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>
26 #include <linux/miscdevice.h>
27 #include <linux/vmalloc.h>
28 #include <linux/reboot.h>
29 #include <linux/debugfs.h>
30 #include <linux/highmem.h>
31 #include <linux/file.h>
32 #include <linux/sysdev.h>
33 #include <linux/cpu.h>
34 #include <linux/sched.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
50 #include <asm/processor.h>
52 #include <asm/uaccess.h>
53 #include <asm/pgtable.h>
54 #include <asm-generic/bitops/le.h>
56 #include "coalesced_mmio.h"
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/kvm.h>
61 MODULE_AUTHOR("Qumranet");
62 MODULE_LICENSE("GPL");
67 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
70 DEFINE_SPINLOCK(kvm_lock);
73 static cpumask_var_t cpus_hardware_enabled;
74 static int kvm_usage_count = 0;
75 static atomic_t hardware_enable_failed;
77 struct kmem_cache *kvm_vcpu_cache;
78 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
80 static __read_mostly struct preempt_ops kvm_preempt_ops;
82 struct dentry *kvm_debugfs_dir;
84 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
86 static int hardware_enable_all(void);
87 static void hardware_disable_all(void);
89 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
91 static bool kvm_rebooting;
93 static bool largepages_enabled = true;
95 inline int kvm_is_mmio_pfn(pfn_t pfn)
98 struct page *page = compound_head(pfn_to_page(pfn));
99 return PageReserved(page);
106 * Switches to specified vcpu, until a matching vcpu_put()
108 void vcpu_load(struct kvm_vcpu *vcpu)
112 mutex_lock(&vcpu->mutex);
114 preempt_notifier_register(&vcpu->preempt_notifier);
115 kvm_arch_vcpu_load(vcpu, cpu);
119 void vcpu_put(struct kvm_vcpu *vcpu)
122 kvm_arch_vcpu_put(vcpu);
123 preempt_notifier_unregister(&vcpu->preempt_notifier);
125 mutex_unlock(&vcpu->mutex);
128 static void ack_flush(void *_completed)
132 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
137 struct kvm_vcpu *vcpu;
139 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
141 raw_spin_lock(&kvm->requests_lock);
142 me = smp_processor_id();
143 kvm_for_each_vcpu(i, vcpu, kvm) {
144 if (test_and_set_bit(req, &vcpu->requests))
147 if (cpus != NULL && cpu != -1 && cpu != me)
148 cpumask_set_cpu(cpu, cpus);
150 if (unlikely(cpus == NULL))
151 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
152 else if (!cpumask_empty(cpus))
153 smp_call_function_many(cpus, ack_flush, NULL, 1);
156 raw_spin_unlock(&kvm->requests_lock);
157 free_cpumask_var(cpus);
161 void kvm_flush_remote_tlbs(struct kvm *kvm)
163 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
164 ++kvm->stat.remote_tlb_flush;
167 void kvm_reload_remote_mmus(struct kvm *kvm)
169 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
172 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
177 mutex_init(&vcpu->mutex);
181 init_waitqueue_head(&vcpu->wq);
183 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
188 vcpu->run = page_address(page);
190 r = kvm_arch_vcpu_init(vcpu);
196 free_page((unsigned long)vcpu->run);
200 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
202 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
204 kvm_arch_vcpu_uninit(vcpu);
205 free_page((unsigned long)vcpu->run);
207 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
209 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
210 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
212 return container_of(mn, struct kvm, mmu_notifier);
215 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
216 struct mm_struct *mm,
217 unsigned long address)
219 struct kvm *kvm = mmu_notifier_to_kvm(mn);
220 int need_tlb_flush, idx;
223 * When ->invalidate_page runs, the linux pte has been zapped
224 * already but the page is still allocated until
225 * ->invalidate_page returns. So if we increase the sequence
226 * here the kvm page fault will notice if the spte can't be
227 * established because the page is going to be freed. If
228 * instead the kvm page fault establishes the spte before
229 * ->invalidate_page runs, kvm_unmap_hva will release it
232 * The sequence increase only need to be seen at spin_unlock
233 * time, and not at spin_lock time.
235 * Increasing the sequence after the spin_unlock would be
236 * unsafe because the kvm page fault could then establish the
237 * pte after kvm_unmap_hva returned, without noticing the page
238 * is going to be freed.
240 idx = srcu_read_lock(&kvm->srcu);
241 spin_lock(&kvm->mmu_lock);
242 kvm->mmu_notifier_seq++;
243 need_tlb_flush = kvm_unmap_hva(kvm, address);
244 spin_unlock(&kvm->mmu_lock);
245 srcu_read_unlock(&kvm->srcu, idx);
247 /* we've to flush the tlb before the pages can be freed */
249 kvm_flush_remote_tlbs(kvm);
253 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
254 struct mm_struct *mm,
255 unsigned long address,
258 struct kvm *kvm = mmu_notifier_to_kvm(mn);
261 idx = srcu_read_lock(&kvm->srcu);
262 spin_lock(&kvm->mmu_lock);
263 kvm->mmu_notifier_seq++;
264 kvm_set_spte_hva(kvm, address, pte);
265 spin_unlock(&kvm->mmu_lock);
266 srcu_read_unlock(&kvm->srcu, idx);
269 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
270 struct mm_struct *mm,
274 struct kvm *kvm = mmu_notifier_to_kvm(mn);
275 int need_tlb_flush = 0, idx;
277 idx = srcu_read_lock(&kvm->srcu);
278 spin_lock(&kvm->mmu_lock);
280 * The count increase must become visible at unlock time as no
281 * spte can be established without taking the mmu_lock and
282 * count is also read inside the mmu_lock critical section.
