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"
23 #include <linux/kvm.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/percpu.h>
27 #include <linux/gfp.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <linux/reboot.h>
32 #include <linux/debugfs.h>
33 #include <linux/highmem.h>
34 #include <linux/file.h>
35 #include <linux/sysdev.h>
36 #include <linux/cpu.h>
37 #include <linux/sched.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
42 #include <asm/processor.h>
45 #include <asm/uaccess.h>
48 MODULE_AUTHOR("Qumranet");
49 MODULE_LICENSE("GPL");
51 static DEFINE_SPINLOCK(kvm_lock);
52 static LIST_HEAD(vm_list);
54 static cpumask_t cpus_hardware_enabled;
56 struct kvm_arch_ops *kvm_arch_ops;
57 struct kmem_cache *kvm_vcpu_cache;
58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
60 static __read_mostly struct preempt_ops kvm_preempt_ops;
62 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
64 static struct kvm_stats_debugfs_item {
67 struct dentry *dentry;
68 } debugfs_entries[] = {
69 { "pf_fixed", STAT_OFFSET(pf_fixed) },
70 { "pf_guest", STAT_OFFSET(pf_guest) },
71 { "tlb_flush", STAT_OFFSET(tlb_flush) },
72 { "invlpg", STAT_OFFSET(invlpg) },
73 { "exits", STAT_OFFSET(exits) },
74 { "io_exits", STAT_OFFSET(io_exits) },
75 { "mmio_exits", STAT_OFFSET(mmio_exits) },
76 { "signal_exits", STAT_OFFSET(signal_exits) },
77 { "irq_window", STAT_OFFSET(irq_window_exits) },
78 { "halt_exits", STAT_OFFSET(halt_exits) },
79 { "request_irq", STAT_OFFSET(request_irq_exits) },
80 { "irq_exits", STAT_OFFSET(irq_exits) },
81 { "light_exits", STAT_OFFSET(light_exits) },
82 { "efer_reload", STAT_OFFSET(efer_reload) },
86 static struct dentry *debugfs_dir;
88 #define MAX_IO_MSRS 256
90 #define CR0_RESERVED_BITS \
91 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
92 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
93 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
94 #define CR4_RESERVED_BITS \
95 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
96 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
97 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
98 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
100 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
101 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
104 // LDT or TSS descriptor in the GDT. 16 bytes.
105 struct segment_descriptor_64 {
106 struct segment_descriptor s;
113 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116 unsigned long segment_base(u16 selector)
118 struct descriptor_table gdt;
119 struct segment_descriptor *d;
120 unsigned long table_base;
121 typedef unsigned long ul;
127 asm ("sgdt %0" : "=m"(gdt));
128 table_base = gdt.base;
130 if (selector & 4) { /* from ldt */
133 asm ("sldt %0" : "=g"(ldt_selector));
134 table_base = segment_base(ldt_selector);
136 d = (struct segment_descriptor *)(table_base + (selector & ~7));
137 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
140 && (d->type == 2 || d->type == 9 || d->type == 11))
141 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
145 EXPORT_SYMBOL_GPL(segment_base);
147 static inline int valid_vcpu(int n)
149 return likely(n >= 0 && n < KVM_MAX_VCPUS);
152 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
154 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
157 vcpu->guest_fpu_loaded = 1;
158 fx_save(&vcpu->host_fx_image);
159 fx_restore(&vcpu->guest_fx_image);
161 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
163 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
165 if (!vcpu->guest_fpu_loaded)
168 vcpu->guest_fpu_loaded = 0;
169 fx_save(&vcpu->guest_fx_image);
170 fx_restore(&vcpu->host_fx_image);
172 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
175 * Switches to specified vcpu, until a matching vcpu_put()
177 static void vcpu_load(struct kvm_vcpu *vcpu)
181 mutex_lock(&vcpu->mutex);
183 preempt_notifier_register(&vcpu->preempt_notifier);
184 kvm_arch_ops->vcpu_load(vcpu, cpu);
188 static void vcpu_put(struct kvm_vcpu *vcpu)
191 kvm_arch_ops->vcpu_put(vcpu);
192 preempt_notifier_unregister(&vcpu->preempt_notifier);
194 mutex_unlock(&vcpu->mutex);
197 static void ack_flush(void *_completed)
199 atomic_t *completed = _completed;
201 atomic_inc(completed);
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
208 struct kvm_vcpu *vcpu;
211 atomic_set(&completed, 0);
214 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
215 vcpu = kvm->vcpus[i];
218 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
221 if (cpu != -1 && cpu != raw_smp_processor_id())
222 if (!cpu_isset(cpu, cpus)) {
229 * We really want smp_call_function_mask() here. But that's not
230 * available, so ipi all cpus in parallel and wait for them
233 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
234 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
235 while (atomic_read(&completed) != needed) {
241 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
246 mutex_init(&vcpu->mutex);
248 vcpu->mmu.root_hpa = INVALID_PAGE;
252 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
257 vcpu->run = page_address(page);
259 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
264 vcpu->pio_data = page_address(page);
266 r = kvm_mmu_create(vcpu);
268 goto fail_free_pio_data;
273 free_page((unsigned long)vcpu->pio_data);
275 free_page((unsigned long)vcpu->run);
279 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
281 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
283 kvm_mmu_destroy(vcpu);
284 free_page((unsigned long)vcpu->pio_data);
285 free_page((unsigned long)vcpu->run);
287 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
289 static struct kvm *kvm_create_vm(void)
291 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
294 return ERR_PTR(-ENOMEM);
296 kvm_io_bus_init(&kvm->pio_bus);
297 mutex_init(&kvm->lock);
298 INIT_LIST_HEAD(&kvm->active_mmu_pages);
299 kvm_io_bus_init(&kvm->mmio_bus);
300 spin_lock(&kvm_lock);
301 list_add(&kvm->vm_list, &vm_list);
302 spin_unlock(&kvm_lock);
307 * Free any memory in @free but not in @dont.
309 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
310 struct kvm_memory_slot *dont)
314 if (!dont || free->phys_mem != dont->phys_mem)
315 if (free->phys_mem) {
316 for (i = 0; i < free->npages; ++i)
317 if (free->phys_mem[i])
318 __free_page(free->phys_mem[i]);
319 vfree(free->phys_mem);
322 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
323 vfree(free->dirty_bitmap);
325 free->phys_mem = NULL;
327 free->dirty_bitmap = NULL;
330 static void kvm_free_physmem(struct kvm *kvm)
334 for (i = 0; i < kvm->nmemslots; ++i)
335 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
338 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
342 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
343 if (vcpu->pio.guest_pages[i]) {
344 __free_page(vcpu->pio.guest_pages[i]);
345 vcpu->pio.guest_pages[i] = NULL;
349 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
352 kvm_mmu_unload(vcpu);
356 static void kvm_free_vcpus(struct kvm *kvm)
361 * Unpin any mmu pages first.
363 for (i = 0; i < KVM_MAX_VCPUS; ++i)
365 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
366 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
368 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
369 kvm->vcpus[i] = NULL;
375 static void kvm_destroy_vm(struct kvm *kvm)
377 spin_lock(&kvm_lock);
378 list_del(&kvm->vm_list);
379 spin_unlock(&kvm_lock);
380 kvm_io_bus_destroy(&kvm->pio_bus);
381 kvm_io_bus_destroy(&kvm->mmio_bus);
384 kvm_free_physmem(kvm);
388 static int kvm_vm_release(struct inode *inode, struct file *filp)
390 struct kvm *kvm = filp->private_data;
396 static void inject_gp(struct kvm_vcpu *vcpu)
398 kvm_arch_ops->inject_gp(vcpu, 0);
402 * Load the pae pdptrs. Return true is they are all valid.
