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 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
96 static inline int valid_vcpu(int n)
98 return likely(n >= 0 && n < KVM_MAX_VCPUS);
101 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
103 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
106 vcpu->guest_fpu_loaded = 1;
107 fx_save(&vcpu->host_fx_image);
108 fx_restore(&vcpu->guest_fx_image);
110 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
112 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
114 if (!vcpu->guest_fpu_loaded)
117 vcpu->guest_fpu_loaded = 0;
118 fx_save(&vcpu->guest_fx_image);
119 fx_restore(&vcpu->host_fx_image);
121 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
124 * Switches to specified vcpu, until a matching vcpu_put()
126 void vcpu_load(struct kvm_vcpu *vcpu)
130 mutex_lock(&vcpu->mutex);
132 preempt_notifier_register(&vcpu->preempt_notifier);
133 kvm_arch_vcpu_load(vcpu, cpu);
137 void vcpu_put(struct kvm_vcpu *vcpu)
140 kvm_arch_vcpu_put(vcpu);
141 preempt_notifier_unregister(&vcpu->preempt_notifier);
143 mutex_unlock(&vcpu->mutex);
146 static void ack_flush(void *_completed)
150 void kvm_flush_remote_tlbs(struct kvm *kvm)
154 struct kvm_vcpu *vcpu;
157 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
158 vcpu = kvm->vcpus[i];
161 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
164 if (cpu != -1 && cpu != raw_smp_processor_id())
167 smp_call_function_mask(cpus, ack_flush, NULL, 1);
170 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
175 mutex_init(&vcpu->mutex);
177 vcpu->mmu.root_hpa = INVALID_PAGE;
180 if (!irqchip_in_kernel(kvm) || id == 0)
181 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
183 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
184 init_waitqueue_head(&vcpu->wq);
186 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
191 vcpu->run = page_address(page);
193 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
198 vcpu->pio_data = page_address(page);
200 r = kvm_mmu_create(vcpu);
202 goto fail_free_pio_data;
204 if (irqchip_in_kernel(kvm)) {
205 r = kvm_create_lapic(vcpu);
207 goto fail_mmu_destroy;
213 kvm_mmu_destroy(vcpu);
215 free_page((unsigned long)vcpu->pio_data);
217 free_page((unsigned long)vcpu->run);
221 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
223 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
225 kvm_free_lapic(vcpu);
226 kvm_mmu_destroy(vcpu);
227 free_page((unsigned long)vcpu->pio_data);
228 free_page((unsigned long)vcpu->run);
230 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
232 static struct kvm *kvm_create_vm(void)
234 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
237 return ERR_PTR(-ENOMEM);
239 kvm_io_bus_init(&kvm->pio_bus);
240 mutex_init(&kvm->lock);
241 INIT_LIST_HEAD(&kvm->active_mmu_pages);
242 kvm_io_bus_init(&kvm->mmio_bus);
243 spin_lock(&kvm_lock);
244 list_add(&kvm->vm_list, &vm_list);
245 spin_unlock(&kvm_lock);
250 * Free any memory in @free but not in @dont.
252 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
253 struct kvm_memory_slot *dont)
255 if (!dont || free->rmap != dont->rmap)
258 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
259 vfree(free->dirty_bitmap);
262 free->dirty_bitmap = NULL;
266 static void kvm_free_physmem(struct kvm *kvm)
270 for (i = 0; i < kvm->nmemslots; ++i)
271 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
274 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
278 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
279 if (vcpu->pio.guest_pages[i]) {
280 kvm_release_page(vcpu->pio.guest_pages[i]);
281 vcpu->pio.guest_pages[i] = NULL;
285 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
288 kvm_mmu_unload(vcpu);
292 static void kvm_free_vcpus(struct kvm *kvm)
297 * Unpin any mmu pages first.
299 for (i = 0; i < KVM_MAX_VCPUS; ++i)
301 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
302 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
304 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
305 kvm->vcpus[i] = NULL;
311 static void kvm_destroy_vm(struct kvm *kvm)
313 spin_lock(&kvm_lock);
314 list_del(&kvm->vm_list);
315 spin_unlock(&kvm_lock);
316 kvm_io_bus_destroy(&kvm->pio_bus);
317 kvm_io_bus_destroy(&kvm->mmio_bus);
321 kvm_free_physmem(kvm);
325 static int kvm_vm_release(struct inode *inode, struct file *filp)
327 struct kvm *kvm = filp->private_data;
333 static void inject_gp(struct kvm_vcpu *vcpu)
335 kvm_x86_ops->inject_gp(vcpu, 0);
338 void fx_init(struct kvm_vcpu *vcpu)
340 unsigned after_mxcsr_mask;
342 /* Initialize guest FPU by resetting ours and saving into guest's */
344 fx_save(&vcpu->host_fx_image);
346 fx_save(&vcpu->guest_fx_image);
347 fx_restore(&vcpu->host_fx_image);
350 vcpu->cr0 |= X86_CR0_ET;
351 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
352 vcpu->guest_fx_image.mxcsr = 0x1f80;
353 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
354 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
356 EXPORT_SYMBOL_GPL(fx_init);
359 * Allocate some memory and give it an address in the guest physical address
362 * Discontiguous memory is allowed, mostly for framebuffers.
364 * Must be called holding kvm->lock.
