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 kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
277 kvm_mmu_unload(vcpu);
281 static void kvm_free_vcpus(struct kvm *kvm)
286 * Unpin any mmu pages first.
288 for (i = 0; i < KVM_MAX_VCPUS; ++i)
290 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
291 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
293 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
294 kvm->vcpus[i] = NULL;
300 static void kvm_destroy_vm(struct kvm *kvm)
302 spin_lock(&kvm_lock);
303 list_del(&kvm->vm_list);
304 spin_unlock(&kvm_lock);
305 kvm_io_bus_destroy(&kvm->pio_bus);
306 kvm_io_bus_destroy(&kvm->mmio_bus);
310 kvm_free_physmem(kvm);
314 static int kvm_vm_release(struct inode *inode, struct file *filp)
316 struct kvm *kvm = filp->private_data;
322 void fx_init(struct kvm_vcpu *vcpu)
324 unsigned after_mxcsr_mask;
326 /* Initialize guest FPU by resetting ours and saving into guest's */
328 fx_save(&vcpu->host_fx_image);
330 fx_save(&vcpu->guest_fx_image);
331 fx_restore(&vcpu->host_fx_image);
334 vcpu->cr0 |= X86_CR0_ET;
335 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
336 vcpu->guest_fx_image.mxcsr = 0x1f80;
337 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
338 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
340 EXPORT_SYMBOL_GPL(fx_init);
343 * Allocate some memory and give it an address in the guest physical address
346 * Discontiguous memory is allowed, mostly for framebuffers.
348 * Must be called holding kvm->lock.
350 int __kvm_set_memory_region(struct kvm *kvm,
351 struct kvm_userspace_memory_region *mem,
356 unsigned long npages;
358 struct kvm_memory_slot *memslot;
359 struct kvm_memory_slot old, new;
362 /* General sanity checks */
363 if (mem->memory_size & (PAGE_SIZE - 1))
365 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
367 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
369 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
372 memslot = &kvm->memslots[mem->slot];
373 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
374 npages = mem->memory_size >> PAGE_SHIFT;
377 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
379 new = old = *memslot;
381 new.base_gfn = base_gfn;
383 new.flags = mem->flags;
385 /* Disallow changing a memory slot's size. */
387 if (npages && old.npages && npages != old.npages)
390 /* Check for overlaps */
392 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
393 struct kvm_memory_slot *s = &kvm->memslots[i];
397 if (!((base_gfn + npages <= s->base_gfn) ||
398 (base_gfn >= s->base_gfn + s->npages)))
402 /* Free page dirty bitmap if unneeded */
403 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
404 new.dirty_bitmap = NULL;
408 /* Allocate if a slot is being created */
409 if (npages && !new.rmap) {
410 new.rmap = vmalloc(npages * sizeof(struct page *));
415 memset(new.rmap, 0, npages * sizeof(*new.rmap));
417 new.user_alloc = user_alloc;
419 new.userspace_addr = mem->userspace_addr;
421 down_write(¤t->mm->mmap_sem);
422 new.userspace_addr = do_mmap(NULL, 0,
424 PROT_READ | PROT_WRITE,
425 MAP_SHARED | MAP_ANONYMOUS,
427 up_write(¤t->mm->mmap_sem);
429 if (IS_ERR((void *)new.userspace_addr))
433 if (!old.user_alloc && old.rmap) {
436 down_write(¤t->mm->mmap_sem);
437 ret = do_munmap(current->mm, old.userspace_addr,
438 old.npages * PAGE_SIZE);
439 up_write(¤t->mm->mmap_sem);
442 "kvm_vm_ioctl_set_memory_region: "
443 "failed to munmap memory\n");
447 /* Allocate page dirty bitmap if needed */
448 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
449 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
451 new.dirty_bitmap = vmalloc(dirty_bytes);
452 if (!new.dirty_bitmap)
454 memset(new.dirty_bitmap, 0, dirty_bytes);
457 if (mem->slot >= kvm->nmemslots)
458 kvm->nmemslots = mem->slot + 1;
460 if (!kvm->n_requested_mmu_pages) {
461 unsigned int n_pages;
464 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
465 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
468 unsigned int nr_mmu_pages;
470 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
471 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
472 nr_mmu_pages = max(nr_mmu_pages,
473 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
474 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
480 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
481 kvm_flush_remote_tlbs(kvm);
483 kvm_free_physmem_slot(&old, &new);
487 kvm_free_physmem_slot(&new, &old);
492 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
494 int kvm_set_memory_region(struct kvm *kvm,
495 struct kvm_userspace_memory_region *mem,
500 mutex_lock(&kvm->lock);
501 r = __kvm_set_memory_region(kvm, mem, user_alloc);
502 mutex_unlock(&kvm->lock);
505 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
507 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
509 kvm_userspace_memory_region *mem,
512 if (mem->slot >= KVM_MEMORY_SLOTS)
514 return kvm_set_memory_region(kvm, mem, user_alloc);
518 * Get (and clear) the dirty memory log for a memory slot.
