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 <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
48 #include <asm/processor.h>
50 #include <asm/uaccess.h>
51 #include <asm/pgtable.h>
52 #include <asm-generic/bitops/le.h>
54 #include "coalesced_mmio.h"
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/kvm.h>
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
65 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
68 DEFINE_SPINLOCK(kvm_lock);
71 static cpumask_var_t cpus_hardware_enabled;
72 static int kvm_usage_count = 0;
73 static atomic_t hardware_enable_failed;
75 struct kmem_cache *kvm_vcpu_cache;
76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
78 static __read_mostly struct preempt_ops kvm_preempt_ops;
80 struct dentry *kvm_debugfs_dir;
82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
84 static int hardware_enable_all(void);
85 static void hardware_disable_all(void);
87 static bool kvm_rebooting;
89 static bool largepages_enabled = true;
91 inline int kvm_is_mmio_pfn(pfn_t pfn)
94 struct page *page = compound_head(pfn_to_page(pfn));
95 return PageReserved(page);
102 * Switches to specified vcpu, until a matching vcpu_put()
104 void vcpu_load(struct kvm_vcpu *vcpu)
108 mutex_lock(&vcpu->mutex);
110 preempt_notifier_register(&vcpu->preempt_notifier);
111 kvm_arch_vcpu_load(vcpu, cpu);
115 void vcpu_put(struct kvm_vcpu *vcpu)
118 kvm_arch_vcpu_put(vcpu);
119 preempt_notifier_unregister(&vcpu->preempt_notifier);
121 mutex_unlock(&vcpu->mutex);
124 static void ack_flush(void *_completed)
128 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
133 struct kvm_vcpu *vcpu;
135 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
137 spin_lock(&kvm->requests_lock);
138 me = smp_processor_id();
139 kvm_for_each_vcpu(i, vcpu, kvm) {
140 if (test_and_set_bit(req, &vcpu->requests))
143 if (cpus != NULL && cpu != -1 && cpu != me)
144 cpumask_set_cpu(cpu, cpus);
146 if (unlikely(cpus == NULL))
147 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
148 else if (!cpumask_empty(cpus))
149 smp_call_function_many(cpus, ack_flush, NULL, 1);
152 spin_unlock(&kvm->requests_lock);
153 free_cpumask_var(cpus);
157 void kvm_flush_remote_tlbs(struct kvm *kvm)
159 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
160 ++kvm->stat.remote_tlb_flush;
163 void kvm_reload_remote_mmus(struct kvm *kvm)
165 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
168 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
173 mutex_init(&vcpu->mutex);
177 init_waitqueue_head(&vcpu->wq);
179 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
184 vcpu->run = page_address(page);
186 r = kvm_arch_vcpu_init(vcpu);
192 free_page((unsigned long)vcpu->run);
196 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
198 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
200 kvm_arch_vcpu_uninit(vcpu);
201 free_page((unsigned long)vcpu->run);
203 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
205 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
206 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
208 return container_of(mn, struct kvm, mmu_notifier);
211 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
212 struct mm_struct *mm,
213 unsigned long address)
215 struct kvm *kvm = mmu_notifier_to_kvm(mn);
219 * When ->invalidate_page runs, the linux pte has been zapped
220 * already but the page is still allocated until
221 * ->invalidate_page returns. So if we increase the sequence
222 * here the kvm page fault will notice if the spte can't be
223 * established because the page is going to be freed. If
224 * instead the kvm page fault establishes the spte before
225 * ->invalidate_page runs, kvm_unmap_hva will release it
228 * The sequence increase only need to be seen at spin_unlock
229 * time, and not at spin_lock time.
231 * Increasing the sequence after the spin_unlock would be
232 * unsafe because the kvm page fault could then establish the
233 * pte after kvm_unmap_hva returned, without noticing the page
234 * is going to be freed.
236 spin_lock(&kvm->mmu_lock);
237 kvm->mmu_notifier_seq++;
238 need_tlb_flush = kvm_unmap_hva(kvm, address);
239 spin_unlock(&kvm->mmu_lock);
241 /* we've to flush the tlb before the pages can be freed */
243 kvm_flush_remote_tlbs(kvm);
247 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
248 struct mm_struct *mm,
249 unsigned long address,
252 struct kvm *kvm = mmu_notifier_to_kvm(mn);
254 spin_lock(&kvm->mmu_lock);
255 kvm->mmu_notifier_seq++;
256 kvm_set_spte_hva(kvm, address, pte);
257 spin_unlock(&kvm->mmu_lock);
260 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
261 struct mm_struct *mm,
265 struct kvm *kvm = mmu_notifier_to_kvm(mn);
266 int need_tlb_flush = 0;
268 spin_lock(&kvm->mmu_lock);
270 * The count increase must become visible at unlock time as no
271 * spte can be established without taking the mmu_lock and
272 * count is also read inside the mmu_lock critical section.