366 int __kvm_set_memory_region(struct kvm *kvm,
367 struct kvm_userspace_memory_region *mem,
372 unsigned long npages;
374 struct kvm_memory_slot *memslot;
375 struct kvm_memory_slot old, new;
378 /* General sanity checks */
379 if (mem->memory_size & (PAGE_SIZE - 1))
381 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
383 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
385 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
388 memslot = &kvm->memslots[mem->slot];
389 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
390 npages = mem->memory_size >> PAGE_SHIFT;
393 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
395 new = old = *memslot;
397 new.base_gfn = base_gfn;
399 new.flags = mem->flags;
401 /* Disallow changing a memory slot's size. */
403 if (npages && old.npages && npages != old.npages)
406 /* Check for overlaps */
408 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
409 struct kvm_memory_slot *s = &kvm->memslots[i];
413 if (!((base_gfn + npages <= s->base_gfn) ||
414 (base_gfn >= s->base_gfn + s->npages)))
418 /* Free page dirty bitmap if unneeded */
419 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
420 new.dirty_bitmap = NULL;
424 /* Allocate if a slot is being created */
425 if (npages && !new.rmap) {
426 new.rmap = vmalloc(npages * sizeof(struct page *));
431 memset(new.rmap, 0, npages * sizeof(*new.rmap));
433 new.user_alloc = user_alloc;
435 new.userspace_addr = mem->userspace_addr;
437 down_write(¤t->mm->mmap_sem);
438 new.userspace_addr = do_mmap(NULL, 0,
440 PROT_READ | PROT_WRITE,
441 MAP_SHARED | MAP_ANONYMOUS,
443 up_write(¤t->mm->mmap_sem);
445 if (IS_ERR((void *)new.userspace_addr))
449 if (!old.user_alloc && old.rmap) {
452 down_write(¤t->mm->mmap_sem);
453 ret = do_munmap(current->mm, old.userspace_addr,
454 old.npages * PAGE_SIZE);
455 up_write(¤t->mm->mmap_sem);
458 "kvm_vm_ioctl_set_memory_region: "
459 "failed to munmap memory\n");
463 /* Allocate page dirty bitmap if needed */
464 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
465 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
467 new.dirty_bitmap = vmalloc(dirty_bytes);
468 if (!new.dirty_bitmap)
470 memset(new.dirty_bitmap, 0, dirty_bytes);
473 if (mem->slot >= kvm->nmemslots)
474 kvm->nmemslots = mem->slot + 1;
476 if (!kvm->n_requested_mmu_pages) {
477 unsigned int n_pages;
480 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
481 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
484 unsigned int nr_mmu_pages;
486 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
487 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
488 nr_mmu_pages = max(nr_mmu_pages,
489 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
490 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
496 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
497 kvm_flush_remote_tlbs(kvm);
499 kvm_free_physmem_slot(&old, &new);
503 kvm_free_physmem_slot(&new, &old);
508 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
510 int kvm_set_memory_region(struct kvm *kvm,
511 struct kvm_userspace_memory_region *mem,
516 mutex_lock(&kvm->lock);
517 r = __kvm_set_memory_region(kvm, mem, user_alloc);
518 mutex_unlock(&kvm->lock);
521 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
523 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
525 kvm_userspace_memory_region *mem,
528 if (mem->slot >= KVM_MEMORY_SLOTS)
530 return kvm_set_memory_region(kvm, mem, user_alloc);
534 * Get (and clear) the dirty memory log for a memory slot.
536 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
537 struct kvm_dirty_log *log)
539 struct kvm_memory_slot *memslot;
542 unsigned long any = 0;
544 mutex_lock(&kvm->lock);
547 if (log->slot >= KVM_MEMORY_SLOTS)
550 memslot = &kvm->memslots[log->slot];
552 if (!memslot->dirty_bitmap)
555 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
557 for (i = 0; !any && i < n/sizeof(long); ++i)
558 any = memslot->dirty_bitmap[i];
561 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
564 /* If nothing is dirty, don't bother messing with page tables. */
566 kvm_mmu_slot_remove_write_access(kvm, log->slot);
567 kvm_flush_remote_tlbs(kvm);
568 memset(memslot->dirty_bitmap, 0, n);
574 mutex_unlock(&kvm->lock);
578 int is_error_page(struct page *page)
580 return page == bad_page;
582 EXPORT_SYMBOL_GPL(is_error_page);
584 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
587 struct kvm_mem_alias *alias;
589 for (i = 0; i < kvm->naliases; ++i) {
590 alias = &kvm->aliases[i];
591 if (gfn >= alias->base_gfn
592 && gfn < alias->base_gfn + alias->npages)
593 return alias->target_gfn + gfn - alias->base_gfn;
598 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
602 for (i = 0; i < kvm->nmemslots; ++i) {
603 struct kvm_memory_slot *memslot = &kvm->memslots[i];
605 if (gfn >= memslot->base_gfn
606 && gfn < memslot->base_gfn + memslot->npages)
612 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
614 gfn = unalias_gfn(kvm, gfn);
615 return __gfn_to_memslot(kvm, gfn);
618 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
622 gfn = unalias_gfn(kvm, gfn);
623 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
624 struct kvm_memory_slot *memslot = &kvm->memslots[i];
626 if (gfn >= memslot->base_gfn
627 && gfn < memslot->base_gfn + memslot->npages)
632 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
635 * Requires current->mm->mmap_sem to be held
637 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
639 struct kvm_memory_slot *slot;
640 struct page *page[1];
645 gfn = unalias_gfn(kvm, gfn);
646 slot = __gfn_to_memslot(kvm, gfn);
652 npages = get_user_pages(current, current->mm,
654 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
664 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
668 down_read(¤t->mm->mmap_sem);
669 page = __gfn_to_page(kvm, gfn);
670 up_read(¤t->mm->mmap_sem);
675 EXPORT_SYMBOL_GPL(gfn_to_page);
677 void kvm_release_page(struct page *page)
679 if (!PageReserved(page))
683 EXPORT_SYMBOL_GPL(kvm_release_page);
685 static int next_segment(unsigned long len, int offset)
687 if (len > PAGE_SIZE - offset)
688 return PAGE_SIZE - offset;
693 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
699 page = gfn_to_page(kvm, gfn);
700 if (is_error_page(page)) {
701 kvm_release_page(page);
704 page_virt = kmap_atomic(page, KM_USER0);
706 memcpy(data, page_virt + offset, len);
708 kunmap_atomic(page_virt, KM_USER0);
709 kvm_release_page(page);
712 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
714 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
716 gfn_t gfn = gpa >> PAGE_SHIFT;
718 int offset = offset_in_page(gpa);
721 while ((seg = next_segment(len, offset)) != 0) {