520 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
521 struct kvm_dirty_log *log)
523 struct kvm_memory_slot *memslot;
526 unsigned long any = 0;
528 mutex_lock(&kvm->lock);
531 if (log->slot >= KVM_MEMORY_SLOTS)
534 memslot = &kvm->memslots[log->slot];
536 if (!memslot->dirty_bitmap)
539 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
541 for (i = 0; !any && i < n/sizeof(long); ++i)
542 any = memslot->dirty_bitmap[i];
545 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
548 /* If nothing is dirty, don't bother messing with page tables. */
550 kvm_mmu_slot_remove_write_access(kvm, log->slot);
551 kvm_flush_remote_tlbs(kvm);
552 memset(memslot->dirty_bitmap, 0, n);
558 mutex_unlock(&kvm->lock);
562 int is_error_page(struct page *page)
564 return page == bad_page;
566 EXPORT_SYMBOL_GPL(is_error_page);
568 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
571 struct kvm_mem_alias *alias;
573 for (i = 0; i < kvm->naliases; ++i) {
574 alias = &kvm->aliases[i];
575 if (gfn >= alias->base_gfn
576 && gfn < alias->base_gfn + alias->npages)
577 return alias->target_gfn + gfn - alias->base_gfn;
582 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
586 for (i = 0; i < kvm->nmemslots; ++i) {
587 struct kvm_memory_slot *memslot = &kvm->memslots[i];
589 if (gfn >= memslot->base_gfn
590 && gfn < memslot->base_gfn + memslot->npages)
596 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
598 gfn = unalias_gfn(kvm, gfn);
599 return __gfn_to_memslot(kvm, gfn);
602 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
606 gfn = unalias_gfn(kvm, gfn);
607 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
608 struct kvm_memory_slot *memslot = &kvm->memslots[i];
610 if (gfn >= memslot->base_gfn
611 && gfn < memslot->base_gfn + memslot->npages)
616 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
619 * Requires current->mm->mmap_sem to be held
621 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
623 struct kvm_memory_slot *slot;
624 struct page *page[1];
629 gfn = unalias_gfn(kvm, gfn);
630 slot = __gfn_to_memslot(kvm, gfn);
636 npages = get_user_pages(current, current->mm,
638 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
648 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
652 down_read(¤t->mm->mmap_sem);
653 page = __gfn_to_page(kvm, gfn);
654 up_read(¤t->mm->mmap_sem);
659 EXPORT_SYMBOL_GPL(gfn_to_page);
661 void kvm_release_page(struct page *page)
663 if (!PageReserved(page))
667 EXPORT_SYMBOL_GPL(kvm_release_page);
669 static int next_segment(unsigned long len, int offset)
671 if (len > PAGE_SIZE - offset)
672 return PAGE_SIZE - offset;
677 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
683 page = gfn_to_page(kvm, gfn);
684 if (is_error_page(page)) {
685 kvm_release_page(page);
688 page_virt = kmap_atomic(page, KM_USER0);
690 memcpy(data, page_virt + offset, len);
692 kunmap_atomic(page_virt, KM_USER0);
693 kvm_release_page(page);
696 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
698 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
700 gfn_t gfn = gpa >> PAGE_SHIFT;
702 int offset = offset_in_page(gpa);
705 while ((seg = next_segment(len, offset)) != 0) {
706 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
716 EXPORT_SYMBOL_GPL(kvm_read_guest);
718 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
724 page = gfn_to_page(kvm, gfn);
725 if (is_error_page(page)) {
726 kvm_release_page(page);
729 page_virt = kmap_atomic(page, KM_USER0);
731 memcpy(page_virt + offset, data, len);
733 kunmap_atomic(page_virt, KM_USER0);
734 mark_page_dirty(kvm, gfn);
735 kvm_release_page(page);
738 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
740 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
743 gfn_t gfn = gpa >> PAGE_SHIFT;
745 int offset = offset_in_page(gpa);
748 while ((seg = next_segment(len, offset)) != 0) {
749 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
760 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
765 page = gfn_to_page(kvm, gfn);
766 if (is_error_page(page)) {
767 kvm_release_page(page);
770 page_virt = kmap_atomic(page, KM_USER0);
772 memset(page_virt + offset, 0, len);
774 kunmap_atomic(page_virt, KM_USER0);
775 kvm_release_page(page);
778 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
780 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
782 gfn_t gfn = gpa >> PAGE_SHIFT;
784 int offset = offset_in_page(gpa);
787 while ((seg = next_segment(len, offset)) != 0) {
788 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