284 kvm->mmu_notifier_count++;
285 for (; start < end; start += PAGE_SIZE)
286 need_tlb_flush |= kvm_unmap_hva(kvm, start);
287 spin_unlock(&kvm->mmu_lock);
288 srcu_read_unlock(&kvm->srcu, idx);
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm);
295 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
296 struct mm_struct *mm,
300 struct kvm *kvm = mmu_notifier_to_kvm(mn);
302 spin_lock(&kvm->mmu_lock);
304 * This sequence increase will notify the kvm page fault that
305 * the page that is going to be mapped in the spte could have
308 kvm->mmu_notifier_seq++;
310 * The above sequence increase must be visible before the
311 * below count decrease but both values are read by the kvm
312 * page fault under mmu_lock spinlock so we don't need to add
313 * a smb_wmb() here in between the two.
315 kvm->mmu_notifier_count--;
316 spin_unlock(&kvm->mmu_lock);
318 BUG_ON(kvm->mmu_notifier_count < 0);
321 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
322 struct mm_struct *mm,
323 unsigned long address)
325 struct kvm *kvm = mmu_notifier_to_kvm(mn);
328 idx = srcu_read_lock(&kvm->srcu);
329 spin_lock(&kvm->mmu_lock);
330 young = kvm_age_hva(kvm, address);
331 spin_unlock(&kvm->mmu_lock);
332 srcu_read_unlock(&kvm->srcu, idx);
335 kvm_flush_remote_tlbs(kvm);
340 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
341 struct mm_struct *mm)
343 struct kvm *kvm = mmu_notifier_to_kvm(mn);
346 idx = srcu_read_lock(&kvm->srcu);
347 kvm_arch_flush_shadow(kvm);
348 srcu_read_unlock(&kvm->srcu, idx);
351 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
352 .invalidate_page = kvm_mmu_notifier_invalidate_page,
353 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
354 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
355 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
356 .change_pte = kvm_mmu_notifier_change_pte,
357 .release = kvm_mmu_notifier_release,
360 static int kvm_init_mmu_notifier(struct kvm *kvm)
362 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
363 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
366 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
368 static int kvm_init_mmu_notifier(struct kvm *kvm)
373 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
375 static struct kvm *kvm_create_vm(void)
378 struct kvm *kvm = kvm_arch_create_vm();
383 r = hardware_enable_all();
385 goto out_err_nodisable;
387 #ifdef CONFIG_HAVE_KVM_IRQCHIP
388 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
389 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
393 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
396 if (init_srcu_struct(&kvm->srcu))
398 for (i = 0; i < KVM_NR_BUSES; i++) {
399 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
401 if (!kvm->buses[i]) {
402 cleanup_srcu_struct(&kvm->srcu);
407 r = kvm_init_mmu_notifier(kvm);
409 cleanup_srcu_struct(&kvm->srcu);
413 kvm->mm = current->mm;
414 atomic_inc(&kvm->mm->mm_count);
415 spin_lock_init(&kvm->mmu_lock);
416 raw_spin_lock_init(&kvm->requests_lock);
417 kvm_eventfd_init(kvm);
418 mutex_init(&kvm->lock);
419 mutex_init(&kvm->irq_lock);
420 mutex_init(&kvm->slots_lock);
421 atomic_set(&kvm->users_count, 1);
422 spin_lock(&kvm_lock);
423 list_add(&kvm->vm_list, &vm_list);
424 spin_unlock(&kvm_lock);
425 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
426 kvm_coalesced_mmio_init(kvm);
432 hardware_disable_all();
434 for (i = 0; i < KVM_NR_BUSES; i++)
435 kfree(kvm->buses[i]);
436 kfree(kvm->memslots);
442 * Free any memory in @free but not in @dont.
444 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
445 struct kvm_memory_slot *dont)
449 if (!dont || free->rmap != dont->rmap)
452 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
453 vfree(free->dirty_bitmap);
456 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
457 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
458 vfree(free->lpage_info[i]);
459 free->lpage_info[i] = NULL;
464 free->dirty_bitmap = NULL;
468 void kvm_free_physmem(struct kvm *kvm)
471 struct kvm_memslots *slots = kvm->memslots;
473 for (i = 0; i < slots->nmemslots; ++i)
474 kvm_free_physmem_slot(&slots->memslots[i], NULL);
476 kfree(kvm->memslots);
479 static void kvm_destroy_vm(struct kvm *kvm)
482 struct mm_struct *mm = kvm->mm;
484 kvm_arch_sync_events(kvm);
485 spin_lock(&kvm_lock);
486 list_del(&kvm->vm_list);
487 spin_unlock(&kvm_lock);
488 kvm_free_irq_routing(kvm);
489 for (i = 0; i < KVM_NR_BUSES; i++)
490 kvm_io_bus_destroy(kvm->buses[i]);
491 kvm_coalesced_mmio_free(kvm);
492 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
493 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
495 kvm_arch_flush_shadow(kvm);
497 kvm_arch_destroy_vm(kvm);
498 hardware_disable_all();
502 void kvm_get_kvm(struct kvm *kvm)
504 atomic_inc(&kvm->users_count);
506 EXPORT_SYMBOL_GPL(kvm_get_kvm);
508 void kvm_put_kvm(struct kvm *kvm)
510 if (atomic_dec_and_test(&kvm->users_count))
513 EXPORT_SYMBOL_GPL(kvm_put_kvm);
516 static int kvm_vm_release(struct inode *inode, struct file *filp)
518 struct kvm *kvm = filp->private_data;
520 kvm_irqfd_release(kvm);
527 * Allocate some memory and give it an address in the guest physical address
530 * Discontiguous memory is allowed, mostly for framebuffers.