722 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
732 EXPORT_SYMBOL_GPL(kvm_read_guest);
734 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
740 page = gfn_to_page(kvm, gfn);
741 if (is_error_page(page)) {
742 kvm_release_page(page);
745 page_virt = kmap_atomic(page, KM_USER0);
747 memcpy(page_virt + offset, data, len);
749 kunmap_atomic(page_virt, KM_USER0);
750 mark_page_dirty(kvm, gfn);
751 kvm_release_page(page);
754 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
756 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
759 gfn_t gfn = gpa >> PAGE_SHIFT;
761 int offset = offset_in_page(gpa);
764 while ((seg = next_segment(len, offset)) != 0) {
765 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
776 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
781 page = gfn_to_page(kvm, gfn);
782 if (is_error_page(page)) {
783 kvm_release_page(page);
786 page_virt = kmap_atomic(page, KM_USER0);
788 memset(page_virt + offset, 0, len);
790 kunmap_atomic(page_virt, KM_USER0);
791 kvm_release_page(page);
794 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
796 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
798 gfn_t gfn = gpa >> PAGE_SHIFT;
800 int offset = offset_in_page(gpa);
803 while ((seg = next_segment(len, offset)) != 0) {
804 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
813 EXPORT_SYMBOL_GPL(kvm_clear_guest);
815 /* WARNING: Does not work on aliased pages. */
816 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
818 struct kvm_memory_slot *memslot;
820 memslot = __gfn_to_memslot(kvm, gfn);
821 if (memslot && memslot->dirty_bitmap) {
822 unsigned long rel_gfn = gfn - memslot->base_gfn;
825 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
826 set_bit(rel_gfn, memslot->dirty_bitmap);
830 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
833 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
837 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
839 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
841 DECLARE_WAITQUEUE(wait, current);
843 add_wait_queue(&vcpu->wq, &wait);
846 * We will block until either an interrupt or a signal wakes us up
848 while (!kvm_cpu_has_interrupt(vcpu)
849 && !signal_pending(current)
850 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
851 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
852 set_current_state(TASK_INTERRUPTIBLE);
858 __set_current_state(TASK_RUNNING);
859 remove_wait_queue(&vcpu->wq, &wait);
862 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
864 ++vcpu->stat.halt_exits;
865 if (irqchip_in_kernel(vcpu->kvm)) {
866 vcpu->mp_state = VCPU_MP_STATE_HALTED;
867 kvm_vcpu_block(vcpu);
868 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
872 vcpu->run->exit_reason = KVM_EXIT_HLT;
876 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
878 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
880 unsigned long nr, a0, a1, a2, a3, ret;
882 kvm_x86_ops->cache_regs(vcpu);
884 nr = vcpu->regs[VCPU_REGS_RAX];
885 a0 = vcpu->regs[VCPU_REGS_RBX];
886 a1 = vcpu->regs[VCPU_REGS_RCX];
887 a2 = vcpu->regs[VCPU_REGS_RDX];
888 a3 = vcpu->regs[VCPU_REGS_RSI];
890 if (!is_long_mode(vcpu)) {
903 vcpu->regs[VCPU_REGS_RAX] = ret;
904 kvm_x86_ops->decache_regs(vcpu);
907 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
909 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
914 mutex_lock(&vcpu->kvm->lock);
917 * Blow out the MMU to ensure that no other VCPU has an active mapping
918 * to ensure that the updated hypercall appears atomically across all
921 kvm_mmu_zap_all(vcpu->kvm);
923 kvm_x86_ops->cache_regs(vcpu);
924 kvm_x86_ops->patch_hypercall(vcpu, instruction);
925 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
929 mutex_unlock(&vcpu->kvm->lock);
934 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
936 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
939 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
941 struct descriptor_table dt = { limit, base };
943 kvm_x86_ops->set_gdt(vcpu, &dt);
946 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
948 struct descriptor_table dt = { limit, base };
950 kvm_x86_ops->set_idt(vcpu, &dt);
953 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
954 unsigned long *rflags)
957 *rflags = kvm_x86_ops->get_rflags(vcpu);
960 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
962 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
973 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
978 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
979 unsigned long *rflags)
983 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
984 *rflags = kvm_x86_ops->get_rflags(vcpu);
993 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
996 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1000 void kvm_resched(struct kvm_vcpu *vcpu)
1002 if (!need_resched())
1006 EXPORT_SYMBOL_GPL(kvm_resched);
1008 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1012 struct kvm_cpuid_entry *e, *best;
1014 kvm_x86_ops->cache_regs(vcpu);
1015 function = vcpu->regs[VCPU_REGS_RAX];
1016 vcpu->regs[VCPU_REGS_RAX] = 0;
1017 vcpu->regs[VCPU_REGS_RBX] = 0;
1018 vcpu->regs[VCPU_REGS_RCX] = 0;
1019 vcpu->regs[VCPU_REGS_RDX] = 0;
1021 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1022 e = &vcpu->cpuid_entries[i];
1023 if (e->function == function) {
1028 * Both basic or both extended?
1030 if (((e->function ^ function) & 0x80000000) == 0)
1031 if (!best || e->function > best->function)
1035 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1036 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1037 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1038 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1040 kvm_x86_ops->decache_regs(vcpu);
1041 kvm_x86_ops->skip_emulated_instruction(vcpu);
1043 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1045 static int pio_copy_data(struct kvm_vcpu *vcpu)
1047 void *p = vcpu->pio_data;
1050 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1052 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1055 free_pio_guest_pages(vcpu);
1058 q += vcpu->pio.guest_page_offset;
1059 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1061 memcpy(q, p, bytes);
1063 memcpy(p, q, bytes);
1064 q -= vcpu->pio.guest_page_offset;
1066 free_pio_guest_pages(vcpu);
1070 static int complete_pio(struct kvm_vcpu *vcpu)
1072 struct kvm_pio_request *io = &vcpu->pio;
1076 kvm_x86_ops->cache_regs(vcpu);
1080 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1084 r = pio_copy_data(vcpu);
1086 kvm_x86_ops->cache_regs(vcpu);
1093 delta *= io->cur_count;
1095 * The size of the register should really depend on
1096 * current address size.