404 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
406 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
407 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
412 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
414 mutex_lock(&vcpu->kvm->lock);
415 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
421 pdpt = kmap_atomic(page, KM_USER0);
422 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
423 kunmap_atomic(pdpt, KM_USER0);
425 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
426 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
433 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
435 mutex_unlock(&vcpu->kvm->lock);
440 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
442 if (cr0 & CR0_RESERVED_BITS) {
443 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
449 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
450 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
455 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
456 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
457 "and a clear PE flag\n");
462 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
464 if ((vcpu->shadow_efer & EFER_LME)) {
468 printk(KERN_DEBUG "set_cr0: #GP, start paging "
469 "in long mode while PAE is disabled\n");
473 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
475 printk(KERN_DEBUG "set_cr0: #GP, start paging "
476 "in long mode while CS.L == 1\n");
483 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
484 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
492 kvm_arch_ops->set_cr0(vcpu, cr0);
495 mutex_lock(&vcpu->kvm->lock);
496 kvm_mmu_reset_context(vcpu);
497 mutex_unlock(&vcpu->kvm->lock);
500 EXPORT_SYMBOL_GPL(set_cr0);
502 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
504 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
506 EXPORT_SYMBOL_GPL(lmsw);
508 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
510 if (cr4 & CR4_RESERVED_BITS) {
511 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
516 if (is_long_mode(vcpu)) {
517 if (!(cr4 & X86_CR4_PAE)) {
518 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
523 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
524 && !load_pdptrs(vcpu, vcpu->cr3)) {
525 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
530 if (cr4 & X86_CR4_VMXE) {
531 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
535 kvm_arch_ops->set_cr4(vcpu, cr4);
536 mutex_lock(&vcpu->kvm->lock);
537 kvm_mmu_reset_context(vcpu);
538 mutex_unlock(&vcpu->kvm->lock);
540 EXPORT_SYMBOL_GPL(set_cr4);
542 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
544 if (is_long_mode(vcpu)) {
545 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
546 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
552 if (cr3 & CR3_PAE_RESERVED_BITS) {
554 "set_cr3: #GP, reserved bits\n");
558 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
559 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
565 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
567 "set_cr3: #GP, reserved bits\n");
574 mutex_lock(&vcpu->kvm->lock);
576 * Does the new cr3 value map to physical memory? (Note, we
577 * catch an invalid cr3 even in real-mode, because it would
578 * cause trouble later on when we turn on paging anyway.)
580 * A real CPU would silently accept an invalid cr3 and would
581 * attempt to use it - with largely undefined (and often hard
582 * to debug) behavior on the guest side.
584 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
588 vcpu->mmu.new_cr3(vcpu);
590 mutex_unlock(&vcpu->kvm->lock);
592 EXPORT_SYMBOL_GPL(set_cr3);
594 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
596 if (cr8 & CR8_RESERVED_BITS) {
597 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
603 EXPORT_SYMBOL_GPL(set_cr8);
605 void fx_init(struct kvm_vcpu *vcpu)
607 unsigned after_mxcsr_mask;
609 /* Initialize guest FPU by resetting ours and saving into guest's */
611 fx_save(&vcpu->host_fx_image);
613 fx_save(&vcpu->guest_fx_image);
614 fx_restore(&vcpu->host_fx_image);
617 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
618 vcpu->guest_fx_image.mxcsr = 0x1f80;
619 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
620 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
622 EXPORT_SYMBOL_GPL(fx_init);
625 * Allocate some memory and give it an address in the guest physical address
628 * Discontiguous memory is allowed, mostly for framebuffers.
630 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
631 struct kvm_memory_region *mem)
635 unsigned long npages;
637 struct kvm_memory_slot *memslot;
638 struct kvm_memory_slot old, new;
639 int memory_config_version;
642 /* General sanity checks */
643 if (mem->memory_size & (PAGE_SIZE - 1))
645 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
647 if (mem->slot >= KVM_MEMORY_SLOTS)
649 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
652 memslot = &kvm->memslots[mem->slot];
653 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
654 npages = mem->memory_size >> PAGE_SHIFT;
657 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
660 mutex_lock(&kvm->lock);
662 memory_config_version = kvm->memory_config_version;
663 new = old = *memslot;
665 new.base_gfn = base_gfn;
667 new.flags = mem->flags;
669 /* Disallow changing a memory slot's size. */
671 if (npages && old.npages && npages != old.npages)
674 /* Check for overlaps */
676 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
677 struct kvm_memory_slot *s = &kvm->memslots[i];
681 if (!((base_gfn + npages <= s->base_gfn) ||
682 (base_gfn >= s->base_gfn + s->npages)))
686 * Do memory allocations outside lock. memory_config_version will
689 mutex_unlock(&kvm->lock);
691 /* Deallocate if slot is being removed */
695 /* Free page dirty bitmap if unneeded */
696 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
697 new.dirty_bitmap = NULL;
701 /* Allocate if a slot is being created */
702 if (npages && !new.phys_mem) {
703 new.phys_mem = vmalloc(npages * sizeof(struct page *));
708 memset(new.phys_mem, 0, npages * sizeof(struct page *));
709 for (i = 0; i < npages; ++i) {
710 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
712 if (!new.phys_mem[i])
714 set_page_private(new.phys_mem[i],0);
718 /* Allocate page dirty bitmap if needed */
719 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
720 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
722 new.dirty_bitmap = vmalloc(dirty_bytes);
723 if (!new.dirty_bitmap)
725 memset(new.dirty_bitmap, 0, dirty_bytes);
728 mutex_lock(&kvm->lock);
730 if (memory_config_version != kvm->memory_config_version) {
731 mutex_unlock(&kvm->lock);
732 kvm_free_physmem_slot(&new, &old);
740 if (mem->slot >= kvm->nmemslots)
741 kvm->nmemslots = mem->slot + 1;
744 ++kvm->memory_config_version;
746 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
747 kvm_flush_remote_tlbs(kvm);
749 mutex_unlock(&kvm->lock);
751 kvm_free_physmem_slot(&old, &new);
755 mutex_unlock(&kvm->lock);
757 kvm_free_physmem_slot(&new, &old);
763 * Get (and clear) the dirty memory log for a memory slot.
