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_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
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;
448 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
450 spin_lock(&vcpu->kvm->lock);
451 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
457 pdpt = kmap_atomic(page, KM_USER0);
458 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
459 kunmap_atomic(pdpt, KM_USER0);
461 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
462 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
469 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
471 spin_unlock(&vcpu->kvm->lock);
476 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
478 if (cr0 & CR0_RESERVED_BITS) {
479 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
485 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
486 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
491 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
492 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
493 "and a clear PE flag\n");
498 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
500 if ((vcpu->shadow_efer & EFER_LME)) {
504 printk(KERN_DEBUG "set_cr0: #GP, start paging "
505 "in long mode while PAE is disabled\n");
509 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
511 printk(KERN_DEBUG "set_cr0: #GP, start paging "
512 "in long mode while CS.L == 1\n");
519 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
520 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
528 kvm_arch_ops->set_cr0(vcpu, cr0);
531 spin_lock(&vcpu->kvm->lock);
532 kvm_mmu_reset_context(vcpu);
533 spin_unlock(&vcpu->kvm->lock);
536 EXPORT_SYMBOL_GPL(set_cr0);
538 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
540 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
542 EXPORT_SYMBOL_GPL(lmsw);
544 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
546 if (cr4 & CR4_RESERVED_BITS) {
547 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
552 if (is_long_mode(vcpu)) {
553 if (!(cr4 & X86_CR4_PAE)) {
554 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
559 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
560 && !load_pdptrs(vcpu, vcpu->cr3)) {
561 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
566 if (cr4 & X86_CR4_VMXE) {
567 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
571 kvm_arch_ops->set_cr4(vcpu, cr4);
572 spin_lock(&vcpu->kvm->lock);
573 kvm_mmu_reset_context(vcpu);
574 spin_unlock(&vcpu->kvm->lock);
576 EXPORT_SYMBOL_GPL(set_cr4);
578 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
580 if (is_long_mode(vcpu)) {
581 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
582 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
588 if (cr3 & CR3_PAE_RESERVED_BITS) {
590 "set_cr3: #GP, reserved bits\n");
594 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
595 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
601 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
603 "set_cr3: #GP, reserved bits\n");
611 spin_lock(&vcpu->kvm->lock);
613 * Does the new cr3 value map to physical memory? (Note, we
614 * catch an invalid cr3 even in real-mode, because it would
615 * cause trouble later on when we turn on paging anyway.)
617 * A real CPU would silently accept an invalid cr3 and would
618 * attempt to use it - with largely undefined (and often hard
619 * to debug) behavior on the guest side.
621 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
624 vcpu->mmu.new_cr3(vcpu);
625 spin_unlock(&vcpu->kvm->lock);
627 EXPORT_SYMBOL_GPL(set_cr3);
629 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
631 if (cr8 & CR8_RESERVED_BITS) {
632 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
638 EXPORT_SYMBOL_GPL(set_cr8);
640 void fx_init(struct kvm_vcpu *vcpu)
642 struct __attribute__ ((__packed__)) fx_image_s {
648 u64 operand;// fpu dp
654 fx_save(vcpu->host_fx_image);
656 fx_save(vcpu->guest_fx_image);
657 fx_restore(vcpu->host_fx_image);
659 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
660 fx_image->mxcsr = 0x1f80;
661 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
662 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
664 EXPORT_SYMBOL_GPL(fx_init);
667 * Allocate some memory and give it an address in the guest physical address
670 * Discontiguous memory is allowed, mostly for framebuffers.
672 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
673 struct kvm_memory_region *mem)
677 unsigned long npages;
679 struct kvm_memory_slot *memslot;
680 struct kvm_memory_slot old, new;
681 int memory_config_version;
684 /* General sanity checks */
685 if (mem->memory_size & (PAGE_SIZE - 1))
687 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
689 if (mem->slot >= KVM_MEMORY_SLOTS)
691 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
694 memslot = &kvm->memslots[mem->slot];
695 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
696 npages = mem->memory_size >> PAGE_SHIFT;
699 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
702 spin_lock(&kvm->lock);
704 memory_config_version = kvm->memory_config_version;
705 new = old = *memslot;
707 new.base_gfn = base_gfn;
709 new.flags = mem->flags;
711 /* Disallow changing a memory slot's size. */
713 if (npages && old.npages && npages != old.npages)
716 /* Check for overlaps */
718 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
719 struct kvm_memory_slot *s = &kvm->memslots[i];
723 if (!((base_gfn + npages <= s->base_gfn) ||
724 (base_gfn >= s->base_gfn + s->npages)))
728 * Do memory allocations outside lock. memory_config_version will
731 spin_unlock(&kvm->lock);
733 /* Deallocate if slot is being removed */
737 /* Free page dirty bitmap if unneeded */
738 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
739 new.dirty_bitmap = NULL;
743 /* Allocate if a slot is being created */
744 if (npages && !new.phys_mem) {
745 new.phys_mem = vmalloc(npages * sizeof(struct page *));
750 memset(new.phys_mem, 0, npages * sizeof(struct page *));
751 for (i = 0; i < npages; ++i) {
752 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
754 if (!new.phys_mem[i])
756 set_page_private(new.phys_mem[i],0);
760 /* Allocate page dirty bitmap if needed */
761 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
762 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
764 new.dirty_bitmap = vmalloc(dirty_bytes);
765 if (!new.dirty_bitmap)
767 memset(new.dirty_bitmap, 0, dirty_bytes);
770 spin_lock(&kvm->lock);
772 if (memory_config_version != kvm->memory_config_version) {
773 spin_unlock(&kvm->lock);
774 kvm_free_physmem_slot(&new, &old);
782 if (mem->slot >= kvm->nmemslots)
783 kvm->nmemslots = mem->slot + 1;
786 ++kvm->memory_config_version;
788 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
789 kvm_flush_remote_tlbs(kvm);
791 spin_unlock(&kvm->lock);
793 kvm_free_physmem_slot(&old, &new);
797 spin_unlock(&kvm->lock);
799 kvm_free_physmem_slot(&new, &old);
805 * Get (and clear) the dirty memory log for a memory slot.