797 EXPORT_SYMBOL_GPL(kvm_clear_guest);
799 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
801 struct kvm_memory_slot *memslot;
803 gfn = unalias_gfn(kvm, gfn);
804 memslot = __gfn_to_memslot(kvm, gfn);
805 if (memslot && memslot->dirty_bitmap) {
806 unsigned long rel_gfn = gfn - memslot->base_gfn;
809 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
810 set_bit(rel_gfn, memslot->dirty_bitmap);
815 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
817 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
819 DECLARE_WAITQUEUE(wait, current);
821 add_wait_queue(&vcpu->wq, &wait);
824 * We will block until either an interrupt or a signal wakes us up
826 while (!kvm_cpu_has_interrupt(vcpu)
827 && !signal_pending(current)
828 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
829 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
830 set_current_state(TASK_INTERRUPTIBLE);
836 __set_current_state(TASK_RUNNING);
837 remove_wait_queue(&vcpu->wq, &wait);
840 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
842 ++vcpu->stat.halt_exits;
843 if (irqchip_in_kernel(vcpu->kvm)) {
844 vcpu->mp_state = VCPU_MP_STATE_HALTED;
845 kvm_vcpu_block(vcpu);
846 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
850 vcpu->run->exit_reason = KVM_EXIT_HLT;
854 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
856 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
858 unsigned long nr, a0, a1, a2, a3, ret;
860 kvm_x86_ops->cache_regs(vcpu);
862 nr = vcpu->regs[VCPU_REGS_RAX];
863 a0 = vcpu->regs[VCPU_REGS_RBX];
864 a1 = vcpu->regs[VCPU_REGS_RCX];
865 a2 = vcpu->regs[VCPU_REGS_RDX];
866 a3 = vcpu->regs[VCPU_REGS_RSI];
868 if (!is_long_mode(vcpu)) {
881 vcpu->regs[VCPU_REGS_RAX] = ret;
882 kvm_x86_ops->decache_regs(vcpu);
885 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
887 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
892 mutex_lock(&vcpu->kvm->lock);
895 * Blow out the MMU to ensure that no other VCPU has an active mapping
896 * to ensure that the updated hypercall appears atomically across all
899 kvm_mmu_zap_all(vcpu->kvm);
901 kvm_x86_ops->cache_regs(vcpu);
902 kvm_x86_ops->patch_hypercall(vcpu, instruction);
903 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
907 mutex_unlock(&vcpu->kvm->lock);
912 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
914 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
917 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
919 struct descriptor_table dt = { limit, base };
921 kvm_x86_ops->set_gdt(vcpu, &dt);
924 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
926 struct descriptor_table dt = { limit, base };
928 kvm_x86_ops->set_idt(vcpu, &dt);
931 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
932 unsigned long *rflags)
935 *rflags = kvm_x86_ops->get_rflags(vcpu);
938 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
940 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
951 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
956 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
957 unsigned long *rflags)
961 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
962 *rflags = kvm_x86_ops->get_rflags(vcpu);
971 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
974 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
978 void kvm_resched(struct kvm_vcpu *vcpu)
984 EXPORT_SYMBOL_GPL(kvm_resched);
986 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
990 struct kvm_cpuid_entry *e, *best;
992 kvm_x86_ops->cache_regs(vcpu);
993 function = vcpu->regs[VCPU_REGS_RAX];
994 vcpu->regs[VCPU_REGS_RAX] = 0;
995 vcpu->regs[VCPU_REGS_RBX] = 0;
996 vcpu->regs[VCPU_REGS_RCX] = 0;
997 vcpu->regs[VCPU_REGS_RDX] = 0;
999 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1000 e = &vcpu->cpuid_entries[i];
1001 if (e->function == function) {
1006 * Both basic or both extended?
1008 if (((e->function ^ function) & 0x80000000) == 0)
1009 if (!best || e->function > best->function)
1013 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1014 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1015 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1016 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1018 kvm_x86_ops->decache_regs(vcpu);
1019 kvm_x86_ops->skip_emulated_instruction(vcpu);
1021 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1024 * Check if userspace requested an interrupt window, and that the
1025 * interrupt window is open.
1027 * No need to exit to userspace if we already have an interrupt queued.