765 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
766 struct kvm_dirty_log *log)
768 struct kvm_memory_slot *memslot;
771 unsigned long any = 0;
773 mutex_lock(&kvm->lock);
776 * Prevent changes to guest memory configuration even while the lock
780 mutex_unlock(&kvm->lock);
782 if (log->slot >= KVM_MEMORY_SLOTS)
785 memslot = &kvm->memslots[log->slot];
787 if (!memslot->dirty_bitmap)
790 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
792 for (i = 0; !any && i < n/sizeof(long); ++i)
793 any = memslot->dirty_bitmap[i];
796 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
799 /* If nothing is dirty, don't bother messing with page tables. */
801 mutex_lock(&kvm->lock);
802 kvm_mmu_slot_remove_write_access(kvm, log->slot);
803 kvm_flush_remote_tlbs(kvm);
804 memset(memslot->dirty_bitmap, 0, n);
805 mutex_unlock(&kvm->lock);
811 mutex_lock(&kvm->lock);
813 mutex_unlock(&kvm->lock);
818 * Set a new alias region. Aliases map a portion of physical memory into
819 * another portion. This is useful for memory windows, for example the PC
822 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
823 struct kvm_memory_alias *alias)
826 struct kvm_mem_alias *p;
829 /* General sanity checks */
830 if (alias->memory_size & (PAGE_SIZE - 1))
832 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
834 if (alias->slot >= KVM_ALIAS_SLOTS)
836 if (alias->guest_phys_addr + alias->memory_size
837 < alias->guest_phys_addr)
839 if (alias->target_phys_addr + alias->memory_size
840 < alias->target_phys_addr)
843 mutex_lock(&kvm->lock);
845 p = &kvm->aliases[alias->slot];
846 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
847 p->npages = alias->memory_size >> PAGE_SHIFT;
848 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
850 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
851 if (kvm->aliases[n - 1].npages)
855 kvm_mmu_zap_all(kvm);
857 mutex_unlock(&kvm->lock);
865 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
868 struct kvm_mem_alias *alias;
870 for (i = 0; i < kvm->naliases; ++i) {
871 alias = &kvm->aliases[i];
872 if (gfn >= alias->base_gfn
873 && gfn < alias->base_gfn + alias->npages)
874 return alias->target_gfn + gfn - alias->base_gfn;
879 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
883 for (i = 0; i < kvm->nmemslots; ++i) {
884 struct kvm_memory_slot *memslot = &kvm->memslots[i];
886 if (gfn >= memslot->base_gfn
887 && gfn < memslot->base_gfn + memslot->npages)
893 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
895 gfn = unalias_gfn(kvm, gfn);
896 return __gfn_to_memslot(kvm, gfn);
899 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
901 struct kvm_memory_slot *slot;
903 gfn = unalias_gfn(kvm, gfn);
904 slot = __gfn_to_memslot(kvm, gfn);
907 return slot->phys_mem[gfn - slot->base_gfn];
909 EXPORT_SYMBOL_GPL(gfn_to_page);
911 /* WARNING: Does not work on aliased pages. */
912 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
914 struct kvm_memory_slot *memslot;
916 memslot = __gfn_to_memslot(kvm, gfn);
917 if (memslot && memslot->dirty_bitmap) {
918 unsigned long rel_gfn = gfn - memslot->base_gfn;
921 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
922 set_bit(rel_gfn, memslot->dirty_bitmap);
926 int emulator_read_std(unsigned long addr,
929 struct kvm_vcpu *vcpu)
934 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
935 unsigned offset = addr & (PAGE_SIZE-1);
936 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
941 if (gpa == UNMAPPED_GVA)
942 return X86EMUL_PROPAGATE_FAULT;
943 pfn = gpa >> PAGE_SHIFT;
944 page = gfn_to_page(vcpu->kvm, pfn);
946 return X86EMUL_UNHANDLEABLE;
947 page_virt = kmap_atomic(page, KM_USER0);
949 memcpy(data, page_virt + offset, tocopy);
951 kunmap_atomic(page_virt, KM_USER0);
958 return X86EMUL_CONTINUE;
960 EXPORT_SYMBOL_GPL(emulator_read_std);
962 static int emulator_write_std(unsigned long addr,
965 struct kvm_vcpu *vcpu)
967 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
968 return X86EMUL_UNHANDLEABLE;
971 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
975 * Note that its important to have this wrapper function because
976 * in the very near future we will be checking for MMIOs against
977 * the LAPIC as well as the general MMIO bus
979 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
982 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
985 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
988 static int emulator_read_emulated(unsigned long addr,
991 struct kvm_vcpu *vcpu)
993 struct kvm_io_device *mmio_dev;
996 if (vcpu->mmio_read_completed) {
997 memcpy(val, vcpu->mmio_data, bytes);
998 vcpu->mmio_read_completed = 0;
999 return X86EMUL_CONTINUE;
1000 } else if (emulator_read_std(addr, val, bytes, vcpu)
1001 == X86EMUL_CONTINUE)
1002 return X86EMUL_CONTINUE;
1004 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1005 if (gpa == UNMAPPED_GVA)
1006 return X86EMUL_PROPAGATE_FAULT;
1009 * Is this MMIO handled locally?
1011 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1013 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1014 return X86EMUL_CONTINUE;
1017 vcpu->mmio_needed = 1;
1018 vcpu->mmio_phys_addr = gpa;
1019 vcpu->mmio_size = bytes;
1020 vcpu->mmio_is_write = 0;
1022 return X86EMUL_UNHANDLEABLE;
1025 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1026 const void *val, int bytes)
1031 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1033 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1036 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1037 virt = kmap_atomic(page, KM_USER0);
1038 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1039 memcpy(virt + offset_in_page(gpa), val, bytes);
1040 kunmap_atomic(virt, KM_USER0);
1044 static int emulator_write_emulated_onepage(unsigned long addr,
1047 struct kvm_vcpu *vcpu)
1049 struct kvm_io_device *mmio_dev;
1050 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1052 if (gpa == UNMAPPED_GVA) {
1053 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1054 return X86EMUL_PROPAGATE_FAULT;
1057 if (emulator_write_phys(vcpu, gpa, val, bytes))
1058 return X86EMUL_CONTINUE;
1061 * Is this MMIO handled locally?
1063 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1065 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1066 return X86EMUL_CONTINUE;
1069 vcpu->mmio_needed = 1;
1070 vcpu->mmio_phys_addr = gpa;
1071 vcpu->mmio_size = bytes;
1072 vcpu->mmio_is_write = 1;
1073 memcpy(vcpu->mmio_data, val, bytes);
1075 return X86EMUL_CONTINUE;
1078 int emulator_write_emulated(unsigned long addr,
1081 struct kvm_vcpu *vcpu)
1083 /* Crossing a page boundary? */
1084 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1087 now = -addr & ~PAGE_MASK;
1088 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1089 if (rc != X86EMUL_CONTINUE)
1095 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1097 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1099 static int emulator_cmpxchg_emulated(unsigned long addr,
1103 struct kvm_vcpu *vcpu)
1105 static int reported;
1109 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1111 return emulator_write_emulated(addr, new, bytes, vcpu);
1114 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1116 return kvm_arch_ops->get_segment_base(vcpu, seg);
1119 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1121 return X86EMUL_CONTINUE;
1124 int emulate_clts(struct kvm_vcpu *vcpu)
1128 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1129 kvm_arch_ops->set_cr0(vcpu, cr0);
1130 return X86EMUL_CONTINUE;
1133 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1135 struct kvm_vcpu *vcpu = ctxt->vcpu;
1139 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1140 return X86EMUL_CONTINUE;
1142 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1143 return X86EMUL_UNHANDLEABLE;
1147 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1149 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1152 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1154 /* FIXME: better handling */
1155 return X86EMUL_UNHANDLEABLE;
1157 return X86EMUL_CONTINUE;
1160 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1162 static int reported;
1164 unsigned long rip = ctxt->vcpu->rip;
1165 unsigned long rip_linear;
1167 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1172 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1174 printk(KERN_ERR "emulation failed but !mmio_needed?"