807 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
808 struct kvm_dirty_log *log)
810 struct kvm_memory_slot *memslot;
813 unsigned long any = 0;
815 spin_lock(&kvm->lock);
818 * Prevent changes to guest memory configuration even while the lock
822 spin_unlock(&kvm->lock);
824 if (log->slot >= KVM_MEMORY_SLOTS)
827 memslot = &kvm->memslots[log->slot];
829 if (!memslot->dirty_bitmap)
832 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
834 for (i = 0; !any && i < n/sizeof(long); ++i)
835 any = memslot->dirty_bitmap[i];
838 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
841 spin_lock(&kvm->lock);
842 kvm_mmu_slot_remove_write_access(kvm, log->slot);
843 kvm_flush_remote_tlbs(kvm);
844 memset(memslot->dirty_bitmap, 0, n);
845 spin_unlock(&kvm->lock);
850 spin_lock(&kvm->lock);
852 spin_unlock(&kvm->lock);
857 * Set a new alias region. Aliases map a portion of physical memory into
858 * another portion. This is useful for memory windows, for example the PC
861 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
862 struct kvm_memory_alias *alias)
865 struct kvm_mem_alias *p;
868 /* General sanity checks */
869 if (alias->memory_size & (PAGE_SIZE - 1))
871 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
873 if (alias->slot >= KVM_ALIAS_SLOTS)
875 if (alias->guest_phys_addr + alias->memory_size
876 < alias->guest_phys_addr)
878 if (alias->target_phys_addr + alias->memory_size
879 < alias->target_phys_addr)
882 spin_lock(&kvm->lock);
884 p = &kvm->aliases[alias->slot];
885 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
886 p->npages = alias->memory_size >> PAGE_SHIFT;
887 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
889 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
890 if (kvm->aliases[n - 1].npages)
894 kvm_mmu_zap_all(kvm);
896 spin_unlock(&kvm->lock);
904 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
907 struct kvm_mem_alias *alias;
909 for (i = 0; i < kvm->naliases; ++i) {
910 alias = &kvm->aliases[i];
911 if (gfn >= alias->base_gfn
912 && gfn < alias->base_gfn + alias->npages)
913 return alias->target_gfn + gfn - alias->base_gfn;
918 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
922 for (i = 0; i < kvm->nmemslots; ++i) {
923 struct kvm_memory_slot *memslot = &kvm->memslots[i];
925 if (gfn >= memslot->base_gfn
926 && gfn < memslot->base_gfn + memslot->npages)
932 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
934 gfn = unalias_gfn(kvm, gfn);
935 return __gfn_to_memslot(kvm, gfn);
938 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
940 struct kvm_memory_slot *slot;
942 gfn = unalias_gfn(kvm, gfn);
943 slot = __gfn_to_memslot(kvm, gfn);
946 return slot->phys_mem[gfn - slot->base_gfn];
948 EXPORT_SYMBOL_GPL(gfn_to_page);
950 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
953 struct kvm_memory_slot *memslot;
954 unsigned long rel_gfn;
956 for (i = 0; i < kvm->nmemslots; ++i) {
957 memslot = &kvm->memslots[i];
959 if (gfn >= memslot->base_gfn
960 && gfn < memslot->base_gfn + memslot->npages) {
962 if (!memslot->dirty_bitmap)
965 rel_gfn = gfn - memslot->base_gfn;
968 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
969 set_bit(rel_gfn, memslot->dirty_bitmap);
975 static int emulator_read_std(unsigned long addr,
978 struct x86_emulate_ctxt *ctxt)
980 struct kvm_vcpu *vcpu = ctxt->vcpu;
984 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
985 unsigned offset = addr & (PAGE_SIZE-1);
986 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
991 if (gpa == UNMAPPED_GVA)
992 return X86EMUL_PROPAGATE_FAULT;
993 pfn = gpa >> PAGE_SHIFT;
994 page = gfn_to_page(vcpu->kvm, pfn);
996 return X86EMUL_UNHANDLEABLE;
997 page_virt = kmap_atomic(page, KM_USER0);
999 memcpy(data, page_virt + offset, tocopy);
1001 kunmap_atomic(page_virt, KM_USER0);
1008 return X86EMUL_CONTINUE;
1011 static int emulator_write_std(unsigned long addr,
1014 struct x86_emulate_ctxt *ctxt)
1016 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1018 return X86EMUL_UNHANDLEABLE;
1021 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1025 * Note that its important to have this wrapper function because
1026 * in the very near future we will be checking for MMIOs against
1027 * the LAPIC as well as the general MMIO bus
1029 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1032 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1035 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1038 static int emulator_read_emulated(unsigned long addr,
1041 struct x86_emulate_ctxt *ctxt)
1043 struct kvm_vcpu *vcpu = ctxt->vcpu;
1044 struct kvm_io_device *mmio_dev;
1047 if (vcpu->mmio_read_completed) {
1048 memcpy(val, vcpu->mmio_data, bytes);
1049 vcpu->mmio_read_completed = 0;
1050 return X86EMUL_CONTINUE;
1051 } else if (emulator_read_std(addr, val, bytes, ctxt)
1052 == X86EMUL_CONTINUE)
1053 return X86EMUL_CONTINUE;
1055 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1056 if (gpa == UNMAPPED_GVA)
1057 return X86EMUL_PROPAGATE_FAULT;
1060 * Is this MMIO handled locally?
1062 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1064 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1065 return X86EMUL_CONTINUE;
1068 vcpu->mmio_needed = 1;
1069 vcpu->mmio_phys_addr = gpa;
1070 vcpu->mmio_size = bytes;
1071 vcpu->mmio_is_write = 0;
1073 return X86EMUL_UNHANDLEABLE;
1076 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1077 const void *val, int bytes)
1082 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1084 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1087 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1088 virt = kmap_atomic(page, KM_USER0);
1089 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1090 memcpy(virt + offset_in_page(gpa), val, bytes);
1091 kunmap_atomic(virt, KM_USER0);
1095 static int emulator_write_emulated_onepage(unsigned long addr,
1098 struct x86_emulate_ctxt *ctxt)
1100 struct kvm_vcpu *vcpu = ctxt->vcpu;
1101 struct kvm_io_device *mmio_dev;
1102 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1104 if (gpa == UNMAPPED_GVA) {
1105 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1106 return X86EMUL_PROPAGATE_FAULT;
1109 if (emulator_write_phys(vcpu, gpa, val, bytes))
1110 return X86EMUL_CONTINUE;
1113 * Is this MMIO handled locally?
