2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include "x86_emulate.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>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
46 #include <asm/processor.h>
49 #include <asm/uaccess.h>
52 MODULE_AUTHOR("Qumranet");
53 MODULE_LICENSE("GPL");
55 static DEFINE_SPINLOCK(kvm_lock);
56 static LIST_HEAD(vm_list);
58 static cpumask_t cpus_hardware_enabled;
60 struct kvm_x86_ops *kvm_x86_ops;
61 struct kmem_cache *kvm_vcpu_cache;
62 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
64 static __read_mostly struct preempt_ops kvm_preempt_ops;
66 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
68 static struct kvm_stats_debugfs_item {
71 struct dentry *dentry;
72 } debugfs_entries[] = {
73 { "pf_fixed", STAT_OFFSET(pf_fixed) },
74 { "pf_guest", STAT_OFFSET(pf_guest) },
75 { "tlb_flush", STAT_OFFSET(tlb_flush) },
76 { "invlpg", STAT_OFFSET(invlpg) },
77 { "exits", STAT_OFFSET(exits) },
78 { "io_exits", STAT_OFFSET(io_exits) },
79 { "mmio_exits", STAT_OFFSET(mmio_exits) },
80 { "signal_exits", STAT_OFFSET(signal_exits) },
81 { "irq_window", STAT_OFFSET(irq_window_exits) },
82 { "halt_exits", STAT_OFFSET(halt_exits) },
83 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
84 { "request_irq", STAT_OFFSET(request_irq_exits) },
85 { "irq_exits", STAT_OFFSET(irq_exits) },
86 { "light_exits", STAT_OFFSET(light_exits) },
87 { "efer_reload", STAT_OFFSET(efer_reload) },
91 static struct dentry *debugfs_dir;
93 #define CR0_RESERVED_BITS \
94 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
95 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
96 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
97 #define CR4_RESERVED_BITS \
98 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
99 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
100 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
101 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
103 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
104 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
106 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
109 static inline int valid_vcpu(int n)
111 return likely(n >= 0 && n < KVM_MAX_VCPUS);
114 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
116 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
119 vcpu->guest_fpu_loaded = 1;
120 fx_save(&vcpu->host_fx_image);
121 fx_restore(&vcpu->guest_fx_image);
123 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
125 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
127 if (!vcpu->guest_fpu_loaded)
130 vcpu->guest_fpu_loaded = 0;
131 fx_save(&vcpu->guest_fx_image);
132 fx_restore(&vcpu->host_fx_image);
134 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
137 * Switches to specified vcpu, until a matching vcpu_put()
139 void vcpu_load(struct kvm_vcpu *vcpu)
143 mutex_lock(&vcpu->mutex);
145 preempt_notifier_register(&vcpu->preempt_notifier);
146 kvm_arch_vcpu_load(vcpu, cpu);
150 void vcpu_put(struct kvm_vcpu *vcpu)
153 kvm_arch_vcpu_put(vcpu);
154 preempt_notifier_unregister(&vcpu->preempt_notifier);
156 mutex_unlock(&vcpu->mutex);
159 static void ack_flush(void *_completed)
163 void kvm_flush_remote_tlbs(struct kvm *kvm)
167 struct kvm_vcpu *vcpu;
170 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
171 vcpu = kvm->vcpus[i];
174 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
177 if (cpu != -1 && cpu != raw_smp_processor_id())
180 smp_call_function_mask(cpus, ack_flush, NULL, 1);
183 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
188 mutex_init(&vcpu->mutex);
190 vcpu->mmu.root_hpa = INVALID_PAGE;
193 if (!irqchip_in_kernel(kvm) || id == 0)
194 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
196 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
197 init_waitqueue_head(&vcpu->wq);
199 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
204 vcpu->run = page_address(page);
206 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
211 vcpu->pio_data = page_address(page);
213 r = kvm_mmu_create(vcpu);
215 goto fail_free_pio_data;
217 if (irqchip_in_kernel(kvm)) {
218 r = kvm_create_lapic(vcpu);
220 goto fail_mmu_destroy;
226 kvm_mmu_destroy(vcpu);
228 free_page((unsigned long)vcpu->pio_data);
230 free_page((unsigned long)vcpu->run);
234 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
236 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
238 kvm_free_lapic(vcpu);
239 kvm_mmu_destroy(vcpu);
240 free_page((unsigned long)vcpu->pio_data);
241 free_page((unsigned long)vcpu->run);
243 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
245 static struct kvm *kvm_create_vm(void)
247 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
250 return ERR_PTR(-ENOMEM);
252 kvm_io_bus_init(&kvm->pio_bus);
253 mutex_init(&kvm->lock);
254 INIT_LIST_HEAD(&kvm->active_mmu_pages);
255 kvm_io_bus_init(&kvm->mmio_bus);
256 spin_lock(&kvm_lock);
257 list_add(&kvm->vm_list, &vm_list);
258 spin_unlock(&kvm_lock);
263 * Free any memory in @free but not in @dont.
265 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
266 struct kvm_memory_slot *dont)
268 if (!dont || free->rmap != dont->rmap)
271 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
272 vfree(free->dirty_bitmap);
275 free->dirty_bitmap = NULL;
279 static void kvm_free_physmem(struct kvm *kvm)
283 for (i = 0; i < kvm->nmemslots; ++i)
284 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
287 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
291 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
292 if (vcpu->pio.guest_pages[i]) {
293 kvm_release_page(vcpu->pio.guest_pages[i]);
294 vcpu->pio.guest_pages[i] = NULL;
298 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
301 kvm_mmu_unload(vcpu);
305 static void kvm_free_vcpus(struct kvm *kvm)
310 * Unpin any mmu pages first.
312 for (i = 0; i < KVM_MAX_VCPUS; ++i)
314 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
315 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
317 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
318 kvm->vcpus[i] = NULL;
324 static void kvm_destroy_vm(struct kvm *kvm)
326 spin_lock(&kvm_lock);
327 list_del(&kvm->vm_list);
328 spin_unlock(&kvm_lock);
329 kvm_io_bus_destroy(&kvm->pio_bus);
330 kvm_io_bus_destroy(&kvm->mmio_bus);
334 kvm_free_physmem(kvm);
338 static int kvm_vm_release(struct inode *inode, struct file *filp)
340 struct kvm *kvm = filp->private_data;
346 static void inject_gp(struct kvm_vcpu *vcpu)
348 kvm_x86_ops->inject_gp(vcpu, 0);
352 * Load the pae pdptrs. Return true is they are all valid.
354 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
356 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
357 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
360 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
362 mutex_lock(&vcpu->kvm->lock);
363 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
364 offset * sizeof(u64), sizeof(pdpte));
369 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
370 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
377 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
379 mutex_unlock(&vcpu->kvm->lock);
384 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
386 if (cr0 & CR0_RESERVED_BITS) {
387 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
393 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
394 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
399 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
400 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
401 "and a clear PE flag\n");
406 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
408 if ((vcpu->shadow_efer & EFER_LME)) {
412 printk(KERN_DEBUG "set_cr0: #GP, start paging "
413 "in long mode while PAE is disabled\n");
417 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
419 printk(KERN_DEBUG "set_cr0: #GP, start paging "
420 "in long mode while CS.L == 1\n");
427 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
428 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
436 kvm_x86_ops->set_cr0(vcpu, cr0);
439 mutex_lock(&vcpu->kvm->lock);
440 kvm_mmu_reset_context(vcpu);
441 mutex_unlock(&vcpu->kvm->lock);
444 EXPORT_SYMBOL_GPL(set_cr0);
446 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
448 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
450 EXPORT_SYMBOL_GPL(lmsw);
452 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
454 if (cr4 & CR4_RESERVED_BITS) {
455 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
460 if (is_long_mode(vcpu)) {
461 if (!(cr4 & X86_CR4_PAE)) {
462 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
467 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
468 && !load_pdptrs(vcpu, vcpu->cr3)) {
469 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
474 if (cr4 & X86_CR4_VMXE) {
475 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
479 kvm_x86_ops->set_cr4(vcpu, cr4);
481 mutex_lock(&vcpu->kvm->lock);
482 kvm_mmu_reset_context(vcpu);
483 mutex_unlock(&vcpu->kvm->lock);
485 EXPORT_SYMBOL_GPL(set_cr4);
487 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
489 if (is_long_mode(vcpu)) {
490 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
491 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
497 if (cr3 & CR3_PAE_RESERVED_BITS) {
499 "set_cr3: #GP, reserved bits\n");
503 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
504 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
511 * We don't check reserved bits in nonpae mode, because
512 * this isn't enforced, and VMware depends on this.