1175 " rip %lx %02x %02x %02x %02x\n",
1176 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1180 struct x86_emulate_ops emulate_ops = {
1181 .read_std = emulator_read_std,
1182 .write_std = emulator_write_std,
1183 .read_emulated = emulator_read_emulated,
1184 .write_emulated = emulator_write_emulated,
1185 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1188 int emulate_instruction(struct kvm_vcpu *vcpu,
1189 struct kvm_run *run,
1193 struct x86_emulate_ctxt emulate_ctxt;
1197 vcpu->mmio_fault_cr2 = cr2;
1198 kvm_arch_ops->cache_regs(vcpu);
1200 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1202 emulate_ctxt.vcpu = vcpu;
1203 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1204 emulate_ctxt.cr2 = cr2;
1205 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1206 ? X86EMUL_MODE_REAL : cs_l
1207 ? X86EMUL_MODE_PROT64 : cs_db
1208 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1210 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1211 emulate_ctxt.cs_base = 0;
1212 emulate_ctxt.ds_base = 0;
1213 emulate_ctxt.es_base = 0;
1214 emulate_ctxt.ss_base = 0;
1216 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1217 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1218 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1219 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1222 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1223 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1225 vcpu->mmio_is_write = 0;
1226 vcpu->pio.string = 0;
1227 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1228 if (vcpu->pio.string)
1229 return EMULATE_DO_MMIO;
1231 if ((r || vcpu->mmio_is_write) && run) {
1232 run->exit_reason = KVM_EXIT_MMIO;
1233 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1234 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1235 run->mmio.len = vcpu->mmio_size;
1236 run->mmio.is_write = vcpu->mmio_is_write;
1240 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1241 return EMULATE_DONE;
1242 if (!vcpu->mmio_needed) {
1243 report_emulation_failure(&emulate_ctxt);
1244 return EMULATE_FAIL;
1246 return EMULATE_DO_MMIO;
1249 kvm_arch_ops->decache_regs(vcpu);
1250 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1252 if (vcpu->mmio_is_write) {
1253 vcpu->mmio_needed = 0;
1254 return EMULATE_DO_MMIO;
1257 return EMULATE_DONE;
1259 EXPORT_SYMBOL_GPL(emulate_instruction);
1261 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1263 if (vcpu->irq_summary ||
1264 (irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu)))
1267 vcpu->run->exit_reason = KVM_EXIT_HLT;
1268 ++vcpu->stat.halt_exits;
1271 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1273 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1275 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1277 kvm_arch_ops->cache_regs(vcpu);
1279 #ifdef CONFIG_X86_64
1280 if (is_long_mode(vcpu)) {
1281 nr = vcpu->regs[VCPU_REGS_RAX];
1282 a0 = vcpu->regs[VCPU_REGS_RDI];
1283 a1 = vcpu->regs[VCPU_REGS_RSI];
1284 a2 = vcpu->regs[VCPU_REGS_RDX];
1285 a3 = vcpu->regs[VCPU_REGS_RCX];
1286 a4 = vcpu->regs[VCPU_REGS_R8];
1287 a5 = vcpu->regs[VCPU_REGS_R9];
1291 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1292 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1293 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1294 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1295 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1296 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1297 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1301 run->hypercall.nr = nr;
1302 run->hypercall.args[0] = a0;
1303 run->hypercall.args[1] = a1;
1304 run->hypercall.args[2] = a2;
1305 run->hypercall.args[3] = a3;
1306 run->hypercall.args[4] = a4;
1307 run->hypercall.args[5] = a5;
1308 run->hypercall.ret = ret;
1309 run->hypercall.longmode = is_long_mode(vcpu);
1310 kvm_arch_ops->decache_regs(vcpu);
1313 vcpu->regs[VCPU_REGS_RAX] = ret;
1314 kvm_arch_ops->decache_regs(vcpu);
1317 EXPORT_SYMBOL_GPL(kvm_hypercall);
1319 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1321 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1324 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1326 struct descriptor_table dt = { limit, base };
1328 kvm_arch_ops->set_gdt(vcpu, &dt);
1331 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1333 struct descriptor_table dt = { limit, base };
1335 kvm_arch_ops->set_idt(vcpu, &dt);
1338 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1339 unsigned long *rflags)
1342 *rflags = kvm_arch_ops->get_rflags(vcpu);
1345 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1347 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1358 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1363 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1364 unsigned long *rflags)
1368 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1369 *rflags = kvm_arch_ops->get_rflags(vcpu);
1378 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1381 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1386 * Register the para guest with the host:
1388 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1390 struct kvm_vcpu_para_state *para_state;
1391 hpa_t para_state_hpa, hypercall_hpa;
1392 struct page *para_state_page;
1393 unsigned char *hypercall;
1394 gpa_t hypercall_gpa;
1396 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1397 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1400 * Needs to be page aligned:
1402 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1405 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1406 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1407 if (is_error_hpa(para_state_hpa))
1410 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1411 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1412 para_state = kmap(para_state_page);
1414 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1415 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1417 para_state->host_version = KVM_PARA_API_VERSION;
1419 * We cannot support guests that try to register themselves
1420 * with a newer API version than the host supports:
1422 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1423 para_state->ret = -KVM_EINVAL;
1424 goto err_kunmap_skip;
1427 hypercall_gpa = para_state->hypercall_gpa;
1428 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1429 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1430 if (is_error_hpa(hypercall_hpa)) {
1431 para_state->ret = -KVM_EINVAL;
1432 goto err_kunmap_skip;
1435 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1436 vcpu->para_state_page = para_state_page;
1437 vcpu->para_state_gpa = para_state_gpa;
1438 vcpu->hypercall_gpa = hypercall_gpa;
1440 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1441 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1442 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1443 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1444 kunmap_atomic(hypercall, KM_USER1);
1446 para_state->ret = 0;
1448 kunmap(para_state_page);
1454 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1459 case 0xc0010010: /* SYSCFG */
1460 case 0xc0010015: /* HWCR */
1461 case MSR_IA32_PLATFORM_ID:
1462 case MSR_IA32_P5_MC_ADDR:
1463 case MSR_IA32_P5_MC_TYPE:
1464 case MSR_IA32_MC0_CTL:
1465 case MSR_IA32_MCG_STATUS:
1466 case MSR_IA32_MCG_CAP:
1467 case MSR_IA32_MC0_MISC:
1468 case MSR_IA32_MC0_MISC+4:
1469 case MSR_IA32_MC0_MISC+8:
1470 case MSR_IA32_MC0_MISC+12:
1471 case MSR_IA32_MC0_MISC+16:
1472 case MSR_IA32_UCODE_REV:
1473 case MSR_IA32_PERF_STATUS:
1474 case MSR_IA32_EBL_CR_POWERON:
1475 /* MTRR registers */
1477 case 0x200 ... 0x2ff:
1480 case 0xcd: /* fsb frequency */
1483 case MSR_IA32_APICBASE:
1484 data = vcpu->apic_base;
1486 case MSR_IA32_MISC_ENABLE:
1487 data = vcpu->ia32_misc_enable_msr;
1489 #ifdef CONFIG_X86_64
1491 data = vcpu->shadow_efer;
1495 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1501 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1504 * Reads an msr value (of 'msr_index') into 'pdata'.
1505 * Returns 0 on success, non-0 otherwise.
1506 * Assumes vcpu_load() was already called.
1508 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1510 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1513 #ifdef CONFIG_X86_64
1515 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1517 if (efer & EFER_RESERVED_BITS) {
1518 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1525 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1526 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1531 kvm_arch_ops->set_efer(vcpu, efer);
1534 efer |= vcpu->shadow_efer & EFER_LMA;
1536 vcpu->shadow_efer = efer;
1541 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1544 #ifdef CONFIG_X86_64
1546 set_efer(vcpu, data);
1549 case MSR_IA32_MC0_STATUS:
1550 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1551 __FUNCTION__, data);
1553 case MSR_IA32_MCG_STATUS:
1554 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1555 __FUNCTION__, data);
1557 case MSR_IA32_UCODE_REV:
1558 case MSR_IA32_UCODE_WRITE:
1559 case 0x200 ... 0x2ff: /* MTRRs */
1561 case MSR_IA32_APICBASE:
1562 vcpu->apic_base = data;
1564 case MSR_IA32_MISC_ENABLE:
1565 vcpu->ia32_misc_enable_msr = data;
1568 * This is the 'probe whether the host is KVM' logic:
1570 case MSR_KVM_API_MAGIC:
1571 return vcpu_register_para(vcpu, data);
1574 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1579 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1582 * Writes msr value into into the appropriate "register".
