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;
56 struct kmem_cache *kvm_vcpu_cache;
57 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
59 static __read_mostly struct preempt_ops kvm_preempt_ops;
61 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
63 static struct kvm_stats_debugfs_item {
66 struct dentry *dentry;
67 } debugfs_entries[] = {
68 { "pf_fixed", STAT_OFFSET(pf_fixed) },
69 { "pf_guest", STAT_OFFSET(pf_guest) },
70 { "tlb_flush", STAT_OFFSET(tlb_flush) },
71 { "invlpg", STAT_OFFSET(invlpg) },
72 { "exits", STAT_OFFSET(exits) },
73 { "io_exits", STAT_OFFSET(io_exits) },
74 { "mmio_exits", STAT_OFFSET(mmio_exits) },
75 { "signal_exits", STAT_OFFSET(signal_exits) },
76 { "irq_window", STAT_OFFSET(irq_window_exits) },
77 { "halt_exits", STAT_OFFSET(halt_exits) },
78 { "request_irq", STAT_OFFSET(request_irq_exits) },
79 { "irq_exits", STAT_OFFSET(irq_exits) },
80 { "light_exits", STAT_OFFSET(light_exits) },
81 { "efer_reload", STAT_OFFSET(efer_reload) },
85 static struct dentry *debugfs_dir;
87 #define MAX_IO_MSRS 256
89 #define CR0_RESERVED_BITS \
90 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
91 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
92 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
93 #define CR4_RESERVED_BITS \
94 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
95 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
96 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
97 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
99 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
100 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
103 // LDT or TSS descriptor in the GDT. 16 bytes.
104 struct segment_descriptor_64 {
105 struct segment_descriptor s;
112 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115 unsigned long segment_base(u16 selector)
117 struct descriptor_table gdt;
118 struct segment_descriptor *d;
119 unsigned long table_base;
120 typedef unsigned long ul;
126 asm ("sgdt %0" : "=m"(gdt));
127 table_base = gdt.base;
129 if (selector & 4) { /* from ldt */
132 asm ("sldt %0" : "=g"(ldt_selector));
133 table_base = segment_base(ldt_selector);
135 d = (struct segment_descriptor *)(table_base + (selector & ~7));
136 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
139 && (d->type == 2 || d->type == 9 || d->type == 11))
140 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
144 EXPORT_SYMBOL_GPL(segment_base);
146 static inline int valid_vcpu(int n)
148 return likely(n >= 0 && n < KVM_MAX_VCPUS);
151 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
153 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
156 vcpu->guest_fpu_loaded = 1;
157 fx_save(&vcpu->host_fx_image);
158 fx_restore(&vcpu->guest_fx_image);
160 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
162 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
164 if (!vcpu->guest_fpu_loaded)
167 vcpu->guest_fpu_loaded = 0;
168 fx_save(&vcpu->guest_fx_image);
169 fx_restore(&vcpu->host_fx_image);
171 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
174 * Switches to specified vcpu, until a matching vcpu_put()
176 static void vcpu_load(struct kvm_vcpu *vcpu)
180 mutex_lock(&vcpu->mutex);
182 preempt_notifier_register(&vcpu->preempt_notifier);
183 kvm_arch_ops->vcpu_load(vcpu, cpu);
187 static void vcpu_put(struct kvm_vcpu *vcpu)
190 kvm_arch_ops->vcpu_put(vcpu);
191 preempt_notifier_unregister(&vcpu->preempt_notifier);
193 mutex_unlock(&vcpu->mutex);
196 static void ack_flush(void *_completed)
198 atomic_t *completed = _completed;
200 atomic_inc(completed);
203 void kvm_flush_remote_tlbs(struct kvm *kvm)
207 struct kvm_vcpu *vcpu;
210 atomic_set(&completed, 0);
213 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
214 vcpu = kvm->vcpus[i];
217 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
220 if (cpu != -1 && cpu != raw_smp_processor_id())
221 if (!cpu_isset(cpu, cpus)) {
228 * We really want smp_call_function_mask() here. But that's not
229 * available, so ipi all cpus in parallel and wait for them
232 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
233 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
234 while (atomic_read(&completed) != needed) {
240 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
245 mutex_init(&vcpu->mutex);
247 vcpu->mmu.root_hpa = INVALID_PAGE;
251 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
256 vcpu->run = page_address(page);
258 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
263 vcpu->pio_data = page_address(page);
265 r = kvm_mmu_create(vcpu);
267 goto fail_free_pio_data;
272 free_page((unsigned long)vcpu->pio_data);
274 free_page((unsigned long)vcpu->run);
278 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
280 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
282 kvm_mmu_destroy(vcpu);
283 free_page((unsigned long)vcpu->pio_data);
284 free_page((unsigned long)vcpu->run);
286 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
288 static struct kvm *kvm_create_vm(void)
290 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
293 return ERR_PTR(-ENOMEM);
295 kvm_io_bus_init(&kvm->pio_bus);
296 mutex_init(&kvm->lock);
297 INIT_LIST_HEAD(&kvm->active_mmu_pages);
298 kvm_io_bus_init(&kvm->mmio_bus);
299 spin_lock(&kvm_lock);
300 list_add(&kvm->vm_list, &vm_list);
301 spin_unlock(&kvm_lock);
306 * Free any memory in @free but not in @dont.
308 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
309 struct kvm_memory_slot *dont)
313 if (!dont || free->phys_mem != dont->phys_mem)
314 if (free->phys_mem) {
315 for (i = 0; i < free->npages; ++i)
316 if (free->phys_mem[i])
317 __free_page(free->phys_mem[i]);
318 vfree(free->phys_mem);
321 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
322 vfree(free->dirty_bitmap);
324 free->phys_mem = NULL;
326 free->dirty_bitmap = NULL;
329 static void kvm_free_physmem(struct kvm *kvm)
333 for (i = 0; i < kvm->nmemslots; ++i)
334 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
337 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
341 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
342 if (vcpu->pio.guest_pages[i]) {
343 __free_page(vcpu->pio.guest_pages[i]);
344 vcpu->pio.guest_pages[i] = NULL;
348 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
351 kvm_mmu_unload(vcpu);
355 static void kvm_free_vcpus(struct kvm *kvm)
360 * Unpin any mmu pages first.
362 for (i = 0; i < KVM_MAX_VCPUS; ++i)
364 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
365 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
367 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
368 kvm->vcpus[i] = NULL;
374 static void kvm_destroy_vm(struct kvm *kvm)
376 spin_lock(&kvm_lock);
377 list_del(&kvm->vm_list);
378 spin_unlock(&kvm_lock);
379 kvm_io_bus_destroy(&kvm->pio_bus);
380 kvm_io_bus_destroy(&kvm->mmio_bus);
382 kvm_free_physmem(kvm);
386 static int kvm_vm_release(struct inode *inode, struct file *filp)
388 struct kvm *kvm = filp->private_data;
394 static void inject_gp(struct kvm_vcpu *vcpu)
396 kvm_arch_ops->inject_gp(vcpu, 0);
400 * Load the pae pdptrs. Return true is they are all valid.
