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.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.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/syscore_ops.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>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
69 static unsigned int halt_poll_ns;
70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
78 DEFINE_SPINLOCK(kvm_lock);
79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
82 static cpumask_var_t cpus_hardware_enabled;
83 static int kvm_usage_count;
84 static atomic_t hardware_enable_failed;
86 struct kmem_cache *kvm_vcpu_cache;
87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
89 static __read_mostly struct preempt_ops kvm_preempt_ops;
91 struct dentry *kvm_debugfs_dir;
93 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
95 #ifdef CONFIG_KVM_COMPAT
96 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
99 static int hardware_enable_all(void);
100 static void hardware_disable_all(void);
102 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
104 static void kvm_release_pfn_dirty(pfn_t pfn);
105 static void mark_page_dirty_in_slot(struct kvm *kvm,
106 struct kvm_memory_slot *memslot, gfn_t gfn);
108 __visible bool kvm_rebooting;
109 EXPORT_SYMBOL_GPL(kvm_rebooting);
111 static bool largepages_enabled = true;
113 bool kvm_is_reserved_pfn(pfn_t pfn)
116 return PageReserved(pfn_to_page(pfn));
122 * Switches to specified vcpu, until a matching vcpu_put()
124 int vcpu_load(struct kvm_vcpu *vcpu)
128 if (mutex_lock_killable(&vcpu->mutex))
131 preempt_notifier_register(&vcpu->preempt_notifier);
132 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 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
155 struct kvm_vcpu *vcpu;
157 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
160 kvm_for_each_vcpu(i, vcpu, kvm) {
161 kvm_make_request(req, vcpu);
164 /* Set ->requests bit before we read ->mode */
167 if (cpus != NULL && cpu != -1 && cpu != me &&
168 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
169 cpumask_set_cpu(cpu, cpus);
171 if (unlikely(cpus == NULL))
172 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
173 else if (!cpumask_empty(cpus))
174 smp_call_function_many(cpus, ack_flush, NULL, 1);
178 free_cpumask_var(cpus);
182 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
183 void kvm_flush_remote_tlbs(struct kvm *kvm)
185 long dirty_count = kvm->tlbs_dirty;
188 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
189 ++kvm->stat.remote_tlb_flush;
190 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
192 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
195 void kvm_reload_remote_mmus(struct kvm *kvm)
197 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
202 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
207 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215 mutex_init(&vcpu->mutex);
220 init_waitqueue_head(&vcpu->wq);
221 kvm_async_pf_vcpu_init(vcpu);
223 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 vcpu->run = page_address(page);
230 kvm_vcpu_set_in_spin_loop(vcpu, false);
231 kvm_vcpu_set_dy_eligible(vcpu, false);
232 vcpu->preempted = false;
234 r = kvm_arch_vcpu_init(vcpu);
240 free_page((unsigned long)vcpu->run);
244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
249 kvm_arch_vcpu_uninit(vcpu);
250 free_page((unsigned long)vcpu->run);
252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
257 return container_of(mn, struct kvm, mmu_notifier);
260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address)
264 struct kvm *kvm = mmu_notifier_to_kvm(mn);
265 int need_tlb_flush, idx;
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
285 idx = srcu_read_lock(&kvm->srcu);
286 spin_lock(&kvm->mmu_lock);
288 kvm->mmu_notifier_seq++;
289 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm);
294 spin_unlock(&kvm->mmu_lock);
296 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
298 srcu_read_unlock(&kvm->srcu, idx);
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