516 mutex_lock(&vcpu->kvm->lock);
518 * Does the new cr3 value map to physical memory? (Note, we
519 * catch an invalid cr3 even in real-mode, because it would
520 * cause trouble later on when we turn on paging anyway.)
522 * A real CPU would silently accept an invalid cr3 and would
523 * attempt to use it - with largely undefined (and often hard
524 * to debug) behavior on the guest side.
526 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
530 vcpu->mmu.new_cr3(vcpu);
532 mutex_unlock(&vcpu->kvm->lock);
534 EXPORT_SYMBOL_GPL(set_cr3);
536 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
538 if (cr8 & CR8_RESERVED_BITS) {
539 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
543 if (irqchip_in_kernel(vcpu->kvm))
544 kvm_lapic_set_tpr(vcpu, cr8);
548 EXPORT_SYMBOL_GPL(set_cr8);
550 unsigned long get_cr8(struct kvm_vcpu *vcpu)
552 if (irqchip_in_kernel(vcpu->kvm))
553 return kvm_lapic_get_cr8(vcpu);
557 EXPORT_SYMBOL_GPL(get_cr8);
559 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
561 if (irqchip_in_kernel(vcpu->kvm))
562 return vcpu->apic_base;
564 return vcpu->apic_base;
566 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
568 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
570 /* TODO: reserve bits check */
571 if (irqchip_in_kernel(vcpu->kvm))
572 kvm_lapic_set_base(vcpu, data);
574 vcpu->apic_base = data;
576 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
578 void fx_init(struct kvm_vcpu *vcpu)
580 unsigned after_mxcsr_mask;
582 /* Initialize guest FPU by resetting ours and saving into guest's */
584 fx_save(&vcpu->host_fx_image);
586 fx_save(&vcpu->guest_fx_image);
587 fx_restore(&vcpu->host_fx_image);
590 vcpu->cr0 |= X86_CR0_ET;
591 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
592 vcpu->guest_fx_image.mxcsr = 0x1f80;
593 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
594 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
596 EXPORT_SYMBOL_GPL(fx_init);
599 * Allocate some memory and give it an address in the guest physical address
602 * Discontiguous memory is allowed, mostly for framebuffers.
604 int kvm_set_memory_region(struct kvm *kvm,
605 struct kvm_userspace_memory_region *mem,
610 unsigned long npages;
612 struct kvm_memory_slot *memslot;
613 struct kvm_memory_slot old, new;
616 /* General sanity checks */
617 if (mem->memory_size & (PAGE_SIZE - 1))
619 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
621 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
623 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
626 memslot = &kvm->memslots[mem->slot];
627 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
628 npages = mem->memory_size >> PAGE_SHIFT;
631 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
633 mutex_lock(&kvm->lock);
635 new = old = *memslot;
637 new.base_gfn = base_gfn;
639 new.flags = mem->flags;
641 /* Disallow changing a memory slot's size. */
643 if (npages && old.npages && npages != old.npages)
646 /* Check for overlaps */
648 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
649 struct kvm_memory_slot *s = &kvm->memslots[i];
653 if (!((base_gfn + npages <= s->base_gfn) ||
654 (base_gfn >= s->base_gfn + s->npages)))
658 /* Free page dirty bitmap if unneeded */
659 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
660 new.dirty_bitmap = NULL;
664 /* Allocate if a slot is being created */
665 if (npages && !new.rmap) {
666 new.rmap = vmalloc(npages * sizeof(struct page *));
671 memset(new.rmap, 0, npages * sizeof(*new.rmap));
673 new.user_alloc = user_alloc;
675 new.userspace_addr = mem->userspace_addr;
677 down_write(¤t->mm->mmap_sem);
678 new.userspace_addr = do_mmap(NULL, 0,
680 PROT_READ | PROT_WRITE,
681 MAP_SHARED | MAP_ANONYMOUS,
683 up_write(¤t->mm->mmap_sem);
685 if (IS_ERR((void *)new.userspace_addr))
689 if (!old.user_alloc && old.rmap) {
692 down_write(¤t->mm->mmap_sem);
693 ret = do_munmap(current->mm, old.userspace_addr,
694 old.npages * PAGE_SIZE);
695 up_write(¤t->mm->mmap_sem);
698 "kvm_vm_ioctl_set_memory_region: "
699 "failed to munmap memory\n");
703 /* Allocate page dirty bitmap if needed */
704 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
705 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
707 new.dirty_bitmap = vmalloc(dirty_bytes);
708 if (!new.dirty_bitmap)
710 memset(new.dirty_bitmap, 0, dirty_bytes);
713 if (mem->slot >= kvm->nmemslots)
714 kvm->nmemslots = mem->slot + 1;
716 if (!kvm->n_requested_mmu_pages) {
717 unsigned int n_pages;
720 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
721 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
724 unsigned int nr_mmu_pages;
726 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
727 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
728 nr_mmu_pages = max(nr_mmu_pages,
729 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
730 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
736 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
737 kvm_flush_remote_tlbs(kvm);
739 mutex_unlock(&kvm->lock);
741 kvm_free_physmem_slot(&old, &new);
745 mutex_unlock(&kvm->lock);
746 kvm_free_physmem_slot(&new, &old);
751 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
753 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
755 kvm_userspace_memory_region *mem,
758 if (mem->slot >= KVM_MEMORY_SLOTS)
760 return kvm_set_memory_region(kvm, mem, user_alloc);
764 * Get (and clear) the dirty memory log for a memory slot.