402 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
404 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
405 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
410 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
412 mutex_lock(&vcpu->kvm->lock);
413 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
419 pdpt = kmap_atomic(page, KM_USER0);
420 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
421 kunmap_atomic(pdpt, KM_USER0);
423 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
424 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
431 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
433 mutex_unlock(&vcpu->kvm->lock);
438 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
440 if (cr0 & CR0_RESERVED_BITS) {
441 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
447 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
448 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
453 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
454 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
455 "and a clear PE flag\n");
460 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
462 if ((vcpu->shadow_efer & EFER_LME)) {
466 printk(KERN_DEBUG "set_cr0: #GP, start paging "
467 "in long mode while PAE is disabled\n");
471 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
473 printk(KERN_DEBUG "set_cr0: #GP, start paging "
474 "in long mode while CS.L == 1\n");
481 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
482 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
490 kvm_arch_ops->set_cr0(vcpu, cr0);
493 mutex_lock(&vcpu->kvm->lock);
494 kvm_mmu_reset_context(vcpu);
495 mutex_unlock(&vcpu->kvm->lock);
498 EXPORT_SYMBOL_GPL(set_cr0);
500 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
502 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
504 EXPORT_SYMBOL_GPL(lmsw);
506 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
508 if (cr4 & CR4_RESERVED_BITS) {
509 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
514 if (is_long_mode(vcpu)) {
515 if (!(cr4 & X86_CR4_PAE)) {
516 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
521 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
522 && !load_pdptrs(vcpu, vcpu->cr3)) {
523 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
528 if (cr4 & X86_CR4_VMXE) {
529 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
533 kvm_arch_ops->set_cr4(vcpu, cr4);
534 mutex_lock(&vcpu->kvm->lock);
535 kvm_mmu_reset_context(vcpu);
536 mutex_unlock(&vcpu->kvm->lock);
538 EXPORT_SYMBOL_GPL(set_cr4);
540 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
542 if (is_long_mode(vcpu)) {
543 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
544 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
550 if (cr3 & CR3_PAE_RESERVED_BITS) {
552 "set_cr3: #GP, reserved bits\n");
556 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
557 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
563 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
565 "set_cr3: #GP, reserved bits\n");
572 mutex_lock(&vcpu->kvm->lock);
574 * Does the new cr3 value map to physical memory? (Note, we
575 * catch an invalid cr3 even in real-mode, because it would
576 * cause trouble later on when we turn on paging anyway.)
578 * A real CPU would silently accept an invalid cr3 and would
579 * attempt to use it - with largely undefined (and often hard
580 * to debug) behavior on the guest side.
582 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
586 vcpu->mmu.new_cr3(vcpu);
588 mutex_unlock(&vcpu->kvm->lock);
590 EXPORT_SYMBOL_GPL(set_cr3);
592 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
594 if (cr8 & CR8_RESERVED_BITS) {
595 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
601 EXPORT_SYMBOL_GPL(set_cr8);
603 void fx_init(struct kvm_vcpu *vcpu)
605 unsigned after_mxcsr_mask;
607 /* Initialize guest FPU by resetting ours and saving into guest's */
609 fx_save(&vcpu->host_fx_image);
611 fx_save(&vcpu->guest_fx_image);
612 fx_restore(&vcpu->host_fx_image);
615 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
616 vcpu->guest_fx_image.mxcsr = 0x1f80;
617 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
618 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
620 EXPORT_SYMBOL_GPL(fx_init);
623 * Allocate some memory and give it an address in the guest physical address
626 * Discontiguous memory is allowed, mostly for framebuffers.
628 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
629 struct kvm_memory_region *mem)
633 unsigned long npages;
635 struct kvm_memory_slot *memslot;
636 struct kvm_memory_slot old, new;
637 int memory_config_version;
640 /* General sanity checks */
641 if (mem->memory_size & (PAGE_SIZE - 1))
643 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
645 if (mem->slot >= KVM_MEMORY_SLOTS)
647 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
650 memslot = &kvm->memslots[mem->slot];
651 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
652 npages = mem->memory_size >> PAGE_SHIFT;
655 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
658 mutex_lock(&kvm->lock);
660 memory_config_version = kvm->memory_config_version;
661 new = old = *memslot;
663 new.base_gfn = base_gfn;
665 new.flags = mem->flags;
667 /* Disallow changing a memory slot's size. */
669 if (npages && old.npages && npages != old.npages)
672 /* Check for overlaps */
674 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
675 struct kvm_memory_slot *s = &kvm->memslots[i];
679 if (!((base_gfn + npages <= s->base_gfn) ||
680 (base_gfn >= s->base_gfn + s->npages)))
684 * Do memory allocations outside lock. memory_config_version will
687 mutex_unlock(&kvm->lock);
689 /* Deallocate if slot is being removed */
693 /* Free page dirty bitmap if unneeded */
694 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
695 new.dirty_bitmap = NULL;
699 /* Allocate if a slot is being created */
700 if (npages && !new.phys_mem) {
701 new.phys_mem = vmalloc(npages * sizeof(struct page *));
706 memset(new.phys_mem, 0, npages * sizeof(struct page *));
707 for (i = 0; i < npages; ++i) {
708 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
710 if (!new.phys_mem[i])
712 set_page_private(new.phys_mem[i],0);
716 /* Allocate page dirty bitmap if needed */
717 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
718 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
720 new.dirty_bitmap = vmalloc(dirty_bytes);
721 if (!new.dirty_bitmap)
723 memset(new.dirty_bitmap, 0, dirty_bytes);
726 mutex_lock(&kvm->lock);
728 if (memory_config_version != kvm->memory_config_version) {
729 mutex_unlock(&kvm->lock);
730 kvm_free_physmem_slot(&new, &old);
738 if (mem->slot >= kvm->nmemslots)
739 kvm->nmemslots = mem->slot + 1;
742 ++kvm->memory_config_version;
744 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
745 kvm_flush_remote_tlbs(kvm);
747 mutex_unlock(&kvm->lock);
749 kvm_free_physmem_slot(&old, &new);
753 mutex_unlock(&kvm->lock);
755 kvm_free_physmem_slot(&new, &old);
761 * Get (and clear) the dirty memory log for a memory slot.