314 srcu_read_unlock(&kvm->srcu, idx);
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int need_tlb_flush = 0, idx;
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
332 kvm->mmu_notifier_count++;
333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334 need_tlb_flush |= kvm->tlbs_dirty;
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 spin_lock(&kvm->mmu_lock);
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
356 kvm->mmu_notifier_seq++;
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
366 BUG_ON(kvm->mmu_notifier_count < 0);
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 idx = srcu_read_lock(&kvm->srcu);
378 spin_lock(&kvm->mmu_lock);
380 young = kvm_age_hva(kvm, start, end);
382 kvm_flush_remote_tlbs(kvm);
384 spin_unlock(&kvm->mmu_lock);
385 srcu_read_unlock(&kvm->srcu, idx);
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391 struct mm_struct *mm,
392 unsigned long address)
394 struct kvm *kvm = mmu_notifier_to_kvm(mn);
397 idx = srcu_read_lock(&kvm->srcu);
398 spin_lock(&kvm->mmu_lock);
399 young = kvm_test_age_hva(kvm, address);
400 spin_unlock(&kvm->mmu_lock);
401 srcu_read_unlock(&kvm->srcu, idx);
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407 struct mm_struct *mm)
409 struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 idx = srcu_read_lock(&kvm->srcu);
413 kvm_arch_flush_shadow_all(kvm);
414 srcu_read_unlock(&kvm->srcu, idx);
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418 .invalidate_page = kvm_mmu_notifier_invalidate_page,
419 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
421 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
422 .test_young = kvm_mmu_notifier_test_young,
423 .change_pte = kvm_mmu_notifier_change_pte,
424 .release = kvm_mmu_notifier_release,
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
433 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442 static void kvm_init_memslots_id(struct kvm *kvm)
445 struct kvm_memslots *slots = kvm->memslots;
447 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
448 slots->id_to_index[i] = slots->memslots[i].id = i;
451 static struct kvm *kvm_create_vm(unsigned long type)
454 struct kvm *kvm = kvm_arch_alloc_vm();
457 return ERR_PTR(-ENOMEM);
459 r = kvm_arch_init_vm(kvm, type);
461 goto out_err_no_disable;
463 r = hardware_enable_all();
465 goto out_err_no_disable;
467 #ifdef CONFIG_HAVE_KVM_IRQFD
468 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
474 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
476 goto out_err_no_srcu;
479 * Init kvm generation close to the maximum to easily test the
480 * code of handling generation number wrap-around.
482 kvm->memslots->generation = -150;
484 kvm_init_memslots_id(kvm);
485 if (init_srcu_struct(&kvm->srcu))
486 goto out_err_no_srcu;
487 if (init_srcu_struct(&kvm->irq_srcu))
488 goto out_err_no_irq_srcu;
489 for (i = 0; i < KVM_NR_BUSES; i++) {
490 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
496 spin_lock_init(&kvm->mmu_lock);
497 kvm->mm = current->mm;
498 atomic_inc(&kvm->mm->mm_count);
499 kvm_eventfd_init(kvm);
500 mutex_init(&kvm->lock);
501 mutex_init(&kvm->irq_lock);
502 mutex_init(&kvm->slots_lock);
503 atomic_set(&kvm->users_count, 1);
504 INIT_LIST_HEAD(&kvm->devices);
506 r = kvm_init_mmu_notifier(kvm);
510 spin_lock(&kvm_lock);
511 list_add(&kvm->vm_list, &vm_list);
512 spin_unlock(&kvm_lock);
517 cleanup_srcu_struct(&kvm->irq_srcu);
519 cleanup_srcu_struct(&kvm->srcu);
521 hardware_disable_all();
523 for (i = 0; i < KVM_NR_BUSES; i++)
524 kfree(kvm->buses[i]);
525 kfree(kvm->memslots);
526 kvm_arch_free_vm(kvm);
531 * Avoid using vmalloc for a small buffer.
532 * Should not be used when the size is statically known.
534 void *kvm_kvzalloc(unsigned long size)
536 if (size > PAGE_SIZE)
537 return vzalloc(size);
539 return kzalloc(size, GFP_KERNEL);
542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
544 if (!memslot->dirty_bitmap)
547 kvfree(memslot->dirty_bitmap);
548 memslot->dirty_bitmap = NULL;
552 * Free any memory in @free but not in @dont.
