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 static struct dentry *debugfs_dir;
68 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
71 static inline int valid_vcpu(int n)
73 return likely(n >= 0 && n < KVM_MAX_VCPUS);
77 * Switches to specified vcpu, until a matching vcpu_put()
79 void vcpu_load(struct kvm_vcpu *vcpu)
83 mutex_lock(&vcpu->mutex);
85 preempt_notifier_register(&vcpu->preempt_notifier);
86 kvm_arch_vcpu_load(vcpu, cpu);
90 void vcpu_put(struct kvm_vcpu *vcpu)
93 kvm_arch_vcpu_put(vcpu);
94 preempt_notifier_unregister(&vcpu->preempt_notifier);
96 mutex_unlock(&vcpu->mutex);
99 static void ack_flush(void *_completed)
103 void kvm_flush_remote_tlbs(struct kvm *kvm)
107 struct kvm_vcpu *vcpu;
110 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
111 vcpu = kvm->vcpus[i];
114 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
117 if (cpu != -1 && cpu != raw_smp_processor_id())
120 smp_call_function_mask(cpus, ack_flush, NULL, 1);
123 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
128 mutex_init(&vcpu->mutex);
130 vcpu->mmu.root_hpa = INVALID_PAGE;
133 if (!irqchip_in_kernel(kvm) || id == 0)
134 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
136 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
137 init_waitqueue_head(&vcpu->wq);
139 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
144 vcpu->run = page_address(page);
146 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
151 vcpu->pio_data = page_address(page);
153 r = kvm_mmu_create(vcpu);
155 goto fail_free_pio_data;
157 if (irqchip_in_kernel(kvm)) {
158 r = kvm_create_lapic(vcpu);
160 goto fail_mmu_destroy;
166 kvm_mmu_destroy(vcpu);
168 free_page((unsigned long)vcpu->pio_data);
170 free_page((unsigned long)vcpu->run);
174 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
176 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
178 kvm_free_lapic(vcpu);
179 kvm_mmu_destroy(vcpu);
180 free_page((unsigned long)vcpu->pio_data);
181 free_page((unsigned long)vcpu->run);
183 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
185 static struct kvm *kvm_create_vm(void)
187 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
190 return ERR_PTR(-ENOMEM);
192 kvm_io_bus_init(&kvm->pio_bus);
193 mutex_init(&kvm->lock);
194 INIT_LIST_HEAD(&kvm->active_mmu_pages);
195 kvm_io_bus_init(&kvm->mmio_bus);
196 spin_lock(&kvm_lock);
197 list_add(&kvm->vm_list, &vm_list);
198 spin_unlock(&kvm_lock);
203 * Free any memory in @free but not in @dont.
205 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
206 struct kvm_memory_slot *dont)
208 if (!dont || free->rmap != dont->rmap)
211 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
212 vfree(free->dirty_bitmap);
215 free->dirty_bitmap = NULL;
219 static void kvm_free_physmem(struct kvm *kvm)
223 for (i = 0; i < kvm->nmemslots; ++i)
224 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
227 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
230 kvm_mmu_unload(vcpu);
234 static void kvm_free_vcpus(struct kvm *kvm)
239 * Unpin any mmu pages first.
241 for (i = 0; i < KVM_MAX_VCPUS; ++i)
243 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
244 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
246 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
247 kvm->vcpus[i] = NULL;
253 static void kvm_destroy_vm(struct kvm *kvm)
255 spin_lock(&kvm_lock);
256 list_del(&kvm->vm_list);
257 spin_unlock(&kvm_lock);
258 kvm_io_bus_destroy(&kvm->pio_bus);
259 kvm_io_bus_destroy(&kvm->mmio_bus);
263 kvm_free_physmem(kvm);
267 static int kvm_vm_release(struct inode *inode, struct file *filp)
269 struct kvm *kvm = filp->private_data;
276 * Allocate some memory and give it an address in the guest physical address
279 * Discontiguous memory is allowed, mostly for framebuffers.
281 * Must be called holding kvm->lock.