532 * Must be called holding mmap_sem for write.
534 int __kvm_set_memory_region(struct kvm *kvm,
535 struct kvm_userspace_memory_region *mem,
538 int r, flush_shadow = 0;
540 unsigned long npages;
542 struct kvm_memory_slot *memslot;
543 struct kvm_memory_slot old, new;
544 struct kvm_memslots *slots, *old_memslots;
547 /* General sanity checks */
548 if (mem->memory_size & (PAGE_SIZE - 1))
550 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
552 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
554 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
556 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
559 memslot = &kvm->memslots->memslots[mem->slot];
560 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
561 npages = mem->memory_size >> PAGE_SHIFT;
564 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
566 new = old = *memslot;
568 new.base_gfn = base_gfn;
570 new.flags = mem->flags;
572 /* Disallow changing a memory slot's size. */
574 if (npages && old.npages && npages != old.npages)
577 /* Check for overlaps */
579 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
580 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
582 if (s == memslot || !s->npages)
584 if (!((base_gfn + npages <= s->base_gfn) ||
585 (base_gfn >= s->base_gfn + s->npages)))
589 /* Free page dirty bitmap if unneeded */
590 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
591 new.dirty_bitmap = NULL;
595 /* Allocate if a slot is being created */
597 if (npages && !new.rmap) {
598 new.rmap = vmalloc(npages * sizeof(struct page *));
603 memset(new.rmap, 0, npages * sizeof(*new.rmap));
605 new.user_alloc = user_alloc;
606 new.userspace_addr = mem->userspace_addr;
611 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
617 /* Avoid unused variable warning if no large pages */
620 if (new.lpage_info[i])
623 lpages = 1 + (base_gfn + npages - 1) /
624 KVM_PAGES_PER_HPAGE(level);
625 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
627 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
629 if (!new.lpage_info[i])
632 memset(new.lpage_info[i], 0,
633 lpages * sizeof(*new.lpage_info[i]));
635 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
636 new.lpage_info[i][0].write_count = 1;
637 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
638 new.lpage_info[i][lpages - 1].write_count = 1;
639 ugfn = new.userspace_addr >> PAGE_SHIFT;
641 * If the gfn and userspace address are not aligned wrt each
642 * other, or if explicitly asked to, disable large page
643 * support for this slot
645 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
647 for (j = 0; j < lpages; ++j)
648 new.lpage_info[i][j].write_count = 1;
653 /* Allocate page dirty bitmap if needed */
654 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
655 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
657 new.dirty_bitmap = vmalloc(dirty_bytes);
658 if (!new.dirty_bitmap)
660 memset(new.dirty_bitmap, 0, dirty_bytes);
661 /* destroy any largepage mappings for dirty tracking */
665 #else /* not defined CONFIG_S390 */
666 new.user_alloc = user_alloc;
668 new.userspace_addr = mem->userspace_addr;
669 #endif /* not defined CONFIG_S390 */
673 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
676 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
677 if (mem->slot >= slots->nmemslots)
678 slots->nmemslots = mem->slot + 1;
679 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
681 old_memslots = kvm->memslots;
682 rcu_assign_pointer(kvm->memslots, slots);
683 synchronize_srcu_expedited(&kvm->srcu);
684 /* From this point no new shadow pages pointing to a deleted
685 * memslot will be created.