274 kvm->mmu_notifier_count++;
275 for (; start < end; start += PAGE_SIZE)
276 need_tlb_flush |= kvm_unmap_hva(kvm, start);
277 spin_unlock(&kvm->mmu_lock);
279 /* we've to flush the tlb before the pages can be freed */
281 kvm_flush_remote_tlbs(kvm);
284 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
285 struct mm_struct *mm,
289 struct kvm *kvm = mmu_notifier_to_kvm(mn);
291 spin_lock(&kvm->mmu_lock);
293 * This sequence increase will notify the kvm page fault that
294 * the page that is going to be mapped in the spte could have
297 kvm->mmu_notifier_seq++;
299 * The above sequence increase must be visible before the
300 * below count decrease but both values are read by the kvm
301 * page fault under mmu_lock spinlock so we don't need to add
302 * a smb_wmb() here in between the two.
304 kvm->mmu_notifier_count--;
305 spin_unlock(&kvm->mmu_lock);
307 BUG_ON(kvm->mmu_notifier_count < 0);
310 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
311 struct mm_struct *mm,
312 unsigned long address)
314 struct kvm *kvm = mmu_notifier_to_kvm(mn);
317 spin_lock(&kvm->mmu_lock);
318 young = kvm_age_hva(kvm, address);
319 spin_unlock(&kvm->mmu_lock);
322 kvm_flush_remote_tlbs(kvm);
327 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
328 struct mm_struct *mm)
330 struct kvm *kvm = mmu_notifier_to_kvm(mn);
331 kvm_arch_flush_shadow(kvm);
334 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
335 .invalidate_page = kvm_mmu_notifier_invalidate_page,
336 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
337 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
338 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
339 .change_pte = kvm_mmu_notifier_change_pte,
340 .release = kvm_mmu_notifier_release,
343 static int kvm_init_mmu_notifier(struct kvm *kvm)
345 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
346 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
349 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
351 static int kvm_init_mmu_notifier(struct kvm *kvm)
356 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
358 static struct kvm *kvm_create_vm(void)
361 struct kvm *kvm = kvm_arch_create_vm();
362 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
369 r = hardware_enable_all();
371 goto out_err_nodisable;
373 #ifdef CONFIG_HAVE_KVM_IRQCHIP
374 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
375 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
379 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
383 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
384 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
388 kvm->coalesced_mmio_ring =
389 (struct kvm_coalesced_mmio_ring *)page_address(page);
392 r = kvm_init_mmu_notifier(kvm);
394 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
400 kvm->mm = current->mm;
401 atomic_inc(&kvm->mm->mm_count);
402 spin_lock_init(&kvm->mmu_lock);
403 spin_lock_init(&kvm->requests_lock);
404 kvm_io_bus_init(&kvm->pio_bus);
405 kvm_eventfd_init(kvm);
406 mutex_init(&kvm->lock);
407 mutex_init(&kvm->irq_lock);
408 kvm_io_bus_init(&kvm->mmio_bus);
409 init_rwsem(&kvm->slots_lock);
410 atomic_set(&kvm->users_count, 1);
411 spin_lock(&kvm_lock);
412 list_add(&kvm->vm_list, &vm_list);
413 spin_unlock(&kvm_lock);
414 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
415 kvm_coalesced_mmio_init(kvm);
421 hardware_disable_all();
423 kfree(kvm->memslots);
429 * Free any memory in @free but not in @dont.
431 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
432 struct kvm_memory_slot *dont)
436 if (!dont || free->rmap != dont->rmap)
439 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
440 vfree(free->dirty_bitmap);
443 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
444 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
445 vfree(free->lpage_info[i]);
446 free->lpage_info[i] = NULL;
451 free->dirty_bitmap = NULL;
455 void kvm_free_physmem(struct kvm *kvm)
458 struct kvm_memslots *slots = kvm->memslots;
460 for (i = 0; i < slots->nmemslots; ++i)
461 kvm_free_physmem_slot(&slots->memslots[i], NULL);
463 kfree(kvm->memslots);
466 static void kvm_destroy_vm(struct kvm *kvm)
468 struct mm_struct *mm = kvm->mm;
470 kvm_arch_sync_events(kvm);
471 spin_lock(&kvm_lock);
472 list_del(&kvm->vm_list);
473 spin_unlock(&kvm_lock);
474 kvm_free_irq_routing(kvm);
475 kvm_io_bus_destroy(&kvm->pio_bus);
476 kvm_io_bus_destroy(&kvm->mmio_bus);
477 kvm_coalesced_mmio_free(kvm);
478 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
479 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
481 kvm_arch_flush_shadow(kvm);
483 kvm_arch_destroy_vm(kvm);
484 hardware_disable_all();
488 void kvm_get_kvm(struct kvm *kvm)
490 atomic_inc(&kvm->users_count);
492 EXPORT_SYMBOL_GPL(kvm_get_kvm);
494 void kvm_put_kvm(struct kvm *kvm)
496 if (atomic_dec_and_test(&kvm->users_count))
499 EXPORT_SYMBOL_GPL(kvm_put_kvm);
502 static int kvm_vm_release(struct inode *inode, struct file *filp)
504 struct kvm *kvm = filp->private_data;
506 kvm_irqfd_release(kvm);
513 * Allocate some memory and give it an address in the guest physical address
516 * Discontiguous memory is allowed, mostly for framebuffers.