766 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
767 struct kvm_dirty_log *log)
769 struct kvm_memory_slot *memslot;
772 unsigned long any = 0;
774 mutex_lock(&kvm->lock);
777 if (log->slot >= KVM_MEMORY_SLOTS)
780 memslot = &kvm->memslots[log->slot];
782 if (!memslot->dirty_bitmap)
785 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
787 for (i = 0; !any && i < n/sizeof(long); ++i)
788 any = memslot->dirty_bitmap[i];
791 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
794 /* If nothing is dirty, don't bother messing with page tables. */
796 kvm_mmu_slot_remove_write_access(kvm, log->slot);
797 kvm_flush_remote_tlbs(kvm);
798 memset(memslot->dirty_bitmap, 0, n);
804 mutex_unlock(&kvm->lock);
808 int is_error_page(struct page *page)
810 return page == bad_page;
812 EXPORT_SYMBOL_GPL(is_error_page);
814 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
817 struct kvm_mem_alias *alias;
819 for (i = 0; i < kvm->naliases; ++i) {
820 alias = &kvm->aliases[i];
821 if (gfn >= alias->base_gfn
822 && gfn < alias->base_gfn + alias->npages)
823 return alias->target_gfn + gfn - alias->base_gfn;
828 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
832 for (i = 0; i < kvm->nmemslots; ++i) {
833 struct kvm_memory_slot *memslot = &kvm->memslots[i];
835 if (gfn >= memslot->base_gfn
836 && gfn < memslot->base_gfn + memslot->npages)
842 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
844 gfn = unalias_gfn(kvm, gfn);
845 return __gfn_to_memslot(kvm, gfn);
848 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
852 gfn = unalias_gfn(kvm, gfn);
853 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
854 struct kvm_memory_slot *memslot = &kvm->memslots[i];
856 if (gfn >= memslot->base_gfn
857 && gfn < memslot->base_gfn + memslot->npages)
862 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
864 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
866 struct kvm_memory_slot *slot;
867 struct page *page[1];
872 gfn = unalias_gfn(kvm, gfn);
873 slot = __gfn_to_memslot(kvm, gfn);
879 down_read(¤t->mm->mmap_sem);
880 npages = get_user_pages(current, current->mm,
882 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
884 up_read(¤t->mm->mmap_sem);
892 EXPORT_SYMBOL_GPL(gfn_to_page);
894 void kvm_release_page(struct page *page)
896 if (!PageReserved(page))
900 EXPORT_SYMBOL_GPL(kvm_release_page);
902 static int next_segment(unsigned long len, int offset)
904 if (len > PAGE_SIZE - offset)
905 return PAGE_SIZE - offset;
910 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
916 page = gfn_to_page(kvm, gfn);
917 if (is_error_page(page)) {
918 kvm_release_page(page);
921 page_virt = kmap_atomic(page, KM_USER0);
923 memcpy(data, page_virt + offset, len);
925 kunmap_atomic(page_virt, KM_USER0);
926 kvm_release_page(page);
929 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
931 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
933 gfn_t gfn = gpa >> PAGE_SHIFT;
935 int offset = offset_in_page(gpa);
938 while ((seg = next_segment(len, offset)) != 0) {
939 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
949 EXPORT_SYMBOL_GPL(kvm_read_guest);
951 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
957 page = gfn_to_page(kvm, gfn);
958 if (is_error_page(page)) {
959 kvm_release_page(page);
962 page_virt = kmap_atomic(page, KM_USER0);
964 memcpy(page_virt + offset, data, len);
966 kunmap_atomic(page_virt, KM_USER0);
967 mark_page_dirty(kvm, gfn);
968 kvm_release_page(page);
971 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
973 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
976 gfn_t gfn = gpa >> PAGE_SHIFT;
978 int offset = offset_in_page(gpa);
981 while ((seg = next_segment(len, offset)) != 0) {
982 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
993 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
998 page = gfn_to_page(kvm, gfn);
999 if (is_error_page(page)) {
1000 kvm_release_page(page);
1003 page_virt = kmap_atomic(page, KM_USER0);
1005 memset(page_virt + offset, 0, len);
1007 kunmap_atomic(page_virt, KM_USER0);
1008 kvm_release_page(page);
1011 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1013 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1015 gfn_t gfn = gpa >> PAGE_SHIFT;
1017 int offset = offset_in_page(gpa);
1020 while ((seg = next_segment(len, offset)) != 0) {
1021 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1030 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1032 /* WARNING: Does not work on aliased pages. */
1033 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1035 struct kvm_memory_slot *memslot;
1037 memslot = __gfn_to_memslot(kvm, gfn);
1038 if (memslot && memslot->dirty_bitmap) {
1039 unsigned long rel_gfn = gfn - memslot->base_gfn;
1042 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1043 set_bit(rel_gfn, memslot->dirty_bitmap);
1047 int emulator_read_std(unsigned long addr,
1050 struct kvm_vcpu *vcpu)
1055 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1056 unsigned offset = addr & (PAGE_SIZE-1);
1057 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1060 if (gpa == UNMAPPED_GVA)
1061 return X86EMUL_PROPAGATE_FAULT;
1062 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1064 return X86EMUL_UNHANDLEABLE;
1071 return X86EMUL_CONTINUE;
1073 EXPORT_SYMBOL_GPL(emulator_read_std);
1075 static int emulator_write_std(unsigned long addr,
1078 struct kvm_vcpu *vcpu)
1080 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1081 return X86EMUL_UNHANDLEABLE;
1085 * Only apic need an MMIO device hook, so shortcut now..
1087 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1090 struct kvm_io_device *dev;
1093 dev = &vcpu->apic->dev;
1094 if (dev->in_range(dev, addr))
1100 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1103 struct kvm_io_device *dev;
1105 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1107 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1111 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1114 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1117 static int emulator_read_emulated(unsigned long addr,
1120 struct kvm_vcpu *vcpu)
1122 struct kvm_io_device *mmio_dev;
1125 if (vcpu->mmio_read_completed) {
1126 memcpy(val, vcpu->mmio_data, bytes);
1127 vcpu->mmio_read_completed = 0;
1128 return X86EMUL_CONTINUE;
1129 } else if (emulator_read_std(addr, val, bytes, vcpu)
1130 == X86EMUL_CONTINUE)
1131 return X86EMUL_CONTINUE;
1133 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1134 if (gpa == UNMAPPED_GVA)
1135 return X86EMUL_PROPAGATE_FAULT;
1138 * Is this MMIO handled locally?
1140 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1142 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1143 return X86EMUL_CONTINUE;
1146 vcpu->mmio_needed = 1;
1147 vcpu->mmio_phys_addr = gpa;
1148 vcpu->mmio_size = bytes;
1149 vcpu->mmio_is_write = 0;
1151 return X86EMUL_UNHANDLEABLE;
1154 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1155 const void *val, int bytes)
1159 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1162 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1166 static int emulator_write_emulated_onepage(unsigned long addr,
1169 struct kvm_vcpu *vcpu)
1171 struct kvm_io_device *mmio_dev;
1172 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1174 if (gpa == UNMAPPED_GVA) {
1175 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1176 return X86EMUL_PROPAGATE_FAULT;
1179 if (emulator_write_phys(vcpu, gpa, val, bytes))
1180 return X86EMUL_CONTINUE;
1183 * Is this MMIO handled locally?