687 * validation of sp->gfn happens in:
688 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
689 * - kvm_is_visible_gfn (mmu_check_roots)
691 kvm_arch_flush_shadow(kvm);
695 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
700 /* map the pages in iommu page table */
702 r = kvm_iommu_map_pages(kvm, &new);
709 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
712 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
713 if (mem->slot >= slots->nmemslots)
714 slots->nmemslots = mem->slot + 1;
716 /* actual memory is freed via old in kvm_free_physmem_slot below */
719 new.dirty_bitmap = NULL;
720 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
721 new.lpage_info[i] = NULL;
724 slots->memslots[mem->slot] = new;
725 old_memslots = kvm->memslots;
726 rcu_assign_pointer(kvm->memslots, slots);
727 synchronize_srcu_expedited(&kvm->srcu);
729 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
731 kvm_free_physmem_slot(&old, &new);
735 kvm_arch_flush_shadow(kvm);
740 kvm_free_physmem_slot(&new, &old);
745 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
747 int kvm_set_memory_region(struct kvm *kvm,
748 struct kvm_userspace_memory_region *mem,
753 mutex_lock(&kvm->slots_lock);
754 r = __kvm_set_memory_region(kvm, mem, user_alloc);
755 mutex_unlock(&kvm->slots_lock);
758 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
760 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
762 kvm_userspace_memory_region *mem,
765 if (mem->slot >= KVM_MEMORY_SLOTS)
767 return kvm_set_memory_region(kvm, mem, user_alloc);
770 int kvm_get_dirty_log(struct kvm *kvm,
771 struct kvm_dirty_log *log, int *is_dirty)
773 struct kvm_memory_slot *memslot;
776 unsigned long any = 0;
779 if (log->slot >= KVM_MEMORY_SLOTS)
782 memslot = &kvm->memslots->memslots[log->slot];
784 if (!memslot->dirty_bitmap)
787 n = kvm_dirty_bitmap_bytes(memslot);
789 for (i = 0; !any && i < n/sizeof(long); ++i)
790 any = memslot->dirty_bitmap[i];
793 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
804 void kvm_disable_largepages(void)
806 largepages_enabled = false;
808 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
810 int is_error_page(struct page *page)
812 return page == bad_page;
814 EXPORT_SYMBOL_GPL(is_error_page);
816 int is_error_pfn(pfn_t pfn)
818 return pfn == bad_pfn;
820 EXPORT_SYMBOL_GPL(is_error_pfn);
822 static inline unsigned long bad_hva(void)
827 int kvm_is_error_hva(unsigned long addr)
829 return addr == bad_hva();
831 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
833 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
836 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
838 for (i = 0; i < slots->nmemslots; ++i) {
839 struct kvm_memory_slot *memslot = &slots->memslots[i];
841 if (gfn >= memslot->base_gfn
842 && gfn < memslot->base_gfn + memslot->npages)
847 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
849 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
851 gfn = unalias_gfn(kvm, gfn);
852 return gfn_to_memslot_unaliased(kvm, gfn);
855 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
858 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
860 gfn = unalias_gfn_instantiation(kvm, gfn);
861 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
862 struct kvm_memory_slot *memslot = &slots->memslots[i];
864 if (memslot->flags & KVM_MEMSLOT_INVALID)
867 if (gfn >= memslot->base_gfn
868 && gfn < memslot->base_gfn + memslot->npages)
873 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
875 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
877 struct vm_area_struct *vma;
878 unsigned long addr, size;
882 addr = gfn_to_hva(kvm, gfn);
883 if (kvm_is_error_hva(addr))
886 down_read(¤t->mm->mmap_sem);
887 vma = find_vma(current->mm, addr);
891 size = vma_kernel_pagesize(vma);
894 up_read(¤t->mm->mmap_sem);
899 int memslot_id(struct kvm *kvm, gfn_t gfn)
902 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
903 struct kvm_memory_slot *memslot = NULL;
905 gfn = unalias_gfn(kvm, gfn);
906 for (i = 0; i < slots->nmemslots; ++i) {
907 memslot = &slots->memslots[i];
909 if (gfn >= memslot->base_gfn
910 && gfn < memslot->base_gfn + memslot->npages)
914 return memslot - slots->memslots;
917 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
919 struct kvm_memory_slot *slot;
921 gfn = unalias_gfn_instantiation(kvm, gfn);
922 slot = gfn_to_memslot_unaliased(kvm, gfn);
923 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
925 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
927 EXPORT_SYMBOL_GPL(gfn_to_hva);
929 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
931 struct page *page[1];
937 npages = get_user_pages_fast(addr, 1, 1, page);
939 if (unlikely(npages != 1)) {
940 struct vm_area_struct *vma;
942 down_read(¤t->mm->mmap_sem);
943 vma = find_vma(current->mm, addr);
945 if (vma == NULL || addr < vma->vm_start ||
946 !(vma->vm_flags & VM_PFNMAP)) {
947 up_read(¤t->mm->mmap_sem);
949 return page_to_pfn(bad_page);
952 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
953 up_read(¤t->mm->mmap_sem);
954 BUG_ON(!kvm_is_mmio_pfn(pfn));
956 pfn = page_to_pfn(page[0]);
961 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
965 addr = gfn_to_hva(kvm, gfn);
966 if (kvm_is_error_hva(addr)) {
968 return page_to_pfn(bad_page);
971 return hva_to_pfn(kvm, addr);
973 EXPORT_SYMBOL_GPL(gfn_to_pfn);
975 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
977 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
980 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
981 struct kvm_memory_slot *slot, gfn_t gfn)
983 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
984 return hva_to_pfn(kvm, addr);
987 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
991 pfn = gfn_to_pfn(kvm, gfn);
992 if (!kvm_is_mmio_pfn(pfn))
993 return pfn_to_page(pfn);
995 WARN_ON(kvm_is_mmio_pfn(pfn));
1001 EXPORT_SYMBOL_GPL(gfn_to_page);
1003 void kvm_release_page_clean(struct page *page)