518 * Must be called holding mmap_sem for write.
520 int __kvm_set_memory_region(struct kvm *kvm,
521 struct kvm_userspace_memory_region *mem,
526 unsigned long npages;
528 struct kvm_memory_slot *memslot;
529 struct kvm_memory_slot old, new;
532 /* General sanity checks */
533 if (mem->memory_size & (PAGE_SIZE - 1))
535 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
537 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
539 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
541 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
544 memslot = &kvm->memslots->memslots[mem->slot];
545 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
546 npages = mem->memory_size >> PAGE_SHIFT;
549 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
551 new = old = *memslot;
553 new.base_gfn = base_gfn;
555 new.flags = mem->flags;
557 /* Disallow changing a memory slot's size. */
559 if (npages && old.npages && npages != old.npages)
562 /* Check for overlaps */
564 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
565 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
567 if (s == memslot || !s->npages)
569 if (!((base_gfn + npages <= s->base_gfn) ||
570 (base_gfn >= s->base_gfn + s->npages)))
574 /* Free page dirty bitmap if unneeded */
575 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
576 new.dirty_bitmap = NULL;
580 /* Allocate if a slot is being created */
582 if (npages && !new.rmap) {
583 new.rmap = vmalloc(npages * sizeof(struct page *));
588 memset(new.rmap, 0, npages * sizeof(*new.rmap));
590 new.user_alloc = user_alloc;
592 * hva_to_rmmap() serialzies with the mmu_lock and to be
593 * safe it has to ignore memslots with !user_alloc &&
597 new.userspace_addr = mem->userspace_addr;
599 new.userspace_addr = 0;
604 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
610 /* Avoid unused variable warning if no large pages */
613 if (new.lpage_info[i])
616 lpages = 1 + (base_gfn + npages - 1) /
617 KVM_PAGES_PER_HPAGE(level);
618 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
620 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
622 if (!new.lpage_info[i])
625 memset(new.lpage_info[i], 0,
626 lpages * sizeof(*new.lpage_info[i]));
628 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
629 new.lpage_info[i][0].write_count = 1;
630 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
631 new.lpage_info[i][lpages - 1].write_count = 1;
632 ugfn = new.userspace_addr >> PAGE_SHIFT;
634 * If the gfn and userspace address are not aligned wrt each
635 * other, or if explicitly asked to, disable large page
636 * support for this slot
638 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
640 for (j = 0; j < lpages; ++j)
641 new.lpage_info[i][j].write_count = 1;
646 /* Allocate page dirty bitmap if needed */
647 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
648 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
650 new.dirty_bitmap = vmalloc(dirty_bytes);
651 if (!new.dirty_bitmap)
653 memset(new.dirty_bitmap, 0, dirty_bytes);
655 kvm_arch_flush_shadow(kvm);
657 #else /* not defined CONFIG_S390 */
658 new.user_alloc = user_alloc;
660 new.userspace_addr = mem->userspace_addr;
661 #endif /* not defined CONFIG_S390 */
664 kvm_arch_flush_shadow(kvm);
666 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
670 spin_lock(&kvm->mmu_lock);
671 if (mem->slot >= kvm->memslots->nmemslots)
672 kvm->memslots->nmemslots = mem->slot + 1;
675 spin_unlock(&kvm->mmu_lock);
677 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
679 kvm_free_physmem_slot(&old, npages ? &new : NULL);
680 /* Slot deletion case: we have to update the current slot */
681 spin_lock(&kvm->mmu_lock);
684 spin_unlock(&kvm->mmu_lock);
686 /* map the pages in iommu page table */
687 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
694 kvm_free_physmem_slot(&new, &old);
699 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
701 int kvm_set_memory_region(struct kvm *kvm,
702 struct kvm_userspace_memory_region *mem,
707 down_write(&kvm->slots_lock);
708 r = __kvm_set_memory_region(kvm, mem, user_alloc);
709 up_write(&kvm->slots_lock);
712 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
714 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
716 kvm_userspace_memory_region *mem,
719 if (mem->slot >= KVM_MEMORY_SLOTS)
721 return kvm_set_memory_region(kvm, mem, user_alloc);
724 int kvm_get_dirty_log(struct kvm *kvm,
725 struct kvm_dirty_log *log, int *is_dirty)
727 struct kvm_memory_slot *memslot;
730 unsigned long any = 0;
733 if (log->slot >= KVM_MEMORY_SLOTS)
736 memslot = &kvm->memslots->memslots[log->slot];
738 if (!memslot->dirty_bitmap)
741 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
743 for (i = 0; !any && i < n/sizeof(long); ++i)
744 any = memslot->dirty_bitmap[i];
747 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
758 void kvm_disable_largepages(void)
760 largepages_enabled = false;
762 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
764 int is_error_page(struct page *page)
766 return page == bad_page;