554 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
555 struct kvm_memory_slot *dont)
557 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
558 kvm_destroy_dirty_bitmap(free);
560 kvm_arch_free_memslot(kvm, free, dont);
565 static void kvm_free_physmem(struct kvm *kvm)
567 struct kvm_memslots *slots = kvm->memslots;
568 struct kvm_memory_slot *memslot;
570 kvm_for_each_memslot(memslot, slots)
571 kvm_free_physmem_slot(kvm, memslot, NULL);
573 kfree(kvm->memslots);
576 static void kvm_destroy_devices(struct kvm *kvm)
578 struct list_head *node, *tmp;
580 list_for_each_safe(node, tmp, &kvm->devices) {
581 struct kvm_device *dev =
582 list_entry(node, struct kvm_device, vm_node);
585 dev->ops->destroy(dev);
589 static void kvm_destroy_vm(struct kvm *kvm)
592 struct mm_struct *mm = kvm->mm;
594 kvm_arch_sync_events(kvm);
595 spin_lock(&kvm_lock);
596 list_del(&kvm->vm_list);
597 spin_unlock(&kvm_lock);
598 kvm_free_irq_routing(kvm);
599 for (i = 0; i < KVM_NR_BUSES; i++)
600 kvm_io_bus_destroy(kvm->buses[i]);
601 kvm_coalesced_mmio_free(kvm);
602 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
603 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
605 kvm_arch_flush_shadow_all(kvm);
607 kvm_arch_destroy_vm(kvm);
608 kvm_destroy_devices(kvm);
609 kvm_free_physmem(kvm);
610 cleanup_srcu_struct(&kvm->irq_srcu);
611 cleanup_srcu_struct(&kvm->srcu);
612 kvm_arch_free_vm(kvm);
613 hardware_disable_all();
617 void kvm_get_kvm(struct kvm *kvm)
619 atomic_inc(&kvm->users_count);
621 EXPORT_SYMBOL_GPL(kvm_get_kvm);
623 void kvm_put_kvm(struct kvm *kvm)
625 if (atomic_dec_and_test(&kvm->users_count))
628 EXPORT_SYMBOL_GPL(kvm_put_kvm);
631 static int kvm_vm_release(struct inode *inode, struct file *filp)
633 struct kvm *kvm = filp->private_data;
635 kvm_irqfd_release(kvm);
642 * Allocation size is twice as large as the actual dirty bitmap size.
643 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
645 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
647 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
649 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
650 if (!memslot->dirty_bitmap)
657 * Insert memslot and re-sort memslots based on their GFN,
658 * so binary search could be used to lookup GFN.
659 * Sorting algorithm takes advantage of having initially
660 * sorted array and known changed memslot position.
662 static void update_memslots(struct kvm_memslots *slots,
663 struct kvm_memory_slot *new)
666 int i = slots->id_to_index[id];
667 struct kvm_memory_slot *mslots = slots->memslots;
669 WARN_ON(mslots[i].id != id);
671 WARN_ON(!mslots[i].npages);
674 if (mslots[i].npages)
677 if (!mslots[i].npages)
681 while (i < KVM_MEM_SLOTS_NUM - 1 &&
682 new->base_gfn <= mslots[i + 1].base_gfn) {
683 if (!mslots[i + 1].npages)
685 mslots[i] = mslots[i + 1];
686 slots->id_to_index[mslots[i].id] = i;
691 * The ">=" is needed when creating a slot with base_gfn == 0,
692 * so that it moves before all those with base_gfn == npages == 0.
694 * On the other hand, if new->npages is zero, the above loop has
695 * already left i pointing to the beginning of the empty part of
696 * mslots, and the ">=" would move the hole backwards in this
697 * case---which is wrong. So skip the loop when deleting a slot.