1005 kvm_release_pfn_clean(page_to_pfn(page));
1007 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1009 void kvm_release_pfn_clean(pfn_t pfn)
1011 if (!kvm_is_mmio_pfn(pfn))
1012 put_page(pfn_to_page(pfn));
1014 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1016 void kvm_release_page_dirty(struct page *page)
1018 kvm_release_pfn_dirty(page_to_pfn(page));
1020 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1022 void kvm_release_pfn_dirty(pfn_t pfn)
1024 kvm_set_pfn_dirty(pfn);
1025 kvm_release_pfn_clean(pfn);
1027 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1029 void kvm_set_page_dirty(struct page *page)
1031 kvm_set_pfn_dirty(page_to_pfn(page));
1033 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1035 void kvm_set_pfn_dirty(pfn_t pfn)
1037 if (!kvm_is_mmio_pfn(pfn)) {
1038 struct page *page = pfn_to_page(pfn);
1039 if (!PageReserved(page))
1043 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1045 void kvm_set_pfn_accessed(pfn_t pfn)
1047 if (!kvm_is_mmio_pfn(pfn))
1048 mark_page_accessed(pfn_to_page(pfn));
1050 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1052 void kvm_get_pfn(pfn_t pfn)
1054 if (!kvm_is_mmio_pfn(pfn))
1055 get_page(pfn_to_page(pfn));
1057 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1059 static int next_segment(unsigned long len, int offset)
1061 if (len > PAGE_SIZE - offset)
1062 return PAGE_SIZE - offset;
1067 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1073 addr = gfn_to_hva(kvm, gfn);
1074 if (kvm_is_error_hva(addr))
1076 r = copy_from_user(data, (void __user *)addr + offset, len);
1081 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1083 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1085 gfn_t gfn = gpa >> PAGE_SHIFT;
1087 int offset = offset_in_page(gpa);
1090 while ((seg = next_segment(len, offset)) != 0) {
1091 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1101 EXPORT_SYMBOL_GPL(kvm_read_guest);
1103 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1108 gfn_t gfn = gpa >> PAGE_SHIFT;
1109 int offset = offset_in_page(gpa);
1111 addr = gfn_to_hva(kvm, gfn);
1112 if (kvm_is_error_hva(addr))
1114 pagefault_disable();
1115 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1121 EXPORT_SYMBOL(kvm_read_guest_atomic);
1123 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1124 int offset, int len)
1129 addr = gfn_to_hva(kvm, gfn);
1130 if (kvm_is_error_hva(addr))
1132 r = copy_to_user((void __user *)addr + offset, data, len);
1135 mark_page_dirty(kvm, gfn);
1138 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1140 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1143 gfn_t gfn = gpa >> PAGE_SHIFT;
1145 int offset = offset_in_page(gpa);
1148 while ((seg = next_segment(len, offset)) != 0) {
1149 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1160 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1162 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1164 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1166 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1168 gfn_t gfn = gpa >> PAGE_SHIFT;
1170 int offset = offset_in_page(gpa);
1173 while ((seg = next_segment(len, offset)) != 0) {
1174 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1183 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1185 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1187 struct kvm_memory_slot *memslot;
1189 gfn = unalias_gfn(kvm, gfn);
1190 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1191 if (memslot && memslot->dirty_bitmap) {
1192 unsigned long rel_gfn = gfn - memslot->base_gfn;
1193 unsigned long *p = memslot->dirty_bitmap +
1194 rel_gfn / BITS_PER_LONG;
1195 int offset = rel_gfn % BITS_PER_LONG;
1198 if (!generic_test_le_bit(offset, p))
1199 generic___set_le_bit(offset, p);
1204 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1206 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1211 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1213 if (kvm_arch_vcpu_runnable(vcpu)) {
1214 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1217 if (kvm_cpu_has_pending_timer(vcpu))
1219 if (signal_pending(current))
1225 finish_wait(&vcpu->wq, &wait);
1228 void kvm_resched(struct kvm_vcpu *vcpu)
1230 if (!need_resched())
1234 EXPORT_SYMBOL_GPL(kvm_resched);
1236 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1241 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1243 /* Sleep for 100 us, and hope lock-holder got scheduled */
1244 expires = ktime_add_ns(ktime_get(), 100000UL);
1245 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1247 finish_wait(&vcpu->wq, &wait);
1249 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1251 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1253 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1256 if (vmf->pgoff == 0)
1257 page = virt_to_page(vcpu->run);
1259 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1260 page = virt_to_page(vcpu->arch.pio_data);
1262 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1263 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1264 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1267 return VM_FAULT_SIGBUS;
1273 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1274 .fault = kvm_vcpu_fault,
1277 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1279 vma->vm_ops = &kvm_vcpu_vm_ops;
1283 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1285 struct kvm_vcpu *vcpu = filp->private_data;
1287 kvm_put_kvm(vcpu->kvm);
1291 static struct file_operations kvm_vcpu_fops = {
1292 .release = kvm_vcpu_release,
1293 .unlocked_ioctl = kvm_vcpu_ioctl,
1294 .compat_ioctl = kvm_vcpu_ioctl,
1295 .mmap = kvm_vcpu_mmap,
1299 * Allocates an inode for the vcpu.
1301 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1303 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1307 * Creates some virtual cpus. Good luck creating more than one.