1185 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1187 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1188 return X86EMUL_CONTINUE;
1191 vcpu->mmio_needed = 1;
1192 vcpu->mmio_phys_addr = gpa;
1193 vcpu->mmio_size = bytes;
1194 vcpu->mmio_is_write = 1;
1195 memcpy(vcpu->mmio_data, val, bytes);
1197 return X86EMUL_CONTINUE;
1200 int emulator_write_emulated(unsigned long addr,
1203 struct kvm_vcpu *vcpu)
1205 /* Crossing a page boundary? */
1206 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1209 now = -addr & ~PAGE_MASK;
1210 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1211 if (rc != X86EMUL_CONTINUE)
1217 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1219 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1221 static int emulator_cmpxchg_emulated(unsigned long addr,
1225 struct kvm_vcpu *vcpu)
1227 static int reported;
1231 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1233 return emulator_write_emulated(addr, new, bytes, vcpu);
1236 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1238 return kvm_x86_ops->get_segment_base(vcpu, seg);
1241 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1243 return X86EMUL_CONTINUE;
1246 int emulate_clts(struct kvm_vcpu *vcpu)
1248 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1249 return X86EMUL_CONTINUE;
1252 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1254 struct kvm_vcpu *vcpu = ctxt->vcpu;
1258 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1259 return X86EMUL_CONTINUE;
1261 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1262 return X86EMUL_UNHANDLEABLE;
1266 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1268 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1271 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1273 /* FIXME: better handling */
1274 return X86EMUL_UNHANDLEABLE;
1276 return X86EMUL_CONTINUE;
1279 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1281 static int reported;
1283 unsigned long rip = vcpu->rip;
1284 unsigned long rip_linear;
1286 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1291 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1293 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1294 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1297 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1299 struct x86_emulate_ops emulate_ops = {
1300 .read_std = emulator_read_std,
1301 .write_std = emulator_write_std,
1302 .read_emulated = emulator_read_emulated,
1303 .write_emulated = emulator_write_emulated,
1304 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1307 int emulate_instruction(struct kvm_vcpu *vcpu,
1308 struct kvm_run *run,
1315 vcpu->mmio_fault_cr2 = cr2;
1316 kvm_x86_ops->cache_regs(vcpu);
1318 vcpu->mmio_is_write = 0;
1319 vcpu->pio.string = 0;
1323 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1325 vcpu->emulate_ctxt.vcpu = vcpu;
1326 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1327 vcpu->emulate_ctxt.cr2 = cr2;
1328 vcpu->emulate_ctxt.mode =
1329 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1330 ? X86EMUL_MODE_REAL : cs_l
1331 ? X86EMUL_MODE_PROT64 : cs_db
1332 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1334 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1335 vcpu->emulate_ctxt.cs_base = 0;
1336 vcpu->emulate_ctxt.ds_base = 0;
1337 vcpu->emulate_ctxt.es_base = 0;
1338 vcpu->emulate_ctxt.ss_base = 0;
1340 vcpu->emulate_ctxt.cs_base =
1341 get_segment_base(vcpu, VCPU_SREG_CS);
1342 vcpu->emulate_ctxt.ds_base =
1343 get_segment_base(vcpu, VCPU_SREG_DS);
1344 vcpu->emulate_ctxt.es_base =
1345 get_segment_base(vcpu, VCPU_SREG_ES);
1346 vcpu->emulate_ctxt.ss_base =
1347 get_segment_base(vcpu, VCPU_SREG_SS);
1350 vcpu->emulate_ctxt.gs_base =
1351 get_segment_base(vcpu, VCPU_SREG_GS);
1352 vcpu->emulate_ctxt.fs_base =
1353 get_segment_base(vcpu, VCPU_SREG_FS);
1355 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1357 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1358 return EMULATE_DONE;
1359 return EMULATE_FAIL;
1363 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1365 if (vcpu->pio.string)
1366 return EMULATE_DO_MMIO;
1368 if ((r || vcpu->mmio_is_write) && run) {
1369 run->exit_reason = KVM_EXIT_MMIO;
1370 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1371 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1372 run->mmio.len = vcpu->mmio_size;
1373 run->mmio.is_write = vcpu->mmio_is_write;
1377 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1378 return EMULATE_DONE;
1379 if (!vcpu->mmio_needed) {
1380 kvm_report_emulation_failure(vcpu, "mmio");
1381 return EMULATE_FAIL;
1383 return EMULATE_DO_MMIO;
1386 kvm_x86_ops->decache_regs(vcpu);
1387 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1389 if (vcpu->mmio_is_write) {
1390 vcpu->mmio_needed = 0;
1391 return EMULATE_DO_MMIO;
1394 return EMULATE_DONE;
1396 EXPORT_SYMBOL_GPL(emulate_instruction);
1399 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1401 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1403 DECLARE_WAITQUEUE(wait, current);
1405 add_wait_queue(&vcpu->wq, &wait);
1408 * We will block until either an interrupt or a signal wakes us up
1410 while (!kvm_cpu_has_interrupt(vcpu)
1411 && !signal_pending(current)
1412 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1413 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1414 set_current_state(TASK_INTERRUPTIBLE);
1420 __set_current_state(TASK_RUNNING);
1421 remove_wait_queue(&vcpu->wq, &wait);
1424 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1426 ++vcpu->stat.halt_exits;
1427 if (irqchip_in_kernel(vcpu->kvm)) {
1428 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1429 kvm_vcpu_block(vcpu);
1430 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1434 vcpu->run->exit_reason = KVM_EXIT_HLT;
1438 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1440 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1442 unsigned long nr, a0, a1, a2, a3, ret;
1444 kvm_x86_ops->cache_regs(vcpu);
1446 nr = vcpu->regs[VCPU_REGS_RAX];
1447 a0 = vcpu->regs[VCPU_REGS_RBX];
1448 a1 = vcpu->regs[VCPU_REGS_RCX];
1449 a2 = vcpu->regs[VCPU_REGS_RDX];
1450 a3 = vcpu->regs[VCPU_REGS_RSI];
1452 if (!is_long_mode(vcpu)) {
1465 vcpu->regs[VCPU_REGS_RAX] = ret;
1466 kvm_x86_ops->decache_regs(vcpu);
1469 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1471 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1473 char instruction[3];
1476 mutex_lock(&vcpu->kvm->lock);
1479 * Blow out the MMU to ensure that no other VCPU has an active mapping
1480 * to ensure that the updated hypercall appears atomically across all
1483 kvm_mmu_zap_all(vcpu->kvm);
1485 kvm_x86_ops->cache_regs(vcpu);
1486 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1487 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1488 != X86EMUL_CONTINUE)
1491 mutex_unlock(&vcpu->kvm->lock);
1496 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1498 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1501 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1503 struct descriptor_table dt = { limit, base };
1505 kvm_x86_ops->set_gdt(vcpu, &dt);
1508 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1510 struct descriptor_table dt = { limit, base };
1512 kvm_x86_ops->set_idt(vcpu, &dt);
1515 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1516 unsigned long *rflags)
1519 *rflags = kvm_x86_ops->get_rflags(vcpu);
1522 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1524 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1535 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1540 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1541 unsigned long *rflags)
1545 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1546 *rflags = kvm_x86_ops->get_rflags(vcpu);
1555 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1558 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1562 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1567 case 0xc0010010: /* SYSCFG */
1568 case 0xc0010015: /* HWCR */
1569 case MSR_IA32_PLATFORM_ID:
1570 case MSR_IA32_P5_MC_ADDR:
1571 case MSR_IA32_P5_MC_TYPE:
1572 case MSR_IA32_MC0_CTL:
1573 case MSR_IA32_MCG_STATUS:
1574 case MSR_IA32_MCG_CAP:
1575 case MSR_IA32_MC0_MISC:
1576 case MSR_IA32_MC0_MISC+4:
1577 case MSR_IA32_MC0_MISC+8:
1578 case MSR_IA32_MC0_MISC+12:
1579 case MSR_IA32_MC0_MISC+16:
1580 case MSR_IA32_UCODE_REV:
1581 case MSR_IA32_PERF_STATUS:
1582 case MSR_IA32_EBL_CR_POWERON:
1583 /* MTRR registers */
1585 case 0x200 ... 0x2ff:
1588 case 0xcd: /* fsb frequency */
1591 case MSR_IA32_APICBASE:
1592 data = kvm_get_apic_base(vcpu);
1594 case MSR_IA32_MISC_ENABLE:
1595 data = vcpu->ia32_misc_enable_msr;
1597 #ifdef CONFIG_X86_64
1599 data = vcpu->shadow_efer;
1603 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1609 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1612 * Reads an msr value (of 'msr_index') into 'pdata'.
1613 * Returns 0 on success, non-0 otherwise.
1614 * Assumes vcpu_load() was already called.
1616 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1618 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1621 #ifdef CONFIG_X86_64
1623 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1625 if (efer & EFER_RESERVED_BITS) {
1626 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1633 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1634 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1639 kvm_x86_ops->set_efer(vcpu, efer);
1642 efer |= vcpu->shadow_efer & EFER_LMA;
1644 vcpu->shadow_efer = efer;
1649 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1652 #ifdef CONFIG_X86_64
1654 set_efer(vcpu, data);
1657 case MSR_IA32_MC0_STATUS:
1658 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1659 __FUNCTION__, data);
1661 case MSR_IA32_MCG_STATUS:
1662 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1663 __FUNCTION__, data);
1665 case MSR_IA32_UCODE_REV:
1666 case MSR_IA32_UCODE_WRITE:
1667 case 0x200 ... 0x2ff: /* MTRRs */
1669 case MSR_IA32_APICBASE:
1670 kvm_set_apic_base(vcpu, data);
1672 case MSR_IA32_MISC_ENABLE:
1673 vcpu->ia32_misc_enable_msr = data;
1676 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1681 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1684 * Writes msr value into into the appropriate "register".