1115 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1117 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1118 return X86EMUL_CONTINUE;
1121 vcpu->mmio_needed = 1;
1122 vcpu->mmio_phys_addr = gpa;
1123 vcpu->mmio_size = bytes;
1124 vcpu->mmio_is_write = 1;
1125 memcpy(vcpu->mmio_data, val, bytes);
1127 return X86EMUL_CONTINUE;
1130 static int emulator_write_emulated(unsigned long addr,
1133 struct x86_emulate_ctxt *ctxt)
1135 /* Crossing a page boundary? */
1136 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1139 now = -addr & ~PAGE_MASK;
1140 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1141 if (rc != X86EMUL_CONTINUE)
1147 return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1150 static int emulator_cmpxchg_emulated(unsigned long addr,
1154 struct x86_emulate_ctxt *ctxt)
1156 static int reported;
1160 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1162 return emulator_write_emulated(addr, new, bytes, ctxt);
1165 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1167 return kvm_arch_ops->get_segment_base(vcpu, seg);
1170 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1172 return X86EMUL_CONTINUE;
1175 int emulate_clts(struct kvm_vcpu *vcpu)
1179 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1180 kvm_arch_ops->set_cr0(vcpu, cr0);
1181 return X86EMUL_CONTINUE;
1184 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1186 struct kvm_vcpu *vcpu = ctxt->vcpu;
1190 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1191 return X86EMUL_CONTINUE;
1193 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1195 return X86EMUL_UNHANDLEABLE;
1199 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1201 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1204 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1206 /* FIXME: better handling */
1207 return X86EMUL_UNHANDLEABLE;
1209 return X86EMUL_CONTINUE;
1212 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1214 static int reported;
1216 unsigned long rip = ctxt->vcpu->rip;
1217 unsigned long rip_linear;
1219 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1224 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1226 printk(KERN_ERR "emulation failed but !mmio_needed?"
1227 " rip %lx %02x %02x %02x %02x\n",
1228 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1232 struct x86_emulate_ops emulate_ops = {
1233 .read_std = emulator_read_std,
1234 .write_std = emulator_write_std,
1235 .read_emulated = emulator_read_emulated,
1236 .write_emulated = emulator_write_emulated,
1237 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1240 int emulate_instruction(struct kvm_vcpu *vcpu,
1241 struct kvm_run *run,
1245 struct x86_emulate_ctxt emulate_ctxt;
1249 vcpu->mmio_fault_cr2 = cr2;
1250 kvm_arch_ops->cache_regs(vcpu);
1252 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1254 emulate_ctxt.vcpu = vcpu;
1255 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1256 emulate_ctxt.cr2 = cr2;
1257 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1258 ? X86EMUL_MODE_REAL : cs_l
1259 ? X86EMUL_MODE_PROT64 : cs_db
1260 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1262 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1263 emulate_ctxt.cs_base = 0;
1264 emulate_ctxt.ds_base = 0;
1265 emulate_ctxt.es_base = 0;
1266 emulate_ctxt.ss_base = 0;
1268 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1269 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1270 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1271 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1274 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1275 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1277 vcpu->mmio_is_write = 0;
1278 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1280 if ((r || vcpu->mmio_is_write) && run) {
1281 run->exit_reason = KVM_EXIT_MMIO;
1282 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1283 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1284 run->mmio.len = vcpu->mmio_size;
1285 run->mmio.is_write = vcpu->mmio_is_write;
1289 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1290 return EMULATE_DONE;
1291 if (!vcpu->mmio_needed) {
1292 report_emulation_failure(&emulate_ctxt);
1293 return EMULATE_FAIL;
1295 return EMULATE_DO_MMIO;
1298 kvm_arch_ops->decache_regs(vcpu);
1299 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1301 if (vcpu->mmio_is_write) {
1302 vcpu->mmio_needed = 0;
1303 return EMULATE_DO_MMIO;
1306 return EMULATE_DONE;
1308 EXPORT_SYMBOL_GPL(emulate_instruction);
1310 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1312 if (vcpu->irq_summary)
1315 vcpu->run->exit_reason = KVM_EXIT_HLT;
1316 ++vcpu->stat.halt_exits;
1319 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1321 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1323 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1325 kvm_arch_ops->cache_regs(vcpu);
1327 #ifdef CONFIG_X86_64
1328 if (is_long_mode(vcpu)) {
1329 nr = vcpu->regs[VCPU_REGS_RAX];
1330 a0 = vcpu->regs[VCPU_REGS_RDI];
1331 a1 = vcpu->regs[VCPU_REGS_RSI];
1332 a2 = vcpu->regs[VCPU_REGS_RDX];
1333 a3 = vcpu->regs[VCPU_REGS_RCX];
1334 a4 = vcpu->regs[VCPU_REGS_R8];
1335 a5 = vcpu->regs[VCPU_REGS_R9];
1339 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1340 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1341 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1342 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1343 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1344 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1345 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1349 run->hypercall.args[0] = a0;
1350 run->hypercall.args[1] = a1;
1351 run->hypercall.args[2] = a2;
1352 run->hypercall.args[3] = a3;
1353 run->hypercall.args[4] = a4;
1354 run->hypercall.args[5] = a5;
1355 run->hypercall.ret = ret;
1356 run->hypercall.longmode = is_long_mode(vcpu);
1357 kvm_arch_ops->decache_regs(vcpu);
1360 vcpu->regs[VCPU_REGS_RAX] = ret;
1361 kvm_arch_ops->decache_regs(vcpu);
1364 EXPORT_SYMBOL_GPL(kvm_hypercall);
1366 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1368 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1371 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1373 struct descriptor_table dt = { limit, base };
1375 kvm_arch_ops->set_gdt(vcpu, &dt);
1378 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1380 struct descriptor_table dt = { limit, base };
1382 kvm_arch_ops->set_idt(vcpu, &dt);
1385 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1386 unsigned long *rflags)
1389 *rflags = kvm_arch_ops->get_rflags(vcpu);
1392 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1394 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1405 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1410 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1411 unsigned long *rflags)