1029 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1030 struct kvm_run *kvm_run)
1032 return (!vcpu->irq_summary &&
1033 kvm_run->request_interrupt_window &&
1034 vcpu->interrupt_window_open &&
1035 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1038 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1039 struct kvm_run *kvm_run)
1041 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1042 kvm_run->cr8 = get_cr8(vcpu);
1043 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1044 if (irqchip_in_kernel(vcpu->kvm))
1045 kvm_run->ready_for_interrupt_injection = 1;
1047 kvm_run->ready_for_interrupt_injection =
1048 (vcpu->interrupt_window_open &&
1049 vcpu->irq_summary == 0);
1052 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1056 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1057 pr_debug("vcpu %d received sipi with vector # %x\n",
1058 vcpu->vcpu_id, vcpu->sipi_vector);
1059 kvm_lapic_reset(vcpu);
1060 r = kvm_x86_ops->vcpu_reset(vcpu);
1063 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1067 if (vcpu->guest_debug.enabled)
1068 kvm_x86_ops->guest_debug_pre(vcpu);
1071 r = kvm_mmu_reload(vcpu);
1075 kvm_inject_pending_timer_irqs(vcpu);
1079 kvm_x86_ops->prepare_guest_switch(vcpu);
1080 kvm_load_guest_fpu(vcpu);
1082 local_irq_disable();
1084 if (signal_pending(current)) {
1088 kvm_run->exit_reason = KVM_EXIT_INTR;
1089 ++vcpu->stat.signal_exits;
1093 if (irqchip_in_kernel(vcpu->kvm))
1094 kvm_x86_ops->inject_pending_irq(vcpu);
1095 else if (!vcpu->mmio_read_completed)
1096 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1098 vcpu->guest_mode = 1;
1102 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1103 kvm_x86_ops->tlb_flush(vcpu);
1105 kvm_x86_ops->run(vcpu, kvm_run);
1107 vcpu->guest_mode = 0;
1113 * We must have an instruction between local_irq_enable() and
1114 * kvm_guest_exit(), so the timer interrupt isn't delayed by
1115 * the interrupt shadow. The stat.exits increment will do nicely.
1116 * But we need to prevent reordering, hence this barrier():
1125 * Profile KVM exit RIPs:
1127 if (unlikely(prof_on == KVM_PROFILING)) {
1128 kvm_x86_ops->cache_regs(vcpu);
1129 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1132 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1135 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1137 kvm_run->exit_reason = KVM_EXIT_INTR;
1138 ++vcpu->stat.request_irq_exits;
1141 if (!need_resched()) {
1142 ++vcpu->stat.light_exits;
1153 post_kvm_run_save(vcpu, kvm_run);
1159 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1166 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1167 kvm_vcpu_block(vcpu);
1172 if (vcpu->sigset_active)
1173 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1175 /* re-sync apic's tpr */
1176 if (!irqchip_in_kernel(vcpu->kvm))
1177 set_cr8(vcpu, kvm_run->cr8);
1179 if (vcpu->pio.cur_count) {
1180 r = complete_pio(vcpu);
1184 #if CONFIG_HAS_IOMEM
1185 if (vcpu->mmio_needed) {
1186 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1187 vcpu->mmio_read_completed = 1;
1188 vcpu->mmio_needed = 0;
1189 r = emulate_instruction(vcpu, kvm_run,
1190 vcpu->mmio_fault_cr2, 0, 1);
1191 if (r == EMULATE_DO_MMIO) {
1193 * Read-modify-write. Back to userspace.
1200 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1201 kvm_x86_ops->cache_regs(vcpu);
1202 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1203 kvm_x86_ops->decache_regs(vcpu);
1206 r = __vcpu_run(vcpu, kvm_run);
1209 if (vcpu->sigset_active)
1210 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1216 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1217 struct kvm_regs *regs)
1221 kvm_x86_ops->cache_regs(vcpu);
1223 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1224 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1225 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1226 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1227 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1228 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1229 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1230 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1231 #ifdef CONFIG_X86_64
1232 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1233 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1234 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1235 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1236 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1237 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1238 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1239 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1242 regs->rip = vcpu->rip;
1243 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
1246 * Don't leak debug flags in case they were set for guest debugging
1248 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1249 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1256 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1257 struct kvm_regs *regs)
1261 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1262 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1263 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1264 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1265 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1266 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1267 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1268 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1269 #ifdef CONFIG_X86_64
1270 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1271 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1272 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1273 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1274 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1275 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1276 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1277 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1280 vcpu->rip = regs->rip;
1281 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
1283 kvm_x86_ops->decache_regs(vcpu);
1290 static void get_segment(struct kvm_vcpu *vcpu,
1291 struct kvm_segment *var, int seg)