1685 * Returns 0 on success, non-0 otherwise.
1686 * Assumes vcpu_load() was already called.
1688 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1690 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1693 void kvm_resched(struct kvm_vcpu *vcpu)
1695 if (!need_resched())
1699 EXPORT_SYMBOL_GPL(kvm_resched);
1701 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1705 struct kvm_cpuid_entry *e, *best;
1707 kvm_x86_ops->cache_regs(vcpu);
1708 function = vcpu->regs[VCPU_REGS_RAX];
1709 vcpu->regs[VCPU_REGS_RAX] = 0;
1710 vcpu->regs[VCPU_REGS_RBX] = 0;
1711 vcpu->regs[VCPU_REGS_RCX] = 0;
1712 vcpu->regs[VCPU_REGS_RDX] = 0;
1714 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1715 e = &vcpu->cpuid_entries[i];
1716 if (e->function == function) {
1721 * Both basic or both extended?
1723 if (((e->function ^ function) & 0x80000000) == 0)
1724 if (!best || e->function > best->function)
1728 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1729 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1730 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1731 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1733 kvm_x86_ops->decache_regs(vcpu);
1734 kvm_x86_ops->skip_emulated_instruction(vcpu);
1736 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1738 static int pio_copy_data(struct kvm_vcpu *vcpu)
1740 void *p = vcpu->pio_data;
1743 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1745 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1748 free_pio_guest_pages(vcpu);
1751 q += vcpu->pio.guest_page_offset;
1752 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1754 memcpy(q, p, bytes);
1756 memcpy(p, q, bytes);
1757 q -= vcpu->pio.guest_page_offset;
1759 free_pio_guest_pages(vcpu);
1763 static int complete_pio(struct kvm_vcpu *vcpu)
1765 struct kvm_pio_request *io = &vcpu->pio;
1769 kvm_x86_ops->cache_regs(vcpu);
1773 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1777 r = pio_copy_data(vcpu);
1779 kvm_x86_ops->cache_regs(vcpu);
1786 delta *= io->cur_count;
1788 * The size of the register should really depend on
1789 * current address size.
1791 vcpu->regs[VCPU_REGS_RCX] -= delta;
1797 vcpu->regs[VCPU_REGS_RDI] += delta;
1799 vcpu->regs[VCPU_REGS_RSI] += delta;
1802 kvm_x86_ops->decache_regs(vcpu);
1804 io->count -= io->cur_count;
1810 static void kernel_pio(struct kvm_io_device *pio_dev,
1811 struct kvm_vcpu *vcpu,
1814 /* TODO: String I/O for in kernel device */
1816 mutex_lock(&vcpu->kvm->lock);
1818 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1822 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1825 mutex_unlock(&vcpu->kvm->lock);
1828 static void pio_string_write(struct kvm_io_device *pio_dev,
1829 struct kvm_vcpu *vcpu)
1831 struct kvm_pio_request *io = &vcpu->pio;
1832 void *pd = vcpu->pio_data;
1835 mutex_lock(&vcpu->kvm->lock);
1836 for (i = 0; i < io->cur_count; i++) {
1837 kvm_iodevice_write(pio_dev, io->port,
1842 mutex_unlock(&vcpu->kvm->lock);
1845 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1846 int size, unsigned port)
1848 struct kvm_io_device *pio_dev;
1850 vcpu->run->exit_reason = KVM_EXIT_IO;
1851 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1852 vcpu->run->io.size = vcpu->pio.size = size;
1853 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1854 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1855 vcpu->run->io.port = vcpu->pio.port = port;
1857 vcpu->pio.string = 0;
1859 vcpu->pio.guest_page_offset = 0;
1862 kvm_x86_ops->cache_regs(vcpu);
1863 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1864 kvm_x86_ops->decache_regs(vcpu);
1866 kvm_x86_ops->skip_emulated_instruction(vcpu);
1868 pio_dev = vcpu_find_pio_dev(vcpu, port);
1870 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1876 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1878 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1879 int size, unsigned long count, int down,
1880 gva_t address, int rep, unsigned port)
1882 unsigned now, in_page;
1886 struct kvm_io_device *pio_dev;
1888 vcpu->run->exit_reason = KVM_EXIT_IO;
1889 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1890 vcpu->run->io.size = vcpu->pio.size = size;
1891 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1892 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1893 vcpu->run->io.port = vcpu->pio.port = port;
1895 vcpu->pio.string = 1;
1896 vcpu->pio.down = down;
1897 vcpu->pio.guest_page_offset = offset_in_page(address);
1898 vcpu->pio.rep = rep;
1901 kvm_x86_ops->skip_emulated_instruction(vcpu);
1906 in_page = PAGE_SIZE - offset_in_page(address);
1908 in_page = offset_in_page(address) + size;
1909 now = min(count, (unsigned long)in_page / size);
1912 * String I/O straddles page boundary. Pin two guest pages
1913 * so that we satisfy atomicity constraints. Do just one
1914 * transaction to avoid complexity.
1921 * String I/O in reverse. Yuck. Kill the guest, fix later.
1923 pr_unimpl(vcpu, "guest string pio down\n");
1927 vcpu->run->io.count = now;
1928 vcpu->pio.cur_count = now;
1930 if (vcpu->pio.cur_count == vcpu->pio.count)
1931 kvm_x86_ops->skip_emulated_instruction(vcpu);
1933 for (i = 0; i < nr_pages; ++i) {
1934 mutex_lock(&vcpu->kvm->lock);
1935 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1936 vcpu->pio.guest_pages[i] = page;
1937 mutex_unlock(&vcpu->kvm->lock);
1940 free_pio_guest_pages(vcpu);
1945 pio_dev = vcpu_find_pio_dev(vcpu, port);
1946 if (!vcpu->pio.in) {
1947 /* string PIO write */
1948 ret = pio_copy_data(vcpu);
1949 if (ret >= 0 && pio_dev) {
1950 pio_string_write(pio_dev, vcpu);
1952 if (vcpu->pio.count == 0)
1956 pr_unimpl(vcpu, "no string pio read support yet, "
1957 "port %x size %d count %ld\n",
1962 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1965 * Check if userspace requested an interrupt window, and that the
1966 * interrupt window is open.
1968 * No need to exit to userspace if we already have an interrupt queued.
1970 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1971 struct kvm_run *kvm_run)
1973 return (!vcpu->irq_summary &&
1974 kvm_run->request_interrupt_window &&
1975 vcpu->interrupt_window_open &&
1976 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1979 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1980 struct kvm_run *kvm_run)
1982 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1983 kvm_run->cr8 = get_cr8(vcpu);
1984 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1985 if (irqchip_in_kernel(vcpu->kvm))
1986 kvm_run->ready_for_interrupt_injection = 1;
1988 kvm_run->ready_for_interrupt_injection =
1989 (vcpu->interrupt_window_open &&
1990 vcpu->irq_summary == 0);
1993 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1997 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1998 pr_debug("vcpu %d received sipi with vector # %x\n",
1999 vcpu->vcpu_id, vcpu->sipi_vector);
2000 kvm_lapic_reset(vcpu);
2001 r = kvm_x86_ops->vcpu_reset(vcpu);
2004 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2008 if (vcpu->guest_debug.enabled)
2009 kvm_x86_ops->guest_debug_pre(vcpu);
2012 r = kvm_mmu_reload(vcpu);
2016 kvm_inject_pending_timer_irqs(vcpu);
2020 kvm_x86_ops->prepare_guest_switch(vcpu);
2021 kvm_load_guest_fpu(vcpu);
2023 local_irq_disable();
2025 if (signal_pending(current)) {
2029 kvm_run->exit_reason = KVM_EXIT_INTR;
2030 ++vcpu->stat.signal_exits;
2034 if (irqchip_in_kernel(vcpu->kvm))
2035 kvm_x86_ops->inject_pending_irq(vcpu);
2036 else if (!vcpu->mmio_read_completed)
2037 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2039 vcpu->guest_mode = 1;
2043 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2044 kvm_x86_ops->tlb_flush(vcpu);
2046 kvm_x86_ops->run(vcpu, kvm_run);
2048 vcpu->guest_mode = 0;
2054 * We must have an instruction between local_irq_enable() and
2055 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2056 * the interrupt shadow. The stat.exits increment will do nicely.