701 new->base_gfn >= mslots[i - 1].base_gfn) {
702 mslots[i] = mslots[i - 1];
703 slots->id_to_index[mslots[i].id] = i;
707 WARN_ON_ONCE(i != slots->used_slots);
710 slots->id_to_index[mslots[i].id] = i;
713 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
715 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
717 #ifdef __KVM_HAVE_READONLY_MEM
718 valid_flags |= KVM_MEM_READONLY;
721 if (mem->flags & ~valid_flags)
727 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
728 struct kvm_memslots *slots)
730 struct kvm_memslots *old_memslots = kvm->memslots;
733 * Set the low bit in the generation, which disables SPTE caching
734 * until the end of synchronize_srcu_expedited.
736 WARN_ON(old_memslots->generation & 1);
737 slots->generation = old_memslots->generation + 1;
739 rcu_assign_pointer(kvm->memslots, slots);
740 synchronize_srcu_expedited(&kvm->srcu);
743 * Increment the new memslot generation a second time. This prevents
744 * vm exits that race with memslot updates from caching a memslot
745 * generation that will (potentially) be valid forever.
749 kvm_arch_memslots_updated(kvm);
755 * Allocate some memory and give it an address in the guest physical address
758 * Discontiguous memory is allowed, mostly for framebuffers.
760 * Must be called holding kvm->slots_lock for write.
762 int __kvm_set_memory_region(struct kvm *kvm,
763 struct kvm_userspace_memory_region *mem)
767 unsigned long npages;
768 struct kvm_memory_slot *slot;
769 struct kvm_memory_slot old, new;
770 struct kvm_memslots *slots = NULL, *old_memslots;
771 enum kvm_mr_change change;
773 r = check_memory_region_flags(mem);
778 /* General sanity checks */
779 if (mem->memory_size & (PAGE_SIZE - 1))
781 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
783 /* We can read the guest memory with __xxx_user() later on. */
784 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
785 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
786 !access_ok(VERIFY_WRITE,
787 (void __user *)(unsigned long)mem->userspace_addr,
790 if (mem->slot >= KVM_MEM_SLOTS_NUM)
792 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
795 slot = id_to_memslot(kvm->memslots, mem->slot);
796 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
797 npages = mem->memory_size >> PAGE_SHIFT;
799 if (npages > KVM_MEM_MAX_NR_PAGES)
803 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
808 new.base_gfn = base_gfn;
810 new.flags = mem->flags;
814 change = KVM_MR_CREATE;
815 else { /* Modify an existing slot. */
816 if ((mem->userspace_addr != old.userspace_addr) ||
817 (npages != old.npages) ||
818 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
821 if (base_gfn != old.base_gfn)
822 change = KVM_MR_MOVE;
823 else if (new.flags != old.flags)
824 change = KVM_MR_FLAGS_ONLY;
825 else { /* Nothing to change. */
830 } else if (old.npages) {
831 change = KVM_MR_DELETE;
832 } else /* Modify a non-existent slot: disallowed. */
835 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
836 /* Check for overlaps */
838 kvm_for_each_memslot(slot, kvm->memslots) {
839 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
840 (slot->id == mem->slot))
842 if (!((base_gfn + npages <= slot->base_gfn) ||
843 (base_gfn >= slot->base_gfn + slot->npages)))
848 /* Free page dirty bitmap if unneeded */
849 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
850 new.dirty_bitmap = NULL;
853 if (change == KVM_MR_CREATE) {
854 new.userspace_addr = mem->userspace_addr;
856 if (kvm_arch_create_memslot(kvm, &new, npages))
860 /* Allocate page dirty bitmap if needed */
861 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
862 if (kvm_create_dirty_bitmap(&new) < 0)
866 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
871 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
872 slot = id_to_memslot(slots, mem->slot);
873 slot->flags |= KVM_MEMSLOT_INVALID;
875 old_memslots = install_new_memslots(kvm, slots);
877 /* slot was deleted or moved, clear iommu mapping */
878 kvm_iommu_unmap_pages(kvm, &old);
879 /* From this point no new shadow pages pointing to a deleted,
880 * or moved, memslot will be created.