1583 * Returns 0 on success, non-0 otherwise.
1584 * Assumes vcpu_load() was already called.
1586 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1588 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1591 void kvm_resched(struct kvm_vcpu *vcpu)
1593 if (!need_resched())
1597 EXPORT_SYMBOL_GPL(kvm_resched);
1599 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1603 struct kvm_cpuid_entry *e, *best;
1605 kvm_arch_ops->cache_regs(vcpu);
1606 function = vcpu->regs[VCPU_REGS_RAX];
1607 vcpu->regs[VCPU_REGS_RAX] = 0;
1608 vcpu->regs[VCPU_REGS_RBX] = 0;
1609 vcpu->regs[VCPU_REGS_RCX] = 0;
1610 vcpu->regs[VCPU_REGS_RDX] = 0;
1612 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1613 e = &vcpu->cpuid_entries[i];
1614 if (e->function == function) {
1619 * Both basic or both extended?
1621 if (((e->function ^ function) & 0x80000000) == 0)
1622 if (!best || e->function > best->function)
1626 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1627 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1628 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1629 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1631 kvm_arch_ops->decache_regs(vcpu);
1632 kvm_arch_ops->skip_emulated_instruction(vcpu);
1634 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1636 static int pio_copy_data(struct kvm_vcpu *vcpu)
1638 void *p = vcpu->pio_data;
1641 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1643 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1646 free_pio_guest_pages(vcpu);
1649 q += vcpu->pio.guest_page_offset;
1650 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1652 memcpy(q, p, bytes);
1654 memcpy(p, q, bytes);
1655 q -= vcpu->pio.guest_page_offset;
1657 free_pio_guest_pages(vcpu);
1661 static int complete_pio(struct kvm_vcpu *vcpu)
1663 struct kvm_pio_request *io = &vcpu->pio;
1667 kvm_arch_ops->cache_regs(vcpu);
1671 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1675 r = pio_copy_data(vcpu);
1677 kvm_arch_ops->cache_regs(vcpu);
1684 delta *= io->cur_count;
1686 * The size of the register should really depend on
1687 * current address size.
1689 vcpu->regs[VCPU_REGS_RCX] -= delta;
1695 vcpu->regs[VCPU_REGS_RDI] += delta;
1697 vcpu->regs[VCPU_REGS_RSI] += delta;
1700 kvm_arch_ops->decache_regs(vcpu);
1702 io->count -= io->cur_count;
1706 kvm_arch_ops->skip_emulated_instruction(vcpu);
1710 static void kernel_pio(struct kvm_io_device *pio_dev,
1711 struct kvm_vcpu *vcpu,
1714 /* TODO: String I/O for in kernel device */
1717 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1721 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1726 static void pio_string_write(struct kvm_io_device *pio_dev,
1727 struct kvm_vcpu *vcpu)
1729 struct kvm_pio_request *io = &vcpu->pio;
1730 void *pd = vcpu->pio_data;
1733 for (i = 0; i < io->cur_count; i++) {
1734 kvm_iodevice_write(pio_dev, io->port,
1741 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1742 int size, unsigned port)
1744 struct kvm_io_device *pio_dev;
1746 vcpu->run->exit_reason = KVM_EXIT_IO;
1747 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1748 vcpu->run->io.size = vcpu->pio.size = size;
1749 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1750 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1751 vcpu->run->io.port = vcpu->pio.port = port;
1753 vcpu->pio.string = 0;
1755 vcpu->pio.guest_page_offset = 0;
1758 kvm_arch_ops->cache_regs(vcpu);
1759 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1760 kvm_arch_ops->decache_regs(vcpu);
1762 pio_dev = vcpu_find_pio_dev(vcpu, port);
1764 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1770 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1772 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1773 int size, unsigned long count, int down,
1774 gva_t address, int rep, unsigned port)
1776 unsigned now, in_page;
1780 struct kvm_io_device *pio_dev;
1782 vcpu->run->exit_reason = KVM_EXIT_IO;
1783 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1784 vcpu->run->io.size = vcpu->pio.size = size;
1785 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1786 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1787 vcpu->run->io.port = vcpu->pio.port = port;
1789 vcpu->pio.string = 1;
1790 vcpu->pio.down = down;
1791 vcpu->pio.guest_page_offset = offset_in_page(address);
1792 vcpu->pio.rep = rep;
1795 kvm_arch_ops->skip_emulated_instruction(vcpu);
1800 in_page = PAGE_SIZE - offset_in_page(address);
1802 in_page = offset_in_page(address) + size;
1803 now = min(count, (unsigned long)in_page / size);
1806 * String I/O straddles page boundary. Pin two guest pages
1807 * so that we satisfy atomicity constraints. Do just one
1808 * transaction to avoid complexity.
1815 * String I/O in reverse. Yuck. Kill the guest, fix later.
1817 pr_unimpl(vcpu, "guest string pio down\n");
1821 vcpu->run->io.count = now;
1822 vcpu->pio.cur_count = now;
1824 for (i = 0; i < nr_pages; ++i) {
1825 mutex_lock(&vcpu->kvm->lock);
1826 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1829 vcpu->pio.guest_pages[i] = page;
1830 mutex_unlock(&vcpu->kvm->lock);
1833 free_pio_guest_pages(vcpu);
1838 pio_dev = vcpu_find_pio_dev(vcpu, port);
1839 if (!vcpu->pio.in) {
1840 /* string PIO write */
1841 ret = pio_copy_data(vcpu);
1842 if (ret >= 0 && pio_dev) {
1843 pio_string_write(pio_dev, vcpu);
1845 if (vcpu->pio.count == 0)
1849 pr_unimpl(vcpu, "no string pio read support yet, "
1850 "port %x size %d count %ld\n",
1855 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1857 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1864 if (vcpu->sigset_active)
1865 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1867 /* re-sync apic's tpr */
1868 vcpu->cr8 = kvm_run->cr8;
1870 if (vcpu->pio.cur_count) {
1871 r = complete_pio(vcpu);
1876 if (vcpu->mmio_needed) {
1877 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1878 vcpu->mmio_read_completed = 1;
1879 vcpu->mmio_needed = 0;
1880 r = emulate_instruction(vcpu, kvm_run,
1881 vcpu->mmio_fault_cr2, 0);
1882 if (r == EMULATE_DO_MMIO) {
1884 * Read-modify-write. Back to userspace.