763 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
764 struct kvm_dirty_log *log)
766 struct kvm_memory_slot *memslot;
769 unsigned long any = 0;
771 mutex_lock(&kvm->lock);
774 * Prevent changes to guest memory configuration even while the lock
778 mutex_unlock(&kvm->lock);
780 if (log->slot >= KVM_MEMORY_SLOTS)
783 memslot = &kvm->memslots[log->slot];
785 if (!memslot->dirty_bitmap)
788 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
790 for (i = 0; !any && i < n/sizeof(long); ++i)
791 any = memslot->dirty_bitmap[i];
794 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
797 /* If nothing is dirty, don't bother messing with page tables. */
799 mutex_lock(&kvm->lock);
800 kvm_mmu_slot_remove_write_access(kvm, log->slot);
801 kvm_flush_remote_tlbs(kvm);
802 memset(memslot->dirty_bitmap, 0, n);
803 mutex_unlock(&kvm->lock);
809 mutex_lock(&kvm->lock);
811 mutex_unlock(&kvm->lock);
816 * Set a new alias region. Aliases map a portion of physical memory into
817 * another portion. This is useful for memory windows, for example the PC
820 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
821 struct kvm_memory_alias *alias)
824 struct kvm_mem_alias *p;
827 /* General sanity checks */
828 if (alias->memory_size & (PAGE_SIZE - 1))
830 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
832 if (alias->slot >= KVM_ALIAS_SLOTS)
834 if (alias->guest_phys_addr + alias->memory_size
835 < alias->guest_phys_addr)
837 if (alias->target_phys_addr + alias->memory_size
838 < alias->target_phys_addr)
841 mutex_lock(&kvm->lock);
843 p = &kvm->aliases[alias->slot];
844 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
845 p->npages = alias->memory_size >> PAGE_SHIFT;
846 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
848 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
849 if (kvm->aliases[n - 1].npages)
853 kvm_mmu_zap_all(kvm);
855 mutex_unlock(&kvm->lock);
863 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
866 struct kvm_mem_alias *alias;
868 for (i = 0; i < kvm->naliases; ++i) {
869 alias = &kvm->aliases[i];
870 if (gfn >= alias->base_gfn
871 && gfn < alias->base_gfn + alias->npages)
872 return alias->target_gfn + gfn - alias->base_gfn;
877 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
881 for (i = 0; i < kvm->nmemslots; ++i) {
882 struct kvm_memory_slot *memslot = &kvm->memslots[i];
884 if (gfn >= memslot->base_gfn
885 && gfn < memslot->base_gfn + memslot->npages)
891 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
893 gfn = unalias_gfn(kvm, gfn);
894 return __gfn_to_memslot(kvm, gfn);
897 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
899 struct kvm_memory_slot *slot;
901 gfn = unalias_gfn(kvm, gfn);
902 slot = __gfn_to_memslot(kvm, gfn);
905 return slot->phys_mem[gfn - slot->base_gfn];
907 EXPORT_SYMBOL_GPL(gfn_to_page);
909 /* WARNING: Does not work on aliased pages. */
910 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
912 struct kvm_memory_slot *memslot;
914 memslot = __gfn_to_memslot(kvm, gfn);
915 if (memslot && memslot->dirty_bitmap) {
916 unsigned long rel_gfn = gfn - memslot->base_gfn;
919 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
920 set_bit(rel_gfn, memslot->dirty_bitmap);
924 int emulator_read_std(unsigned long addr,
927 struct kvm_vcpu *vcpu)
932 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
933 unsigned offset = addr & (PAGE_SIZE-1);
934 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
939 if (gpa == UNMAPPED_GVA)
940 return X86EMUL_PROPAGATE_FAULT;
941 pfn = gpa >> PAGE_SHIFT;
942 page = gfn_to_page(vcpu->kvm, pfn);
944 return X86EMUL_UNHANDLEABLE;
945 page_virt = kmap_atomic(page, KM_USER0);
947 memcpy(data, page_virt + offset, tocopy);
949 kunmap_atomic(page_virt, KM_USER0);
956 return X86EMUL_CONTINUE;
958 EXPORT_SYMBOL_GPL(emulator_read_std);
960 static int emulator_write_std(unsigned long addr,
963 struct kvm_vcpu *vcpu)
965 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
967 return X86EMUL_UNHANDLEABLE;
970 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
974 * Note that its important to have this wrapper function because
975 * in the very near future we will be checking for MMIOs against
976 * the LAPIC as well as the general MMIO bus
978 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
981 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
984 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
987 static int emulator_read_emulated(unsigned long addr,
990 struct kvm_vcpu *vcpu)
992 struct kvm_io_device *mmio_dev;
995 if (vcpu->mmio_read_completed) {
996 memcpy(val, vcpu->mmio_data, bytes);
997 vcpu->mmio_read_completed = 0;
998 return X86EMUL_CONTINUE;
999 } else if (emulator_read_std(addr, val, bytes, vcpu)
1000 == X86EMUL_CONTINUE)
1001 return X86EMUL_CONTINUE;
1003 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1004 if (gpa == UNMAPPED_GVA)
1005 return X86EMUL_PROPAGATE_FAULT;
1008 * Is this MMIO handled locally?
1010 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1012 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1013 return X86EMUL_CONTINUE;
1016 vcpu->mmio_needed = 1;
1017 vcpu->mmio_phys_addr = gpa;
1018 vcpu->mmio_size = bytes;
1019 vcpu->mmio_is_write = 0;
1021 return X86EMUL_UNHANDLEABLE;
1024 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1025 const void *val, int bytes)
1030 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1032 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1035 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1036 virt = kmap_atomic(page, KM_USER0);
1037 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1038 memcpy(virt + offset_in_page(gpa), val, bytes);
1039 kunmap_atomic(virt, KM_USER0);
1043 static int emulator_write_emulated_onepage(unsigned long addr,
1046 struct kvm_vcpu *vcpu)
1048 struct kvm_io_device *mmio_dev;
1049 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1051 if (gpa == UNMAPPED_GVA) {
1052 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1053 return X86EMUL_PROPAGATE_FAULT;
1056 if (emulator_write_phys(vcpu, gpa, val, bytes))
1057 return X86EMUL_CONTINUE;
1060 * Is this MMIO handled locally?