768 EXPORT_SYMBOL_GPL(is_error_page);
770 int is_error_pfn(pfn_t pfn)
772 return pfn == bad_pfn;
774 EXPORT_SYMBOL_GPL(is_error_pfn);
776 static inline unsigned long bad_hva(void)
781 int kvm_is_error_hva(unsigned long addr)
783 return addr == bad_hva();
785 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
787 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
790 struct kvm_memslots *slots = kvm->memslots;
792 for (i = 0; i < slots->nmemslots; ++i) {
793 struct kvm_memory_slot *memslot = &slots->memslots[i];
795 if (gfn >= memslot->base_gfn
796 && gfn < memslot->base_gfn + memslot->npages)
801 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
803 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
805 gfn = unalias_gfn(kvm, gfn);
806 return gfn_to_memslot_unaliased(kvm, gfn);
809 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
812 struct kvm_memslots *slots = kvm->memslots;
814 gfn = unalias_gfn(kvm, gfn);
815 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
816 struct kvm_memory_slot *memslot = &slots->memslots[i];
818 if (gfn >= memslot->base_gfn
819 && gfn < memslot->base_gfn + memslot->npages)
824 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
826 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
828 struct kvm_memory_slot *slot;
830 gfn = unalias_gfn(kvm, gfn);
831 slot = gfn_to_memslot_unaliased(kvm, gfn);
834 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
836 EXPORT_SYMBOL_GPL(gfn_to_hva);
838 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
840 struct page *page[1];
847 addr = gfn_to_hva(kvm, gfn);
848 if (kvm_is_error_hva(addr)) {
850 return page_to_pfn(bad_page);
853 npages = get_user_pages_fast(addr, 1, 1, page);
855 if (unlikely(npages != 1)) {
856 struct vm_area_struct *vma;
858 down_read(¤t->mm->mmap_sem);
859 vma = find_vma(current->mm, addr);
861 if (vma == NULL || addr < vma->vm_start ||
862 !(vma->vm_flags & VM_PFNMAP)) {
863 up_read(¤t->mm->mmap_sem);
865 return page_to_pfn(bad_page);
868 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
869 up_read(¤t->mm->mmap_sem);
870 BUG_ON(!kvm_is_mmio_pfn(pfn));
872 pfn = page_to_pfn(page[0]);
877 EXPORT_SYMBOL_GPL(gfn_to_pfn);
879 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
883 pfn = gfn_to_pfn(kvm, gfn);
884 if (!kvm_is_mmio_pfn(pfn))
885 return pfn_to_page(pfn);
887 WARN_ON(kvm_is_mmio_pfn(pfn));
893 EXPORT_SYMBOL_GPL(gfn_to_page);
895 void kvm_release_page_clean(struct page *page)
897 kvm_release_pfn_clean(page_to_pfn(page));
899 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
901 void kvm_release_pfn_clean(pfn_t pfn)
903 if (!kvm_is_mmio_pfn(pfn))
904 put_page(pfn_to_page(pfn));
906 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
908 void kvm_release_page_dirty(struct page *page)
910 kvm_release_pfn_dirty(page_to_pfn(page));
912 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
914 void kvm_release_pfn_dirty(pfn_t pfn)
916 kvm_set_pfn_dirty(pfn);
917 kvm_release_pfn_clean(pfn);
919 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
921 void kvm_set_page_dirty(struct page *page)
923 kvm_set_pfn_dirty(page_to_pfn(page));
925 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
927 void kvm_set_pfn_dirty(pfn_t pfn)
929 if (!kvm_is_mmio_pfn(pfn)) {
930 struct page *page = pfn_to_page(pfn);
931 if (!PageReserved(page))
935 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
937 void kvm_set_pfn_accessed(pfn_t pfn)
939 if (!kvm_is_mmio_pfn(pfn))
940 mark_page_accessed(pfn_to_page(pfn));
942 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
944 void kvm_get_pfn(pfn_t pfn)
946 if (!kvm_is_mmio_pfn(pfn))
947 get_page(pfn_to_page(pfn));
949 EXPORT_SYMBOL_GPL(kvm_get_pfn);
951 static int next_segment(unsigned long len, int offset)
953 if (len > PAGE_SIZE - offset)
954 return PAGE_SIZE - offset;
959 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
965 addr = gfn_to_hva(kvm, gfn);
966 if (kvm_is_error_hva(addr))
968 r = copy_from_user(data, (void __user *)addr + offset, len);
973 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
975 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
977 gfn_t gfn = gpa >> PAGE_SHIFT;
979 int offset = offset_in_page(gpa);
982 while ((seg = next_segment(len, offset)) != 0) {
983 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
993 EXPORT_SYMBOL_GPL(kvm_read_guest);
995 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1000 gfn_t gfn = gpa >> PAGE_SHIFT;
1001 int offset = offset_in_page(gpa);
1003 addr = gfn_to_hva(kvm, gfn);
1004 if (kvm_is_error_hva(addr))
1006 pagefault_disable();
1007 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1013 EXPORT_SYMBOL(kvm_read_guest_atomic);
1015 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1016 int offset, int len)
1021 addr = gfn_to_hva(kvm, gfn);
1022 if (kvm_is_error_hva(addr))
1024 r = copy_to_user((void __user *)addr + offset, data, len);
1027 mark_page_dirty(kvm, gfn);
1030 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1032 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1035 gfn_t gfn = gpa >> PAGE_SHIFT;