1891 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1892 kvm_arch_ops->cache_regs(vcpu);
1893 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1894 kvm_arch_ops->decache_regs(vcpu);
1897 r = kvm_arch_ops->run(vcpu, kvm_run);
1900 if (vcpu->sigset_active)
1901 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1907 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1908 struct kvm_regs *regs)
1912 kvm_arch_ops->cache_regs(vcpu);
1914 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1915 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1916 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1917 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1918 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1919 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1920 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1921 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1922 #ifdef CONFIG_X86_64
1923 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1924 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1925 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1926 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1927 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1928 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1929 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1930 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1933 regs->rip = vcpu->rip;
1934 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1937 * Don't leak debug flags in case they were set for guest debugging
1939 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1940 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1947 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1948 struct kvm_regs *regs)
1952 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1953 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1954 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1955 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1956 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1957 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1958 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1959 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1960 #ifdef CONFIG_X86_64
1961 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1962 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1963 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1964 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1965 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1966 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1967 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1968 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1971 vcpu->rip = regs->rip;
1972 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1974 kvm_arch_ops->decache_regs(vcpu);
1981 static void get_segment(struct kvm_vcpu *vcpu,
1982 struct kvm_segment *var, int seg)
1984 return kvm_arch_ops->get_segment(vcpu, var, seg);
1987 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1988 struct kvm_sregs *sregs)
1990 struct descriptor_table dt;
1994 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1995 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1996 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1997 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1998 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1999 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2001 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2002 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2004 kvm_arch_ops->get_idt(vcpu, &dt);
2005 sregs->idt.limit = dt.limit;
2006 sregs->idt.base = dt.base;
2007 kvm_arch_ops->get_gdt(vcpu, &dt);
2008 sregs->gdt.limit = dt.limit;
2009 sregs->gdt.base = dt.base;
2011 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2012 sregs->cr0 = vcpu->cr0;
2013 sregs->cr2 = vcpu->cr2;
2014 sregs->cr3 = vcpu->cr3;
2015 sregs->cr4 = vcpu->cr4;
2016 sregs->cr8 = vcpu->cr8;
2017 sregs->efer = vcpu->shadow_efer;
2018 sregs->apic_base = vcpu->apic_base;
2020 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2021 sizeof sregs->interrupt_bitmap);
2028 static void set_segment(struct kvm_vcpu *vcpu,
2029 struct kvm_segment *var, int seg)
2031 return kvm_arch_ops->set_segment(vcpu, var, seg);
2034 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2035 struct kvm_sregs *sregs)
2037 int mmu_reset_needed = 0;
2039 struct descriptor_table dt;
2043 dt.limit = sregs->idt.limit;
2044 dt.base = sregs->idt.base;
2045 kvm_arch_ops->set_idt(vcpu, &dt);
2046 dt.limit = sregs->gdt.limit;
2047 dt.base = sregs->gdt.base;
2048 kvm_arch_ops->set_gdt(vcpu, &dt);
2050 vcpu->cr2 = sregs->cr2;
2051 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2052 vcpu->cr3 = sregs->cr3;
2054 vcpu->cr8 = sregs->cr8;
2056 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2057 #ifdef CONFIG_X86_64
2058 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2060 vcpu->apic_base = sregs->apic_base;
2062 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2064 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2065 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2067 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2068 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2069 if (!is_long_mode(vcpu) && is_pae(vcpu))
2070 load_pdptrs(vcpu, vcpu->cr3);
2072 if (mmu_reset_needed)
2073 kvm_mmu_reset_context(vcpu);
2075 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2076 sizeof vcpu->irq_pending);
2077 vcpu->irq_summary = 0;
2078 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2079 if (vcpu->irq_pending[i])
2080 __set_bit(i, &vcpu->irq_summary);
2082 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2083 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2084 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2085 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2086 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2087 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2089 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2090 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2098 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2099 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2101 * This list is modified at module load time to reflect the
2102 * capabilities of the host cpu.
2104 static u32 msrs_to_save[] = {
2105 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2107 #ifdef CONFIG_X86_64
2108 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2110 MSR_IA32_TIME_STAMP_COUNTER,
2113 static unsigned num_msrs_to_save;
2115 static u32 emulated_msrs[] = {
2116 MSR_IA32_MISC_ENABLE,
2119 static __init void kvm_init_msr_list(void)
2124 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2125 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2128 msrs_to_save[j] = msrs_to_save[i];
2131 num_msrs_to_save = j;
2135 * Adapt set_msr() to msr_io()'s calling convention
2137 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2139 return kvm_set_msr(vcpu, index, *data);
2143 * Read or write a bunch of msrs. All parameters are kernel addresses.
2145 * @return number of msrs set successfully.
2147 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2148 struct kvm_msr_entry *entries,
2149 int (*do_msr)(struct kvm_vcpu *vcpu,
2150 unsigned index, u64 *data))
2156 for (i = 0; i < msrs->nmsrs; ++i)
2157 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2166 * Read or write a bunch of msrs. Parameters are user addresses.
2168 * @return number of msrs set successfully.
2170 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2171 int (*do_msr)(struct kvm_vcpu *vcpu,
2172 unsigned index, u64 *data),
2175 struct kvm_msrs msrs;
2176 struct kvm_msr_entry *entries;
2181 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2185 if (msrs.nmsrs >= MAX_IO_MSRS)
2189 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2190 entries = vmalloc(size);
2195 if (copy_from_user(entries, user_msrs->entries, size))
2198 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2203 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2215 * Translate a guest virtual address to a guest physical address.
2217 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2218 struct kvm_translation *tr)
2220 unsigned long vaddr = tr->linear_address;
2224 mutex_lock(&vcpu->kvm->lock);
2225 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2226 tr->physical_address = gpa;
2227 tr->valid = gpa != UNMAPPED_GVA;
2230 mutex_unlock(&vcpu->kvm->lock);
2236 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2237 struct kvm_interrupt *irq)
2239 if (irq->irq < 0 || irq->irq >= 256)
2243 set_bit(irq->irq, vcpu->irq_pending);
2244 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2251 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2252 struct kvm_debug_guest *dbg)
2258 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2265 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2266 unsigned long address,
2269 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2270 unsigned long pgoff;
2273 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2275 page = virt_to_page(vcpu->run);
2276 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2277 page = virt_to_page(vcpu->pio_data);
2279 return NOPAGE_SIGBUS;
2282 *type = VM_FAULT_MINOR;
2287 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2288 .nopage = kvm_vcpu_nopage,
2291 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2293 vma->vm_ops = &kvm_vcpu_vm_ops;
2297 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2299 struct kvm_vcpu *vcpu = filp->private_data;
2301 fput(vcpu->kvm->filp);
2305 static struct file_operations kvm_vcpu_fops = {
2306 .release = kvm_vcpu_release,
2307 .unlocked_ioctl = kvm_vcpu_ioctl,
2308 .compat_ioctl = kvm_vcpu_ioctl,
2309 .mmap = kvm_vcpu_mmap,
2313 * Allocates an inode for the vcpu.
2315 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2318 struct inode *inode;
2321 r = anon_inode_getfd(&fd, &inode, &file,
2322 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2325 atomic_inc(&vcpu->kvm->filp->f_count);
2330 * Creates some virtual cpus. Good luck creating more than one.