283 int __kvm_set_memory_region(struct kvm *kvm,
284 struct kvm_userspace_memory_region *mem,
289 unsigned long npages;
291 struct kvm_memory_slot *memslot;
292 struct kvm_memory_slot old, new;
295 /* General sanity checks */
296 if (mem->memory_size & (PAGE_SIZE - 1))
298 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
300 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
302 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
305 memslot = &kvm->memslots[mem->slot];
306 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
307 npages = mem->memory_size >> PAGE_SHIFT;
310 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
312 new = old = *memslot;
314 new.base_gfn = base_gfn;
316 new.flags = mem->flags;
318 /* Disallow changing a memory slot's size. */
320 if (npages && old.npages && npages != old.npages)
323 /* Check for overlaps */
325 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
326 struct kvm_memory_slot *s = &kvm->memslots[i];
330 if (!((base_gfn + npages <= s->base_gfn) ||
331 (base_gfn >= s->base_gfn + s->npages)))
335 /* Free page dirty bitmap if unneeded */
336 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
337 new.dirty_bitmap = NULL;
341 /* Allocate if a slot is being created */
342 if (npages && !new.rmap) {
343 new.rmap = vmalloc(npages * sizeof(struct page *));
348 memset(new.rmap, 0, npages * sizeof(*new.rmap));
350 new.user_alloc = user_alloc;
352 new.userspace_addr = mem->userspace_addr;
354 down_write(¤t->mm->mmap_sem);
355 new.userspace_addr = do_mmap(NULL, 0,
357 PROT_READ | PROT_WRITE,
358 MAP_SHARED | MAP_ANONYMOUS,
360 up_write(¤t->mm->mmap_sem);
362 if (IS_ERR((void *)new.userspace_addr))
366 if (!old.user_alloc && old.rmap) {
369 down_write(¤t->mm->mmap_sem);
370 ret = do_munmap(current->mm, old.userspace_addr,
371 old.npages * PAGE_SIZE);
372 up_write(¤t->mm->mmap_sem);
375 "kvm_vm_ioctl_set_memory_region: "
376 "failed to munmap memory\n");
380 /* Allocate page dirty bitmap if needed */
381 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
382 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
384 new.dirty_bitmap = vmalloc(dirty_bytes);
385 if (!new.dirty_bitmap)
387 memset(new.dirty_bitmap, 0, dirty_bytes);
390 if (mem->slot >= kvm->nmemslots)
391 kvm->nmemslots = mem->slot + 1;
393 if (!kvm->n_requested_mmu_pages) {
394 unsigned int n_pages;
397 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
398 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
401 unsigned int nr_mmu_pages;
403 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
404 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
405 nr_mmu_pages = max(nr_mmu_pages,
406 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
407 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
413 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
414 kvm_flush_remote_tlbs(kvm);
416 kvm_free_physmem_slot(&old, &new);
420 kvm_free_physmem_slot(&new, &old);
425 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
427 int kvm_set_memory_region(struct kvm *kvm,
428 struct kvm_userspace_memory_region *mem,
433 mutex_lock(&kvm->lock);
434 r = __kvm_set_memory_region(kvm, mem, user_alloc);
435 mutex_unlock(&kvm->lock);
438 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
440 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
442 kvm_userspace_memory_region *mem,
445 if (mem->slot >= KVM_MEMORY_SLOTS)
447 return kvm_set_memory_region(kvm, mem, user_alloc);
451 * Get (and clear) the dirty memory log for a memory slot.
453 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
454 struct kvm_dirty_log *log)
456 struct kvm_memory_slot *memslot;
459 unsigned long any = 0;
461 mutex_lock(&kvm->lock);
464 if (log->slot >= KVM_MEMORY_SLOTS)
467 memslot = &kvm->memslots[log->slot];
469 if (!memslot->dirty_bitmap)
472 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
474 for (i = 0; !any && i < n/sizeof(long); ++i)
475 any = memslot->dirty_bitmap[i];
478 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
481 /* If nothing is dirty, don't bother messing with page tables. */
483 kvm_mmu_slot_remove_write_access(kvm, log->slot);
484 kvm_flush_remote_tlbs(kvm);
485 memset(memslot->dirty_bitmap, 0, n);
491 mutex_unlock(&kvm->lock);
495 int is_error_page(struct page *page)
497 return page == bad_page;
499 EXPORT_SYMBOL_GPL(is_error_page);
501 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
504 struct kvm_mem_alias *alias;
506 for (i = 0; i < kvm->naliases; ++i) {
507 alias = &kvm->aliases[i];
508 if (gfn >= alias->base_gfn
509 && gfn < alias->base_gfn + alias->npages)
510 return alias->target_gfn + gfn - alias->base_gfn;