1037 int offset = offset_in_page(gpa);
1040 while ((seg = next_segment(len, offset)) != 0) {
1041 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1052 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1054 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1056 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1058 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1060 gfn_t gfn = gpa >> PAGE_SHIFT;
1062 int offset = offset_in_page(gpa);
1065 while ((seg = next_segment(len, offset)) != 0) {
1066 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1075 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1077 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1079 struct kvm_memory_slot *memslot;
1081 gfn = unalias_gfn(kvm, gfn);
1082 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1083 if (memslot && memslot->dirty_bitmap) {
1084 unsigned long rel_gfn = gfn - memslot->base_gfn;
1087 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1088 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1093 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1095 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1100 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1102 if (kvm_arch_vcpu_runnable(vcpu)) {
1103 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1106 if (kvm_cpu_has_pending_timer(vcpu))
1108 if (signal_pending(current))
1114 finish_wait(&vcpu->wq, &wait);
1117 void kvm_resched(struct kvm_vcpu *vcpu)
1119 if (!need_resched())
1123 EXPORT_SYMBOL_GPL(kvm_resched);
1125 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1130 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1132 /* Sleep for 100 us, and hope lock-holder got scheduled */
1133 expires = ktime_add_ns(ktime_get(), 100000UL);
1134 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1136 finish_wait(&vcpu->wq, &wait);
1138 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1140 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1142 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1145 if (vmf->pgoff == 0)
1146 page = virt_to_page(vcpu->run);
1148 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1149 page = virt_to_page(vcpu->arch.pio_data);
1151 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1152 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1153 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1156 return VM_FAULT_SIGBUS;
1162 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1163 .fault = kvm_vcpu_fault,
1166 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1168 vma->vm_ops = &kvm_vcpu_vm_ops;
1172 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1174 struct kvm_vcpu *vcpu = filp->private_data;
1176 kvm_put_kvm(vcpu->kvm);
1180 static struct file_operations kvm_vcpu_fops = {
1181 .release = kvm_vcpu_release,
1182 .unlocked_ioctl = kvm_vcpu_ioctl,
1183 .compat_ioctl = kvm_vcpu_ioctl,
1184 .mmap = kvm_vcpu_mmap,
1188 * Allocates an inode for the vcpu.
1190 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1192 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1196 * Creates some virtual cpus. Good luck creating more than one.
1198 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1201 struct kvm_vcpu *vcpu, *v;
1203 vcpu = kvm_arch_vcpu_create(kvm, id);
1205 return PTR_ERR(vcpu);
1207 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1209 r = kvm_arch_vcpu_setup(vcpu);
1213 mutex_lock(&kvm->lock);
1214 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1219 kvm_for_each_vcpu(r, v, kvm)
1220 if (v->vcpu_id == id) {
1225 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1227 /* Now it's all set up, let userspace reach it */
1229 r = create_vcpu_fd(vcpu);
1235 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1237 atomic_inc(&kvm->online_vcpus);
1239 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1240 if (kvm->bsp_vcpu_id == id)
1241 kvm->bsp_vcpu = vcpu;
1243 mutex_unlock(&kvm->lock);
1247 mutex_unlock(&kvm->lock);
1248 kvm_arch_vcpu_destroy(vcpu);
1252 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1255 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1256 vcpu->sigset_active = 1;
1257 vcpu->sigset = *sigset;
1259 vcpu->sigset_active = 0;
1263 static long kvm_vcpu_ioctl(struct file *filp,
1264 unsigned int ioctl, unsigned long arg)
1266 struct kvm_vcpu *vcpu = filp->private_data;
1267 void __user *argp = (void __user *)arg;
1269 struct kvm_fpu *fpu = NULL;
1270 struct kvm_sregs *kvm_sregs = NULL;
1272 if (vcpu->kvm->mm != current->mm)
1279 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1281 case KVM_GET_REGS: {
1282 struct kvm_regs *kvm_regs;
1285 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1288 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1292 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1299 case KVM_SET_REGS: {
1300 struct kvm_regs *kvm_regs;
1303 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1307 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1309 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1317 case KVM_GET_SREGS: {