1062 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1064 kvm_iodevice_write(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 = 1;
1072 memcpy(vcpu->mmio_data, val, bytes);
1074 return X86EMUL_CONTINUE;
1077 int emulator_write_emulated(unsigned long addr,
1080 struct kvm_vcpu *vcpu)
1082 /* Crossing a page boundary? */
1083 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1086 now = -addr & ~PAGE_MASK;
1087 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1088 if (rc != X86EMUL_CONTINUE)
1094 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1096 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1098 static int emulator_cmpxchg_emulated(unsigned long addr,
1102 struct kvm_vcpu *vcpu)
1104 static int reported;
1108 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1110 return emulator_write_emulated(addr, new, bytes, vcpu);
1113 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1115 return kvm_arch_ops->get_segment_base(vcpu, seg);
1118 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1120 return X86EMUL_CONTINUE;
1123 int emulate_clts(struct kvm_vcpu *vcpu)
1127 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1128 kvm_arch_ops->set_cr0(vcpu, cr0);
1129 return X86EMUL_CONTINUE;
1132 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1134 struct kvm_vcpu *vcpu = ctxt->vcpu;
1138 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1139 return X86EMUL_CONTINUE;
1141 printk(KERN_DEBUG "%s: unexpected dr %u\n",
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 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1228 if ((r || vcpu->mmio_is_write) && run) {
1229 run->exit_reason = KVM_EXIT_MMIO;
1230 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1231 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1232 run->mmio.len = vcpu->mmio_size;
1233 run->mmio.is_write = vcpu->mmio_is_write;
1237 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1238 return EMULATE_DONE;
1239 if (!vcpu->mmio_needed) {
1240 report_emulation_failure(&emulate_ctxt);
1241 return EMULATE_FAIL;
1243 return EMULATE_DO_MMIO;
1246 kvm_arch_ops->decache_regs(vcpu);
1247 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1249 if (vcpu->mmio_is_write) {
1250 vcpu->mmio_needed = 0;
1251 return EMULATE_DO_MMIO;
1254 return EMULATE_DONE;
1256 EXPORT_SYMBOL_GPL(emulate_instruction);
1258 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1260 if (vcpu->irq_summary)
1263 vcpu->run->exit_reason = KVM_EXIT_HLT;
1264 ++vcpu->stat.halt_exits;
1267 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1269 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1271 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1273 kvm_arch_ops->cache_regs(vcpu);
1275 #ifdef CONFIG_X86_64
1276 if (is_long_mode(vcpu)) {
1277 nr = vcpu->regs[VCPU_REGS_RAX];
1278 a0 = vcpu->regs[VCPU_REGS_RDI];
1279 a1 = vcpu->regs[VCPU_REGS_RSI];
1280 a2 = vcpu->regs[VCPU_REGS_RDX];
1281 a3 = vcpu->regs[VCPU_REGS_RCX];
1282 a4 = vcpu->regs[VCPU_REGS_R8];
1283 a5 = vcpu->regs[VCPU_REGS_R9];
1287 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1288 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1289 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1290 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1291 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1292 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1293 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1297 run->hypercall.nr = nr;
1298 run->hypercall.args[0] = a0;
1299 run->hypercall.args[1] = a1;
1300 run->hypercall.args[2] = a2;
1301 run->hypercall.args[3] = a3;
1302 run->hypercall.args[4] = a4;
1303 run->hypercall.args[5] = a5;
1304 run->hypercall.ret = ret;
1305 run->hypercall.longmode = is_long_mode(vcpu);
1306 kvm_arch_ops->decache_regs(vcpu);
1309 vcpu->regs[VCPU_REGS_RAX] = ret;
1310 kvm_arch_ops->decache_regs(vcpu);
1313 EXPORT_SYMBOL_GPL(kvm_hypercall);
1315 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1317 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1320 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1322 struct descriptor_table dt = { limit, base };
1324 kvm_arch_ops->set_gdt(vcpu, &dt);
1327 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1329 struct descriptor_table dt = { limit, base };
1331 kvm_arch_ops->set_idt(vcpu, &dt);
1334 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1335 unsigned long *rflags)
1338 *rflags = kvm_arch_ops->get_rflags(vcpu);
1341 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1343 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1354 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1359 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1360 unsigned long *rflags)
1364 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1365 *rflags = kvm_arch_ops->get_rflags(vcpu);
1374 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1377 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1382 * Register the para guest with the host:
1384 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1386 struct kvm_vcpu_para_state *para_state;
1387 hpa_t para_state_hpa, hypercall_hpa;
1388 struct page *para_state_page;
1389 unsigned char *hypercall;
1390 gpa_t hypercall_gpa;
1392 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1393 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1396 * Needs to be page aligned:
1398 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1401 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1402 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1403 if (is_error_hpa(para_state_hpa))
1406 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1407 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1408 para_state = kmap(para_state_page);
1410 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1411 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1413 para_state->host_version = KVM_PARA_API_VERSION;
1415 * We cannot support guests that try to register themselves
1416 * with a newer API version than the host supports:
1418 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1419 para_state->ret = -KVM_EINVAL;
1420 goto err_kunmap_skip;
1423 hypercall_gpa = para_state->hypercall_gpa;
1424 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1425 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1426 if (is_error_hpa(hypercall_hpa)) {
1427 para_state->ret = -KVM_EINVAL;
1428 goto err_kunmap_skip;
1431 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1432 vcpu->para_state_page = para_state_page;
1433 vcpu->para_state_gpa = para_state_gpa;
1434 vcpu->hypercall_gpa = hypercall_gpa;
1436 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1437 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1438 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1439 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1440 kunmap_atomic(hypercall, KM_USER1);
1442 para_state->ret = 0;
1444 kunmap(para_state_page);
1450 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1455 case 0xc0010010: /* SYSCFG */
1456 case 0xc0010015: /* HWCR */
1457 case MSR_IA32_PLATFORM_ID:
1458 case MSR_IA32_P5_MC_ADDR:
1459 case MSR_IA32_P5_MC_TYPE:
1460 case MSR_IA32_MC0_CTL:
1461 case MSR_IA32_MCG_STATUS:
1462 case MSR_IA32_MCG_CAP:
1463 case MSR_IA32_MC0_MISC:
1464 case MSR_IA32_MC0_MISC+4:
1465 case MSR_IA32_MC0_MISC+8:
1466 case MSR_IA32_MC0_MISC+12:
1467 case MSR_IA32_MC0_MISC+16:
1468 case MSR_IA32_UCODE_REV:
1469 case MSR_IA32_PERF_STATUS:
1470 case MSR_IA32_EBL_CR_POWERON:
1471 /* MTRR registers */
1473 case 0x200 ... 0x2ff:
1476 case 0xcd: /* fsb frequency */
1479 case MSR_IA32_APICBASE:
1480 data = vcpu->apic_base;
1482 case MSR_IA32_MISC_ENABLE:
1483 data = vcpu->ia32_misc_enable_msr;
1485 #ifdef CONFIG_X86_64
1487 data = vcpu->shadow_efer;
1491 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1497 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1500 * Reads an msr value (of 'msr_index') into 'pdata'.
1501 * Returns 0 on success, non-0 otherwise.
1502 * Assumes vcpu_load() was already called.
1504 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1506 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1509 #ifdef CONFIG_X86_64
1511 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1513 if (efer & EFER_RESERVED_BITS) {
1514 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1521 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1522 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1527 kvm_arch_ops->set_efer(vcpu, efer);
1530 efer |= vcpu->shadow_efer & EFER_LMA;
1532 vcpu->shadow_efer = efer;
1537 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1540 #ifdef CONFIG_X86_64
1542 set_efer(vcpu, data);
1545 case MSR_IA32_MC0_STATUS:
1546 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1547 __FUNCTION__, data);
1549 case MSR_IA32_MCG_STATUS:
1550 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1551 __FUNCTION__, data);
1553 case MSR_IA32_UCODE_REV:
1554 case MSR_IA32_UCODE_WRITE:
1555 case 0x200 ... 0x2ff: /* MTRRs */
1557 case MSR_IA32_APICBASE:
1558 vcpu->apic_base = data;
1560 case MSR_IA32_MISC_ENABLE:
1561 vcpu->ia32_misc_enable_msr = data;
1564 * This is the 'probe whether the host is KVM' logic:
1566 case MSR_KVM_API_MAGIC:
1567 return vcpu_register_para(vcpu, data);
1570 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1575 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1578 * Writes msr value into into the appropriate "register".
1579 * Returns 0 on success, non-0 otherwise.
1580 * Assumes vcpu_load() was already called.