882 * validation of sp->gfn happens in:
883 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
884 * - kvm_is_visible_gfn (mmu_check_roots)
886 kvm_arch_flush_shadow_memslot(kvm, slot);
889 * We can re-use the old_memslots from above, the only difference
890 * from the currently installed memslots is the invalid flag. This
891 * will get overwritten by update_memslots anyway.
893 slots = old_memslots;
896 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
900 /* actual memory is freed via old in kvm_free_physmem_slot below */
901 if (change == KVM_MR_DELETE) {
902 new.dirty_bitmap = NULL;
903 memset(&new.arch, 0, sizeof(new.arch));
906 update_memslots(slots, &new);
907 old_memslots = install_new_memslots(kvm, slots);
909 kvm_arch_commit_memory_region(kvm, mem, &old, change);
911 kvm_free_physmem_slot(kvm, &old, &new);
915 * IOMMU mapping: New slots need to be mapped. Old slots need to be
916 * un-mapped and re-mapped if their base changes. Since base change
917 * unmapping is handled above with slot deletion, mapping alone is
918 * needed here. Anything else the iommu might care about for existing
919 * slots (size changes, userspace addr changes and read-only flag
920 * changes) is disallowed above, so any other attribute changes getting
921 * here can be skipped.
923 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
924 r = kvm_iommu_map_pages(kvm, &new);
933 kvm_free_physmem_slot(kvm, &new, &old);
937 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
939 int kvm_set_memory_region(struct kvm *kvm,
940 struct kvm_userspace_memory_region *mem)
944 mutex_lock(&kvm->slots_lock);
945 r = __kvm_set_memory_region(kvm, mem);
946 mutex_unlock(&kvm->slots_lock);
949 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
951 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
952 struct kvm_userspace_memory_region *mem)
954 if (mem->slot >= KVM_USER_MEM_SLOTS)
956 return kvm_set_memory_region(kvm, mem);
959 int kvm_get_dirty_log(struct kvm *kvm,
960 struct kvm_dirty_log *log, int *is_dirty)
962 struct kvm_memory_slot *memslot;
965 unsigned long any = 0;
968 if (log->slot >= KVM_USER_MEM_SLOTS)
971 memslot = id_to_memslot(kvm->memslots, log->slot);
973 if (!memslot->dirty_bitmap)
976 n = kvm_dirty_bitmap_bytes(memslot);
978 for (i = 0; !any && i < n/sizeof(long); ++i)
979 any = memslot->dirty_bitmap[i];
982 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
992 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
994 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
996 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
997 * are dirty write protect them for next write.
998 * @kvm: pointer to kvm instance
999 * @log: slot id and address to which we copy the log
1000 * @is_dirty: flag set if any page is dirty
1002 * We need to keep it in mind that VCPU threads can write to the bitmap
1003 * concurrently. So, to avoid losing track of dirty pages we keep the
1006 * 1. Take a snapshot of the bit and clear it if needed.
1007 * 2. Write protect the corresponding page.
1008 * 3. Copy the snapshot to the userspace.
1009 * 4. Upon return caller flushes TLB's if needed.
1011 * Between 2 and 4, the guest may write to the page using the remaining TLB
1012 * entry. This is not a problem because the page is reported dirty using
1013 * the snapshot taken before and step 4 ensures that writes done after
1014 * exiting to userspace will be logged for the next call.