515 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
519 for (i = 0; i < kvm->nmemslots; ++i) {
520 struct kvm_memory_slot *memslot = &kvm->memslots[i];
522 if (gfn >= memslot->base_gfn
523 && gfn < memslot->base_gfn + memslot->npages)
529 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
531 gfn = unalias_gfn(kvm, gfn);
532 return __gfn_to_memslot(kvm, gfn);
535 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
539 gfn = unalias_gfn(kvm, gfn);
540 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
541 struct kvm_memory_slot *memslot = &kvm->memslots[i];
543 if (gfn >= memslot->base_gfn
544 && gfn < memslot->base_gfn + memslot->npages)
549 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
552 * Requires current->mm->mmap_sem to be held
554 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
556 struct kvm_memory_slot *slot;
557 struct page *page[1];
562 gfn = unalias_gfn(kvm, gfn);
563 slot = __gfn_to_memslot(kvm, gfn);
569 npages = get_user_pages(current, current->mm,
571 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
581 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
585 down_read(¤t->mm->mmap_sem);
586 page = __gfn_to_page(kvm, gfn);
587 up_read(¤t->mm->mmap_sem);
592 EXPORT_SYMBOL_GPL(gfn_to_page);
594 void kvm_release_page(struct page *page)
596 if (!PageReserved(page))
600 EXPORT_SYMBOL_GPL(kvm_release_page);
602 static int next_segment(unsigned long len, int offset)
604 if (len > PAGE_SIZE - offset)
605 return PAGE_SIZE - offset;
610 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
616 page = gfn_to_page(kvm, gfn);
617 if (is_error_page(page)) {
618 kvm_release_page(page);
621 page_virt = kmap_atomic(page, KM_USER0);
623 memcpy(data, page_virt + offset, len);
625 kunmap_atomic(page_virt, KM_USER0);
626 kvm_release_page(page);
629 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
631 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
633 gfn_t gfn = gpa >> PAGE_SHIFT;
635 int offset = offset_in_page(gpa);
638 while ((seg = next_segment(len, offset)) != 0) {
639 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
649 EXPORT_SYMBOL_GPL(kvm_read_guest);
651 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
657 page = gfn_to_page(kvm, gfn);
658 if (is_error_page(page)) {
659 kvm_release_page(page);
662 page_virt = kmap_atomic(page, KM_USER0);
664 memcpy(page_virt + offset, data, len);
666 kunmap_atomic(page_virt, KM_USER0);
667 mark_page_dirty(kvm, gfn);
668 kvm_release_page(page);
671 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
673 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
676 gfn_t gfn = gpa >> PAGE_SHIFT;
678 int offset = offset_in_page(gpa);
681 while ((seg = next_segment(len, offset)) != 0) {
682 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
693 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
698 page = gfn_to_page(kvm, gfn);
699 if (is_error_page(page)) {
700 kvm_release_page(page);
703 page_virt = kmap_atomic(page, KM_USER0);
705 memset(page_virt + offset, 0, len);
707 kunmap_atomic(page_virt, KM_USER0);
708 kvm_release_page(page);
711 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
713 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
715 gfn_t gfn = gpa >> PAGE_SHIFT;
717 int offset = offset_in_page(gpa);
720 while ((seg = next_segment(len, offset)) != 0) {
721 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
730 EXPORT_SYMBOL_GPL(kvm_clear_guest);
732 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
734 struct kvm_memory_slot *memslot;
736 gfn = unalias_gfn(kvm, gfn);
737 memslot = __gfn_to_memslot(kvm, gfn);
738 if (memslot && memslot->dirty_bitmap) {
739 unsigned long rel_gfn = gfn - memslot->base_gfn;
742 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
743 set_bit(rel_gfn, memslot->dirty_bitmap);
748 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
750 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
752 DECLARE_WAITQUEUE(wait, current);
754 add_wait_queue(&vcpu->wq, &wait);
757 * We will block until either an interrupt or a signal wakes us up
759 while (!kvm_cpu_has_interrupt(vcpu)
760 && !signal_pending(current)
761 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
762 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
763 set_current_state(TASK_INTERRUPTIBLE);
769 __set_current_state(TASK_RUNNING);
770 remove_wait_queue(&vcpu->wq, &wait);
773 void kvm_resched(struct kvm_vcpu *vcpu)
779 EXPORT_SYMBOL_GPL(kvm_resched);
782 * Check if userspace requested an interrupt window, and that the
783 * interrupt window is open.
785 * No need to exit to userspace if we already have an interrupt queued.