1582 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1584 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1587 void kvm_resched(struct kvm_vcpu *vcpu)
1589 if (!need_resched())
1593 EXPORT_SYMBOL_GPL(kvm_resched);
1595 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1599 struct kvm_cpuid_entry *e, *best;
1601 kvm_arch_ops->cache_regs(vcpu);
1602 function = vcpu->regs[VCPU_REGS_RAX];
1603 vcpu->regs[VCPU_REGS_RAX] = 0;
1604 vcpu->regs[VCPU_REGS_RBX] = 0;
1605 vcpu->regs[VCPU_REGS_RCX] = 0;
1606 vcpu->regs[VCPU_REGS_RDX] = 0;
1608 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1609 e = &vcpu->cpuid_entries[i];
1610 if (e->function == function) {
1615 * Both basic or both extended?
1617 if (((e->function ^ function) & 0x80000000) == 0)
1618 if (!best || e->function > best->function)
1622 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1623 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1624 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1625 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1627 kvm_arch_ops->decache_regs(vcpu);
1628 kvm_arch_ops->skip_emulated_instruction(vcpu);
1630 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1632 static int pio_copy_data(struct kvm_vcpu *vcpu)
1634 void *p = vcpu->pio_data;
1637 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1639 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1642 free_pio_guest_pages(vcpu);
1645 q += vcpu->pio.guest_page_offset;
1646 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1648 memcpy(q, p, bytes);
1650 memcpy(p, q, bytes);
1651 q -= vcpu->pio.guest_page_offset;
1653 free_pio_guest_pages(vcpu);
1657 static int complete_pio(struct kvm_vcpu *vcpu)
1659 struct kvm_pio_request *io = &vcpu->pio;
1663 kvm_arch_ops->cache_regs(vcpu);
1667 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1671 r = pio_copy_data(vcpu);
1673 kvm_arch_ops->cache_regs(vcpu);
1680 delta *= io->cur_count;
1682 * The size of the register should really depend on
1683 * current address size.
1685 vcpu->regs[VCPU_REGS_RCX] -= delta;
1691 vcpu->regs[VCPU_REGS_RDI] += delta;
1693 vcpu->regs[VCPU_REGS_RSI] += delta;
1696 kvm_arch_ops->decache_regs(vcpu);
1698 io->count -= io->cur_count;
1702 kvm_arch_ops->skip_emulated_instruction(vcpu);
1706 static void kernel_pio(struct kvm_io_device *pio_dev,
1707 struct kvm_vcpu *vcpu,
1710 /* TODO: String I/O for in kernel device */
1713 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1717 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1722 static void pio_string_write(struct kvm_io_device *pio_dev,
1723 struct kvm_vcpu *vcpu)
1725 struct kvm_pio_request *io = &vcpu->pio;
1726 void *pd = vcpu->pio_data;
1729 for (i = 0; i < io->cur_count; i++) {
1730 kvm_iodevice_write(pio_dev, io->port,
1737 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1738 int size, unsigned long count, int string, int down,
1739 gva_t address, int rep, unsigned port)
1741 unsigned now, in_page;
1745 struct kvm_io_device *pio_dev;
1747 vcpu->run->exit_reason = KVM_EXIT_IO;
1748 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1749 vcpu->run->io.size = size;
1750 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1751 vcpu->run->io.count = count;
1752 vcpu->run->io.port = port;
1753 vcpu->pio.count = count;
1754 vcpu->pio.cur_count = count;
1755 vcpu->pio.size = size;
1757 vcpu->pio.port = port;
1758 vcpu->pio.string = string;
1759 vcpu->pio.down = down;
1760 vcpu->pio.guest_page_offset = offset_in_page(address);
1761 vcpu->pio.rep = rep;
1763 pio_dev = vcpu_find_pio_dev(vcpu, port);
1765 kvm_arch_ops->cache_regs(vcpu);
1766 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1767 kvm_arch_ops->decache_regs(vcpu);
1769 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1777 kvm_arch_ops->skip_emulated_instruction(vcpu);
1781 now = min(count, PAGE_SIZE / size);
1784 in_page = PAGE_SIZE - offset_in_page(address);
1786 in_page = offset_in_page(address) + size;
1787 now = min(count, (unsigned long)in_page / size);
1790 * String I/O straddles page boundary. Pin two guest pages
1791 * so that we satisfy atomicity constraints. Do just one
1792 * transaction to avoid complexity.
1799 * String I/O in reverse. Yuck. Kill the guest, fix later.
1801 printk(KERN_ERR "kvm: guest string pio down\n");
1805 vcpu->run->io.count = now;
1806 vcpu->pio.cur_count = now;
1808 for (i = 0; i < nr_pages; ++i) {
1809 mutex_lock(&vcpu->kvm->lock);
1810 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1813 vcpu->pio.guest_pages[i] = page;
1814 mutex_unlock(&vcpu->kvm->lock);
1817 free_pio_guest_pages(vcpu);
1822 if (!vcpu->pio.in) {
1823 /* string PIO write */
1824 ret = pio_copy_data(vcpu);
1825 if (ret >= 0 && pio_dev) {
1826 pio_string_write(pio_dev, vcpu);
1828 if (vcpu->pio.count == 0)
1832 printk(KERN_ERR "no string pio read support yet, "
1833 "port %x size %d count %ld\n",
1838 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1840 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1847 if (vcpu->sigset_active)
1848 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1850 /* re-sync apic's tpr */
1851 vcpu->cr8 = kvm_run->cr8;
1853 if (vcpu->pio.cur_count) {
1854 r = complete_pio(vcpu);
1859 if (vcpu->mmio_needed) {
1860 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1861 vcpu->mmio_read_completed = 1;
1862 vcpu->mmio_needed = 0;
1863 r = emulate_instruction(vcpu, kvm_run,
1864 vcpu->mmio_fault_cr2, 0);
1865 if (r == EMULATE_DO_MMIO) {
1867 * Read-modify-write. Back to userspace.