1293 return kvm_x86_ops->get_segment(vcpu, var, seg);
1296 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1297 struct kvm_sregs *sregs)
1299 struct descriptor_table dt;
1304 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1305 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1306 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1307 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1308 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1309 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1311 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1312 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1314 kvm_x86_ops->get_idt(vcpu, &dt);
1315 sregs->idt.limit = dt.limit;
1316 sregs->idt.base = dt.base;
1317 kvm_x86_ops->get_gdt(vcpu, &dt);
1318 sregs->gdt.limit = dt.limit;
1319 sregs->gdt.base = dt.base;
1321 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1322 sregs->cr0 = vcpu->cr0;
1323 sregs->cr2 = vcpu->cr2;
1324 sregs->cr3 = vcpu->cr3;
1325 sregs->cr4 = vcpu->cr4;
1326 sregs->cr8 = get_cr8(vcpu);
1327 sregs->efer = vcpu->shadow_efer;
1328 sregs->apic_base = kvm_get_apic_base(vcpu);
1330 if (irqchip_in_kernel(vcpu->kvm)) {
1331 memset(sregs->interrupt_bitmap, 0,
1332 sizeof sregs->interrupt_bitmap);
1333 pending_vec = kvm_x86_ops->get_irq(vcpu);
1334 if (pending_vec >= 0)
1335 set_bit(pending_vec,
1336 (unsigned long *)sregs->interrupt_bitmap);
1338 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1339 sizeof sregs->interrupt_bitmap);
1346 static void set_segment(struct kvm_vcpu *vcpu,
1347 struct kvm_segment *var, int seg)
1349 return kvm_x86_ops->set_segment(vcpu, var, seg);
1352 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1353 struct kvm_sregs *sregs)
1355 int mmu_reset_needed = 0;
1356 int i, pending_vec, max_bits;
1357 struct descriptor_table dt;
1361 dt.limit = sregs->idt.limit;
1362 dt.base = sregs->idt.base;
1363 kvm_x86_ops->set_idt(vcpu, &dt);
1364 dt.limit = sregs->gdt.limit;
1365 dt.base = sregs->gdt.base;
1366 kvm_x86_ops->set_gdt(vcpu, &dt);
1368 vcpu->cr2 = sregs->cr2;
1369 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1370 vcpu->cr3 = sregs->cr3;
1372 set_cr8(vcpu, sregs->cr8);
1374 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1375 #ifdef CONFIG_X86_64
1376 kvm_x86_ops->set_efer(vcpu, sregs->efer);
1378 kvm_set_apic_base(vcpu, sregs->apic_base);
1380 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1382 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1383 vcpu->cr0 = sregs->cr0;
1384 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
1386 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1387 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1388 if (!is_long_mode(vcpu) && is_pae(vcpu))
1389 load_pdptrs(vcpu, vcpu->cr3);
1391 if (mmu_reset_needed)
1392 kvm_mmu_reset_context(vcpu);
1394 if (!irqchip_in_kernel(vcpu->kvm)) {
1395 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1396 sizeof vcpu->irq_pending);
1397 vcpu->irq_summary = 0;
1398 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
1399 if (vcpu->irq_pending[i])
1400 __set_bit(i, &vcpu->irq_summary);
1402 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
1403 pending_vec = find_first_bit(
1404 (const unsigned long *)sregs->interrupt_bitmap,
1406 /* Only pending external irq is handled here */
1407 if (pending_vec < max_bits) {
1408 kvm_x86_ops->set_irq(vcpu, pending_vec);
1409 pr_debug("Set back pending irq %d\n",
1414 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1415 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1416 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1417 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1418 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1419 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1421 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1422 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1429 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1431 struct kvm_segment cs;
1433 get_segment(vcpu, &cs, VCPU_SREG_CS);
1437 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
1440 * Translate a guest virtual address to a guest physical address.
1442 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1443 struct kvm_translation *tr)
1445 unsigned long vaddr = tr->linear_address;
1449 mutex_lock(&vcpu->kvm->lock);
1450 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1451 tr->physical_address = gpa;
1452 tr->valid = gpa != UNMAPPED_GVA;
1455 mutex_unlock(&vcpu->kvm->lock);
1461 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1462 struct kvm_interrupt *irq)
1464 if (irq->irq < 0 || irq->irq >= 256)
1466 if (irqchip_in_kernel(vcpu->kvm))
1470 set_bit(irq->irq, vcpu->irq_pending);
1471 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1478 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1479 struct kvm_debug_guest *dbg)
1485 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
1492 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1493 unsigned long address,
1496 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1497 unsigned long pgoff;
1500 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1502 page = virt_to_page(vcpu->run);
1503 else if (pgoff == KVM_PIO_PAGE_OFFSET)
1504 page = virt_to_page(vcpu->pio_data);
1506 return NOPAGE_SIGBUS;
1509 *type = VM_FAULT_MINOR;
1514 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1515 .nopage = kvm_vcpu_nopage,
1518 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1520 vma->vm_ops = &kvm_vcpu_vm_ops;
1524 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1526 struct kvm_vcpu *vcpu = filp->private_data;
1528 fput(vcpu->kvm->filp);
1532 static struct file_operations kvm_vcpu_fops = {
1533 .release = kvm_vcpu_release,
1534 .unlocked_ioctl = kvm_vcpu_ioctl,
1535 .compat_ioctl = kvm_vcpu_ioctl,
1536 .mmap = kvm_vcpu_mmap,
1540 * Allocates an inode for the vcpu.
1542 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1545 struct inode *inode;
1548 r = anon_inode_getfd(&fd, &inode, &file,
1549 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
1552 atomic_inc(&vcpu->kvm->filp->f_count);
1557 * Creates some virtual cpus. Good luck creating more than one.