1309 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1312 struct kvm_vcpu *vcpu, *v;
1314 vcpu = kvm_arch_vcpu_create(kvm, id);
1316 return PTR_ERR(vcpu);
1318 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1320 r = kvm_arch_vcpu_setup(vcpu);
1324 mutex_lock(&kvm->lock);
1325 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1330 kvm_for_each_vcpu(r, v, kvm)
1331 if (v->vcpu_id == id) {
1336 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1338 /* Now it's all set up, let userspace reach it */
1340 r = create_vcpu_fd(vcpu);
1346 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1348 atomic_inc(&kvm->online_vcpus);
1350 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1351 if (kvm->bsp_vcpu_id == id)
1352 kvm->bsp_vcpu = vcpu;
1354 mutex_unlock(&kvm->lock);
1358 mutex_unlock(&kvm->lock);
1359 kvm_arch_vcpu_destroy(vcpu);
1363 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1366 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1367 vcpu->sigset_active = 1;
1368 vcpu->sigset = *sigset;
1370 vcpu->sigset_active = 0;
1374 static long kvm_vcpu_ioctl(struct file *filp,
1375 unsigned int ioctl, unsigned long arg)
1377 struct kvm_vcpu *vcpu = filp->private_data;
1378 void __user *argp = (void __user *)arg;
1380 struct kvm_fpu *fpu = NULL;
1381 struct kvm_sregs *kvm_sregs = NULL;
1383 if (vcpu->kvm->mm != current->mm)
1390 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1392 case KVM_GET_REGS: {
1393 struct kvm_regs *kvm_regs;
1396 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1399 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1403 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1410 case KVM_SET_REGS: {
1411 struct kvm_regs *kvm_regs;
1414 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1418 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1420 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1428 case KVM_GET_SREGS: {
1429 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1433 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1437 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1442 case KVM_SET_SREGS: {
1443 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1448 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1450 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1456 case KVM_GET_MP_STATE: {
1457 struct kvm_mp_state mp_state;
1459 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1463 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1468 case KVM_SET_MP_STATE: {
1469 struct kvm_mp_state mp_state;
1472 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1474 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1480 case KVM_TRANSLATE: {
1481 struct kvm_translation tr;
1484 if (copy_from_user(&tr, argp, sizeof tr))
1486 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1490 if (copy_to_user(argp, &tr, sizeof tr))
1495 case KVM_SET_GUEST_DEBUG: {
1496 struct kvm_guest_debug dbg;
1499 if (copy_from_user(&dbg, argp, sizeof dbg))
1501 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1507 case KVM_SET_SIGNAL_MASK: {
1508 struct kvm_signal_mask __user *sigmask_arg = argp;
1509 struct kvm_signal_mask kvm_sigmask;
1510 sigset_t sigset, *p;
1515 if (copy_from_user(&kvm_sigmask, argp,
1516 sizeof kvm_sigmask))
1519 if (kvm_sigmask.len != sizeof sigset)
1522 if (copy_from_user(&sigset, sigmask_arg->sigset,
1527 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1531 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1535 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1539 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1545 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1550 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1552 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1559 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1567 static long kvm_vm_ioctl(struct file *filp,
1568 unsigned int ioctl, unsigned long arg)
1570 struct kvm *kvm = filp->private_data;
1571 void __user *argp = (void __user *)arg;
1574 if (kvm->mm != current->mm)
1577 case KVM_CREATE_VCPU:
1578 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1582 case KVM_SET_USER_MEMORY_REGION: {
1583 struct kvm_userspace_memory_region kvm_userspace_mem;
1586 if (copy_from_user(&kvm_userspace_mem, argp,
1587 sizeof kvm_userspace_mem))
1590 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1595 case KVM_GET_DIRTY_LOG: {
1596 struct kvm_dirty_log log;
1599 if (copy_from_user(&log, argp, sizeof log))
1601 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1606 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1607 case KVM_REGISTER_COALESCED_MMIO: {
1608 struct kvm_coalesced_mmio_zone zone;
1610 if (copy_from_user(&zone, argp, sizeof zone))
1613 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1619 case KVM_UNREGISTER_COALESCED_MMIO: {
1620 struct kvm_coalesced_mmio_zone zone;
1622 if (copy_from_user(&zone, argp, sizeof zone))
1625 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1633 struct kvm_irqfd data;
1636 if (copy_from_user(&data, argp, sizeof data))
1638 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1641 case KVM_IOEVENTFD: {
1642 struct kvm_ioeventfd data;
1645 if (copy_from_user(&data, argp, sizeof data))
1647 r = kvm_ioeventfd(kvm, &data);
1650 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1651 case KVM_SET_BOOT_CPU_ID:
1653 mutex_lock(&kvm->lock);