1874 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1875 kvm_arch_ops->cache_regs(vcpu);
1876 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1877 kvm_arch_ops->decache_regs(vcpu);
1880 r = kvm_arch_ops->run(vcpu, kvm_run);
1883 if (vcpu->sigset_active)
1884 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1890 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1891 struct kvm_regs *regs)
1895 kvm_arch_ops->cache_regs(vcpu);
1897 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1898 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1899 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1900 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1901 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1902 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1903 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1904 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1905 #ifdef CONFIG_X86_64
1906 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1907 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1908 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1909 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1910 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1911 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1912 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1913 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1916 regs->rip = vcpu->rip;
1917 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1920 * Don't leak debug flags in case they were set for guest debugging
1922 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1923 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1930 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1931 struct kvm_regs *regs)
1935 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1936 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1937 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1938 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1939 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1940 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1941 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1942 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1943 #ifdef CONFIG_X86_64
1944 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1945 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1946 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1947 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1948 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1949 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1950 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1951 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1954 vcpu->rip = regs->rip;
1955 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1957 kvm_arch_ops->decache_regs(vcpu);
1964 static void get_segment(struct kvm_vcpu *vcpu,
1965 struct kvm_segment *var, int seg)
1967 return kvm_arch_ops->get_segment(vcpu, var, seg);
1970 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1971 struct kvm_sregs *sregs)
1973 struct descriptor_table dt;
1977 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1978 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1979 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1980 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1981 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1982 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1984 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1985 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1987 kvm_arch_ops->get_idt(vcpu, &dt);
1988 sregs->idt.limit = dt.limit;
1989 sregs->idt.base = dt.base;
1990 kvm_arch_ops->get_gdt(vcpu, &dt);
1991 sregs->gdt.limit = dt.limit;
1992 sregs->gdt.base = dt.base;
1994 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1995 sregs->cr0 = vcpu->cr0;
1996 sregs->cr2 = vcpu->cr2;
1997 sregs->cr3 = vcpu->cr3;
1998 sregs->cr4 = vcpu->cr4;
1999 sregs->cr8 = vcpu->cr8;
2000 sregs->efer = vcpu->shadow_efer;
2001 sregs->apic_base = vcpu->apic_base;
2003 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2004 sizeof sregs->interrupt_bitmap);
2011 static void set_segment(struct kvm_vcpu *vcpu,
2012 struct kvm_segment *var, int seg)
2014 return kvm_arch_ops->set_segment(vcpu, var, seg);
2017 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2018 struct kvm_sregs *sregs)
2020 int mmu_reset_needed = 0;
2022 struct descriptor_table dt;
2026 dt.limit = sregs->idt.limit;
2027 dt.base = sregs->idt.base;
2028 kvm_arch_ops->set_idt(vcpu, &dt);
2029 dt.limit = sregs->gdt.limit;
2030 dt.base = sregs->gdt.base;
2031 kvm_arch_ops->set_gdt(vcpu, &dt);
2033 vcpu->cr2 = sregs->cr2;
2034 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2035 vcpu->cr3 = sregs->cr3;
2037 vcpu->cr8 = sregs->cr8;
2039 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2040 #ifdef CONFIG_X86_64
2041 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2043 vcpu->apic_base = sregs->apic_base;
2045 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2047 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2048 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2050 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2051 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2052 if (!is_long_mode(vcpu) && is_pae(vcpu))
2053 load_pdptrs(vcpu, vcpu->cr3);
2055 if (mmu_reset_needed)
2056 kvm_mmu_reset_context(vcpu);
2058 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2059 sizeof vcpu->irq_pending);
2060 vcpu->irq_summary = 0;
2061 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2062 if (vcpu->irq_pending[i])
2063 __set_bit(i, &vcpu->irq_summary);
2065 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2066 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2067 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2068 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2069 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2070 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2072 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2073 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2081 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2082 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2084 * This list is modified at module load time to reflect the
2085 * capabilities of the host cpu.
2087 static u32 msrs_to_save[] = {
2088 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2090 #ifdef CONFIG_X86_64
2091 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2093 MSR_IA32_TIME_STAMP_COUNTER,
2096 static unsigned num_msrs_to_save;
2098 static u32 emulated_msrs[] = {
2099 MSR_IA32_MISC_ENABLE,
2102 static __init void kvm_init_msr_list(void)
2107 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2108 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2111 msrs_to_save[j] = msrs_to_save[i];
2114 num_msrs_to_save = j;
2118 * Adapt set_msr() to msr_io()'s calling convention
2120 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2122 return kvm_set_msr(vcpu, index, *data);
2126 * Read or write a bunch of msrs. All parameters are kernel addresses.
2128 * @return number of msrs set successfully.
2130 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2131 struct kvm_msr_entry *entries,
2132 int (*do_msr)(struct kvm_vcpu *vcpu,
2133 unsigned index, u64 *data))
2139 for (i = 0; i < msrs->nmsrs; ++i)
2140 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2149 * Read or write a bunch of msrs. Parameters are user addresses.
2151 * @return number of msrs set successfully.
2153 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2154 int (*do_msr)(struct kvm_vcpu *vcpu,
2155 unsigned index, u64 *data),
2158 struct kvm_msrs msrs;
2159 struct kvm_msr_entry *entries;
2164 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2168 if (msrs.nmsrs >= MAX_IO_MSRS)
2172 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2173 entries = vmalloc(size);
2178 if (copy_from_user(entries, user_msrs->entries, size))
2181 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2186 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2198 * Translate a guest virtual address to a guest physical address.
2200 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2201 struct kvm_translation *tr)
2203 unsigned long vaddr = tr->linear_address;
2207 mutex_lock(&vcpu->kvm->lock);
2208 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2209 tr->physical_address = gpa;
2210 tr->valid = gpa != UNMAPPED_GVA;
2213 mutex_unlock(&vcpu->kvm->lock);
2219 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2220 struct kvm_interrupt *irq)
2222 if (irq->irq < 0 || irq->irq >= 256)
2226 set_bit(irq->irq, vcpu->irq_pending);
2227 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2234 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2235 struct kvm_debug_guest *dbg)
2241 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2248 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2249 unsigned long address,
2252 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2253 unsigned long pgoff;
2256 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2258 page = virt_to_page(vcpu->run);
2259 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2260 page = virt_to_page(vcpu->pio_data);
2262 return NOPAGE_SIGBUS;
2265 *type = VM_FAULT_MINOR;
2270 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2271 .nopage = kvm_vcpu_nopage,
2274 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2276 vma->vm_ops = &kvm_vcpu_vm_ops;
2280 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2282 struct kvm_vcpu *vcpu = filp->private_data;
2284 fput(vcpu->kvm->filp);
2288 static struct file_operations kvm_vcpu_fops = {
2289 .release = kvm_vcpu_release,
2290 .unlocked_ioctl = kvm_vcpu_ioctl,
2291 .compat_ioctl = kvm_vcpu_ioctl,
2292 .mmap = kvm_vcpu_mmap,
2296 * Allocates an inode for the vcpu.
2298 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2301 struct inode *inode;
2304 r = anon_inode_getfd(&fd, &inode, &file,
2305 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2308 atomic_inc(&vcpu->kvm->filp->f_count);
2313 * Creates some virtual cpus. Good luck creating more than one.