1318 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1322 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1326 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1331 case KVM_SET_SREGS: {
1332 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1337 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1339 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1345 case KVM_GET_MP_STATE: {
1346 struct kvm_mp_state mp_state;
1348 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1352 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1357 case KVM_SET_MP_STATE: {
1358 struct kvm_mp_state mp_state;
1361 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1363 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1369 case KVM_TRANSLATE: {
1370 struct kvm_translation tr;
1373 if (copy_from_user(&tr, argp, sizeof tr))
1375 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1379 if (copy_to_user(argp, &tr, sizeof tr))
1384 case KVM_SET_GUEST_DEBUG: {
1385 struct kvm_guest_debug dbg;
1388 if (copy_from_user(&dbg, argp, sizeof dbg))
1390 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1396 case KVM_SET_SIGNAL_MASK: {
1397 struct kvm_signal_mask __user *sigmask_arg = argp;
1398 struct kvm_signal_mask kvm_sigmask;
1399 sigset_t sigset, *p;
1404 if (copy_from_user(&kvm_sigmask, argp,
1405 sizeof kvm_sigmask))
1408 if (kvm_sigmask.len != sizeof sigset)
1411 if (copy_from_user(&sigset, sigmask_arg->sigset,
1416 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1420 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1424 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1428 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1434 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1439 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1441 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1448 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1456 static long kvm_vm_ioctl(struct file *filp,
1457 unsigned int ioctl, unsigned long arg)
1459 struct kvm *kvm = filp->private_data;
1460 void __user *argp = (void __user *)arg;
1463 if (kvm->mm != current->mm)
1466 case KVM_CREATE_VCPU:
1467 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1471 case KVM_SET_USER_MEMORY_REGION: {
1472 struct kvm_userspace_memory_region kvm_userspace_mem;
1475 if (copy_from_user(&kvm_userspace_mem, argp,
1476 sizeof kvm_userspace_mem))
1479 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1484 case KVM_GET_DIRTY_LOG: {
1485 struct kvm_dirty_log log;
1488 if (copy_from_user(&log, argp, sizeof log))
1490 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1495 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1496 case KVM_REGISTER_COALESCED_MMIO: {
1497 struct kvm_coalesced_mmio_zone zone;
1499 if (copy_from_user(&zone, argp, sizeof zone))
1502 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1508 case KVM_UNREGISTER_COALESCED_MMIO: {
1509 struct kvm_coalesced_mmio_zone zone;
1511 if (copy_from_user(&zone, argp, sizeof zone))
1514 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1522 struct kvm_irqfd data;
1525 if (copy_from_user(&data, argp, sizeof data))
1527 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1530 case KVM_IOEVENTFD: {
1531 struct kvm_ioeventfd data;
1534 if (copy_from_user(&data, argp, sizeof data))
1536 r = kvm_ioeventfd(kvm, &data);
1539 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1540 case KVM_SET_BOOT_CPU_ID:
1542 mutex_lock(&kvm->lock);
1543 if (atomic_read(&kvm->online_vcpus) != 0)
1546 kvm->bsp_vcpu_id = arg;
1547 mutex_unlock(&kvm->lock);
1551 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1553 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1559 #ifdef CONFIG_COMPAT
1560 struct compat_kvm_dirty_log {
1564 compat_uptr_t dirty_bitmap; /* one bit per page */
1569 static long kvm_vm_compat_ioctl(struct file *filp,
1570 unsigned int ioctl, unsigned long arg)
1572 struct kvm *kvm = filp->private_data;
1575 if (kvm->mm != current->mm)
1578 case KVM_GET_DIRTY_LOG: {
1579 struct compat_kvm_dirty_log compat_log;
1580 struct kvm_dirty_log log;
1583 if (copy_from_user(&compat_log, (void __user *)arg,
1584 sizeof(compat_log)))
1586 log.slot = compat_log.slot;
1587 log.padding1 = compat_log.padding1;
1588 log.padding2 = compat_log.padding2;
1589 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1591 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1597 r = kvm_vm_ioctl(filp, ioctl, arg);
1605 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1607 struct page *page[1];
1610 gfn_t gfn = vmf->pgoff;
1611 struct kvm *kvm = vma->vm_file->private_data;
1613 addr = gfn_to_hva(kvm, gfn);
1614 if (kvm_is_error_hva(addr))
1615 return VM_FAULT_SIGBUS;
1617 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1619 if (unlikely(npages != 1))
1620 return VM_FAULT_SIGBUS;
1622 vmf->page = page[0];
1626 static const struct vm_operations_struct kvm_vm_vm_ops = {
1627 .fault = kvm_vm_fault,
1630 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1632 vma->vm_ops = &kvm_vm_vm_ops;