1415 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1416 *rflags = kvm_arch_ops->get_rflags(vcpu);
1425 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1428 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1433 * Register the para guest with the host:
1435 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1437 struct kvm_vcpu_para_state *para_state;
1438 hpa_t para_state_hpa, hypercall_hpa;
1439 struct page *para_state_page;
1440 unsigned char *hypercall;
1441 gpa_t hypercall_gpa;
1443 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1444 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1447 * Needs to be page aligned:
1449 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1452 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1453 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1454 if (is_error_hpa(para_state_hpa))
1457 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1458 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1459 para_state = kmap(para_state_page);
1461 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1462 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1464 para_state->host_version = KVM_PARA_API_VERSION;
1466 * We cannot support guests that try to register themselves
1467 * with a newer API version than the host supports:
1469 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1470 para_state->ret = -KVM_EINVAL;
1471 goto err_kunmap_skip;
1474 hypercall_gpa = para_state->hypercall_gpa;
1475 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1476 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1477 if (is_error_hpa(hypercall_hpa)) {
1478 para_state->ret = -KVM_EINVAL;
1479 goto err_kunmap_skip;
1482 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1483 vcpu->para_state_page = para_state_page;
1484 vcpu->para_state_gpa = para_state_gpa;
1485 vcpu->hypercall_gpa = hypercall_gpa;
1487 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1488 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1489 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1490 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1491 kunmap_atomic(hypercall, KM_USER1);
1493 para_state->ret = 0;
1495 kunmap(para_state_page);
1501 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1506 case 0xc0010010: /* SYSCFG */
1507 case 0xc0010015: /* HWCR */
1508 case MSR_IA32_PLATFORM_ID:
1509 case MSR_IA32_P5_MC_ADDR:
1510 case MSR_IA32_P5_MC_TYPE:
1511 case MSR_IA32_MC0_CTL:
1512 case MSR_IA32_MCG_STATUS:
1513 case MSR_IA32_MCG_CAP:
1514 case MSR_IA32_MC0_MISC:
1515 case MSR_IA32_MC0_MISC+4:
1516 case MSR_IA32_MC0_MISC+8:
1517 case MSR_IA32_MC0_MISC+12:
1518 case MSR_IA32_MC0_MISC+16:
1519 case MSR_IA32_UCODE_REV:
1520 case MSR_IA32_PERF_STATUS:
1521 case MSR_IA32_EBL_CR_POWERON:
1522 /* MTRR registers */
1524 case 0x200 ... 0x2ff:
1527 case 0xcd: /* fsb frequency */
1530 case MSR_IA32_APICBASE:
1531 data = vcpu->apic_base;
1533 case MSR_IA32_MISC_ENABLE:
1534 data = vcpu->ia32_misc_enable_msr;
1536 #ifdef CONFIG_X86_64
1538 data = vcpu->shadow_efer;
1542 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1548 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1551 * Reads an msr value (of 'msr_index') into 'pdata'.
1552 * Returns 0 on success, non-0 otherwise.
1553 * Assumes vcpu_load() was already called.
1555 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1557 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1560 #ifdef CONFIG_X86_64
1562 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1564 if (efer & EFER_RESERVED_BITS) {
1565 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1572 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1573 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1578 kvm_arch_ops->set_efer(vcpu, efer);
1581 efer |= vcpu->shadow_efer & EFER_LMA;
1583 vcpu->shadow_efer = efer;
1588 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1591 #ifdef CONFIG_X86_64
1593 set_efer(vcpu, data);
1596 case MSR_IA32_MC0_STATUS:
1597 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1598 __FUNCTION__, data);
1600 case MSR_IA32_MCG_STATUS:
1601 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1602 __FUNCTION__, data);
1604 case MSR_IA32_UCODE_REV:
1605 case MSR_IA32_UCODE_WRITE:
1606 case 0x200 ... 0x2ff: /* MTRRs */
1608 case MSR_IA32_APICBASE:
1609 vcpu->apic_base = data;
1611 case MSR_IA32_MISC_ENABLE:
1612 vcpu->ia32_misc_enable_msr = data;
1615 * This is the 'probe whether the host is KVM' logic:
1617 case MSR_KVM_API_MAGIC:
1618 return vcpu_register_para(vcpu, data);
1621 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1626 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1629 * Writes msr value into into the appropriate "register".
1630 * Returns 0 on success, non-0 otherwise.
1631 * Assumes vcpu_load() was already called.
1633 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1635 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1638 void kvm_resched(struct kvm_vcpu *vcpu)
1640 if (!need_resched())
1646 EXPORT_SYMBOL_GPL(kvm_resched);
1648 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1652 struct kvm_cpuid_entry *e, *best;
1654 kvm_arch_ops->cache_regs(vcpu);
1655 function = vcpu->regs[VCPU_REGS_RAX];
1656 vcpu->regs[VCPU_REGS_RAX] = 0;
1657 vcpu->regs[VCPU_REGS_RBX] = 0;
1658 vcpu->regs[VCPU_REGS_RCX] = 0;
1659 vcpu->regs[VCPU_REGS_RDX] = 0;
1661 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1662 e = &vcpu->cpuid_entries[i];
1663 if (e->function == function) {
1668 * Both basic or both extended?
1670 if (((e->function ^ function) & 0x80000000) == 0)
1671 if (!best || e->function > best->function)
1675 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1676 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1677 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1678 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1680 kvm_arch_ops->decache_regs(vcpu);
1681 kvm_arch_ops->skip_emulated_instruction(vcpu);
1683 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1685 static int pio_copy_data(struct kvm_vcpu *vcpu)
1687 void *p = vcpu->pio_data;
1690 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1692 kvm_arch_ops->vcpu_put(vcpu);
1693 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1696 kvm_arch_ops->vcpu_load(vcpu);
1697 free_pio_guest_pages(vcpu);
1700 q += vcpu->pio.guest_page_offset;
1701 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1703 memcpy(q, p, bytes);
1705 memcpy(p, q, bytes);
1706 q -= vcpu->pio.guest_page_offset;
1708 kvm_arch_ops->vcpu_load(vcpu);
1709 free_pio_guest_pages(vcpu);
1713 static int complete_pio(struct kvm_vcpu *vcpu)
1715 struct kvm_pio_request *io = &vcpu->pio;
1719 kvm_arch_ops->cache_regs(vcpu);
1723 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1727 r = pio_copy_data(vcpu);
1729 kvm_arch_ops->cache_regs(vcpu);
1736 delta *= io->cur_count;
1738 * The size of the register should really depend on
1739 * current address size.