2332 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2335 struct kvm_vcpu *vcpu;
2340 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2342 return PTR_ERR(vcpu);
2344 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2346 /* We do fxsave: this must be aligned. */
2347 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2350 r = kvm_mmu_setup(vcpu);
2355 mutex_lock(&kvm->lock);
2356 if (kvm->vcpus[n]) {
2358 mutex_unlock(&kvm->lock);
2361 kvm->vcpus[n] = vcpu;
2362 mutex_unlock(&kvm->lock);
2364 /* Now it's all set up, let userspace reach it */
2365 r = create_vcpu_fd(vcpu);
2371 mutex_lock(&kvm->lock);
2372 kvm->vcpus[n] = NULL;
2373 mutex_unlock(&kvm->lock);
2377 kvm_mmu_unload(vcpu);
2381 kvm_arch_ops->vcpu_free(vcpu);
2385 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2389 struct kvm_cpuid_entry *e, *entry;
2391 rdmsrl(MSR_EFER, efer);
2393 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2394 e = &vcpu->cpuid_entries[i];
2395 if (e->function == 0x80000001) {
2400 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2401 entry->edx &= ~(1 << 20);
2402 printk(KERN_INFO "kvm: guest NX capability removed\n");
2406 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2407 struct kvm_cpuid *cpuid,
2408 struct kvm_cpuid_entry __user *entries)
2413 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2416 if (copy_from_user(&vcpu->cpuid_entries, entries,
2417 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2419 vcpu->cpuid_nent = cpuid->nent;
2420 cpuid_fix_nx_cap(vcpu);
2427 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2430 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2431 vcpu->sigset_active = 1;
2432 vcpu->sigset = *sigset;
2434 vcpu->sigset_active = 0;
2439 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2440 * we have asm/x86/processor.h
2451 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2452 #ifdef CONFIG_X86_64
2453 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2455 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2459 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2461 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2465 memcpy(fpu->fpr, fxsave->st_space, 128);
2466 fpu->fcw = fxsave->cwd;
2467 fpu->fsw = fxsave->swd;
2468 fpu->ftwx = fxsave->twd;
2469 fpu->last_opcode = fxsave->fop;
2470 fpu->last_ip = fxsave->rip;
2471 fpu->last_dp = fxsave->rdp;
2472 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2479 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2481 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2485 memcpy(fxsave->st_space, fpu->fpr, 128);
2486 fxsave->cwd = fpu->fcw;
2487 fxsave->swd = fpu->fsw;
2488 fxsave->twd = fpu->ftwx;
2489 fxsave->fop = fpu->last_opcode;
2490 fxsave->rip = fpu->last_ip;
2491 fxsave->rdp = fpu->last_dp;
2492 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2499 static long kvm_vcpu_ioctl(struct file *filp,
2500 unsigned int ioctl, unsigned long arg)
2502 struct kvm_vcpu *vcpu = filp->private_data;
2503 void __user *argp = (void __user *)arg;
2511 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2513 case KVM_GET_REGS: {
2514 struct kvm_regs kvm_regs;
2516 memset(&kvm_regs, 0, sizeof kvm_regs);
2517 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2521 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2526 case KVM_SET_REGS: {
2527 struct kvm_regs kvm_regs;
2530 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2532 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2538 case KVM_GET_SREGS: {
2539 struct kvm_sregs kvm_sregs;
2541 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2542 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2546 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2551 case KVM_SET_SREGS: {
2552 struct kvm_sregs kvm_sregs;
2555 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2557 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2563 case KVM_TRANSLATE: {
2564 struct kvm_translation tr;
2567 if (copy_from_user(&tr, argp, sizeof tr))
2569 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2573 if (copy_to_user(argp, &tr, sizeof tr))
2578 case KVM_INTERRUPT: {
2579 struct kvm_interrupt irq;
2582 if (copy_from_user(&irq, argp, sizeof irq))
2584 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2590 case KVM_DEBUG_GUEST: {
2591 struct kvm_debug_guest dbg;
2594 if (copy_from_user(&dbg, argp, sizeof dbg))
2596 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2603 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2606 r = msr_io(vcpu, argp, do_set_msr, 0);
2608 case KVM_SET_CPUID: {
2609 struct kvm_cpuid __user *cpuid_arg = argp;
2610 struct kvm_cpuid cpuid;
2613 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2615 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2620 case KVM_SET_SIGNAL_MASK: {
2621 struct kvm_signal_mask __user *sigmask_arg = argp;
2622 struct kvm_signal_mask kvm_sigmask;
2623 sigset_t sigset, *p;
2628 if (copy_from_user(&kvm_sigmask, argp,
2629 sizeof kvm_sigmask))
2632 if (kvm_sigmask.len != sizeof sigset)
2635 if (copy_from_user(&sigset, sigmask_arg->sigset,
2640 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2646 memset(&fpu, 0, sizeof fpu);
2647 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2651 if (copy_to_user(argp, &fpu, sizeof fpu))
2660 if (copy_from_user(&fpu, argp, sizeof fpu))
2662 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2675 static long kvm_vm_ioctl(struct file *filp,
2676 unsigned int ioctl, unsigned long arg)
2678 struct kvm *kvm = filp->private_data;
2679 void __user *argp = (void __user *)arg;
2683 case KVM_CREATE_VCPU:
2684 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2688 case KVM_SET_MEMORY_REGION: {
2689 struct kvm_memory_region kvm_mem;
2692 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2694 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2699 case KVM_GET_DIRTY_LOG: {
2700 struct kvm_dirty_log log;
2703 if (copy_from_user(&log, argp, sizeof log))
2705 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2710 case KVM_SET_MEMORY_ALIAS: {
2711 struct kvm_memory_alias alias;
2714 if (copy_from_user(&alias, argp, sizeof alias))
2716 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2721 case KVM_CREATE_IRQCHIP:
2723 kvm->vpic = kvm_create_pic(kvm);
2729 case KVM_IRQ_LINE: {
2730 struct kvm_irq_level irq_event;
2733 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2735 if (irqchip_in_kernel(kvm)) {
2736 if (irq_event.irq < 16)
2737 kvm_pic_set_irq(pic_irqchip(kvm),
2752 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2753 unsigned long address,
2756 struct kvm *kvm = vma->vm_file->private_data;
2757 unsigned long pgoff;
2760 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2761 page = gfn_to_page(kvm, pgoff);
2763 return NOPAGE_SIGBUS;
2766 *type = VM_FAULT_MINOR;
2771 static struct vm_operations_struct kvm_vm_vm_ops = {
2772 .nopage = kvm_vm_nopage,
2775 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2777 vma->vm_ops = &kvm_vm_vm_ops;
2781 static struct file_operations kvm_vm_fops = {
2782 .release = kvm_vm_release,
2783 .unlocked_ioctl = kvm_vm_ioctl,
2784 .compat_ioctl = kvm_vm_ioctl,
2785 .mmap = kvm_vm_mmap,
2788 static int kvm_dev_ioctl_create_vm(void)
2791 struct inode *inode;
2795 kvm = kvm_create_vm();
2797 return PTR_ERR(kvm);
2798 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2800 kvm_destroy_vm(kvm);
2809 static long kvm_dev_ioctl(struct file *filp,
2810 unsigned int ioctl, unsigned long arg)
2812 void __user *argp = (void __user *)arg;
2816 case KVM_GET_API_VERSION:
2820 r = KVM_API_VERSION;
2826 r = kvm_dev_ioctl_create_vm();
2828 case KVM_GET_MSR_INDEX_LIST: {
2829 struct kvm_msr_list __user *user_msr_list = argp;