1636 static struct file_operations kvm_vm_fops = {
1637 .release = kvm_vm_release,
1638 .unlocked_ioctl = kvm_vm_ioctl,
1639 #ifdef CONFIG_COMPAT
1640 .compat_ioctl = kvm_vm_compat_ioctl,
1642 .mmap = kvm_vm_mmap,
1645 static int kvm_dev_ioctl_create_vm(void)
1650 kvm = kvm_create_vm();
1652 return PTR_ERR(kvm);
1653 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1660 static long kvm_dev_ioctl_check_extension_generic(long arg)
1663 case KVM_CAP_USER_MEMORY:
1664 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1665 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1666 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1667 case KVM_CAP_SET_BOOT_CPU_ID:
1669 case KVM_CAP_INTERNAL_ERROR_DATA:
1671 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1672 case KVM_CAP_IRQ_ROUTING:
1673 return KVM_MAX_IRQ_ROUTES;
1678 return kvm_dev_ioctl_check_extension(arg);
1681 static long kvm_dev_ioctl(struct file *filp,
1682 unsigned int ioctl, unsigned long arg)
1687 case KVM_GET_API_VERSION:
1691 r = KVM_API_VERSION;
1697 r = kvm_dev_ioctl_create_vm();
1699 case KVM_CHECK_EXTENSION:
1700 r = kvm_dev_ioctl_check_extension_generic(arg);
1702 case KVM_GET_VCPU_MMAP_SIZE:
1706 r = PAGE_SIZE; /* struct kvm_run */
1708 r += PAGE_SIZE; /* pio data page */
1710 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1711 r += PAGE_SIZE; /* coalesced mmio ring page */
1714 case KVM_TRACE_ENABLE:
1715 case KVM_TRACE_PAUSE:
1716 case KVM_TRACE_DISABLE:
1720 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1726 static struct file_operations kvm_chardev_ops = {
1727 .unlocked_ioctl = kvm_dev_ioctl,
1728 .compat_ioctl = kvm_dev_ioctl,
1731 static struct miscdevice kvm_dev = {
1737 static void hardware_enable(void *junk)
1739 int cpu = raw_smp_processor_id();
1742 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1745 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1747 r = kvm_arch_hardware_enable(NULL);
1750 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1751 atomic_inc(&hardware_enable_failed);
1752 printk(KERN_INFO "kvm: enabling virtualization on "
1753 "CPU%d failed\n", cpu);
1757 static void hardware_disable(void *junk)
1759 int cpu = raw_smp_processor_id();
1761 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1763 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1764 kvm_arch_hardware_disable(NULL);
1767 static void hardware_disable_all_nolock(void)
1769 BUG_ON(!kvm_usage_count);
1772 if (!kvm_usage_count)
1773 on_each_cpu(hardware_disable, NULL, 1);
1776 static void hardware_disable_all(void)
1778 spin_lock(&kvm_lock);
1779 hardware_disable_all_nolock();
1780 spin_unlock(&kvm_lock);
1783 static int hardware_enable_all(void)
1787 spin_lock(&kvm_lock);
1790 if (kvm_usage_count == 1) {
1791 atomic_set(&hardware_enable_failed, 0);
1792 on_each_cpu(hardware_enable, NULL, 1);
1794 if (atomic_read(&hardware_enable_failed)) {
1795 hardware_disable_all_nolock();
1800 spin_unlock(&kvm_lock);
1805 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1810 if (!kvm_usage_count)
1813 val &= ~CPU_TASKS_FROZEN;
1816 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1818 hardware_disable(NULL);
1820 case CPU_UP_CANCELED:
1821 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1823 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1826 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1828 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1835 asmlinkage void kvm_handle_fault_on_reboot(void)
1838 /* spin while reset goes on */
1841 /* Fault while not rebooting. We want the trace. */
1844 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1846 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1850 * Some (well, at least mine) BIOSes hang on reboot if
1853 * And Intel TXT required VMX off for all cpu when system shutdown.
1855 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1856 kvm_rebooting = true;
1857 on_each_cpu(hardware_disable, NULL, 1);
1861 static struct notifier_block kvm_reboot_notifier = {
1862 .notifier_call = kvm_reboot,
1866 void kvm_io_bus_init(struct kvm_io_bus *bus)
1868 memset(bus, 0, sizeof(*bus));
1871 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1875 for (i = 0; i < bus->dev_count; i++) {
1876 struct kvm_io_device *pos = bus->devs[i];
1878 kvm_iodevice_destructor(pos);
1882 /* kvm_io_bus_write - called under kvm->slots_lock */
1883 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1884 int len, const void *val)
1887 for (i = 0; i < bus->dev_count; i++)
1888 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1893 /* kvm_io_bus_read - called under kvm->slots_lock */
1894 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1897 for (i = 0; i < bus->dev_count; i++)
1898 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1903 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1904 struct kvm_io_device *dev)
1908 down_write(&kvm->slots_lock);
1909 ret = __kvm_io_bus_register_dev(bus, dev);
1910 up_write(&kvm->slots_lock);