1741 vcpu->regs[VCPU_REGS_RCX] -= delta;
1747 vcpu->regs[VCPU_REGS_RDI] += delta;
1749 vcpu->regs[VCPU_REGS_RSI] += delta;
1752 kvm_arch_ops->decache_regs(vcpu);
1754 io->count -= io->cur_count;
1758 kvm_arch_ops->skip_emulated_instruction(vcpu);
1762 static void kernel_pio(struct kvm_io_device *pio_dev,
1763 struct kvm_vcpu *vcpu,
1766 /* TODO: String I/O for in kernel device */
1769 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1773 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1778 static void pio_string_write(struct kvm_io_device *pio_dev,
1779 struct kvm_vcpu *vcpu)
1781 struct kvm_pio_request *io = &vcpu->pio;
1782 void *pd = vcpu->pio_data;
1785 for (i = 0; i < io->cur_count; i++) {
1786 kvm_iodevice_write(pio_dev, io->port,
1793 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1794 int size, unsigned long count, int string, int down,
1795 gva_t address, int rep, unsigned port)
1797 unsigned now, in_page;
1801 struct kvm_io_device *pio_dev;
1803 vcpu->run->exit_reason = KVM_EXIT_IO;
1804 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1805 vcpu->run->io.size = size;
1806 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1807 vcpu->run->io.count = count;
1808 vcpu->run->io.port = port;
1809 vcpu->pio.count = count;
1810 vcpu->pio.cur_count = count;
1811 vcpu->pio.size = size;
1813 vcpu->pio.port = port;
1814 vcpu->pio.string = string;
1815 vcpu->pio.down = down;
1816 vcpu->pio.guest_page_offset = offset_in_page(address);
1817 vcpu->pio.rep = rep;
1819 pio_dev = vcpu_find_pio_dev(vcpu, port);
1821 kvm_arch_ops->cache_regs(vcpu);
1822 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1823 kvm_arch_ops->decache_regs(vcpu);
1825 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1833 kvm_arch_ops->skip_emulated_instruction(vcpu);
1837 now = min(count, PAGE_SIZE / size);
1840 in_page = PAGE_SIZE - offset_in_page(address);
1842 in_page = offset_in_page(address) + size;
1843 now = min(count, (unsigned long)in_page / size);
1846 * String I/O straddles page boundary. Pin two guest pages
1847 * so that we satisfy atomicity constraints. Do just one
1848 * transaction to avoid complexity.
1855 * String I/O in reverse. Yuck. Kill the guest, fix later.
1857 printk(KERN_ERR "kvm: guest string pio down\n");
1861 vcpu->run->io.count = now;
1862 vcpu->pio.cur_count = now;
1864 for (i = 0; i < nr_pages; ++i) {
1865 spin_lock(&vcpu->kvm->lock);
1866 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1869 vcpu->pio.guest_pages[i] = page;
1870 spin_unlock(&vcpu->kvm->lock);
1873 free_pio_guest_pages(vcpu);
1878 if (!vcpu->pio.in) {
1879 /* string PIO write */
1880 ret = pio_copy_data(vcpu);
1881 if (ret >= 0 && pio_dev) {
1882 pio_string_write(pio_dev, vcpu);
1884 if (vcpu->pio.count == 0)
1888 printk(KERN_ERR "no string pio read support yet, "
1889 "port %x size %d count %ld\n",
1894 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1896 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1903 if (vcpu->sigset_active)
1904 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1906 /* re-sync apic's tpr */
1907 vcpu->cr8 = kvm_run->cr8;
1909 if (vcpu->pio.cur_count) {
1910 r = complete_pio(vcpu);
1915 if (vcpu->mmio_needed) {
1916 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1917 vcpu->mmio_read_completed = 1;
1918 vcpu->mmio_needed = 0;
1919 r = emulate_instruction(vcpu, kvm_run,
1920 vcpu->mmio_fault_cr2, 0);
1921 if (r == EMULATE_DO_MMIO) {
1923 * Read-modify-write. Back to userspace.
1930 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1931 kvm_arch_ops->cache_regs(vcpu);
1932 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1933 kvm_arch_ops->decache_regs(vcpu);
1936 r = kvm_arch_ops->run(vcpu, kvm_run);
1939 if (vcpu->sigset_active)
1940 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1946 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1947 struct kvm_regs *regs)
1951 kvm_arch_ops->cache_regs(vcpu);
1953 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1954 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1955 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1956 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1957 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1958 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1959 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1960 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1961 #ifdef CONFIG_X86_64
1962 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1963 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1964 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1965 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1966 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1967 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1968 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1969 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1972 regs->rip = vcpu->rip;
1973 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1976 * Don't leak debug flags in case they were set for guest debugging
1978 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1979 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1986 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1987 struct kvm_regs *regs)
1991 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1992 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1993 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1994 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1995 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1996 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1997 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1998 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1999 #ifdef CONFIG_X86_64
2000 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2001 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2002 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2003 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2004 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2005 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2006 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2007 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2010 vcpu->rip = regs->rip;
2011 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2013 kvm_arch_ops->decache_regs(vcpu);
2020 static void get_segment(struct kvm_vcpu *vcpu,
2021 struct kvm_segment *var, int seg)
2023 return kvm_arch_ops->get_segment(vcpu, var, seg);
2026 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2027 struct kvm_sregs *sregs)
2029 struct descriptor_table dt;
2033 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2034 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2035 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2036 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2037 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2038 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2040 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2041 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2043 kvm_arch_ops->get_idt(vcpu, &dt);
2044 sregs->idt.limit = dt.limit;
2045 sregs->idt.base = dt.base;
2046 kvm_arch_ops->get_gdt(vcpu, &dt);
2047 sregs->gdt.limit = dt.limit;
2048 sregs->gdt.base = dt.base;
2050 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2051 sregs->cr0 = vcpu->cr0;
2052 sregs->cr2 = vcpu->cr2;
2053 sregs->cr3 = vcpu->cr3;
2054 sregs->cr4 = vcpu->cr4;
2055 sregs->cr8 = vcpu->cr8;
2056 sregs->efer = vcpu->shadow_efer;
2057 sregs->apic_base = vcpu->apic_base;
2059 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2060 sizeof sregs->interrupt_bitmap);
2067 static void set_segment(struct kvm_vcpu *vcpu,
2068 struct kvm_segment *var, int seg)
2070 return kvm_arch_ops->set_segment(vcpu, var, seg);
2073 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2074 struct kvm_sregs *sregs)
2076 int mmu_reset_needed = 0;
2078 struct descriptor_table dt;
2082 dt.limit = sregs->idt.limit;
2083 dt.base = sregs->idt.base;
2084 kvm_arch_ops->set_idt(vcpu, &dt);
2085 dt.limit = sregs->gdt.limit;
2086 dt.base = sregs->gdt.base;
2087 kvm_arch_ops->set_gdt(vcpu, &dt);
2089 vcpu->cr2 = sregs->cr2;
2090 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2091 vcpu->cr3 = sregs->cr3;
2093 vcpu->cr8 = sregs->cr8;
2095 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2096 #ifdef CONFIG_X86_64
2097 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2099 vcpu->apic_base = sregs->apic_base;
2101 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2103 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2104 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2106 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2107 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2108 if (!is_long_mode(vcpu) && is_pae(vcpu))
2109 load_pdptrs(vcpu, vcpu->cr3);
2111 if (mmu_reset_needed)
2112 kvm_mmu_reset_context(vcpu);
2114 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2115 sizeof vcpu->irq_pending);
2116 vcpu->irq_summary = 0;
2117 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2118 if (vcpu->irq_pending[i])
2119 __set_bit(i, &vcpu->irq_summary);
2121 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2122 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2123 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2124 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2125 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2126 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2128 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2129 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2137 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2138 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2140 * This list is modified at module load time to reflect the
2141 * capabilities of the host cpu.