1098 vcpu->regs[VCPU_REGS_RCX] -= delta;
1104 vcpu->regs[VCPU_REGS_RDI] += delta;
1106 vcpu->regs[VCPU_REGS_RSI] += delta;
1109 kvm_x86_ops->decache_regs(vcpu);
1111 io->count -= io->cur_count;
1117 static void kernel_pio(struct kvm_io_device *pio_dev,
1118 struct kvm_vcpu *vcpu,
1121 /* TODO: String I/O for in kernel device */
1123 mutex_lock(&vcpu->kvm->lock);
1125 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1129 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1132 mutex_unlock(&vcpu->kvm->lock);
1135 static void pio_string_write(struct kvm_io_device *pio_dev,
1136 struct kvm_vcpu *vcpu)
1138 struct kvm_pio_request *io = &vcpu->pio;
1139 void *pd = vcpu->pio_data;
1142 mutex_lock(&vcpu->kvm->lock);
1143 for (i = 0; i < io->cur_count; i++) {
1144 kvm_iodevice_write(pio_dev, io->port,
1149 mutex_unlock(&vcpu->kvm->lock);
1152 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1153 int size, unsigned port)
1155 struct kvm_io_device *pio_dev;
1157 vcpu->run->exit_reason = KVM_EXIT_IO;
1158 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1159 vcpu->run->io.size = vcpu->pio.size = size;
1160 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1161 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1162 vcpu->run->io.port = vcpu->pio.port = port;
1164 vcpu->pio.string = 0;
1166 vcpu->pio.guest_page_offset = 0;
1169 kvm_x86_ops->cache_regs(vcpu);
1170 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1171 kvm_x86_ops->decache_regs(vcpu);
1173 kvm_x86_ops->skip_emulated_instruction(vcpu);
1175 pio_dev = vcpu_find_pio_dev(vcpu, port);
1177 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1183 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1185 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1186 int size, unsigned long count, int down,
1187 gva_t address, int rep, unsigned port)
1189 unsigned now, in_page;
1193 struct kvm_io_device *pio_dev;
1195 vcpu->run->exit_reason = KVM_EXIT_IO;
1196 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1197 vcpu->run->io.size = vcpu->pio.size = size;
1198 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1199 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1200 vcpu->run->io.port = vcpu->pio.port = port;
1202 vcpu->pio.string = 1;
1203 vcpu->pio.down = down;
1204 vcpu->pio.guest_page_offset = offset_in_page(address);
1205 vcpu->pio.rep = rep;
1208 kvm_x86_ops->skip_emulated_instruction(vcpu);
1213 in_page = PAGE_SIZE - offset_in_page(address);
1215 in_page = offset_in_page(address) + size;
1216 now = min(count, (unsigned long)in_page / size);
1219 * String I/O straddles page boundary. Pin two guest pages
1220 * so that we satisfy atomicity constraints. Do just one
1221 * transaction to avoid complexity.
1228 * String I/O in reverse. Yuck. Kill the guest, fix later.
1230 pr_unimpl(vcpu, "guest string pio down\n");
1234 vcpu->run->io.count = now;
1235 vcpu->pio.cur_count = now;
1237 if (vcpu->pio.cur_count == vcpu->pio.count)
1238 kvm_x86_ops->skip_emulated_instruction(vcpu);
1240 for (i = 0; i < nr_pages; ++i) {
1241 mutex_lock(&vcpu->kvm->lock);
1242 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1243 vcpu->pio.guest_pages[i] = page;
1244 mutex_unlock(&vcpu->kvm->lock);
1247 free_pio_guest_pages(vcpu);
1252 pio_dev = vcpu_find_pio_dev(vcpu, port);
1253 if (!vcpu->pio.in) {
1254 /* string PIO write */
1255 ret = pio_copy_data(vcpu);
1256 if (ret >= 0 && pio_dev) {
1257 pio_string_write(pio_dev, vcpu);
1259 if (vcpu->pio.count == 0)
1263 pr_unimpl(vcpu, "no string pio read support yet, "
1264 "port %x size %d count %ld\n",
1269 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1272 * Check if userspace requested an interrupt window, and that the
1273 * interrupt window is open.
1275 * No need to exit to userspace if we already have an interrupt queued.
1277 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1278 struct kvm_run *kvm_run)
1280 return (!vcpu->irq_summary &&
1281 kvm_run->request_interrupt_window &&
1282 vcpu->interrupt_window_open &&
1283 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1286 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1287 struct kvm_run *kvm_run)
1289 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1290 kvm_run->cr8 = get_cr8(vcpu);
1291 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1292 if (irqchip_in_kernel(vcpu->kvm))
1293 kvm_run->ready_for_interrupt_injection = 1;
1295 kvm_run->ready_for_interrupt_injection =
1296 (vcpu->interrupt_window_open &&
1297 vcpu->irq_summary == 0);
1300 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1304 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1305 pr_debug("vcpu %d received sipi with vector # %x\n",
1306 vcpu->vcpu_id, vcpu->sipi_vector);
1307 kvm_lapic_reset(vcpu);
1308 r = kvm_x86_ops->vcpu_reset(vcpu);
1311 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1315 if (vcpu->guest_debug.enabled)
1316 kvm_x86_ops->guest_debug_pre(vcpu);
1319 r = kvm_mmu_reload(vcpu);
1323 kvm_inject_pending_timer_irqs(vcpu);
1327 kvm_x86_ops->prepare_guest_switch(vcpu);
1328 kvm_load_guest_fpu(vcpu);
1330 local_irq_disable();
1332 if (signal_pending(current)) {
1336 kvm_run->exit_reason = KVM_EXIT_INTR;
1337 ++vcpu->stat.signal_exits;
1341 if (irqchip_in_kernel(vcpu->kvm))
1342 kvm_x86_ops->inject_pending_irq(vcpu);
1343 else if (!vcpu->mmio_read_completed)
1344 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1346 vcpu->guest_mode = 1;
1350 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1351 kvm_x86_ops->tlb_flush(vcpu);
1353 kvm_x86_ops->run(vcpu, kvm_run);
1355 vcpu->guest_mode = 0;
1361 * We must have an instruction between local_irq_enable() and
1362 * kvm_guest_exit(), so the timer interrupt isn't delayed by
1363 * the interrupt shadow. The stat.exits increment will do nicely.