1559 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1562 struct kvm_vcpu *vcpu;
1567 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
1569 return PTR_ERR(vcpu);
1571 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1573 /* We do fxsave: this must be aligned. */
1574 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
1577 r = kvm_x86_ops->vcpu_reset(vcpu);
1579 r = kvm_mmu_setup(vcpu);
1584 mutex_lock(&kvm->lock);
1585 if (kvm->vcpus[n]) {
1587 mutex_unlock(&kvm->lock);
1590 kvm->vcpus[n] = vcpu;
1591 mutex_unlock(&kvm->lock);
1593 /* Now it's all set up, let userspace reach it */
1594 r = create_vcpu_fd(vcpu);
1600 mutex_lock(&kvm->lock);
1601 kvm->vcpus[n] = NULL;
1602 mutex_unlock(&kvm->lock);
1606 kvm_mmu_unload(vcpu);
1610 kvm_x86_ops->vcpu_free(vcpu);
1614 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1617 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1618 vcpu->sigset_active = 1;
1619 vcpu->sigset = *sigset;
1621 vcpu->sigset_active = 0;
1626 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
1627 * we have asm/x86/processor.h
1638 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
1639 #ifdef CONFIG_X86_64
1640 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
1642 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
1646 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1648 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
1652 memcpy(fpu->fpr, fxsave->st_space, 128);
1653 fpu->fcw = fxsave->cwd;
1654 fpu->fsw = fxsave->swd;
1655 fpu->ftwx = fxsave->twd;
1656 fpu->last_opcode = fxsave->fop;
1657 fpu->last_ip = fxsave->rip;
1658 fpu->last_dp = fxsave->rdp;
1659 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
1666 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1668 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
1672 memcpy(fxsave->st_space, fpu->fpr, 128);
1673 fxsave->cwd = fpu->fcw;
1674 fxsave->swd = fpu->fsw;
1675 fxsave->twd = fpu->ftwx;
1676 fxsave->fop = fpu->last_opcode;
1677 fxsave->rip = fpu->last_ip;
1678 fxsave->rdp = fpu->last_dp;
1679 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
1686 static long kvm_vcpu_ioctl(struct file *filp,
1687 unsigned int ioctl, unsigned long arg)
1689 struct kvm_vcpu *vcpu = filp->private_data;
1690 void __user *argp = (void __user *)arg;
1698 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
1700 case KVM_GET_REGS: {
1701 struct kvm_regs kvm_regs;
1703 memset(&kvm_regs, 0, sizeof kvm_regs);
1704 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
1708 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1713 case KVM_SET_REGS: {
1714 struct kvm_regs kvm_regs;
1717 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1719 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
1725 case KVM_GET_SREGS: {
1726 struct kvm_sregs kvm_sregs;
1728 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1729 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1733 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1738 case KVM_SET_SREGS: {
1739 struct kvm_sregs kvm_sregs;
1742 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1744 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1750 case KVM_TRANSLATE: {
1751 struct kvm_translation tr;
1754 if (copy_from_user(&tr, argp, sizeof tr))
1756 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
1760 if (copy_to_user(argp, &tr, sizeof tr))
1765 case KVM_INTERRUPT: {
1766 struct kvm_interrupt irq;
1769 if (copy_from_user(&irq, argp, sizeof irq))
1771 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1777 case KVM_DEBUG_GUEST: {
1778 struct kvm_debug_guest dbg;
1781 if (copy_from_user(&dbg, argp, sizeof dbg))
1783 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
1789 case KVM_SET_SIGNAL_MASK: {
1790 struct kvm_signal_mask __user *sigmask_arg = argp;
1791 struct kvm_signal_mask kvm_sigmask;
1792 sigset_t sigset, *p;
1797 if (copy_from_user(&kvm_sigmask, argp,
1798 sizeof kvm_sigmask))
1801 if (kvm_sigmask.len != sizeof sigset)
1804 if (copy_from_user(&sigset, sigmask_arg->sigset,
1809 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1815 memset(&fpu, 0, sizeof fpu);
1816 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
1820 if (copy_to_user(argp, &fpu, sizeof fpu))
1829 if (copy_from_user(&fpu, argp, sizeof fpu))
1831 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
1838 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1844 static long kvm_vm_ioctl(struct file *filp,
1845 unsigned int ioctl, unsigned long arg)
1847 struct kvm *kvm = filp->private_data;