2143 static u32 msrs_to_save[] = {
2144 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2146 #ifdef CONFIG_X86_64
2147 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2149 MSR_IA32_TIME_STAMP_COUNTER,
2152 static unsigned num_msrs_to_save;
2154 static u32 emulated_msrs[] = {
2155 MSR_IA32_MISC_ENABLE,
2158 static __init void kvm_init_msr_list(void)
2163 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2164 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2167 msrs_to_save[j] = msrs_to_save[i];
2170 num_msrs_to_save = j;
2174 * Adapt set_msr() to msr_io()'s calling convention
2176 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2178 return kvm_set_msr(vcpu, index, *data);
2182 * Read or write a bunch of msrs. All parameters are kernel addresses.
2184 * @return number of msrs set successfully.
2186 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2187 struct kvm_msr_entry *entries,
2188 int (*do_msr)(struct kvm_vcpu *vcpu,
2189 unsigned index, u64 *data))
2195 for (i = 0; i < msrs->nmsrs; ++i)
2196 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2205 * Read or write a bunch of msrs. Parameters are user addresses.
2207 * @return number of msrs set successfully.
2209 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2210 int (*do_msr)(struct kvm_vcpu *vcpu,
2211 unsigned index, u64 *data),
2214 struct kvm_msrs msrs;
2215 struct kvm_msr_entry *entries;
2220 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2224 if (msrs.nmsrs >= MAX_IO_MSRS)
2228 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2229 entries = vmalloc(size);
2234 if (copy_from_user(entries, user_msrs->entries, size))
2237 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2242 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2254 * Translate a guest virtual address to a guest physical address.
2256 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2257 struct kvm_translation *tr)
2259 unsigned long vaddr = tr->linear_address;
2263 spin_lock(&vcpu->kvm->lock);
2264 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2265 tr->physical_address = gpa;
2266 tr->valid = gpa != UNMAPPED_GVA;
2269 spin_unlock(&vcpu->kvm->lock);
2275 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2276 struct kvm_interrupt *irq)
2278 if (irq->irq < 0 || irq->irq >= 256)
2282 set_bit(irq->irq, vcpu->irq_pending);
2283 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2290 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2291 struct kvm_debug_guest *dbg)
2297 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2304 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2305 unsigned long address,
2308 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2309 unsigned long pgoff;
2312 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2314 page = virt_to_page(vcpu->run);
2315 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2316 page = virt_to_page(vcpu->pio_data);
2318 return NOPAGE_SIGBUS;
2321 *type = VM_FAULT_MINOR;
2326 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2327 .nopage = kvm_vcpu_nopage,
2330 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2332 vma->vm_ops = &kvm_vcpu_vm_ops;
2336 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2338 struct kvm_vcpu *vcpu = filp->private_data;
2340 fput(vcpu->kvm->filp);
2344 static struct file_operations kvm_vcpu_fops = {
2345 .release = kvm_vcpu_release,
2346 .unlocked_ioctl = kvm_vcpu_ioctl,
2347 .compat_ioctl = kvm_vcpu_ioctl,
2348 .mmap = kvm_vcpu_mmap,
2352 * Allocates an inode for the vcpu.
2354 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2357 struct inode *inode;
2360 r = anon_inode_getfd(&fd, &inode, &file,
2361 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2364 atomic_inc(&vcpu->kvm->filp->f_count);
2369 * Creates some virtual cpus. Good luck creating more than one.
2371 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2374 struct kvm_vcpu *vcpu;
2381 vcpu = &kvm->vcpus[n];
2384 mutex_lock(&vcpu->mutex);
2387 mutex_unlock(&vcpu->mutex);
2391 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2395 vcpu->run = page_address(page);
2397 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2401 vcpu->pio_data = page_address(page);
2403 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2405 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2408 r = kvm_arch_ops->vcpu_create(vcpu);
2410 goto out_free_vcpus;
2412 r = kvm_mmu_create(vcpu);
2414 goto out_free_vcpus;
2416 kvm_arch_ops->vcpu_load(vcpu);
2417 r = kvm_mmu_setup(vcpu);
2419 r = kvm_arch_ops->vcpu_setup(vcpu);
2423 goto out_free_vcpus;
2425 r = create_vcpu_fd(vcpu);
2427 goto out_free_vcpus;
2429 spin_lock(&kvm_lock);
2430 if (n >= kvm->nvcpus)
2431 kvm->nvcpus = n + 1;
2432 spin_unlock(&kvm_lock);
2439 kvm_free_vcpu(vcpu);
2441 free_page((unsigned long)vcpu->run);
2444 mutex_unlock(&vcpu->mutex);
2449 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2453 struct kvm_cpuid_entry *e, *entry;
2455 rdmsrl(MSR_EFER, efer);
2457 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2458 e = &vcpu->cpuid_entries[i];
2459 if (e->function == 0x80000001) {
2464 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2465 entry->edx &= ~(1 << 20);
2466 printk(KERN_INFO "kvm: guest NX capability removed\n");
2470 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2471 struct kvm_cpuid *cpuid,
2472 struct kvm_cpuid_entry __user *entries)
2477 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2480 if (copy_from_user(&vcpu->cpuid_entries, entries,
2481 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2483 vcpu->cpuid_nent = cpuid->nent;
2484 cpuid_fix_nx_cap(vcpu);
2491 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2494 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2495 vcpu->sigset_active = 1;
2496 vcpu->sigset = *sigset;
2498 vcpu->sigset_active = 0;
2503 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2504 * we have asm/x86/processor.h
2515 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2516 #ifdef CONFIG_X86_64