787 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
788 struct kvm_run *kvm_run)
790 return (!vcpu->irq_summary &&
791 kvm_run->request_interrupt_window &&
792 vcpu->interrupt_window_open &&
793 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
796 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
797 struct kvm_run *kvm_run)
799 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
800 kvm_run->cr8 = get_cr8(vcpu);
801 kvm_run->apic_base = kvm_get_apic_base(vcpu);
802 if (irqchip_in_kernel(vcpu->kvm))
803 kvm_run->ready_for_interrupt_injection = 1;
805 kvm_run->ready_for_interrupt_injection =
806 (vcpu->interrupt_window_open &&
807 vcpu->irq_summary == 0);
810 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
814 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
815 pr_debug("vcpu %d received sipi with vector # %x\n",
816 vcpu->vcpu_id, vcpu->sipi_vector);
817 kvm_lapic_reset(vcpu);
818 r = kvm_x86_ops->vcpu_reset(vcpu);
821 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
825 if (vcpu->guest_debug.enabled)
826 kvm_x86_ops->guest_debug_pre(vcpu);
829 r = kvm_mmu_reload(vcpu);
833 kvm_inject_pending_timer_irqs(vcpu);
837 kvm_x86_ops->prepare_guest_switch(vcpu);
838 kvm_load_guest_fpu(vcpu);
842 if (signal_pending(current)) {
846 kvm_run->exit_reason = KVM_EXIT_INTR;
847 ++vcpu->stat.signal_exits;
851 if (irqchip_in_kernel(vcpu->kvm))
852 kvm_x86_ops->inject_pending_irq(vcpu);
853 else if (!vcpu->mmio_read_completed)
854 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
856 vcpu->guest_mode = 1;
860 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
861 kvm_x86_ops->tlb_flush(vcpu);
863 kvm_x86_ops->run(vcpu, kvm_run);
865 vcpu->guest_mode = 0;
871 * We must have an instruction between local_irq_enable() and
872 * kvm_guest_exit(), so the timer interrupt isn't delayed by
873 * the interrupt shadow. The stat.exits increment will do nicely.
874 * But we need to prevent reordering, hence this barrier():
883 * Profile KVM exit RIPs:
885 if (unlikely(prof_on == KVM_PROFILING)) {
886 kvm_x86_ops->cache_regs(vcpu);
887 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
890 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
893 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
895 kvm_run->exit_reason = KVM_EXIT_INTR;
896 ++vcpu->stat.request_irq_exits;
899 if (!need_resched()) {
900 ++vcpu->stat.light_exits;
911 post_kvm_run_save(vcpu, kvm_run);
917 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
924 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
925 kvm_vcpu_block(vcpu);
930 if (vcpu->sigset_active)
931 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
933 /* re-sync apic's tpr */
934 if (!irqchip_in_kernel(vcpu->kvm))
935 set_cr8(vcpu, kvm_run->cr8);
937 if (vcpu->pio.cur_count) {
938 r = complete_pio(vcpu);
943 if (vcpu->mmio_needed) {
944 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
945 vcpu->mmio_read_completed = 1;
946 vcpu->mmio_needed = 0;
947 r = emulate_instruction(vcpu, kvm_run,
948 vcpu->mmio_fault_cr2, 0, 1);
949 if (r == EMULATE_DO_MMIO) {
951 * Read-modify-write. Back to userspace.
958 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
959 kvm_x86_ops->cache_regs(vcpu);
960 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
961 kvm_x86_ops->decache_regs(vcpu);
964 r = __vcpu_run(vcpu, kvm_run);
967 if (vcpu->sigset_active)
968 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
974 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
975 struct kvm_regs *regs)
979 kvm_x86_ops->cache_regs(vcpu);
981 regs->rax = vcpu->regs[VCPU_REGS_RAX];
982 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
983 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
984 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
985 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
986 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
987 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
988 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
990 regs->r8 = vcpu->regs[VCPU_REGS_R8];
991 regs->r9 = vcpu->regs[VCPU_REGS_R9];
992 regs->r10 = vcpu->regs[VCPU_REGS_R10];
993 regs->r11 = vcpu->regs[VCPU_REGS_R11];
994 regs->r12 = vcpu->regs[VCPU_REGS_R12];
995 regs->r13 = vcpu->regs[VCPU_REGS_R13];
996 regs->r14 = vcpu->regs[VCPU_REGS_R14];
997 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1000 regs->rip = vcpu->rip;
1001 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
1004 * Don't leak debug flags in case they were set for guest debugging
1006 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1007 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1014 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1015 struct kvm_regs *regs)
1019 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1020 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1021 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1022 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1023 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1024 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1025 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1026 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1027 #ifdef CONFIG_X86_64
1028 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1029 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1030 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1031 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1032 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1033 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1034 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1035 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1038 vcpu->rip = regs->rip;
1039 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
1041 kvm_x86_ops->decache_regs(vcpu);
1048 static void get_segment(struct kvm_vcpu *vcpu,
1049 struct kvm_segment *var, int seg)