1848 void __user *argp = (void __user *)arg;
1852 case KVM_CREATE_VCPU:
1853 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1857 case KVM_SET_USER_MEMORY_REGION: {
1858 struct kvm_userspace_memory_region kvm_userspace_mem;
1861 if (copy_from_user(&kvm_userspace_mem, argp,
1862 sizeof kvm_userspace_mem))
1865 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1870 case KVM_GET_DIRTY_LOG: {
1871 struct kvm_dirty_log log;
1874 if (copy_from_user(&log, argp, sizeof log))
1876 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1882 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1888 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1889 unsigned long address,
1892 struct kvm *kvm = vma->vm_file->private_data;
1893 unsigned long pgoff;
1896 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1897 if (!kvm_is_visible_gfn(kvm, pgoff))
1898 return NOPAGE_SIGBUS;
1899 /* current->mm->mmap_sem is already held so call lockless version */
1900 page = __gfn_to_page(kvm, pgoff);
1901 if (is_error_page(page)) {
1902 kvm_release_page(page);
1903 return NOPAGE_SIGBUS;
1906 *type = VM_FAULT_MINOR;
1911 static struct vm_operations_struct kvm_vm_vm_ops = {
1912 .nopage = kvm_vm_nopage,
1915 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1917 vma->vm_ops = &kvm_vm_vm_ops;
1921 static struct file_operations kvm_vm_fops = {
1922 .release = kvm_vm_release,
1923 .unlocked_ioctl = kvm_vm_ioctl,
1924 .compat_ioctl = kvm_vm_ioctl,
1925 .mmap = kvm_vm_mmap,
1928 static int kvm_dev_ioctl_create_vm(void)
1931 struct inode *inode;
1935 kvm = kvm_create_vm();
1937 return PTR_ERR(kvm);
1938 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1940 kvm_destroy_vm(kvm);
1949 static long kvm_dev_ioctl(struct file *filp,
1950 unsigned int ioctl, unsigned long arg)
1952 void __user *argp = (void __user *)arg;
1956 case KVM_GET_API_VERSION:
1960 r = KVM_API_VERSION;
1966 r = kvm_dev_ioctl_create_vm();
1968 case KVM_CHECK_EXTENSION: {
1969 int ext = (long)argp;
1972 case KVM_CAP_IRQCHIP:
1974 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1975 case KVM_CAP_USER_MEMORY:
1976 case KVM_CAP_SET_TSS_ADDR:
1985 case KVM_GET_VCPU_MMAP_SIZE:
1992 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1998 static struct file_operations kvm_chardev_ops = {
1999 .unlocked_ioctl = kvm_dev_ioctl,
2000 .compat_ioctl = kvm_dev_ioctl,
2003 static struct miscdevice kvm_dev = {
2010 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2013 static void decache_vcpus_on_cpu(int cpu)
2016 struct kvm_vcpu *vcpu;
2019 spin_lock(&kvm_lock);
2020 list_for_each_entry(vm, &vm_list, vm_list)
2021 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2022 vcpu = vm->vcpus[i];
2026 * If the vcpu is locked, then it is running on some
2027 * other cpu and therefore it is not cached on the
2030 * If it's not locked, check the last cpu it executed
2033 if (mutex_trylock(&vcpu->mutex)) {
2034 if (vcpu->cpu == cpu) {
2035 kvm_x86_ops->vcpu_decache(vcpu);
2038 mutex_unlock(&vcpu->mutex);
2041 spin_unlock(&kvm_lock);
2044 static void hardware_enable(void *junk)
2046 int cpu = raw_smp_processor_id();
2048 if (cpu_isset(cpu, cpus_hardware_enabled))
2050 cpu_set(cpu, cpus_hardware_enabled);
2051 kvm_x86_ops->hardware_enable(NULL);
2054 static void hardware_disable(void *junk)
2056 int cpu = raw_smp_processor_id();
2058 if (!cpu_isset(cpu, cpus_hardware_enabled))
2060 cpu_clear(cpu, cpus_hardware_enabled);
2061 decache_vcpus_on_cpu(cpu);
2062 kvm_x86_ops->hardware_disable(NULL);
2065 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2072 case CPU_DYING_FROZEN:
2073 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2075 hardware_disable(NULL);
2077 case CPU_UP_CANCELED:
2078 case CPU_UP_CANCELED_FROZEN:
2079 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2081 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2084 case CPU_ONLINE_FROZEN:
2085 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2087 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2093 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2096 if (val == SYS_RESTART) {
2098 * Some (well, at least mine) BIOSes hang on reboot if
2101 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2102 on_each_cpu(hardware_disable, NULL, 0, 1);
2107 static struct notifier_block kvm_reboot_notifier = {
2108 .notifier_call = kvm_reboot,
2112 void kvm_io_bus_init(struct kvm_io_bus *bus)
2114 memset(bus, 0, sizeof(*bus));
2117 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2121 for (i = 0; i < bus->dev_count; i++) {
2122 struct kvm_io_device *pos = bus->devs[i];