1654 if (atomic_read(&kvm->online_vcpus) != 0)
1657 kvm->bsp_vcpu_id = arg;
1658 mutex_unlock(&kvm->lock);
1662 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1664 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1670 #ifdef CONFIG_COMPAT
1671 struct compat_kvm_dirty_log {
1675 compat_uptr_t dirty_bitmap; /* one bit per page */
1680 static long kvm_vm_compat_ioctl(struct file *filp,
1681 unsigned int ioctl, unsigned long arg)
1683 struct kvm *kvm = filp->private_data;
1686 if (kvm->mm != current->mm)
1689 case KVM_GET_DIRTY_LOG: {
1690 struct compat_kvm_dirty_log compat_log;
1691 struct kvm_dirty_log log;
1694 if (copy_from_user(&compat_log, (void __user *)arg,
1695 sizeof(compat_log)))
1697 log.slot = compat_log.slot;
1698 log.padding1 = compat_log.padding1;
1699 log.padding2 = compat_log.padding2;
1700 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1702 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1708 r = kvm_vm_ioctl(filp, ioctl, arg);
1716 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1718 struct page *page[1];
1721 gfn_t gfn = vmf->pgoff;
1722 struct kvm *kvm = vma->vm_file->private_data;
1724 addr = gfn_to_hva(kvm, gfn);
1725 if (kvm_is_error_hva(addr))
1726 return VM_FAULT_SIGBUS;
1728 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1730 if (unlikely(npages != 1))
1731 return VM_FAULT_SIGBUS;
1733 vmf->page = page[0];
1737 static const struct vm_operations_struct kvm_vm_vm_ops = {
1738 .fault = kvm_vm_fault,
1741 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1743 vma->vm_ops = &kvm_vm_vm_ops;
1747 static struct file_operations kvm_vm_fops = {
1748 .release = kvm_vm_release,
1749 .unlocked_ioctl = kvm_vm_ioctl,
1750 #ifdef CONFIG_COMPAT
1751 .compat_ioctl = kvm_vm_compat_ioctl,
1753 .mmap = kvm_vm_mmap,
1756 static int kvm_dev_ioctl_create_vm(void)
1761 kvm = kvm_create_vm();
1763 return PTR_ERR(kvm);
1764 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1771 static long kvm_dev_ioctl_check_extension_generic(long arg)
1774 case KVM_CAP_USER_MEMORY:
1775 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1776 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1777 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1778 case KVM_CAP_SET_BOOT_CPU_ID:
1780 case KVM_CAP_INTERNAL_ERROR_DATA:
1782 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1783 case KVM_CAP_IRQ_ROUTING:
1784 return KVM_MAX_IRQ_ROUTES;
1789 return kvm_dev_ioctl_check_extension(arg);
1792 static long kvm_dev_ioctl(struct file *filp,
1793 unsigned int ioctl, unsigned long arg)
1798 case KVM_GET_API_VERSION:
1802 r = KVM_API_VERSION;
1808 r = kvm_dev_ioctl_create_vm();
1810 case KVM_CHECK_EXTENSION:
1811 r = kvm_dev_ioctl_check_extension_generic(arg);
1813 case KVM_GET_VCPU_MMAP_SIZE:
1817 r = PAGE_SIZE; /* struct kvm_run */
1819 r += PAGE_SIZE; /* pio data page */
1821 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1822 r += PAGE_SIZE; /* coalesced mmio ring page */
1825 case KVM_TRACE_ENABLE:
1826 case KVM_TRACE_PAUSE:
1827 case KVM_TRACE_DISABLE:
1831 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1837 static struct file_operations kvm_chardev_ops = {
1838 .unlocked_ioctl = kvm_dev_ioctl,
1839 .compat_ioctl = kvm_dev_ioctl,
1842 static struct miscdevice kvm_dev = {
1848 static void hardware_enable(void *junk)
1850 int cpu = raw_smp_processor_id();
1853 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1856 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1858 r = kvm_arch_hardware_enable(NULL);
1861 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1862 atomic_inc(&hardware_enable_failed);
1863 printk(KERN_INFO "kvm: enabling virtualization on "
1864 "CPU%d failed\n", cpu);
1868 static void hardware_disable(void *junk)
1870 int cpu = raw_smp_processor_id();
1872 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1874 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1875 kvm_arch_hardware_disable(NULL);
1878 static void hardware_disable_all_nolock(void)
1880 BUG_ON(!kvm_usage_count);
1883 if (!kvm_usage_count)
1884 on_each_cpu(hardware_disable, NULL, 1);
1887 static void hardware_disable_all(void)
1889 spin_lock(&kvm_lock);
1890 hardware_disable_all_nolock();
1891 spin_unlock(&kvm_lock);
1894 static int hardware_enable_all(void)
1898 spin_lock(&kvm_lock);
1901 if (kvm_usage_count == 1) {
1902 atomic_set(&hardware_enable_failed, 0);
1903 on_each_cpu(hardware_enable, NULL, 1);
1905 if (atomic_read(&hardware_enable_failed)) {
1906 hardware_disable_all_nolock();
1911 spin_unlock(&kvm_lock);
1916 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1921 if (!kvm_usage_count)
1924 val &= ~CPU_TASKS_FROZEN;
1927 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1929 hardware_disable(NULL);
1931 case CPU_UP_CANCELED:
1932 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1934 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1937 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1939 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1946 asmlinkage void kvm_handle_fault_on_reboot(void)
1949 /* spin while reset goes on */
1952 /* Fault while not rebooting. We want the trace. */
1955 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1957 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1961 * Some (well, at least mine) BIOSes hang on reboot if
1964 * And Intel TXT required VMX off for all cpu when system shutdown.