2830 struct kvm_msr_list msr_list;
2834 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2837 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2838 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2841 if (n < num_msrs_to_save)
2844 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2845 num_msrs_to_save * sizeof(u32)))
2847 if (copy_to_user(user_msr_list->indices
2848 + num_msrs_to_save * sizeof(u32),
2850 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2855 case KVM_CHECK_EXTENSION: {
2856 int ext = (long)argp;
2859 case KVM_CAP_IRQCHIP:
2868 case KVM_GET_VCPU_MMAP_SIZE:
2881 static struct file_operations kvm_chardev_ops = {
2882 .unlocked_ioctl = kvm_dev_ioctl,
2883 .compat_ioctl = kvm_dev_ioctl,
2886 static struct miscdevice kvm_dev = {
2893 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2896 static void decache_vcpus_on_cpu(int cpu)
2899 struct kvm_vcpu *vcpu;
2902 spin_lock(&kvm_lock);
2903 list_for_each_entry(vm, &vm_list, vm_list)
2904 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2905 vcpu = vm->vcpus[i];
2909 * If the vcpu is locked, then it is running on some
2910 * other cpu and therefore it is not cached on the
2913 * If it's not locked, check the last cpu it executed
2916 if (mutex_trylock(&vcpu->mutex)) {
2917 if (vcpu->cpu == cpu) {
2918 kvm_arch_ops->vcpu_decache(vcpu);
2921 mutex_unlock(&vcpu->mutex);
2924 spin_unlock(&kvm_lock);
2927 static void hardware_enable(void *junk)
2929 int cpu = raw_smp_processor_id();
2931 if (cpu_isset(cpu, cpus_hardware_enabled))
2933 cpu_set(cpu, cpus_hardware_enabled);
2934 kvm_arch_ops->hardware_enable(NULL);
2937 static void hardware_disable(void *junk)
2939 int cpu = raw_smp_processor_id();
2941 if (!cpu_isset(cpu, cpus_hardware_enabled))
2943 cpu_clear(cpu, cpus_hardware_enabled);
2944 decache_vcpus_on_cpu(cpu);
2945 kvm_arch_ops->hardware_disable(NULL);
2948 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2955 case CPU_DYING_FROZEN:
2956 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2958 hardware_disable(NULL);
2960 case CPU_UP_CANCELED:
2961 case CPU_UP_CANCELED_FROZEN:
2962 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2964 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2967 case CPU_ONLINE_FROZEN:
2968 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2970 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2976 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2979 if (val == SYS_RESTART) {
2981 * Some (well, at least mine) BIOSes hang on reboot if
2984 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2985 on_each_cpu(hardware_disable, NULL, 0, 1);
2990 static struct notifier_block kvm_reboot_notifier = {
2991 .notifier_call = kvm_reboot,
2995 void kvm_io_bus_init(struct kvm_io_bus *bus)
2997 memset(bus, 0, sizeof(*bus));
3000 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3004 for (i = 0; i < bus->dev_count; i++) {
3005 struct kvm_io_device *pos = bus->devs[i];
3007 kvm_iodevice_destructor(pos);
3011 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3015 for (i = 0; i < bus->dev_count; i++) {
3016 struct kvm_io_device *pos = bus->devs[i];
3018 if (pos->in_range(pos, addr))
3025 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3027 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3029 bus->devs[bus->dev_count++] = dev;
3032 static struct notifier_block kvm_cpu_notifier = {
3033 .notifier_call = kvm_cpu_hotplug,
3034 .priority = 20, /* must be > scheduler priority */
3037 static u64 stat_get(void *_offset)
3039 unsigned offset = (long)_offset;
3042 struct kvm_vcpu *vcpu;
3045 spin_lock(&kvm_lock);
3046 list_for_each_entry(kvm, &vm_list, vm_list)
3047 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3048 vcpu = kvm->vcpus[i];
3050 total += *(u32 *)((void *)vcpu + offset);
3052 spin_unlock(&kvm_lock);
3056 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3058 static __init void kvm_init_debug(void)
3060 struct kvm_stats_debugfs_item *p;
3062 debugfs_dir = debugfs_create_dir("kvm", NULL);
3063 for (p = debugfs_entries; p->name; ++p)
3064 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3065 (void *)(long)p->offset,
3069 static void kvm_exit_debug(void)
3071 struct kvm_stats_debugfs_item *p;
3073 for (p = debugfs_entries; p->name; ++p)
3074 debugfs_remove(p->dentry);
3075 debugfs_remove(debugfs_dir);
3078 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3080 hardware_disable(NULL);
3084 static int kvm_resume(struct sys_device *dev)
3086 hardware_enable(NULL);
3090 static struct sysdev_class kvm_sysdev_class = {
3091 set_kset_name("kvm"),
3092 .suspend = kvm_suspend,
3093 .resume = kvm_resume,
3096 static struct sys_device kvm_sysdev = {
3098 .cls = &kvm_sysdev_class,
3101 hpa_t bad_page_address;
3104 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3106 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3109 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3111 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3113 kvm_arch_ops->vcpu_load(vcpu, cpu);
3116 static void kvm_sched_out(struct preempt_notifier *pn,
3117 struct task_struct *next)
3119 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3121 kvm_arch_ops->vcpu_put(vcpu);
3124 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3125 struct module *module)
3131 printk(KERN_ERR "kvm: already loaded the other module\n");
3135 if (!ops->cpu_has_kvm_support()) {
3136 printk(KERN_ERR "kvm: no hardware support\n");
3139 if (ops->disabled_by_bios()) {
3140 printk(KERN_ERR "kvm: disabled by bios\n");
3146 r = kvm_arch_ops->hardware_setup();
3150 for_each_online_cpu(cpu) {
3151 smp_call_function_single(cpu,
3152 kvm_arch_ops->check_processor_compatibility,
3158 on_each_cpu(hardware_enable, NULL, 0, 1);
3159 r = register_cpu_notifier(&kvm_cpu_notifier);
3162 register_reboot_notifier(&kvm_reboot_notifier);
3164 r = sysdev_class_register(&kvm_sysdev_class);
3168 r = sysdev_register(&kvm_sysdev);
3172 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3173 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3174 __alignof__(struct kvm_vcpu), 0, 0);
3175 if (!kvm_vcpu_cache) {
3180 kvm_chardev_ops.owner = module;
3182 r = misc_register(&kvm_dev);
3184 printk (KERN_ERR "kvm: misc device register failed\n");
3188 kvm_preempt_ops.sched_in = kvm_sched_in;
3189 kvm_preempt_ops.sched_out = kvm_sched_out;
3194 kmem_cache_destroy(kvm_vcpu_cache);
3196 sysdev_unregister(&kvm_sysdev);
3198 sysdev_class_unregister(&kvm_sysdev_class);
3200 unregister_reboot_notifier(&kvm_reboot_notifier);
3201 unregister_cpu_notifier(&kvm_cpu_notifier);
3203 on_each_cpu(hardware_disable, NULL, 0, 1);
3205 kvm_arch_ops->hardware_unsetup();
3207 kvm_arch_ops = NULL;
3211 void kvm_exit_arch(void)
3213 misc_deregister(&kvm_dev);
3214 kmem_cache_destroy(kvm_vcpu_cache);
3215 sysdev_unregister(&kvm_sysdev);
3216 sysdev_class_unregister(&kvm_sysdev_class);
3217 unregister_reboot_notifier(&kvm_reboot_notifier);
3218 unregister_cpu_notifier(&kvm_cpu_notifier);
3219 on_each_cpu(hardware_disable, NULL, 0, 1);
3220 kvm_arch_ops->hardware_unsetup();
3221 kvm_arch_ops = NULL;
3224 static __init int kvm_init(void)
3226 static struct page *bad_page;
3229 r = kvm_mmu_module_init();
3235 kvm_init_msr_list();
3237 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3242 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3243 memset(__va(bad_page_address), 0, PAGE_SIZE);
3249 kvm_mmu_module_exit();
3254 static __exit void kvm_exit(void)
3257 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3258 kvm_mmu_module_exit();
3261 module_init(kvm_init)
3262 module_exit(kvm_exit)
3264 EXPORT_SYMBOL_GPL(kvm_init_arch);
3265 EXPORT_SYMBOL_GPL(kvm_exit_arch);