1915 /* An unlocked version. Caller must have write lock on slots_lock. */
1916 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1917 struct kvm_io_device *dev)
1919 if (bus->dev_count > NR_IOBUS_DEVS-1)
1922 bus->devs[bus->dev_count++] = dev;
1927 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1928 struct kvm_io_bus *bus,
1929 struct kvm_io_device *dev)
1931 down_write(&kvm->slots_lock);
1932 __kvm_io_bus_unregister_dev(bus, dev);
1933 up_write(&kvm->slots_lock);
1936 /* An unlocked version. Caller must have write lock on slots_lock. */
1937 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1938 struct kvm_io_device *dev)
1942 for (i = 0; i < bus->dev_count; i++)
1943 if (bus->devs[i] == dev) {
1944 bus->devs[i] = bus->devs[--bus->dev_count];
1949 static struct notifier_block kvm_cpu_notifier = {
1950 .notifier_call = kvm_cpu_hotplug,
1951 .priority = 20, /* must be > scheduler priority */
1954 static int vm_stat_get(void *_offset, u64 *val)
1956 unsigned offset = (long)_offset;
1960 spin_lock(&kvm_lock);
1961 list_for_each_entry(kvm, &vm_list, vm_list)
1962 *val += *(u32 *)((void *)kvm + offset);
1963 spin_unlock(&kvm_lock);
1967 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1969 static int vcpu_stat_get(void *_offset, u64 *val)
1971 unsigned offset = (long)_offset;
1973 struct kvm_vcpu *vcpu;
1977 spin_lock(&kvm_lock);
1978 list_for_each_entry(kvm, &vm_list, vm_list)
1979 kvm_for_each_vcpu(i, vcpu, kvm)
1980 *val += *(u32 *)((void *)vcpu + offset);
1982 spin_unlock(&kvm_lock);
1986 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1988 static const struct file_operations *stat_fops[] = {
1989 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1990 [KVM_STAT_VM] = &vm_stat_fops,
1993 static void kvm_init_debug(void)
1995 struct kvm_stats_debugfs_item *p;
1997 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1998 for (p = debugfs_entries; p->name; ++p)
1999 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2000 (void *)(long)p->offset,
2001 stat_fops[p->kind]);
2004 static void kvm_exit_debug(void)
2006 struct kvm_stats_debugfs_item *p;
2008 for (p = debugfs_entries; p->name; ++p)
2009 debugfs_remove(p->dentry);
2010 debugfs_remove(kvm_debugfs_dir);
2013 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2015 if (kvm_usage_count)
2016 hardware_disable(NULL);
2020 static int kvm_resume(struct sys_device *dev)
2022 if (kvm_usage_count)
2023 hardware_enable(NULL);
2027 static struct sysdev_class kvm_sysdev_class = {
2029 .suspend = kvm_suspend,
2030 .resume = kvm_resume,
2033 static struct sys_device kvm_sysdev = {
2035 .cls = &kvm_sysdev_class,
2038 struct page *bad_page;
2042 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2044 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2047 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2049 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2051 kvm_arch_vcpu_load(vcpu, cpu);
2054 static void kvm_sched_out(struct preempt_notifier *pn,
2055 struct task_struct *next)
2057 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2059 kvm_arch_vcpu_put(vcpu);
2062 int kvm_init(void *opaque, unsigned int vcpu_size,
2063 struct module *module)
2068 r = kvm_arch_init(opaque);
2072 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2074 if (bad_page == NULL) {
2079 bad_pfn = page_to_pfn(bad_page);
2081 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2086 r = kvm_arch_hardware_setup();
2090 for_each_online_cpu(cpu) {
2091 smp_call_function_single(cpu,
2092 kvm_arch_check_processor_compat,
2098 r = register_cpu_notifier(&kvm_cpu_notifier);
2101 register_reboot_notifier(&kvm_reboot_notifier);
2103 r = sysdev_class_register(&kvm_sysdev_class);
2107 r = sysdev_register(&kvm_sysdev);
2111 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2112 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2113 __alignof__(struct kvm_vcpu),
2115 if (!kvm_vcpu_cache) {
2120 kvm_chardev_ops.owner = module;
2121 kvm_vm_fops.owner = module;
2122 kvm_vcpu_fops.owner = module;
2124 r = misc_register(&kvm_dev);
2126 printk(KERN_ERR "kvm: misc device register failed\n");
2130 kvm_preempt_ops.sched_in = kvm_sched_in;
2131 kvm_preempt_ops.sched_out = kvm_sched_out;
2138 kmem_cache_destroy(kvm_vcpu_cache);
2140 sysdev_unregister(&kvm_sysdev);
2142 sysdev_class_unregister(&kvm_sysdev_class);
2144 unregister_reboot_notifier(&kvm_reboot_notifier);
2145 unregister_cpu_notifier(&kvm_cpu_notifier);
2148 kvm_arch_hardware_unsetup();
2150 free_cpumask_var(cpus_hardware_enabled);
2152 __free_page(bad_page);
2158 EXPORT_SYMBOL_GPL(kvm_init);
2162 tracepoint_synchronize_unregister();
2164 misc_deregister(&kvm_dev);
2165 kmem_cache_destroy(kvm_vcpu_cache);
2166 sysdev_unregister(&kvm_sysdev);
2167 sysdev_class_unregister(&kvm_sysdev_class);
2168 unregister_reboot_notifier(&kvm_reboot_notifier);
2169 unregister_cpu_notifier(&kvm_cpu_notifier);
2170 on_each_cpu(hardware_disable, NULL, 1);
2171 kvm_arch_hardware_unsetup();
2173 free_cpumask_var(cpus_hardware_enabled);
2174 __free_page(bad_page);
2176 EXPORT_SYMBOL_GPL(kvm_exit);