1051 return kvm_x86_ops->get_segment(vcpu, var, seg);
1054 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1055 struct kvm_sregs *sregs)
1057 struct descriptor_table dt;
1062 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1063 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1064 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1065 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1066 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1067 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1069 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1070 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1072 kvm_x86_ops->get_idt(vcpu, &dt);
1073 sregs->idt.limit = dt.limit;
1074 sregs->idt.base = dt.base;
1075 kvm_x86_ops->get_gdt(vcpu, &dt);
1076 sregs->gdt.limit = dt.limit;
1077 sregs->gdt.base = dt.base;
1079 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1080 sregs->cr0 = vcpu->cr0;
1081 sregs->cr2 = vcpu->cr2;
1082 sregs->cr3 = vcpu->cr3;
1083 sregs->cr4 = vcpu->cr4;
1084 sregs->cr8 = get_cr8(vcpu);
1085 sregs->efer = vcpu->shadow_efer;
1086 sregs->apic_base = kvm_get_apic_base(vcpu);
1088 if (irqchip_in_kernel(vcpu->kvm)) {
1089 memset(sregs->interrupt_bitmap, 0,
1090 sizeof sregs->interrupt_bitmap);
1091 pending_vec = kvm_x86_ops->get_irq(vcpu);
1092 if (pending_vec >= 0)
1093 set_bit(pending_vec,
1094 (unsigned long *)sregs->interrupt_bitmap);
1096 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1097 sizeof sregs->interrupt_bitmap);
1104 static void set_segment(struct kvm_vcpu *vcpu,
1105 struct kvm_segment *var, int seg)
1107 return kvm_x86_ops->set_segment(vcpu, var, seg);
1110 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1111 struct kvm_sregs *sregs)
1113 int mmu_reset_needed = 0;
1114 int i, pending_vec, max_bits;
1115 struct descriptor_table dt;
1119 dt.limit = sregs->idt.limit;
1120 dt.base = sregs->idt.base;
1121 kvm_x86_ops->set_idt(vcpu, &dt);
1122 dt.limit = sregs->gdt.limit;
1123 dt.base = sregs->gdt.base;
1124 kvm_x86_ops->set_gdt(vcpu, &dt);
1126 vcpu->cr2 = sregs->cr2;
1127 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1128 vcpu->cr3 = sregs->cr3;
1130 set_cr8(vcpu, sregs->cr8);
1132 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1133 #ifdef CONFIG_X86_64
1134 kvm_x86_ops->set_efer(vcpu, sregs->efer);
1136 kvm_set_apic_base(vcpu, sregs->apic_base);
1138 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1140 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1141 vcpu->cr0 = sregs->cr0;
1142 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
1144 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1145 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1146 if (!is_long_mode(vcpu) && is_pae(vcpu))
1147 load_pdptrs(vcpu, vcpu->cr3);
1149 if (mmu_reset_needed)
1150 kvm_mmu_reset_context(vcpu);
1152 if (!irqchip_in_kernel(vcpu->kvm)) {
1153 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1154 sizeof vcpu->irq_pending);
1155 vcpu->irq_summary = 0;
1156 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
1157 if (vcpu->irq_pending[i])
1158 __set_bit(i, &vcpu->irq_summary);
1160 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
1161 pending_vec = find_first_bit(
1162 (const unsigned long *)sregs->interrupt_bitmap,
1164 /* Only pending external irq is handled here */
1165 if (pending_vec < max_bits) {
1166 kvm_x86_ops->set_irq(vcpu, pending_vec);
1167 pr_debug("Set back pending irq %d\n",
1172 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1173 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1174 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1175 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1176 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1177 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1179 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1180 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1187 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1189 struct kvm_segment cs;
1191 get_segment(vcpu, &cs, VCPU_SREG_CS);
1195 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
1198 * Translate a guest virtual address to a guest physical address.
1200 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1201 struct kvm_translation *tr)
1203 unsigned long vaddr = tr->linear_address;
1207 mutex_lock(&vcpu->kvm->lock);
1208 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1209 tr->physical_address = gpa;
1210 tr->valid = gpa != UNMAPPED_GVA;
1213 mutex_unlock(&vcpu->kvm->lock);
1219 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1220 struct kvm_interrupt *irq)
1222 if (irq->irq < 0 || irq->irq >= 256)
1224 if (irqchip_in_kernel(vcpu->kvm))
1228 set_bit(irq->irq, vcpu->irq_pending);
1229 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1236 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1237 struct kvm_debug_guest *dbg)
1243 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
1250 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1251 unsigned long address,
1254 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1255 unsigned long pgoff;
1258 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1260 page = virt_to_page(vcpu->run);
1261 else if (pgoff == KVM_PIO_PAGE_OFFSET)
1262 page = virt_to_page(vcpu->pio_data);
1264 return NOPAGE_SIGBUS;
1267 *type = VM_FAULT_MINOR;
1272 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1273 .nopage = kvm_vcpu_nopage,
1276 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1278 vma->vm_ops = &kvm_vcpu_vm_ops;
1282 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1284 struct kvm_vcpu *vcpu = filp->private_data;
1286 fput(vcpu->kvm->filp);
1290 static struct file_operations kvm_vcpu_fops = {
1291 .release = kvm_vcpu_release,
1292 .unlocked_ioctl = kvm_vcpu_ioctl,
1293 .compat_ioctl = kvm_vcpu_ioctl,
1294 .mmap = kvm_vcpu_mmap,
1298 * Allocates an inode for the vcpu.