1017 int kvm_get_dirty_log_protect(struct kvm *kvm,
1018 struct kvm_dirty_log *log, bool *is_dirty)
1020 struct kvm_memory_slot *memslot;
1023 unsigned long *dirty_bitmap;
1024 unsigned long *dirty_bitmap_buffer;
1027 if (log->slot >= KVM_USER_MEM_SLOTS)
1030 memslot = id_to_memslot(kvm->memslots, log->slot);
1032 dirty_bitmap = memslot->dirty_bitmap;
1037 n = kvm_dirty_bitmap_bytes(memslot);
1039 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1040 memset(dirty_bitmap_buffer, 0, n);
1042 spin_lock(&kvm->mmu_lock);
1044 for (i = 0; i < n / sizeof(long); i++) {
1048 if (!dirty_bitmap[i])
1053 mask = xchg(&dirty_bitmap[i], 0);
1054 dirty_bitmap_buffer[i] = mask;
1057 offset = i * BITS_PER_LONG;
1058 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1063 spin_unlock(&kvm->mmu_lock);
1066 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1073 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1076 bool kvm_largepages_enabled(void)
1078 return largepages_enabled;
1081 void kvm_disable_largepages(void)
1083 largepages_enabled = false;
1085 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1087 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1089 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1091 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1093 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1095 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1097 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1098 memslot->flags & KVM_MEMSLOT_INVALID)
1103 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1105 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1107 struct vm_area_struct *vma;
1108 unsigned long addr, size;
1112 addr = gfn_to_hva(kvm, gfn);
1113 if (kvm_is_error_hva(addr))
1116 down_read(¤t->mm->mmap_sem);
1117 vma = find_vma(current->mm, addr);
1121 size = vma_kernel_pagesize(vma);
1124 up_read(¤t->mm->mmap_sem);
1129 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1131 return slot->flags & KVM_MEM_READONLY;
1134 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1135 gfn_t *nr_pages, bool write)
1137 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1138 return KVM_HVA_ERR_BAD;
1140 if (memslot_is_readonly(slot) && write)
1141 return KVM_HVA_ERR_RO_BAD;
1144 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1146 return __gfn_to_hva_memslot(slot, gfn);
1149 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1152 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1155 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1158 return gfn_to_hva_many(slot, gfn, NULL);
1160 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1162 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1164 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1166 EXPORT_SYMBOL_GPL(gfn_to_hva);
1169 * If writable is set to false, the hva returned by this function is only
1170 * allowed to be read.
1172 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1173 gfn_t gfn, bool *writable)
1175 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1177 if (!kvm_is_error_hva(hva) && writable)
1178 *writable = !memslot_is_readonly(slot);
1183 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1185 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1187 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1190 static int kvm_read_hva(void *data, void __user *hva, int len)
1192 return __copy_from_user(data, hva, len);
1195 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1197 return __copy_from_user_inatomic(data, hva, len);
1200 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1201 unsigned long start, int write, struct page **page)
1203 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1206 flags |= FOLL_WRITE;
1208 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1211 static inline int check_user_page_hwpoison(unsigned long addr)
1213 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1215 rc = __get_user_pages(current, current->mm, addr, 1,
1216 flags, NULL, NULL, NULL);
1217 return rc == -EHWPOISON;
1221 * The atomic path to get the writable pfn which will be stored in @pfn,
1222 * true indicates success, otherwise false is returned.
1224 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1225 bool write_fault, bool *writable, pfn_t *pfn)
1227 struct page *page[1];
1230 if (!(async || atomic))
1234 * Fast pin a writable pfn only if it is a write fault request
1235 * or the caller allows to map a writable pfn for a read fault
1238 if (!(write_fault || writable))
1241 npages = __get_user_pages_fast(addr, 1, 1, page);
1243 *pfn = page_to_pfn(page[0]);
1254 * The slow path to get the pfn of the specified host virtual address,
1255 * 1 indicates success, -errno is returned if error is detected.
1257 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1258 bool *writable, pfn_t *pfn)
1260 struct page *page[1];
1266 *writable = write_fault;
1269 down_read(¤t->mm->mmap_sem);
1270 npages = get_user_page_nowait(current, current->mm,
1271 addr, write_fault, page);
1272 up_read(¤t->mm->mmap_sem);
1274 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1275 write_fault, 0, page,
1276 FOLL_TOUCH|FOLL_HWPOISON);
1280 /* map read fault as writable if possible */
1281 if (unlikely(!write_fault) && writable) {
1282 struct page *wpage[1];
1284 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1293 *pfn = page_to_pfn(page[0]);
1297 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1299 if (unlikely(!(vma->vm_flags & VM_READ)))
1302 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1309 * Pin guest page in memory and return its pfn.