1364 * But we need to prevent reordering, hence this barrier():
1373 * Profile KVM exit RIPs:
1375 if (unlikely(prof_on == KVM_PROFILING)) {
1376 kvm_x86_ops->cache_regs(vcpu);
1377 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1380 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1383 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1385 kvm_run->exit_reason = KVM_EXIT_INTR;
1386 ++vcpu->stat.request_irq_exits;
1389 if (!need_resched()) {
1390 ++vcpu->stat.light_exits;
1401 post_kvm_run_save(vcpu, kvm_run);
1407 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1414 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1415 kvm_vcpu_block(vcpu);
1420 if (vcpu->sigset_active)
1421 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1423 /* re-sync apic's tpr */
1424 if (!irqchip_in_kernel(vcpu->kvm))
1425 set_cr8(vcpu, kvm_run->cr8);
1427 if (vcpu->pio.cur_count) {
1428 r = complete_pio(vcpu);
1432 #if CONFIG_HAS_IOMEM
1433 if (vcpu->mmio_needed) {
1434 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1435 vcpu->mmio_read_completed = 1;
1436 vcpu->mmio_needed = 0;
1437 r = emulate_instruction(vcpu, kvm_run,
1438 vcpu->mmio_fault_cr2, 0, 1);
1439 if (r == EMULATE_DO_MMIO) {
1441 * Read-modify-write. Back to userspace.
1448 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1449 kvm_x86_ops->cache_regs(vcpu);
1450 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1451 kvm_x86_ops->decache_regs(vcpu);
1454 r = __vcpu_run(vcpu, kvm_run);
1457 if (vcpu->sigset_active)
1458 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1464 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1465 struct kvm_regs *regs)
1469 kvm_x86_ops->cache_regs(vcpu);
1471 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1472 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1473 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1474 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1475 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1476 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1477 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1478 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1479 #ifdef CONFIG_X86_64
1480 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1481 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1482 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1483 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1484 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1485 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1486 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1487 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1490 regs->rip = vcpu->rip;
1491 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
1494 * Don't leak debug flags in case they were set for guest debugging
1496 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1497 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1504 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1505 struct kvm_regs *regs)
1509 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1510 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1511 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1512 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1513 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1514 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1515 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1516 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1517 #ifdef CONFIG_X86_64
1518 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1519 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1520 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1521 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1522 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1523 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1524 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1525 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1528 vcpu->rip = regs->rip;
1529 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
1531 kvm_x86_ops->decache_regs(vcpu);
1538 static void get_segment(struct kvm_vcpu *vcpu,
1539 struct kvm_segment *var, int seg)
1541 return kvm_x86_ops->get_segment(vcpu, var, seg);
1544 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1545 struct kvm_sregs *sregs)
1547 struct descriptor_table dt;
1552 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1553 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1554 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1555 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1556 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1557 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1559 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1560 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1562 kvm_x86_ops->get_idt(vcpu, &dt);
1563 sregs->idt.limit = dt.limit;
1564 sregs->idt.base = dt.base;
1565 kvm_x86_ops->get_gdt(vcpu, &dt);
1566 sregs->gdt.limit = dt.limit;
1567 sregs->gdt.base = dt.base;
1569 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1570 sregs->cr0 = vcpu->cr0;
1571 sregs->cr2 = vcpu->cr2;
1572 sregs->cr3 = vcpu->cr3;
1573 sregs->cr4 = vcpu->cr4;
1574 sregs->cr8 = get_cr8(vcpu);
1575 sregs->efer = vcpu->shadow_efer;
1576 sregs->apic_base = kvm_get_apic_base(vcpu);
1578 if (irqchip_in_kernel(vcpu->kvm)) {
1579 memset(sregs->interrupt_bitmap, 0,
1580 sizeof sregs->interrupt_bitmap);
1581 pending_vec = kvm_x86_ops->get_irq(vcpu);
1582 if (pending_vec >= 0)
1583 set_bit(pending_vec,
1584 (unsigned long *)sregs->interrupt_bitmap);
1586 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1587 sizeof sregs->interrupt_bitmap);
1594 static void set_segment(struct kvm_vcpu *vcpu,
1595 struct kvm_segment *var, int seg)
1597 return kvm_x86_ops->set_segment(vcpu, var, seg);
1600 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1601 struct kvm_sregs *sregs)
1603 int mmu_reset_needed = 0;
1604 int i, pending_vec, max_bits;
1605 struct descriptor_table dt;
1609 dt.limit = sregs->idt.limit;
1610 dt.base = sregs->idt.base;
1611 kvm_x86_ops->set_idt(vcpu, &dt);
1612 dt.limit = sregs->gdt.limit;
1613 dt.base = sregs->gdt.base;
1614 kvm_x86_ops->set_gdt(vcpu, &dt);
1616 vcpu->cr2 = sregs->cr2;
1617 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1618 vcpu->cr3 = sregs->cr3;
1620 set_cr8(vcpu, sregs->cr8);
1622 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1623 #ifdef CONFIG_X86_64
1624 kvm_x86_ops->set_efer(vcpu, sregs->efer);
1626 kvm_set_apic_base(vcpu, sregs->apic_base);
1628 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1630 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1631 vcpu->cr0 = sregs->cr0;
1632 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
1634 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1635 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1636 if (!is_long_mode(vcpu) && is_pae(vcpu))
1637 load_pdptrs(vcpu, vcpu->cr3);
1639 if (mmu_reset_needed)
1640 kvm_mmu_reset_context(vcpu);
1642 if (!irqchip_in_kernel(vcpu->kvm)) {
1643 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1644 sizeof vcpu->irq_pending);
1645 vcpu->irq_summary = 0;
1646 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
1647 if (vcpu->irq_pending[i])
1648 __set_bit(i, &vcpu->irq_summary);
1650 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
1651 pending_vec = find_first_bit(
1652 (const unsigned long *)sregs->interrupt_bitmap,
1654 /* Only pending external irq is handled here */
1655 if (pending_vec < max_bits) {
1656 kvm_x86_ops->set_irq(vcpu, pending_vec);
1657 pr_debug("Set back pending irq %d\n",
1662 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1663 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1664 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1665 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1666 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1667 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1669 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1670 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1677 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1679 struct kvm_segment cs;
1681 get_segment(vcpu, &cs, VCPU_SREG_CS);
1685 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
1688 * Translate a guest virtual address to a guest physical address.