2517 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2519 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2523 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2525 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2529 memcpy(fpu->fpr, fxsave->st_space, 128);
2530 fpu->fcw = fxsave->cwd;
2531 fpu->fsw = fxsave->swd;
2532 fpu->ftwx = fxsave->twd;
2533 fpu->last_opcode = fxsave->fop;
2534 fpu->last_ip = fxsave->rip;
2535 fpu->last_dp = fxsave->rdp;
2536 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2543 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2545 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2549 memcpy(fxsave->st_space, fpu->fpr, 128);
2550 fxsave->cwd = fpu->fcw;
2551 fxsave->swd = fpu->fsw;
2552 fxsave->twd = fpu->ftwx;
2553 fxsave->fop = fpu->last_opcode;
2554 fxsave->rip = fpu->last_ip;
2555 fxsave->rdp = fpu->last_dp;
2556 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2563 static long kvm_vcpu_ioctl(struct file *filp,
2564 unsigned int ioctl, unsigned long arg)
2566 struct kvm_vcpu *vcpu = filp->private_data;
2567 void __user *argp = (void __user *)arg;
2575 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2577 case KVM_GET_REGS: {
2578 struct kvm_regs kvm_regs;
2580 memset(&kvm_regs, 0, sizeof kvm_regs);
2581 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2585 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2590 case KVM_SET_REGS: {
2591 struct kvm_regs kvm_regs;
2594 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2596 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2602 case KVM_GET_SREGS: {
2603 struct kvm_sregs kvm_sregs;
2605 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2606 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2610 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2615 case KVM_SET_SREGS: {
2616 struct kvm_sregs kvm_sregs;
2619 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2621 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2627 case KVM_TRANSLATE: {
2628 struct kvm_translation tr;
2631 if (copy_from_user(&tr, argp, sizeof tr))
2633 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2637 if (copy_to_user(argp, &tr, sizeof tr))
2642 case KVM_INTERRUPT: {
2643 struct kvm_interrupt irq;
2646 if (copy_from_user(&irq, argp, sizeof irq))
2648 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2654 case KVM_DEBUG_GUEST: {
2655 struct kvm_debug_guest dbg;
2658 if (copy_from_user(&dbg, argp, sizeof dbg))
2660 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2667 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2670 r = msr_io(vcpu, argp, do_set_msr, 0);
2672 case KVM_SET_CPUID: {
2673 struct kvm_cpuid __user *cpuid_arg = argp;
2674 struct kvm_cpuid cpuid;
2677 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2679 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2684 case KVM_SET_SIGNAL_MASK: {
2685 struct kvm_signal_mask __user *sigmask_arg = argp;
2686 struct kvm_signal_mask kvm_sigmask;
2687 sigset_t sigset, *p;
2692 if (copy_from_user(&kvm_sigmask, argp,
2693 sizeof kvm_sigmask))
2696 if (kvm_sigmask.len != sizeof sigset)
2699 if (copy_from_user(&sigset, sigmask_arg->sigset,
2704 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2710 memset(&fpu, 0, sizeof fpu);
2711 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2715 if (copy_to_user(argp, &fpu, sizeof fpu))
2724 if (copy_from_user(&fpu, argp, sizeof fpu))
2726 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2739 static long kvm_vm_ioctl(struct file *filp,
2740 unsigned int ioctl, unsigned long arg)
2742 struct kvm *kvm = filp->private_data;
2743 void __user *argp = (void __user *)arg;
2747 case KVM_CREATE_VCPU:
2748 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2752 case KVM_SET_MEMORY_REGION: {
2753 struct kvm_memory_region kvm_mem;
2756 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2758 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2763 case KVM_GET_DIRTY_LOG: {
2764 struct kvm_dirty_log log;
2767 if (copy_from_user(&log, argp, sizeof log))
2769 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2774 case KVM_SET_MEMORY_ALIAS: {
2775 struct kvm_memory_alias alias;
2778 if (copy_from_user(&alias, argp, sizeof alias))
2780 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2792 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2793 unsigned long address,
2796 struct kvm *kvm = vma->vm_file->private_data;
2797 unsigned long pgoff;
2800 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2801 page = gfn_to_page(kvm, pgoff);
2803 return NOPAGE_SIGBUS;
2806 *type = VM_FAULT_MINOR;
2811 static struct vm_operations_struct kvm_vm_vm_ops = {
2812 .nopage = kvm_vm_nopage,
2815 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2817 vma->vm_ops = &kvm_vm_vm_ops;
2821 static struct file_operations kvm_vm_fops = {
2822 .release = kvm_vm_release,
2823 .unlocked_ioctl = kvm_vm_ioctl,
2824 .compat_ioctl = kvm_vm_ioctl,
2825 .mmap = kvm_vm_mmap,
2828 static int kvm_dev_ioctl_create_vm(void)
2831 struct inode *inode;
2835 kvm = kvm_create_vm();
2837 return PTR_ERR(kvm);
2838 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2840 kvm_destroy_vm(kvm);
2849 static long kvm_dev_ioctl(struct file *filp,
2850 unsigned int ioctl, unsigned long arg)
2852 void __user *argp = (void __user *)arg;
2856 case KVM_GET_API_VERSION:
2860 r = KVM_API_VERSION;
2866 r = kvm_dev_ioctl_create_vm();
2868 case KVM_GET_MSR_INDEX_LIST: {
2869 struct kvm_msr_list __user *user_msr_list = argp;
2870 struct kvm_msr_list msr_list;
2874 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2877 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2878 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2881 if (n < num_msrs_to_save)
2884 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2885 num_msrs_to_save * sizeof(u32)))
2887 if (copy_to_user(user_msr_list->indices
2888 + num_msrs_to_save * sizeof(u32),
2890 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2895 case KVM_CHECK_EXTENSION:
2897 * No extensions defined at present.