2315 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2318 struct kvm_vcpu *vcpu;
2323 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2325 return PTR_ERR(vcpu);
2327 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2329 /* We do fxsave: this must be aligned. */
2330 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2333 r = kvm_mmu_setup(vcpu);
2338 mutex_lock(&kvm->lock);
2339 if (kvm->vcpus[n]) {
2341 mutex_unlock(&kvm->lock);
2344 kvm->vcpus[n] = vcpu;
2345 mutex_unlock(&kvm->lock);
2347 /* Now it's all set up, let userspace reach it */
2348 r = create_vcpu_fd(vcpu);
2354 mutex_lock(&kvm->lock);
2355 kvm->vcpus[n] = NULL;
2356 mutex_unlock(&kvm->lock);
2360 kvm_mmu_unload(vcpu);
2364 kvm_arch_ops->vcpu_free(vcpu);
2368 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2372 struct kvm_cpuid_entry *e, *entry;
2374 rdmsrl(MSR_EFER, efer);
2376 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2377 e = &vcpu->cpuid_entries[i];
2378 if (e->function == 0x80000001) {
2383 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2384 entry->edx &= ~(1 << 20);
2385 printk(KERN_INFO "kvm: guest NX capability removed\n");
2389 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2390 struct kvm_cpuid *cpuid,
2391 struct kvm_cpuid_entry __user *entries)
2396 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2399 if (copy_from_user(&vcpu->cpuid_entries, entries,
2400 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2402 vcpu->cpuid_nent = cpuid->nent;
2403 cpuid_fix_nx_cap(vcpu);
2410 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2413 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2414 vcpu->sigset_active = 1;
2415 vcpu->sigset = *sigset;
2417 vcpu->sigset_active = 0;
2422 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2423 * we have asm/x86/processor.h
2434 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2435 #ifdef CONFIG_X86_64
2436 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2438 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2442 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2444 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2448 memcpy(fpu->fpr, fxsave->st_space, 128);
2449 fpu->fcw = fxsave->cwd;
2450 fpu->fsw = fxsave->swd;
2451 fpu->ftwx = fxsave->twd;
2452 fpu->last_opcode = fxsave->fop;
2453 fpu->last_ip = fxsave->rip;
2454 fpu->last_dp = fxsave->rdp;
2455 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2462 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2464 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2468 memcpy(fxsave->st_space, fpu->fpr, 128);
2469 fxsave->cwd = fpu->fcw;
2470 fxsave->swd = fpu->fsw;
2471 fxsave->twd = fpu->ftwx;
2472 fxsave->fop = fpu->last_opcode;
2473 fxsave->rip = fpu->last_ip;
2474 fxsave->rdp = fpu->last_dp;
2475 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2482 static long kvm_vcpu_ioctl(struct file *filp,
2483 unsigned int ioctl, unsigned long arg)
2485 struct kvm_vcpu *vcpu = filp->private_data;
2486 void __user *argp = (void __user *)arg;
2494 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2496 case KVM_GET_REGS: {
2497 struct kvm_regs kvm_regs;
2499 memset(&kvm_regs, 0, sizeof kvm_regs);
2500 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2504 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2509 case KVM_SET_REGS: {
2510 struct kvm_regs kvm_regs;
2513 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2515 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2521 case KVM_GET_SREGS: {
2522 struct kvm_sregs kvm_sregs;
2524 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2525 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2529 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2534 case KVM_SET_SREGS: {
2535 struct kvm_sregs kvm_sregs;
2538 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2540 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2546 case KVM_TRANSLATE: {
2547 struct kvm_translation tr;
2550 if (copy_from_user(&tr, argp, sizeof tr))
2552 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2556 if (copy_to_user(argp, &tr, sizeof tr))
2561 case KVM_INTERRUPT: {
2562 struct kvm_interrupt irq;
2565 if (copy_from_user(&irq, argp, sizeof irq))
2567 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2573 case KVM_DEBUG_GUEST: {
2574 struct kvm_debug_guest dbg;
2577 if (copy_from_user(&dbg, argp, sizeof dbg))
2579 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2586 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2589 r = msr_io(vcpu, argp, do_set_msr, 0);
2591 case KVM_SET_CPUID: {
2592 struct kvm_cpuid __user *cpuid_arg = argp;
2593 struct kvm_cpuid cpuid;
2596 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2598 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2603 case KVM_SET_SIGNAL_MASK: {
2604 struct kvm_signal_mask __user *sigmask_arg = argp;
2605 struct kvm_signal_mask kvm_sigmask;
2606 sigset_t sigset, *p;
2611 if (copy_from_user(&kvm_sigmask, argp,
2612 sizeof kvm_sigmask))
2615 if (kvm_sigmask.len != sizeof sigset)
2618 if (copy_from_user(&sigset, sigmask_arg->sigset,
2623 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2629 memset(&fpu, 0, sizeof fpu);
2630 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2634 if (copy_to_user(argp, &fpu, sizeof fpu))
2643 if (copy_from_user(&fpu, argp, sizeof fpu))
2645 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2658 static long kvm_vm_ioctl(struct file *filp,
2659 unsigned int ioctl, unsigned long arg)
2661 struct kvm *kvm = filp->private_data;
2662 void __user *argp = (void __user *)arg;
2666 case KVM_CREATE_VCPU:
2667 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2671 case KVM_SET_MEMORY_REGION: {
2672 struct kvm_memory_region kvm_mem;
2675 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2677 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2682 case KVM_GET_DIRTY_LOG: {
2683 struct kvm_dirty_log log;
2686 if (copy_from_user(&log, argp, sizeof log))
2688 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2693 case KVM_SET_MEMORY_ALIAS: {
2694 struct kvm_memory_alias alias;
2697 if (copy_from_user(&alias, argp, sizeof alias))
2699 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2711 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2712 unsigned long address,
2715 struct kvm *kvm = vma->vm_file->private_data;
2716 unsigned long pgoff;
2719 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2720 page = gfn_to_page(kvm, pgoff);
2722 return NOPAGE_SIGBUS;
2725 *type = VM_FAULT_MINOR;
2730 static struct vm_operations_struct kvm_vm_vm_ops = {
2731 .nopage = kvm_vm_nopage,
2734 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2736 vma->vm_ops = &kvm_vm_vm_ops;
2740 static struct file_operations kvm_vm_fops = {
2741 .release = kvm_vm_release,
2742 .unlocked_ioctl = kvm_vm_ioctl,
2743 .compat_ioctl = kvm_vm_ioctl,
2744 .mmap = kvm_vm_mmap,
2747 static int kvm_dev_ioctl_create_vm(void)
2750 struct inode *inode;
2754 kvm = kvm_create_vm();
2756 return PTR_ERR(kvm);
2757 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2759 kvm_destroy_vm(kvm);
2768 static long kvm_dev_ioctl(struct file *filp,
2769 unsigned int ioctl, unsigned long arg)
2771 void __user *argp = (void __user *)arg;
2775 case KVM_GET_API_VERSION:
2779 r = KVM_API_VERSION;
2785 r = kvm_dev_ioctl_create_vm();
2787 case KVM_GET_MSR_INDEX_LIST: {
2788 struct kvm_msr_list __user *user_msr_list = argp;
2789 struct kvm_msr_list msr_list;
2793 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2796 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2797 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2800 if (n < num_msrs_to_save)
2803 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2804 num_msrs_to_save * sizeof(u32)))
2806 if (copy_to_user(user_msr_list->indices
2807 + num_msrs_to_save * sizeof(u32),
2809 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2814 case KVM_CHECK_EXTENSION:
2816 * No extensions defined at present.