1690 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1691 struct kvm_translation *tr)
1693 unsigned long vaddr = tr->linear_address;
1697 mutex_lock(&vcpu->kvm->lock);
1698 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1699 tr->physical_address = gpa;
1700 tr->valid = gpa != UNMAPPED_GVA;
1703 mutex_unlock(&vcpu->kvm->lock);
1709 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1710 struct kvm_interrupt *irq)
1712 if (irq->irq < 0 || irq->irq >= 256)
1714 if (irqchip_in_kernel(vcpu->kvm))
1718 set_bit(irq->irq, vcpu->irq_pending);
1719 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1726 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1727 struct kvm_debug_guest *dbg)
1733 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
1740 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1741 unsigned long address,
1744 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1745 unsigned long pgoff;
1748 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1750 page = virt_to_page(vcpu->run);
1751 else if (pgoff == KVM_PIO_PAGE_OFFSET)
1752 page = virt_to_page(vcpu->pio_data);
1754 return NOPAGE_SIGBUS;
1757 *type = VM_FAULT_MINOR;
1762 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1763 .nopage = kvm_vcpu_nopage,
1766 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1768 vma->vm_ops = &kvm_vcpu_vm_ops;
1772 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1774 struct kvm_vcpu *vcpu = filp->private_data;
1776 fput(vcpu->kvm->filp);
1780 static struct file_operations kvm_vcpu_fops = {
1781 .release = kvm_vcpu_release,
1782 .unlocked_ioctl = kvm_vcpu_ioctl,
1783 .compat_ioctl = kvm_vcpu_ioctl,
1784 .mmap = kvm_vcpu_mmap,
1788 * Allocates an inode for the vcpu.
1790 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1793 struct inode *inode;
1796 r = anon_inode_getfd(&fd, &inode, &file,
1797 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
1800 atomic_inc(&vcpu->kvm->filp->f_count);
1805 * Creates some virtual cpus. Good luck creating more than one.
1807 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1810 struct kvm_vcpu *vcpu;
1815 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
1817 return PTR_ERR(vcpu);
1819 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1821 /* We do fxsave: this must be aligned. */
1822 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
1825 r = kvm_x86_ops->vcpu_reset(vcpu);
1827 r = kvm_mmu_setup(vcpu);
1832 mutex_lock(&kvm->lock);
1833 if (kvm->vcpus[n]) {
1835 mutex_unlock(&kvm->lock);
1838 kvm->vcpus[n] = vcpu;
1839 mutex_unlock(&kvm->lock);
1841 /* Now it's all set up, let userspace reach it */
1842 r = create_vcpu_fd(vcpu);
1848 mutex_lock(&kvm->lock);
1849 kvm->vcpus[n] = NULL;
1850 mutex_unlock(&kvm->lock);
1854 kvm_mmu_unload(vcpu);
1858 kvm_x86_ops->vcpu_free(vcpu);
1862 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1865 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1866 vcpu->sigset_active = 1;
1867 vcpu->sigset = *sigset;
1869 vcpu->sigset_active = 0;
1874 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
1875 * we have asm/x86/processor.h
1886 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
1887 #ifdef CONFIG_X86_64
1888 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
1890 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
1894 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1896 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
1900 memcpy(fpu->fpr, fxsave->st_space, 128);
1901 fpu->fcw = fxsave->cwd;
1902 fpu->fsw = fxsave->swd;
1903 fpu->ftwx = fxsave->twd;
1904 fpu->last_opcode = fxsave->fop;
1905 fpu->last_ip = fxsave->rip;
1906 fpu->last_dp = fxsave->rdp;
1907 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
1914 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1916 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
1920 memcpy(fxsave->st_space, fpu->fpr, 128);
1921 fxsave->cwd = fpu->fcw;
1922 fxsave->swd = fpu->fsw;
1923 fxsave->twd = fpu->ftwx;
1924 fxsave->fop = fpu->last_opcode;
1925 fxsave->rip = fpu->last_ip;
1926 fxsave->rdp = fpu->last_dp;
1927 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
1934 static long kvm_vcpu_ioctl(struct file *filp,
1935 unsigned int ioctl, unsigned long arg)
1937 struct kvm_vcpu *vcpu = filp->private_data;
1938 void __user *argp = (void __user *)arg;
1946 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
1948 case KVM_GET_REGS: {
1949 struct kvm_regs kvm_regs;
1951 memset(&kvm_regs, 0, sizeof kvm_regs);
1952 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
1956 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1961 case KVM_SET_REGS: {
1962 struct kvm_regs kvm_regs;
1965 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1967 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
1973 case KVM_GET_SREGS: {
1974 struct kvm_sregs kvm_sregs;
1976 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1977 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1981 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1986 case KVM_SET_SREGS: {
1987 struct kvm_sregs kvm_sregs;
1990 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1992 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1998 case KVM_TRANSLATE: {