2057 * But we need to prevent reordering, hence this barrier():
2066 * Profile KVM exit RIPs:
2068 if (unlikely(prof_on == KVM_PROFILING)) {
2069 kvm_x86_ops->cache_regs(vcpu);
2070 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2073 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2076 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2078 kvm_run->exit_reason = KVM_EXIT_INTR;
2079 ++vcpu->stat.request_irq_exits;
2082 if (!need_resched()) {
2083 ++vcpu->stat.light_exits;
2094 post_kvm_run_save(vcpu, kvm_run);
2100 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2107 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2108 kvm_vcpu_block(vcpu);
2113 if (vcpu->sigset_active)
2114 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2116 /* re-sync apic's tpr */
2117 if (!irqchip_in_kernel(vcpu->kvm))
2118 set_cr8(vcpu, kvm_run->cr8);
2120 if (vcpu->pio.cur_count) {
2121 r = complete_pio(vcpu);
2125 #if CONFIG_HAS_IOMEM
2126 if (vcpu->mmio_needed) {
2127 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2128 vcpu->mmio_read_completed = 1;
2129 vcpu->mmio_needed = 0;
2130 r = emulate_instruction(vcpu, kvm_run,
2131 vcpu->mmio_fault_cr2, 0, 1);
2132 if (r == EMULATE_DO_MMIO) {
2134 * Read-modify-write. Back to userspace.
2141 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2142 kvm_x86_ops->cache_regs(vcpu);
2143 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2144 kvm_x86_ops->decache_regs(vcpu);
2147 r = __vcpu_run(vcpu, kvm_run);
2150 if (vcpu->sigset_active)
2151 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2157 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2158 struct kvm_regs *regs)
2162 kvm_x86_ops->cache_regs(vcpu);
2164 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2165 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2166 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2167 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2168 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2169 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2170 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2171 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2172 #ifdef CONFIG_X86_64
2173 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2174 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2175 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2176 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2177 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2178 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2179 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2180 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2183 regs->rip = vcpu->rip;
2184 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2187 * Don't leak debug flags in case they were set for guest debugging
2189 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2190 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2197 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2198 struct kvm_regs *regs)
2202 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2203 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2204 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2205 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2206 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2207 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2208 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2209 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2210 #ifdef CONFIG_X86_64
2211 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2212 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2213 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2214 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2215 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2216 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2217 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2218 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2221 vcpu->rip = regs->rip;
2222 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2224 kvm_x86_ops->decache_regs(vcpu);
2231 static void get_segment(struct kvm_vcpu *vcpu,
2232 struct kvm_segment *var, int seg)
2234 return kvm_x86_ops->get_segment(vcpu, var, seg);
2237 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2238 struct kvm_sregs *sregs)
2240 struct descriptor_table dt;
2245 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2246 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2247 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2248 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2249 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2250 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2252 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2253 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2255 kvm_x86_ops->get_idt(vcpu, &dt);
2256 sregs->idt.limit = dt.limit;
2257 sregs->idt.base = dt.base;
2258 kvm_x86_ops->get_gdt(vcpu, &dt);
2259 sregs->gdt.limit = dt.limit;
2260 sregs->gdt.base = dt.base;
2262 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2263 sregs->cr0 = vcpu->cr0;
2264 sregs->cr2 = vcpu->cr2;
2265 sregs->cr3 = vcpu->cr3;
2266 sregs->cr4 = vcpu->cr4;
2267 sregs->cr8 = get_cr8(vcpu);
2268 sregs->efer = vcpu->shadow_efer;
2269 sregs->apic_base = kvm_get_apic_base(vcpu);
2271 if (irqchip_in_kernel(vcpu->kvm)) {
2272 memset(sregs->interrupt_bitmap, 0,
2273 sizeof sregs->interrupt_bitmap);
2274 pending_vec = kvm_x86_ops->get_irq(vcpu);
2275 if (pending_vec >= 0)
2276 set_bit(pending_vec,
2277 (unsigned long *)sregs->interrupt_bitmap);
2279 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2280 sizeof sregs->interrupt_bitmap);
2287 static void set_segment(struct kvm_vcpu *vcpu,
2288 struct kvm_segment *var, int seg)
2290 return kvm_x86_ops->set_segment(vcpu, var, seg);
2293 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2294 struct kvm_sregs *sregs)
2296 int mmu_reset_needed = 0;
2297 int i, pending_vec, max_bits;
2298 struct descriptor_table dt;
2302 dt.limit = sregs->idt.limit;
2303 dt.base = sregs->idt.base;
2304 kvm_x86_ops->set_idt(vcpu, &dt);
2305 dt.limit = sregs->gdt.limit;
2306 dt.base = sregs->gdt.base;
2307 kvm_x86_ops->set_gdt(vcpu, &dt);
2309 vcpu->cr2 = sregs->cr2;
2310 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2311 vcpu->cr3 = sregs->cr3;
2313 set_cr8(vcpu, sregs->cr8);
2315 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2316 #ifdef CONFIG_X86_64
2317 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2319 kvm_set_apic_base(vcpu, sregs->apic_base);
2321 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2323 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2324 vcpu->cr0 = sregs->cr0;
2325 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2327 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2328 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2329 if (!is_long_mode(vcpu) && is_pae(vcpu))
2330 load_pdptrs(vcpu, vcpu->cr3);
2332 if (mmu_reset_needed)
2333 kvm_mmu_reset_context(vcpu);
2335 if (!irqchip_in_kernel(vcpu->kvm)) {
2336 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2337 sizeof vcpu->irq_pending);
2338 vcpu->irq_summary = 0;
2339 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2340 if (vcpu->irq_pending[i])
2341 __set_bit(i, &vcpu->irq_summary);
2343 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2344 pending_vec = find_first_bit(
2345 (const unsigned long *)sregs->interrupt_bitmap,
2347 /* Only pending external irq is handled here */
2348 if (pending_vec < max_bits) {
2349 kvm_x86_ops->set_irq(vcpu, pending_vec);
2350 pr_debug("Set back pending irq %d\n",
2355 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2356 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2357 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2358 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2359 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2360 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2362 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2363 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2370 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2372 struct kvm_segment cs;
2374 get_segment(vcpu, &cs, VCPU_SREG_CS);
2378 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2381 * Translate a guest virtual address to a guest physical address.