1300 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1303 struct inode *inode;
1306 r = anon_inode_getfd(&fd, &inode, &file,
1307 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
1310 atomic_inc(&vcpu->kvm->filp->f_count);
1315 * Creates some virtual cpus. Good luck creating more than one.
1317 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1320 struct kvm_vcpu *vcpu;
1325 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
1327 return PTR_ERR(vcpu);
1329 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1331 /* We do fxsave: this must be aligned. */
1332 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
1335 r = kvm_x86_ops->vcpu_reset(vcpu);
1337 r = kvm_mmu_setup(vcpu);
1342 mutex_lock(&kvm->lock);
1343 if (kvm->vcpus[n]) {
1345 mutex_unlock(&kvm->lock);
1348 kvm->vcpus[n] = vcpu;
1349 mutex_unlock(&kvm->lock);
1351 /* Now it's all set up, let userspace reach it */
1352 r = create_vcpu_fd(vcpu);
1358 mutex_lock(&kvm->lock);
1359 kvm->vcpus[n] = NULL;
1360 mutex_unlock(&kvm->lock);
1364 kvm_mmu_unload(vcpu);
1368 kvm_x86_ops->vcpu_free(vcpu);
1372 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1375 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1376 vcpu->sigset_active = 1;
1377 vcpu->sigset = *sigset;
1379 vcpu->sigset_active = 0;
1383 static long kvm_vcpu_ioctl(struct file *filp,
1384 unsigned int ioctl, unsigned long arg)
1386 struct kvm_vcpu *vcpu = filp->private_data;
1387 void __user *argp = (void __user *)arg;
1395 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
1397 case KVM_GET_REGS: {
1398 struct kvm_regs kvm_regs;
1400 memset(&kvm_regs, 0, sizeof kvm_regs);
1401 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
1405 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1410 case KVM_SET_REGS: {
1411 struct kvm_regs kvm_regs;
1414 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1416 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
1422 case KVM_GET_SREGS: {
1423 struct kvm_sregs kvm_sregs;
1425 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1426 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1430 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1435 case KVM_SET_SREGS: {
1436 struct kvm_sregs kvm_sregs;
1439 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1441 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1447 case KVM_TRANSLATE: {
1448 struct kvm_translation tr;
1451 if (copy_from_user(&tr, argp, sizeof tr))
1453 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
1457 if (copy_to_user(argp, &tr, sizeof tr))
1462 case KVM_INTERRUPT: {
1463 struct kvm_interrupt irq;
1466 if (copy_from_user(&irq, argp, sizeof irq))
1468 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1474 case KVM_DEBUG_GUEST: {
1475 struct kvm_debug_guest dbg;
1478 if (copy_from_user(&dbg, argp, sizeof dbg))
1480 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
1486 case KVM_SET_SIGNAL_MASK: {
1487 struct kvm_signal_mask __user *sigmask_arg = argp;
1488 struct kvm_signal_mask kvm_sigmask;
1489 sigset_t sigset, *p;
1494 if (copy_from_user(&kvm_sigmask, argp,
1495 sizeof kvm_sigmask))
1498 if (kvm_sigmask.len != sizeof sigset)
1501 if (copy_from_user(&sigset, sigmask_arg->sigset,
1506 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1512 memset(&fpu, 0, sizeof fpu);
1513 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1517 if (copy_to_user(argp, &fpu, sizeof fpu))
1526 if (copy_from_user(&fpu, argp, sizeof fpu))
1528 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1535 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1541 static long kvm_vm_ioctl(struct file *filp,
1542 unsigned int ioctl, unsigned long arg)
1544 struct kvm *kvm = filp->private_data;
1545 void __user *argp = (void __user *)arg;
1549 case KVM_CREATE_VCPU:
1550 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1554 case KVM_SET_USER_MEMORY_REGION: {
1555 struct kvm_userspace_memory_region kvm_userspace_mem;
1558 if (copy_from_user(&kvm_userspace_mem, argp,
1559 sizeof kvm_userspace_mem))
1562 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1567 case KVM_GET_DIRTY_LOG: {
1568 struct kvm_dirty_log log;
1571 if (copy_from_user(&log, argp, sizeof log))