1999 struct kvm_translation tr;
2002 if (copy_from_user(&tr, argp, sizeof tr))
2004 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2008 if (copy_to_user(argp, &tr, sizeof tr))
2013 case KVM_INTERRUPT: {
2014 struct kvm_interrupt irq;
2017 if (copy_from_user(&irq, argp, sizeof irq))
2019 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2025 case KVM_DEBUG_GUEST: {
2026 struct kvm_debug_guest dbg;
2029 if (copy_from_user(&dbg, argp, sizeof dbg))
2031 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2037 case KVM_SET_SIGNAL_MASK: {
2038 struct kvm_signal_mask __user *sigmask_arg = argp;
2039 struct kvm_signal_mask kvm_sigmask;
2040 sigset_t sigset, *p;
2045 if (copy_from_user(&kvm_sigmask, argp,
2046 sizeof kvm_sigmask))
2049 if (kvm_sigmask.len != sizeof sigset)
2052 if (copy_from_user(&sigset, sigmask_arg->sigset,
2057 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2063 memset(&fpu, 0, sizeof fpu);
2064 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2068 if (copy_to_user(argp, &fpu, sizeof fpu))
2077 if (copy_from_user(&fpu, argp, sizeof fpu))
2079 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2086 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2092 static long kvm_vm_ioctl(struct file *filp,
2093 unsigned int ioctl, unsigned long arg)
2095 struct kvm *kvm = filp->private_data;
2096 void __user *argp = (void __user *)arg;
2100 case KVM_CREATE_VCPU:
2101 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2105 case KVM_SET_USER_MEMORY_REGION: {
2106 struct kvm_userspace_memory_region kvm_userspace_mem;
2109 if (copy_from_user(&kvm_userspace_mem, argp,
2110 sizeof kvm_userspace_mem))
2113 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2118 case KVM_GET_DIRTY_LOG: {
2119 struct kvm_dirty_log log;
2122 if (copy_from_user(&log, argp, sizeof log))
2124 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2130 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2136 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2137 unsigned long address,
2140 struct kvm *kvm = vma->vm_file->private_data;
2141 unsigned long pgoff;
2144 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2145 if (!kvm_is_visible_gfn(kvm, pgoff))
2146 return NOPAGE_SIGBUS;
2147 /* current->mm->mmap_sem is already held so call lockless version */
2148 page = __gfn_to_page(kvm, pgoff);
2149 if (is_error_page(page)) {
2150 kvm_release_page(page);
2151 return NOPAGE_SIGBUS;
2154 *type = VM_FAULT_MINOR;
2159 static struct vm_operations_struct kvm_vm_vm_ops = {
2160 .nopage = kvm_vm_nopage,
2163 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2165 vma->vm_ops = &kvm_vm_vm_ops;
2169 static struct file_operations kvm_vm_fops = {
2170 .release = kvm_vm_release,
2171 .unlocked_ioctl = kvm_vm_ioctl,
2172 .compat_ioctl = kvm_vm_ioctl,
2173 .mmap = kvm_vm_mmap,
2176 static int kvm_dev_ioctl_create_vm(void)
2179 struct inode *inode;
2183 kvm = kvm_create_vm();
2185 return PTR_ERR(kvm);
2186 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2188 kvm_destroy_vm(kvm);
2197 static long kvm_dev_ioctl(struct file *filp,
2198 unsigned int ioctl, unsigned long arg)
2200 void __user *argp = (void __user *)arg;
2204 case KVM_GET_API_VERSION:
2208 r = KVM_API_VERSION;
2214 r = kvm_dev_ioctl_create_vm();
2216 case KVM_CHECK_EXTENSION: {
2217 int ext = (long)argp;
2220 case KVM_CAP_IRQCHIP:
2222 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2223 case KVM_CAP_USER_MEMORY:
2224 case KVM_CAP_SET_TSS_ADDR:
2233 case KVM_GET_VCPU_MMAP_SIZE:
2240 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2246 static struct file_operations kvm_chardev_ops = {
2247 .unlocked_ioctl = kvm_dev_ioctl,
2248 .compat_ioctl = kvm_dev_ioctl,
2251 static struct miscdevice kvm_dev = {
2258 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2261 static void decache_vcpus_on_cpu(int cpu)
2264 struct kvm_vcpu *vcpu;
2267 spin_lock(&kvm_lock);
2268 list_for_each_entry(vm, &vm_list, vm_list)
2269 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2270 vcpu = vm->vcpus[i];
2274 * If the vcpu is locked, then it is running on some
2275 * other cpu and therefore it is not cached on the
2278 * If it's not locked, check the last cpu it executed
2281 if (mutex_trylock(&vcpu->mutex)) {
2282 if (vcpu->cpu == cpu) {
2283 kvm_x86_ops->vcpu_decache(vcpu);
2286 mutex_unlock(&vcpu->mutex);
2289 spin_unlock(&kvm_lock);
2292 static void hardware_enable(void *junk)
2294 int cpu = raw_smp_processor_id();
2296 if (cpu_isset(cpu, cpus_hardware_enabled))
2298 cpu_set(cpu, cpus_hardware_enabled);
2299 kvm_x86_ops->hardware_enable(NULL);
2302 static void hardware_disable(void *junk)
2304 int cpu = raw_smp_processor_id();
2306 if (!cpu_isset(cpu, cpus_hardware_enabled))
2308 cpu_clear(cpu, cpus_hardware_enabled);
2309 decache_vcpus_on_cpu(cpu);
2310 kvm_x86_ops->hardware_disable(NULL);
2313 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2320 case CPU_DYING_FROZEN:
2321 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2323 hardware_disable(NULL);
2325 case CPU_UP_CANCELED:
2326 case CPU_UP_CANCELED_FROZEN:
2327 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2329 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2332 case CPU_ONLINE_FROZEN:
2333 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2335 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2341 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2344 if (val == SYS_RESTART) {
2346 * Some (well, at least mine) BIOSes hang on reboot if
2349 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2350 on_each_cpu(hardware_disable, NULL, 0, 1);
2355 static struct notifier_block kvm_reboot_notifier = {
2356 .notifier_call = kvm_reboot,
2360 void kvm_io_bus_init(struct kvm_io_bus *bus)
2362 memset(bus, 0, sizeof(*bus));
2365 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2369 for (i = 0; i < bus->dev_count; i++) {
2370 struct kvm_io_device *pos = bus->devs[i];
2372 kvm_iodevice_destructor(pos);
2376 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2380 for (i = 0; i < bus->dev_count; i++) {
2381 struct kvm_io_device *pos = bus->devs[i];
2383 if (pos->in_range(pos, addr))
2390 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2392 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2394 bus->devs[bus->dev_count++] = dev;
2397 static struct notifier_block kvm_cpu_notifier = {
2398 .notifier_call = kvm_cpu_hotplug,
2399 .priority = 20, /* must be > scheduler priority */
2402 static u64 stat_get(void *_offset)
2404 unsigned offset = (long)_offset;
2407 struct kvm_vcpu *vcpu;
2410 spin_lock(&kvm_lock);
2411 list_for_each_entry(kvm, &vm_list, vm_list)
2412 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2413 vcpu = kvm->vcpus[i];
2415 total += *(u32 *)((void *)vcpu + offset);
2417 spin_unlock(&kvm_lock);
2421 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
2423 static __init void kvm_init_debug(void)
2425 struct kvm_stats_debugfs_item *p;
2427 debugfs_dir = debugfs_create_dir("kvm", NULL);
2428 for (p = debugfs_entries; p->name; ++p)
2429 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
2430 (void *)(long)p->offset,
2434 static void kvm_exit_debug(void)
2436 struct kvm_stats_debugfs_item *p;
2438 for (p = debugfs_entries; p->name; ++p)
2439 debugfs_remove(p->dentry);
2440 debugfs_remove(debugfs_dir);
2443 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2445 hardware_disable(NULL);
2449 static int kvm_resume(struct sys_device *dev)
2451 hardware_enable(NULL);
2455 static struct sysdev_class kvm_sysdev_class = {
2457 .suspend = kvm_suspend,
2458 .resume = kvm_resume,
2461 static struct sys_device kvm_sysdev = {
2463 .cls = &kvm_sysdev_class,
2466 struct page *bad_page;
2469 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2471 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2474 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2476 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2478 kvm_x86_ops->vcpu_load(vcpu, cpu);
2481 static void kvm_sched_out(struct preempt_notifier *pn,
2482 struct task_struct *next)
2484 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2486 kvm_x86_ops->vcpu_put(vcpu);
2489 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
2490 struct module *module)
2496 printk(KERN_ERR "kvm: already loaded the other module\n");
2500 if (!ops->cpu_has_kvm_support()) {
2501 printk(KERN_ERR "kvm: no hardware support\n");
2504 if (ops->disabled_by_bios()) {
2505 printk(KERN_ERR "kvm: disabled by bios\n");
2511 r = kvm_x86_ops->hardware_setup();
2515 for_each_online_cpu(cpu) {
2516 smp_call_function_single(cpu,
2517 kvm_x86_ops->check_processor_compatibility,
2523 on_each_cpu(hardware_enable, NULL, 0, 1);
2524 r = register_cpu_notifier(&kvm_cpu_notifier);
2527 register_reboot_notifier(&kvm_reboot_notifier);
2529 r = sysdev_class_register(&kvm_sysdev_class);
2533 r = sysdev_register(&kvm_sysdev);
2537 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2538 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2539 __alignof__(struct kvm_vcpu), 0, 0);
2540 if (!kvm_vcpu_cache) {
2545 kvm_chardev_ops.owner = module;
2547 r = misc_register(&kvm_dev);
2549 printk(KERN_ERR "kvm: misc device register failed\n");
2553 kvm_preempt_ops.sched_in = kvm_sched_in;
2554 kvm_preempt_ops.sched_out = kvm_sched_out;
2556 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2561 kmem_cache_destroy(kvm_vcpu_cache);
2563 sysdev_unregister(&kvm_sysdev);
2565 sysdev_class_unregister(&kvm_sysdev_class);
2567 unregister_reboot_notifier(&kvm_reboot_notifier);
2568 unregister_cpu_notifier(&kvm_cpu_notifier);
2570 on_each_cpu(hardware_disable, NULL, 0, 1);
2572 kvm_x86_ops->hardware_unsetup();
2577 EXPORT_SYMBOL_GPL(kvm_init_x86);
2579 void kvm_exit_x86(void)
2581 misc_deregister(&kvm_dev);
2582 kmem_cache_destroy(kvm_vcpu_cache);
2583 sysdev_unregister(&kvm_sysdev);
2584 sysdev_class_unregister(&kvm_sysdev_class);
2585 unregister_reboot_notifier(&kvm_reboot_notifier);
2586 unregister_cpu_notifier(&kvm_cpu_notifier);
2587 on_each_cpu(hardware_disable, NULL, 0, 1);
2588 kvm_x86_ops->hardware_unsetup();
2591 EXPORT_SYMBOL_GPL(kvm_exit_x86);
2593 static __init int kvm_init(void)
2597 r = kvm_mmu_module_init();
2605 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2607 if (bad_page == NULL) {
2616 kvm_mmu_module_exit();
2621 static __exit void kvm_exit(void)
2624 __free_page(bad_page);
2625 kvm_mmu_module_exit();
2628 module_init(kvm_init)
2629 module_exit(kvm_exit)