2820 case KVM_GET_VCPU_MMAP_SIZE:
2833 static struct file_operations kvm_chardev_ops = {
2834 .unlocked_ioctl = kvm_dev_ioctl,
2835 .compat_ioctl = kvm_dev_ioctl,
2838 static struct miscdevice kvm_dev = {
2845 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2848 static void decache_vcpus_on_cpu(int cpu)
2851 struct kvm_vcpu *vcpu;
2854 spin_lock(&kvm_lock);
2855 list_for_each_entry(vm, &vm_list, vm_list)
2856 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2857 vcpu = vm->vcpus[i];
2861 * If the vcpu is locked, then it is running on some
2862 * other cpu and therefore it is not cached on the
2865 * If it's not locked, check the last cpu it executed
2868 if (mutex_trylock(&vcpu->mutex)) {
2869 if (vcpu->cpu == cpu) {
2870 kvm_arch_ops->vcpu_decache(vcpu);
2873 mutex_unlock(&vcpu->mutex);
2876 spin_unlock(&kvm_lock);
2879 static void hardware_enable(void *junk)
2881 int cpu = raw_smp_processor_id();
2883 if (cpu_isset(cpu, cpus_hardware_enabled))
2885 cpu_set(cpu, cpus_hardware_enabled);
2886 kvm_arch_ops->hardware_enable(NULL);
2889 static void hardware_disable(void *junk)
2891 int cpu = raw_smp_processor_id();
2893 if (!cpu_isset(cpu, cpus_hardware_enabled))
2895 cpu_clear(cpu, cpus_hardware_enabled);
2896 decache_vcpus_on_cpu(cpu);
2897 kvm_arch_ops->hardware_disable(NULL);
2900 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2907 case CPU_DYING_FROZEN:
2908 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2910 hardware_disable(NULL);
2912 case CPU_UP_CANCELED:
2913 case CPU_UP_CANCELED_FROZEN:
2914 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2916 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2919 case CPU_ONLINE_FROZEN:
2920 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2922 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2928 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2931 if (val == SYS_RESTART) {
2933 * Some (well, at least mine) BIOSes hang on reboot if
2936 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2937 on_each_cpu(hardware_disable, NULL, 0, 1);
2942 static struct notifier_block kvm_reboot_notifier = {
2943 .notifier_call = kvm_reboot,
2947 void kvm_io_bus_init(struct kvm_io_bus *bus)
2949 memset(bus, 0, sizeof(*bus));
2952 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2956 for (i = 0; i < bus->dev_count; i++) {
2957 struct kvm_io_device *pos = bus->devs[i];
2959 kvm_iodevice_destructor(pos);
2963 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2967 for (i = 0; i < bus->dev_count; i++) {
2968 struct kvm_io_device *pos = bus->devs[i];
2970 if (pos->in_range(pos, addr))
2977 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2979 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2981 bus->devs[bus->dev_count++] = dev;
2984 static struct notifier_block kvm_cpu_notifier = {
2985 .notifier_call = kvm_cpu_hotplug,
2986 .priority = 20, /* must be > scheduler priority */
2989 static u64 stat_get(void *_offset)
2991 unsigned offset = (long)_offset;
2994 struct kvm_vcpu *vcpu;
2997 spin_lock(&kvm_lock);
2998 list_for_each_entry(kvm, &vm_list, vm_list)
2999 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3000 vcpu = kvm->vcpus[i];
3002 total += *(u32 *)((void *)vcpu + offset);
3004 spin_unlock(&kvm_lock);
3008 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3010 static __init void kvm_init_debug(void)
3012 struct kvm_stats_debugfs_item *p;
3014 debugfs_dir = debugfs_create_dir("kvm", NULL);
3015 for (p = debugfs_entries; p->name; ++p)
3016 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3017 (void *)(long)p->offset,
3021 static void kvm_exit_debug(void)
3023 struct kvm_stats_debugfs_item *p;
3025 for (p = debugfs_entries; p->name; ++p)
3026 debugfs_remove(p->dentry);
3027 debugfs_remove(debugfs_dir);
3030 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3032 hardware_disable(NULL);
3036 static int kvm_resume(struct sys_device *dev)
3038 hardware_enable(NULL);
3042 static struct sysdev_class kvm_sysdev_class = {
3043 set_kset_name("kvm"),
3044 .suspend = kvm_suspend,
3045 .resume = kvm_resume,
3048 static struct sys_device kvm_sysdev = {
3050 .cls = &kvm_sysdev_class,
3053 hpa_t bad_page_address;
3056 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3058 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3061 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3063 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3065 kvm_arch_ops->vcpu_load(vcpu, cpu);
3068 static void kvm_sched_out(struct preempt_notifier *pn,
3069 struct task_struct *next)
3071 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3073 kvm_arch_ops->vcpu_put(vcpu);
3076 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3077 struct module *module)
3083 printk(KERN_ERR "kvm: already loaded the other module\n");
3087 if (!ops->cpu_has_kvm_support()) {
3088 printk(KERN_ERR "kvm: no hardware support\n");
3091 if (ops->disabled_by_bios()) {
3092 printk(KERN_ERR "kvm: disabled by bios\n");
3098 r = kvm_arch_ops->hardware_setup();
3102 for_each_online_cpu(cpu) {
3103 smp_call_function_single(cpu,
3104 kvm_arch_ops->check_processor_compatibility,
3110 on_each_cpu(hardware_enable, NULL, 0, 1);
3111 r = register_cpu_notifier(&kvm_cpu_notifier);
3114 register_reboot_notifier(&kvm_reboot_notifier);
3116 r = sysdev_class_register(&kvm_sysdev_class);
3120 r = sysdev_register(&kvm_sysdev);
3124 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3125 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3126 __alignof__(struct kvm_vcpu), 0, 0);
3127 if (!kvm_vcpu_cache) {
3132 kvm_chardev_ops.owner = module;
3134 r = misc_register(&kvm_dev);
3136 printk (KERN_ERR "kvm: misc device register failed\n");
3140 kvm_preempt_ops.sched_in = kvm_sched_in;
3141 kvm_preempt_ops.sched_out = kvm_sched_out;
3146 kmem_cache_destroy(kvm_vcpu_cache);
3148 sysdev_unregister(&kvm_sysdev);
3150 sysdev_class_unregister(&kvm_sysdev_class);
3152 unregister_reboot_notifier(&kvm_reboot_notifier);
3153 unregister_cpu_notifier(&kvm_cpu_notifier);
3155 on_each_cpu(hardware_disable, NULL, 0, 1);
3157 kvm_arch_ops->hardware_unsetup();
3159 kvm_arch_ops = NULL;
3163 void kvm_exit_arch(void)
3165 misc_deregister(&kvm_dev);
3166 kmem_cache_destroy(kvm_vcpu_cache);
3167 sysdev_unregister(&kvm_sysdev);
3168 sysdev_class_unregister(&kvm_sysdev_class);
3169 unregister_reboot_notifier(&kvm_reboot_notifier);
3170 unregister_cpu_notifier(&kvm_cpu_notifier);
3171 on_each_cpu(hardware_disable, NULL, 0, 1);
3172 kvm_arch_ops->hardware_unsetup();
3173 kvm_arch_ops = NULL;
3176 static __init int kvm_init(void)
3178 static struct page *bad_page;
3181 r = kvm_mmu_module_init();
3187 kvm_init_msr_list();
3189 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3194 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3195 memset(__va(bad_page_address), 0, PAGE_SIZE);
3201 kvm_mmu_module_exit();
3206 static __exit void kvm_exit(void)
3209 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3210 kvm_mmu_module_exit();
3213 module_init(kvm_init)
3214 module_exit(kvm_exit)
3216 EXPORT_SYMBOL_GPL(kvm_init_arch);
3217 EXPORT_SYMBOL_GPL(kvm_exit_arch);