1573 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1579 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1585 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1586 unsigned long address,
1589 struct kvm *kvm = vma->vm_file->private_data;
1590 unsigned long pgoff;
1593 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1594 if (!kvm_is_visible_gfn(kvm, pgoff))
1595 return NOPAGE_SIGBUS;
1596 /* current->mm->mmap_sem is already held so call lockless version */
1597 page = __gfn_to_page(kvm, pgoff);
1598 if (is_error_page(page)) {
1599 kvm_release_page(page);
1600 return NOPAGE_SIGBUS;
1603 *type = VM_FAULT_MINOR;
1608 static struct vm_operations_struct kvm_vm_vm_ops = {
1609 .nopage = kvm_vm_nopage,
1612 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1614 vma->vm_ops = &kvm_vm_vm_ops;
1618 static struct file_operations kvm_vm_fops = {
1619 .release = kvm_vm_release,
1620 .unlocked_ioctl = kvm_vm_ioctl,
1621 .compat_ioctl = kvm_vm_ioctl,
1622 .mmap = kvm_vm_mmap,
1625 static int kvm_dev_ioctl_create_vm(void)
1628 struct inode *inode;
1632 kvm = kvm_create_vm();
1634 return PTR_ERR(kvm);
1635 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1637 kvm_destroy_vm(kvm);
1646 static long kvm_dev_ioctl(struct file *filp,
1647 unsigned int ioctl, unsigned long arg)
1649 void __user *argp = (void __user *)arg;
1653 case KVM_GET_API_VERSION:
1657 r = KVM_API_VERSION;
1663 r = kvm_dev_ioctl_create_vm();
1665 case KVM_CHECK_EXTENSION: {
1666 int ext = (long)argp;
1669 case KVM_CAP_IRQCHIP:
1671 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1672 case KVM_CAP_USER_MEMORY:
1673 case KVM_CAP_SET_TSS_ADDR:
1682 case KVM_GET_VCPU_MMAP_SIZE:
1689 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1695 static struct file_operations kvm_chardev_ops = {
1696 .unlocked_ioctl = kvm_dev_ioctl,
1697 .compat_ioctl = kvm_dev_ioctl,
1700 static struct miscdevice kvm_dev = {
1707 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
1710 static void decache_vcpus_on_cpu(int cpu)
1713 struct kvm_vcpu *vcpu;
1716 spin_lock(&kvm_lock);
1717 list_for_each_entry(vm, &vm_list, vm_list)
1718 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1719 vcpu = vm->vcpus[i];
1723 * If the vcpu is locked, then it is running on some
1724 * other cpu and therefore it is not cached on the
1727 * If it's not locked, check the last cpu it executed
1730 if (mutex_trylock(&vcpu->mutex)) {
1731 if (vcpu->cpu == cpu) {
1732 kvm_x86_ops->vcpu_decache(vcpu);
1735 mutex_unlock(&vcpu->mutex);
1738 spin_unlock(&kvm_lock);
1741 static void hardware_enable(void *junk)
1743 int cpu = raw_smp_processor_id();
1745 if (cpu_isset(cpu, cpus_hardware_enabled))
1747 cpu_set(cpu, cpus_hardware_enabled);
1748 kvm_x86_ops->hardware_enable(NULL);
1751 static void hardware_disable(void *junk)
1753 int cpu = raw_smp_processor_id();
1755 if (!cpu_isset(cpu, cpus_hardware_enabled))
1757 cpu_clear(cpu, cpus_hardware_enabled);
1758 decache_vcpus_on_cpu(cpu);
1759 kvm_x86_ops->hardware_disable(NULL);
1762 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1769 case CPU_DYING_FROZEN:
1770 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1772 hardware_disable(NULL);
1774 case CPU_UP_CANCELED:
1775 case CPU_UP_CANCELED_FROZEN:
1776 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1778 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1781 case CPU_ONLINE_FROZEN:
1782 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1784 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1790 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1793 if (val == SYS_RESTART) {
1795 * Some (well, at least mine) BIOSes hang on reboot if
1798 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1799 on_each_cpu(hardware_disable, NULL, 0, 1);
1804 static struct notifier_block kvm_reboot_notifier = {
1805 .notifier_call = kvm_reboot,
1809 void kvm_io_bus_init(struct kvm_io_bus *bus)
1811 memset(bus, 0, sizeof(*bus));
1814 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1818 for (i = 0; i < bus->dev_count; i++) {
1819 struct kvm_io_device *pos = bus->devs[i];
1821 kvm_iodevice_destructor(pos);
1825 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1829 for (i = 0; i < bus->dev_count; i++) {
1830 struct kvm_io_device *pos = bus->devs[i];
1832 if (pos->in_range(pos, addr))