2383 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2384 struct kvm_translation *tr)
2386 unsigned long vaddr = tr->linear_address;
2390 mutex_lock(&vcpu->kvm->lock);
2391 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2392 tr->physical_address = gpa;
2393 tr->valid = gpa != UNMAPPED_GVA;
2396 mutex_unlock(&vcpu->kvm->lock);
2402 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2403 struct kvm_interrupt *irq)
2405 if (irq->irq < 0 || irq->irq >= 256)
2407 if (irqchip_in_kernel(vcpu->kvm))
2411 set_bit(irq->irq, vcpu->irq_pending);
2412 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2419 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2420 struct kvm_debug_guest *dbg)
2426 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2433 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2434 unsigned long address,
2437 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2438 unsigned long pgoff;
2441 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2443 page = virt_to_page(vcpu->run);
2444 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2445 page = virt_to_page(vcpu->pio_data);
2447 return NOPAGE_SIGBUS;
2450 *type = VM_FAULT_MINOR;
2455 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2456 .nopage = kvm_vcpu_nopage,
2459 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2461 vma->vm_ops = &kvm_vcpu_vm_ops;
2465 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2467 struct kvm_vcpu *vcpu = filp->private_data;
2469 fput(vcpu->kvm->filp);
2473 static struct file_operations kvm_vcpu_fops = {
2474 .release = kvm_vcpu_release,
2475 .unlocked_ioctl = kvm_vcpu_ioctl,
2476 .compat_ioctl = kvm_vcpu_ioctl,
2477 .mmap = kvm_vcpu_mmap,
2481 * Allocates an inode for the vcpu.
2483 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2486 struct inode *inode;
2489 r = anon_inode_getfd(&fd, &inode, &file,
2490 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2493 atomic_inc(&vcpu->kvm->filp->f_count);
2498 * Creates some virtual cpus. Good luck creating more than one.
2500 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2503 struct kvm_vcpu *vcpu;
2508 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2510 return PTR_ERR(vcpu);
2512 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2514 /* We do fxsave: this must be aligned. */
2515 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2518 r = kvm_x86_ops->vcpu_reset(vcpu);
2520 r = kvm_mmu_setup(vcpu);
2525 mutex_lock(&kvm->lock);
2526 if (kvm->vcpus[n]) {
2528 mutex_unlock(&kvm->lock);
2531 kvm->vcpus[n] = vcpu;
2532 mutex_unlock(&kvm->lock);
2534 /* Now it's all set up, let userspace reach it */
2535 r = create_vcpu_fd(vcpu);
2541 mutex_lock(&kvm->lock);
2542 kvm->vcpus[n] = NULL;
2543 mutex_unlock(&kvm->lock);
2547 kvm_mmu_unload(vcpu);
2551 kvm_x86_ops->vcpu_free(vcpu);
2555 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2558 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2559 vcpu->sigset_active = 1;
2560 vcpu->sigset = *sigset;
2562 vcpu->sigset_active = 0;
2567 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2568 * we have asm/x86/processor.h
2579 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2580 #ifdef CONFIG_X86_64
2581 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2583 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2587 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2589 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2593 memcpy(fpu->fpr, fxsave->st_space, 128);
2594 fpu->fcw = fxsave->cwd;
2595 fpu->fsw = fxsave->swd;
2596 fpu->ftwx = fxsave->twd;
2597 fpu->last_opcode = fxsave->fop;
2598 fpu->last_ip = fxsave->rip;
2599 fpu->last_dp = fxsave->rdp;
2600 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2607 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2609 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2613 memcpy(fxsave->st_space, fpu->fpr, 128);
2614 fxsave->cwd = fpu->fcw;
2615 fxsave->swd = fpu->fsw;
2616 fxsave->twd = fpu->ftwx;
2617 fxsave->fop = fpu->last_opcode;
2618 fxsave->rip = fpu->last_ip;
2619 fxsave->rdp = fpu->last_dp;
2620 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2627 static long kvm_vcpu_ioctl(struct file *filp,
2628 unsigned int ioctl, unsigned long arg)
2630 struct kvm_vcpu *vcpu = filp->private_data;
2631 void __user *argp = (void __user *)arg;
2639 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2641 case KVM_GET_REGS: {
2642 struct kvm_regs kvm_regs;
2644 memset(&kvm_regs, 0, sizeof kvm_regs);
2645 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2649 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2654 case KVM_SET_REGS: {
2655 struct kvm_regs kvm_regs;
2658 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2660 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2666 case KVM_GET_SREGS: {
2667 struct kvm_sregs kvm_sregs;
2669 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2670 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2674 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2679 case KVM_SET_SREGS: {
2680 struct kvm_sregs kvm_sregs;
2683 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2685 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2691 case KVM_TRANSLATE: {
2692 struct kvm_translation tr;
2695 if (copy_from_user(&tr, argp, sizeof tr))
2697 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2701 if (copy_to_user(argp, &tr, sizeof tr))
2706 case KVM_INTERRUPT: {
2707 struct kvm_interrupt irq;
2710 if (copy_from_user(&irq, argp, sizeof irq))
2712 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2718 case KVM_DEBUG_GUEST: {
2719 struct kvm_debug_guest dbg;
2722 if (copy_from_user(&dbg, argp, sizeof dbg))
2724 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2730 case KVM_SET_SIGNAL_MASK: {
2731 struct kvm_signal_mask __user *sigmask_arg = argp;
2732 struct kvm_signal_mask kvm_sigmask;
2733 sigset_t sigset, *p;
2738 if (copy_from_user(&kvm_sigmask, argp,
2739 sizeof kvm_sigmask))
2742 if (kvm_sigmask.len != sizeof sigset)
2745 if (copy_from_user(&sigset, sigmask_arg->sigset,
2750 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2756 memset(&fpu, 0, sizeof fpu);
2757 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2761 if (copy_to_user(argp, &fpu, sizeof fpu))
2770 if (copy_from_user(&fpu, argp, sizeof fpu))
2772 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2779 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2785 static long kvm_vm_ioctl(struct file *filp,
2786 unsigned int ioctl, unsigned long arg)
2788 struct kvm *kvm = filp->private_data;
2789 void __user *argp = (void __user *)arg;
2793 case KVM_CREATE_VCPU:
2794 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2798 case KVM_SET_USER_MEMORY_REGION: {
2799 struct kvm_userspace_memory_region kvm_userspace_mem;
2802 if (copy_from_user(&kvm_userspace_mem, argp,
2803 sizeof kvm_userspace_mem))
2806 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2811 case KVM_GET_DIRTY_LOG: {
2812 struct kvm_dirty_log log;
2815 if (copy_from_user(&log, argp, sizeof log))
2817 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2823 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2829 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2830 unsigned long address,
2833 struct kvm *kvm = vma->vm_file->private_data;
2834 unsigned long pgoff;
2837 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2838 if (!kvm_is_visible_gfn(kvm, pgoff))
2839 return NOPAGE_SIGBUS;
2840 page = gfn_to_page(kvm, pgoff);
2841 if (is_error_page(page)) {
2842 kvm_release_page(page);
2843 return NOPAGE_SIGBUS;
2846 *type = VM_FAULT_MINOR;
2851 static struct vm_operations_struct kvm_vm_vm_ops = {
2852 .nopage = kvm_vm_nopage,
2855 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2857 vma->vm_ops = &kvm_vm_vm_ops;