2124 kvm_iodevice_destructor(pos);
2128 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2132 for (i = 0; i < bus->dev_count; i++) {
2133 struct kvm_io_device *pos = bus->devs[i];
2135 if (pos->in_range(pos, addr))
2142 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2144 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2146 bus->devs[bus->dev_count++] = dev;
2149 static struct notifier_block kvm_cpu_notifier = {
2150 .notifier_call = kvm_cpu_hotplug,
2151 .priority = 20, /* must be > scheduler priority */
2154 static u64 stat_get(void *_offset)
2156 unsigned offset = (long)_offset;
2159 struct kvm_vcpu *vcpu;
2162 spin_lock(&kvm_lock);
2163 list_for_each_entry(kvm, &vm_list, vm_list)
2164 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2165 vcpu = kvm->vcpus[i];
2167 total += *(u32 *)((void *)vcpu + offset);
2169 spin_unlock(&kvm_lock);
2173 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
2175 static __init void kvm_init_debug(void)
2177 struct kvm_stats_debugfs_item *p;
2179 debugfs_dir = debugfs_create_dir("kvm", NULL);
2180 for (p = debugfs_entries; p->name; ++p)
2181 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
2182 (void *)(long)p->offset,
2186 static void kvm_exit_debug(void)
2188 struct kvm_stats_debugfs_item *p;
2190 for (p = debugfs_entries; p->name; ++p)
2191 debugfs_remove(p->dentry);
2192 debugfs_remove(debugfs_dir);
2195 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2197 hardware_disable(NULL);
2201 static int kvm_resume(struct sys_device *dev)
2203 hardware_enable(NULL);
2207 static struct sysdev_class kvm_sysdev_class = {
2209 .suspend = kvm_suspend,
2210 .resume = kvm_resume,
2213 static struct sys_device kvm_sysdev = {
2215 .cls = &kvm_sysdev_class,
2218 struct page *bad_page;
2221 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2223 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2226 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2228 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2230 kvm_x86_ops->vcpu_load(vcpu, cpu);
2233 static void kvm_sched_out(struct preempt_notifier *pn,
2234 struct task_struct *next)
2236 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2238 kvm_x86_ops->vcpu_put(vcpu);
2241 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
2242 struct module *module)
2248 printk(KERN_ERR "kvm: already loaded the other module\n");
2252 if (!ops->cpu_has_kvm_support()) {
2253 printk(KERN_ERR "kvm: no hardware support\n");
2256 if (ops->disabled_by_bios()) {
2257 printk(KERN_ERR "kvm: disabled by bios\n");
2263 r = kvm_x86_ops->hardware_setup();
2267 for_each_online_cpu(cpu) {
2268 smp_call_function_single(cpu,
2269 kvm_x86_ops->check_processor_compatibility,
2275 on_each_cpu(hardware_enable, NULL, 0, 1);
2276 r = register_cpu_notifier(&kvm_cpu_notifier);
2279 register_reboot_notifier(&kvm_reboot_notifier);
2281 r = sysdev_class_register(&kvm_sysdev_class);
2285 r = sysdev_register(&kvm_sysdev);
2289 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2290 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2291 __alignof__(struct kvm_vcpu), 0, 0);
2292 if (!kvm_vcpu_cache) {
2297 kvm_chardev_ops.owner = module;
2299 r = misc_register(&kvm_dev);
2301 printk(KERN_ERR "kvm: misc device register failed\n");
2305 kvm_preempt_ops.sched_in = kvm_sched_in;
2306 kvm_preempt_ops.sched_out = kvm_sched_out;
2308 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2313 kmem_cache_destroy(kvm_vcpu_cache);
2315 sysdev_unregister(&kvm_sysdev);
2317 sysdev_class_unregister(&kvm_sysdev_class);
2319 unregister_reboot_notifier(&kvm_reboot_notifier);
2320 unregister_cpu_notifier(&kvm_cpu_notifier);
2322 on_each_cpu(hardware_disable, NULL, 0, 1);
2324 kvm_x86_ops->hardware_unsetup();
2329 EXPORT_SYMBOL_GPL(kvm_init_x86);
2331 void kvm_exit_x86(void)
2333 misc_deregister(&kvm_dev);
2334 kmem_cache_destroy(kvm_vcpu_cache);
2335 sysdev_unregister(&kvm_sysdev);
2336 sysdev_class_unregister(&kvm_sysdev_class);
2337 unregister_reboot_notifier(&kvm_reboot_notifier);
2338 unregister_cpu_notifier(&kvm_cpu_notifier);
2339 on_each_cpu(hardware_disable, NULL, 0, 1);
2340 kvm_x86_ops->hardware_unsetup();
2343 EXPORT_SYMBOL_GPL(kvm_exit_x86);
2345 static __init int kvm_init(void)
2349 r = kvm_mmu_module_init();
2357 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2359 if (bad_page == NULL) {
2368 kvm_mmu_module_exit();
2373 static __exit void kvm_exit(void)
2376 __free_page(bad_page);
2377 kvm_mmu_module_exit();
2380 module_init(kvm_init)
2381 module_exit(kvm_exit)