printk(KERN_ERR "kvm: disabled by bios\n");
3173 r = kvm_arch_ops->hardware_setup();
3177 on_each_cpu(hardware_enable, NULL, 0, 1);
3178 r = register_cpu_notifier(&kvm_cpu_notifier);
3181 register_reboot_notifier(&kvm_reboot_notifier);
3183 r = sysdev_class_register(&kvm_sysdev_class);
3187 r = sysdev_register(&kvm_sysdev);
3191 kvm_chardev_ops.owner = module;
3193 r = misc_register(&kvm_dev);
3195 printk (KERN_ERR "kvm: misc device register failed\n");
3202 sysdev_unregister(&kvm_sysdev);
3204 sysdev_class_unregister(&kvm_sysdev_class);
3206 unregister_reboot_notifier(&kvm_reboot_notifier);
3207 unregister_cpu_notifier(&kvm_cpu_notifier);
3209 on_each_cpu(hardware_disable, NULL, 0, 1);
3210 kvm_arch_ops->hardware_unsetup();
3212 kvm_arch_ops = NULL;
3216 void kvm_exit_arch(void)
3218 misc_deregister(&kvm_dev);
3219 sysdev_unregister(&kvm_sysdev);
3220 sysdev_class_unregister(&kvm_sysdev_class);
3221 unregister_reboot_notifier(&kvm_reboot_notifier);
3222 unregister_cpu_notifier(&kvm_cpu_notifier);
3223 on_each_cpu(hardware_disable, NULL, 0, 1);
3224 kvm_arch_ops->hardware_unsetup();
3225 kvm_arch_ops = NULL;
3228 static __init int kvm_init(void)
3230 static struct page *bad_page;
3233 r = kvm_mmu_module_init();
3239 kvm_init_msr_list();
3241 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3246 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3247 memset(__va(bad_page_address), 0, PAGE_SIZE);
3253 kvm_mmu_module_exit();
3258 static __exit void kvm_exit(void)
3261 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3262 kvm_mmu_module_exit();
3265 module_init(kvm_init)
3266 module_exit(kvm_exit)
3268 EXPORT_SYMBOL_GPL(kvm_init_arch);
3269 EXPORT_SYMBOL_GPL(kvm_exit_arch);