1966 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1967 kvm_rebooting = true;
1968 on_each_cpu(hardware_disable, NULL, 1);
1972 static struct notifier_block kvm_reboot_notifier = {
1973 .notifier_call = kvm_reboot,
1977 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1981 for (i = 0; i < bus->dev_count; i++) {
1982 struct kvm_io_device *pos = bus->devs[i];
1984 kvm_iodevice_destructor(pos);
1989 /* kvm_io_bus_write - called under kvm->slots_lock */
1990 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1991 int len, const void *val)
1994 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1995 for (i = 0; i < bus->dev_count; i++)
1996 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2001 /* kvm_io_bus_read - called under kvm->slots_lock */
2002 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2006 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2008 for (i = 0; i < bus->dev_count; i++)
2009 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2014 /* Caller must hold slots_lock. */
2015 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2016 struct kvm_io_device *dev)
2018 struct kvm_io_bus *new_bus, *bus;
2020 bus = kvm->buses[bus_idx];
2021 if (bus->dev_count > NR_IOBUS_DEVS-1)
2024 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2027 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2028 new_bus->devs[new_bus->dev_count++] = dev;
2029 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2030 synchronize_srcu_expedited(&kvm->srcu);
2036 /* Caller must hold slots_lock. */
2037 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2038 struct kvm_io_device *dev)
2041 struct kvm_io_bus *new_bus, *bus;
2043 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2047 bus = kvm->buses[bus_idx];
2048 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2051 for (i = 0; i < new_bus->dev_count; i++)
2052 if (new_bus->devs[i] == dev) {
2054 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2063 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2064 synchronize_srcu_expedited(&kvm->srcu);
2069 static struct notifier_block kvm_cpu_notifier = {
2070 .notifier_call = kvm_cpu_hotplug,
2071 .priority = 20, /* must be > scheduler priority */
2074 static int vm_stat_get(void *_offset, u64 *val)
2076 unsigned offset = (long)_offset;
2080 spin_lock(&kvm_lock);
2081 list_for_each_entry(kvm, &vm_list, vm_list)
2082 *val += *(u32 *)((void *)kvm + offset);
2083 spin_unlock(&kvm_lock);
2087 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2089 static int vcpu_stat_get(void *_offset, u64 *val)
2091 unsigned offset = (long)_offset;
2093 struct kvm_vcpu *vcpu;
2097 spin_lock(&kvm_lock);
2098 list_for_each_entry(kvm, &vm_list, vm_list)
2099 kvm_for_each_vcpu(i, vcpu, kvm)
2100 *val += *(u32 *)((void *)vcpu + offset);
2102 spin_unlock(&kvm_lock);
2106 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2108 static const struct file_operations *stat_fops[] = {
2109 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2110 [KVM_STAT_VM] = &vm_stat_fops,
2113 static void kvm_init_debug(void)
2115 struct kvm_stats_debugfs_item *p;
2117 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2118 for (p = debugfs_entries; p->name; ++p)
2119 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2120 (void *)(long)p->offset,
2121 stat_fops[p->kind]);
2124 static void kvm_exit_debug(void)
2126 struct kvm_stats_debugfs_item *p;
2128 for (p = debugfs_entries; p->name; ++p)
2129 debugfs_remove(p->dentry);
2130 debugfs_remove(kvm_debugfs_dir);
2133 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2135 if (kvm_usage_count)
2136 hardware_disable(NULL);
2140 static int kvm_resume(struct sys_device *dev)
2142 if (kvm_usage_count)
2143 hardware_enable(NULL);
2147 static struct sysdev_class kvm_sysdev_class = {
2149 .suspend = kvm_suspend,
2150 .resume = kvm_resume,
2153 static struct sys_device kvm_sysdev = {
2155 .cls = &kvm_sysdev_class,
2158 struct page *bad_page;
2162 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2164 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2167 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2169 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2171 kvm_arch_vcpu_load(vcpu, cpu);
2174 static void kvm_sched_out(struct preempt_notifier *pn,
2175 struct task_struct *next)
2177 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2179 kvm_arch_vcpu_put(vcpu);
2182 int kvm_init(void *opaque, unsigned int vcpu_size,
2183 struct module *module)
2188 r = kvm_arch_init(opaque);
2192 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2194 if (bad_page == NULL) {
2199 bad_pfn = page_to_pfn(bad_page);
2201 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2206 r = kvm_arch_hardware_setup();
2210 for_each_online_cpu(cpu) {
2211 smp_call_function_single(cpu,
2212 kvm_arch_check_processor_compat,
2218 r = register_cpu_notifier(&kvm_cpu_notifier);
2221 register_reboot_notifier(&kvm_reboot_notifier);
2223 r = sysdev_class_register(&kvm_sysdev_class);
2227 r = sysdev_register(&kvm_sysdev);
2231 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2232 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2233 __alignof__(struct kvm_vcpu),
2235 if (!kvm_vcpu_cache) {
2240 kvm_chardev_ops.owner = module;
2241 kvm_vm_fops.owner = module;
2242 kvm_vcpu_fops.owner = module;
2244 r = misc_register(&kvm_dev);
2246 printk(KERN_ERR "kvm: misc device register failed\n");
2250 kvm_preempt_ops.sched_in = kvm_sched_in;
2251 kvm_preempt_ops.sched_out = kvm_sched_out;
2258 kmem_cache_destroy(kvm_vcpu_cache);
2260 sysdev_unregister(&kvm_sysdev);
2262 sysdev_class_unregister(&kvm_sysdev_class);
2264 unregister_reboot_notifier(&kvm_reboot_notifier);
2265 unregister_cpu_notifier(&kvm_cpu_notifier);
2268 kvm_arch_hardware_unsetup();
2270 free_cpumask_var(cpus_hardware_enabled);
2272 __free_page(bad_page);
2278 EXPORT_SYMBOL_GPL(kvm_init);
2282 tracepoint_synchronize_unregister();
2284 misc_deregister(&kvm_dev);
2285 kmem_cache_destroy(kvm_vcpu_cache);
2286 sysdev_unregister(&kvm_sysdev);
2287 sysdev_class_unregister(&kvm_sysdev_class);
2288 unregister_reboot_notifier(&kvm_reboot_notifier);
2289 unregister_cpu_notifier(&kvm_cpu_notifier);
2290 on_each_cpu(hardware_disable, NULL, 1);
2291 kvm_arch_hardware_unsetup();
2293 free_cpumask_var(cpus_hardware_enabled);
2294 __free_page(bad_page);
2296 EXPORT_SYMBOL_GPL(kvm_exit);