2901 case KVM_GET_VCPU_MMAP_SIZE:
2914 static struct file_operations kvm_chardev_ops = {
2915 .open = kvm_dev_open,
2916 .release = kvm_dev_release,
2917 .unlocked_ioctl = kvm_dev_ioctl,
2918 .compat_ioctl = kvm_dev_ioctl,
2921 static struct miscdevice kvm_dev = {
2928 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2931 static void decache_vcpus_on_cpu(int cpu)
2934 struct kvm_vcpu *vcpu;
2937 spin_lock(&kvm_lock);
2938 list_for_each_entry(vm, &vm_list, vm_list)
2939 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2940 vcpu = &vm->vcpus[i];
2942 * If the vcpu is locked, then it is running on some
2943 * other cpu and therefore it is not cached on the
2946 * If it's not locked, check the last cpu it executed
2949 if (mutex_trylock(&vcpu->mutex)) {
2950 if (vcpu->cpu == cpu) {
2951 kvm_arch_ops->vcpu_decache(vcpu);
2954 mutex_unlock(&vcpu->mutex);
2957 spin_unlock(&kvm_lock);
2960 static void hardware_enable(void *junk)
2962 int cpu = raw_smp_processor_id();
2964 if (cpu_isset(cpu, cpus_hardware_enabled))
2966 cpu_set(cpu, cpus_hardware_enabled);
2967 kvm_arch_ops->hardware_enable(NULL);
2970 static void hardware_disable(void *junk)
2972 int cpu = raw_smp_processor_id();
2974 if (!cpu_isset(cpu, cpus_hardware_enabled))
2976 cpu_clear(cpu, cpus_hardware_enabled);
2977 decache_vcpus_on_cpu(cpu);
2978 kvm_arch_ops->hardware_disable(NULL);
2981 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2988 case CPU_DYING_FROZEN:
2989 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2991 hardware_disable(NULL);
2993 case CPU_UP_CANCELED:
2994 case CPU_UP_CANCELED_FROZEN:
2995 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2997 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3000 case CPU_ONLINE_FROZEN:
3001 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3003 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3009 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3012 if (val == SYS_RESTART) {
3014 * Some (well, at least mine) BIOSes hang on reboot if
3017 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3018 on_each_cpu(hardware_disable, NULL, 0, 1);
3023 static struct notifier_block kvm_reboot_notifier = {
3024 .notifier_call = kvm_reboot,
3028 void kvm_io_bus_init(struct kvm_io_bus *bus)
3030 memset(bus, 0, sizeof(*bus));
3033 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3037 for (i = 0; i < bus->dev_count; i++) {
3038 struct kvm_io_device *pos = bus->devs[i];
3040 kvm_iodevice_destructor(pos);
3044 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3048 for (i = 0; i < bus->dev_count; i++) {
3049 struct kvm_io_device *pos = bus->devs[i];
3051 if (pos->in_range(pos, addr))
3058 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3060 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3062 bus->devs[bus->dev_count++] = dev;
3065 static struct notifier_block kvm_cpu_notifier = {
3066 .notifier_call = kvm_cpu_hotplug,
3067 .priority = 20, /* must be > scheduler priority */
3070 static u64 stat_get(void *_offset)
3072 unsigned offset = (long)_offset;
3075 struct kvm_vcpu *vcpu;
3078 spin_lock(&kvm_lock);
3079 list_for_each_entry(kvm, &vm_list, vm_list)
3080 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3081 vcpu = &kvm->vcpus[i];
3082 total += *(u32 *)((void *)vcpu + offset);
3084 spin_unlock(&kvm_lock);
3088 static void stat_set(void *offset, u64 val)
3092 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3094 static __init void kvm_init_debug(void)
3096 struct kvm_stats_debugfs_item *p;
3098 debugfs_dir = debugfs_create_dir("kvm", NULL);
3099 for (p = debugfs_entries; p->name; ++p)
3100 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3101 (void *)(long)p->offset,
3105 static void kvm_exit_debug(void)
3107 struct kvm_stats_debugfs_item *p;
3109 for (p = debugfs_entries; p->name; ++p)
3110 debugfs_remove(p->dentry);
3111 debugfs_remove(debugfs_dir);
3114 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3116 hardware_disable(NULL);
3120 static int kvm_resume(struct sys_device *dev)
3122 hardware_enable(NULL);
3126 static struct sysdev_class kvm_sysdev_class = {
3127 set_kset_name("kvm"),
3128 .suspend = kvm_suspend,
3129 .resume = kvm_resume,
3132 static struct sys_device kvm_sysdev = {
3134 .cls = &kvm_sysdev_class,
3137 hpa_t bad_page_address;
3139 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3144 printk(KERN_ERR "kvm: already loaded the other module\n");
3148 if (!ops->cpu_has_kvm_support()) {
3149 printk(KERN_ERR "kvm: no hardware support\n");
3152 if (ops->disabled_by_bios()) {
3153 printk(KERN_ERR "kvm: disabled by bios\n");
3159 r = kvm_arch_ops->hardware_setup();
3163 on_each_cpu(hardware_enable, NULL, 0, 1);
3164 r = register_cpu_notifier(&kvm_cpu_notifier);
3167 register_reboot_notifier(&kvm_reboot_notifier);
3169 r = sysdev_class_register(&kvm_sysdev_class);
3173 r = sysdev_register(&kvm_sysdev);
3177 kvm_chardev_ops.owner = module;
3179 r = misc_register(&kvm_dev);
3181 printk (KERN_ERR "kvm: misc device register failed\n");
3188 sysdev_unregister(&kvm_sysdev);
3190 sysdev_class_unregister(&kvm_sysdev_class);
3192 unregister_reboot_notifier(&kvm_reboot_notifier);
3193 unregister_cpu_notifier(&kvm_cpu_notifier);
3195 on_each_cpu(hardware_disable, NULL, 0, 1);
3196 kvm_arch_ops->hardware_unsetup();
3198 kvm_arch_ops = NULL;
3202 void kvm_exit_arch(void)
3204 misc_deregister(&kvm_dev);
3205 sysdev_unregister(&kvm_sysdev);
3206 sysdev_class_unregister(&kvm_sysdev_class);
3207 unregister_reboot_notifier(&kvm_reboot_notifier);
3208 unregister_cpu_notifier(&kvm_cpu_notifier);
3209 on_each_cpu(hardware_disable, NULL, 0, 1);
3210 kvm_arch_ops->hardware_unsetup();
3211 kvm_arch_ops = NULL;
3214 static __init int kvm_init(void)
3216 static struct page *bad_page;
3219 r = kvm_mmu_module_init();
3225 kvm_init_msr_list();
3227 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3232 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3233 memset(__va(bad_page_address), 0, PAGE_SIZE);
3239 kvm_mmu_module_exit();
3244 static __exit void kvm_exit(void)
3247 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3248 kvm_mmu_module_exit();
3251 module_init(kvm_init)
3252 module_exit(kvm_exit)
3254 EXPORT_SYMBOL_GPL(kvm_init_arch);
3255 EXPORT_SYMBOL_GPL(kvm_exit_arch);