1839 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1841 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1843 bus->devs[bus->dev_count++] = dev;
1846 static struct notifier_block kvm_cpu_notifier = {
1847 .notifier_call = kvm_cpu_hotplug,
1848 .priority = 20, /* must be > scheduler priority */
1851 static u64 stat_get(void *_offset)
1853 unsigned offset = (long)_offset;
1856 struct kvm_vcpu *vcpu;
1859 spin_lock(&kvm_lock);
1860 list_for_each_entry(kvm, &vm_list, vm_list)
1861 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1862 vcpu = kvm->vcpus[i];
1864 total += *(u32 *)((void *)vcpu + offset);
1866 spin_unlock(&kvm_lock);
1870 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1872 static __init void kvm_init_debug(void)
1874 struct kvm_stats_debugfs_item *p;
1876 debugfs_dir = debugfs_create_dir("kvm", NULL);
1877 for (p = debugfs_entries; p->name; ++p)
1878 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1879 (void *)(long)p->offset,
1883 static void kvm_exit_debug(void)
1885 struct kvm_stats_debugfs_item *p;
1887 for (p = debugfs_entries; p->name; ++p)
1888 debugfs_remove(p->dentry);
1889 debugfs_remove(debugfs_dir);
1892 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1894 hardware_disable(NULL);
1898 static int kvm_resume(struct sys_device *dev)
1900 hardware_enable(NULL);
1904 static struct sysdev_class kvm_sysdev_class = {
1906 .suspend = kvm_suspend,
1907 .resume = kvm_resume,
1910 static struct sys_device kvm_sysdev = {
1912 .cls = &kvm_sysdev_class,
1915 struct page *bad_page;
1918 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1920 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1923 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1925 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1927 kvm_x86_ops->vcpu_load(vcpu, cpu);
1930 static void kvm_sched_out(struct preempt_notifier *pn,
1931 struct task_struct *next)
1933 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1935 kvm_x86_ops->vcpu_put(vcpu);
1938 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
1939 struct module *module)
1945 printk(KERN_ERR "kvm: already loaded the other module\n");
1949 if (!ops->cpu_has_kvm_support()) {
1950 printk(KERN_ERR "kvm: no hardware support\n");
1953 if (ops->disabled_by_bios()) {
1954 printk(KERN_ERR "kvm: disabled by bios\n");
1960 r = kvm_x86_ops->hardware_setup();
1964 for_each_online_cpu(cpu) {
1965 smp_call_function_single(cpu,
1966 kvm_x86_ops->check_processor_compatibility,
1972 on_each_cpu(hardware_enable, NULL, 0, 1);
1973 r = register_cpu_notifier(&kvm_cpu_notifier);
1976 register_reboot_notifier(&kvm_reboot_notifier);
1978 r = sysdev_class_register(&kvm_sysdev_class);
1982 r = sysdev_register(&kvm_sysdev);
1986 /* A kmem cache lets us meet the alignment requirements of fx_save. */
1987 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1988 __alignof__(struct kvm_vcpu), 0, 0);
1989 if (!kvm_vcpu_cache) {
1994 kvm_chardev_ops.owner = module;
1996 r = misc_register(&kvm_dev);
1998 printk(KERN_ERR "kvm: misc device register failed\n");
2002 kvm_preempt_ops.sched_in = kvm_sched_in;
2003 kvm_preempt_ops.sched_out = kvm_sched_out;
2005 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2010 kmem_cache_destroy(kvm_vcpu_cache);
2012 sysdev_unregister(&kvm_sysdev);
2014 sysdev_class_unregister(&kvm_sysdev_class);
2016 unregister_reboot_notifier(&kvm_reboot_notifier);
2017 unregister_cpu_notifier(&kvm_cpu_notifier);
2019 on_each_cpu(hardware_disable, NULL, 0, 1);
2021 kvm_x86_ops->hardware_unsetup();
2026 EXPORT_SYMBOL_GPL(kvm_init_x86);
2028 void kvm_exit_x86(void)
2030 misc_deregister(&kvm_dev);
2031 kmem_cache_destroy(kvm_vcpu_cache);
2032 sysdev_unregister(&kvm_sysdev);
2033 sysdev_class_unregister(&kvm_sysdev_class);
2034 unregister_reboot_notifier(&kvm_reboot_notifier);
2035 unregister_cpu_notifier(&kvm_cpu_notifier);
2036 on_each_cpu(hardware_disable, NULL, 0, 1);
2037 kvm_x86_ops->hardware_unsetup();
2040 EXPORT_SYMBOL_GPL(kvm_exit_x86);
2042 static __init int kvm_init(void)
2046 r = kvm_mmu_module_init();
2054 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2056 if (bad_page == NULL) {
2065 kvm_mmu_module_exit();
2070 static __exit void kvm_exit(void)
2073 __free_page(bad_page);
2074 kvm_mmu_module_exit();
2077 module_init(kvm_init)
2078 module_exit(kvm_exit)