2861 static struct file_operations kvm_vm_fops = {
2862 .release = kvm_vm_release,
2863 .unlocked_ioctl = kvm_vm_ioctl,
2864 .compat_ioctl = kvm_vm_ioctl,
2865 .mmap = kvm_vm_mmap,
2868 static int kvm_dev_ioctl_create_vm(void)
2871 struct inode *inode;
2875 kvm = kvm_create_vm();
2877 return PTR_ERR(kvm);
2878 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2880 kvm_destroy_vm(kvm);
2889 static long kvm_dev_ioctl(struct file *filp,
2890 unsigned int ioctl, unsigned long arg)
2892 void __user *argp = (void __user *)arg;
2896 case KVM_GET_API_VERSION:
2900 r = KVM_API_VERSION;
2906 r = kvm_dev_ioctl_create_vm();
2908 case KVM_CHECK_EXTENSION: {
2909 int ext = (long)argp;
2912 case KVM_CAP_IRQCHIP:
2914 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2915 case KVM_CAP_USER_MEMORY:
2916 case KVM_CAP_SET_TSS_ADDR:
2925 case KVM_GET_VCPU_MMAP_SIZE:
2932 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2938 static struct file_operations kvm_chardev_ops = {
2939 .unlocked_ioctl = kvm_dev_ioctl,
2940 .compat_ioctl = kvm_dev_ioctl,
2943 static struct miscdevice kvm_dev = {
2950 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2953 static void decache_vcpus_on_cpu(int cpu)
2956 struct kvm_vcpu *vcpu;
2959 spin_lock(&kvm_lock);
2960 list_for_each_entry(vm, &vm_list, vm_list)
2961 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2962 vcpu = vm->vcpus[i];
2966 * If the vcpu is locked, then it is running on some
2967 * other cpu and therefore it is not cached on the
2970 * If it's not locked, check the last cpu it executed
2973 if (mutex_trylock(&vcpu->mutex)) {
2974 if (vcpu->cpu == cpu) {
2975 kvm_x86_ops->vcpu_decache(vcpu);
2978 mutex_unlock(&vcpu->mutex);
2981 spin_unlock(&kvm_lock);
2984 static void hardware_enable(void *junk)
2986 int cpu = raw_smp_processor_id();
2988 if (cpu_isset(cpu, cpus_hardware_enabled))
2990 cpu_set(cpu, cpus_hardware_enabled);
2991 kvm_x86_ops->hardware_enable(NULL);
2994 static void hardware_disable(void *junk)
2996 int cpu = raw_smp_processor_id();
2998 if (!cpu_isset(cpu, cpus_hardware_enabled))
3000 cpu_clear(cpu, cpus_hardware_enabled);
3001 decache_vcpus_on_cpu(cpu);
3002 kvm_x86_ops->hardware_disable(NULL);
3005 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3012 case CPU_DYING_FROZEN:
3013 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3015 hardware_disable(NULL);
3017 case CPU_UP_CANCELED:
3018 case CPU_UP_CANCELED_FROZEN:
3019 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3021 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3024 case CPU_ONLINE_FROZEN:
3025 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3027 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3033 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3036 if (val == SYS_RESTART) {
3038 * Some (well, at least mine) BIOSes hang on reboot if
3041 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3042 on_each_cpu(hardware_disable, NULL, 0, 1);
3047 static struct notifier_block kvm_reboot_notifier = {
3048 .notifier_call = kvm_reboot,
3052 void kvm_io_bus_init(struct kvm_io_bus *bus)
3054 memset(bus, 0, sizeof(*bus));
3057 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3061 for (i = 0; i < bus->dev_count; i++) {
3062 struct kvm_io_device *pos = bus->devs[i];
3064 kvm_iodevice_destructor(pos);
3068 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3072 for (i = 0; i < bus->dev_count; i++) {
3073 struct kvm_io_device *pos = bus->devs[i];
3075 if (pos->in_range(pos, addr))
3082 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3084 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3086 bus->devs[bus->dev_count++] = dev;
3089 static struct notifier_block kvm_cpu_notifier = {
3090 .notifier_call = kvm_cpu_hotplug,
3091 .priority = 20, /* must be > scheduler priority */
3094 static u64 stat_get(void *_offset)
3096 unsigned offset = (long)_offset;
3099 struct kvm_vcpu *vcpu;
3102 spin_lock(&kvm_lock);
3103 list_for_each_entry(kvm, &vm_list, vm_list)
3104 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3105 vcpu = kvm->vcpus[i];
3107 total += *(u32 *)((void *)vcpu + offset);
3109 spin_unlock(&kvm_lock);
3113 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3115 static __init void kvm_init_debug(void)
3117 struct kvm_stats_debugfs_item *p;
3119 debugfs_dir = debugfs_create_dir("kvm", NULL);
3120 for (p = debugfs_entries; p->name; ++p)
3121 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3122 (void *)(long)p->offset,
3126 static void kvm_exit_debug(void)
3128 struct kvm_stats_debugfs_item *p;
3130 for (p = debugfs_entries; p->name; ++p)
3131 debugfs_remove(p->dentry);
3132 debugfs_remove(debugfs_dir);
3135 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3137 hardware_disable(NULL);
3141 static int kvm_resume(struct sys_device *dev)
3143 hardware_enable(NULL);
3147 static struct sysdev_class kvm_sysdev_class = {
3149 .suspend = kvm_suspend,
3150 .resume = kvm_resume,
3153 static struct sys_device kvm_sysdev = {
3155 .cls = &kvm_sysdev_class,
3158 struct page *bad_page;
3161 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3163 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3166 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3168 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3170 kvm_x86_ops->vcpu_load(vcpu, cpu);
3173 static void kvm_sched_out(struct preempt_notifier *pn,
3174 struct task_struct *next)
3176 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3178 kvm_x86_ops->vcpu_put(vcpu);
3181 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3182 struct module *module)
3188 printk(KERN_ERR "kvm: already loaded the other module\n");
3192 if (!ops->cpu_has_kvm_support()) {
3193 printk(KERN_ERR "kvm: no hardware support\n");
3196 if (ops->disabled_by_bios()) {
3197 printk(KERN_ERR "kvm: disabled by bios\n");
3203 r = kvm_x86_ops->hardware_setup();
3207 for_each_online_cpu(cpu) {
3208 smp_call_function_single(cpu,
3209 kvm_x86_ops->check_processor_compatibility,
3215 on_each_cpu(hardware_enable, NULL, 0, 1);
3216 r = register_cpu_notifier(&kvm_cpu_notifier);
3219 register_reboot_notifier(&kvm_reboot_notifier);
3221 r = sysdev_class_register(&kvm_sysdev_class);
3225 r = sysdev_register(&kvm_sysdev);
3229 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3230 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3231 __alignof__(struct kvm_vcpu), 0, 0);
3232 if (!kvm_vcpu_cache) {
3237 kvm_chardev_ops.owner = module;
3239 r = misc_register(&kvm_dev);
3241 printk(KERN_ERR "kvm: misc device register failed\n");
3245 kvm_preempt_ops.sched_in = kvm_sched_in;
3246 kvm_preempt_ops.sched_out = kvm_sched_out;
3248 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3253 kmem_cache_destroy(kvm_vcpu_cache);
3255 sysdev_unregister(&kvm_sysdev);
3257 sysdev_class_unregister(&kvm_sysdev_class);
3259 unregister_reboot_notifier(&kvm_reboot_notifier);
3260 unregister_cpu_notifier(&kvm_cpu_notifier);
3262 on_each_cpu(hardware_disable, NULL, 0, 1);
3264 kvm_x86_ops->hardware_unsetup();
3269 EXPORT_SYMBOL_GPL(kvm_init_x86);
3271 void kvm_exit_x86(void)
3273 misc_deregister(&kvm_dev);
3274 kmem_cache_destroy(kvm_vcpu_cache);
3275 sysdev_unregister(&kvm_sysdev);
3276 sysdev_class_unregister(&kvm_sysdev_class);
3277 unregister_reboot_notifier(&kvm_reboot_notifier);
3278 unregister_cpu_notifier(&kvm_cpu_notifier);
3279 on_each_cpu(hardware_disable, NULL, 0, 1);
3280 kvm_x86_ops->hardware_unsetup();
3283 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3285 static __init int kvm_init(void)
3289 r = kvm_mmu_module_init();
3297 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3299 if (bad_page == NULL) {
3308 kvm_mmu_module_exit();
3313 static __exit void kvm_exit(void)
3316 __free_page(bad_page);
3317 kvm_mmu_module_exit();
3320 module_init(kvm_init)
3321 module_exit(kvm_exit)