1310 * @addr: host virtual address which maps memory to the guest
1311 * @atomic: whether this function can sleep
1312 * @async: whether this function need to wait IO complete if the
1313 * host page is not in the memory
1314 * @write_fault: whether we should get a writable host page
1315 * @writable: whether it allows to map a writable host page for !@write_fault
1317 * The function will map a writable host page for these two cases:
1318 * 1): @write_fault = true
1319 * 2): @write_fault = false && @writable, @writable will tell the caller
1320 * whether the mapping is writable.
1322 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1323 bool write_fault, bool *writable)
1325 struct vm_area_struct *vma;
1329 /* we can do it either atomically or asynchronously, not both */
1330 BUG_ON(atomic && async);
1332 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1336 return KVM_PFN_ERR_FAULT;
1338 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1342 down_read(¤t->mm->mmap_sem);
1343 if (npages == -EHWPOISON ||
1344 (!async && check_user_page_hwpoison(addr))) {
1345 pfn = KVM_PFN_ERR_HWPOISON;
1349 vma = find_vma_intersection(current->mm, addr, addr + 1);
1352 pfn = KVM_PFN_ERR_FAULT;
1353 else if ((vma->vm_flags & VM_PFNMAP)) {
1354 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1356 BUG_ON(!kvm_is_reserved_pfn(pfn));
1358 if (async && vma_is_valid(vma, write_fault))
1360 pfn = KVM_PFN_ERR_FAULT;
1363 up_read(¤t->mm->mmap_sem);
1368 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1369 bool *async, bool write_fault, bool *writable)
1371 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1373 if (addr == KVM_HVA_ERR_RO_BAD)
1374 return KVM_PFN_ERR_RO_FAULT;
1376 if (kvm_is_error_hva(addr))
1377 return KVM_PFN_NOSLOT;
1379 /* Do not map writable pfn in the readonly memslot. */
1380 if (writable && memslot_is_readonly(slot)) {
1385 return hva_to_pfn(addr, atomic, async, write_fault,
1389 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1390 bool write_fault, bool *writable)
1392 struct kvm_memory_slot *slot;
1397 slot = gfn_to_memslot(kvm, gfn);
1399 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1403 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1405 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1407 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1409 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1410 bool write_fault, bool *writable)
1412 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1414 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1416 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1418 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1420 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1422 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1425 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1427 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1429 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1431 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1434 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1436 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1438 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1440 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1446 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1447 if (kvm_is_error_hva(addr))
1450 if (entry < nr_pages)
1453 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1455 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1457 static struct page *kvm_pfn_to_page(pfn_t pfn)
1459 if (is_error_noslot_pfn(pfn))
1460 return KVM_ERR_PTR_BAD_PAGE;
1462 if (kvm_is_reserved_pfn(pfn)) {
1464 return KVM_ERR_PTR_BAD_PAGE;
1467 return pfn_to_page(pfn);
1470 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1474 pfn = gfn_to_pfn(kvm, gfn);
1476 return kvm_pfn_to_page(pfn);
1478 EXPORT_SYMBOL_GPL(gfn_to_page);
1480 void kvm_release_page_clean(struct page *page)
1482 WARN_ON(is_error_page(page));
1484 kvm_release_pfn_clean(page_to_pfn(page));
1486 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1488 void kvm_release_pfn_clean(pfn_t pfn)
1490 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1491 put_page(pfn_to_page(pfn));
1493 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1495 void kvm_release_page_dirty(struct page *page)
1497 WARN_ON(is_error_page(page));
1499 kvm_release_pfn_dirty(page_to_pfn(page));
1501 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1503 static void kvm_release_pfn_dirty(pfn_t pfn)
1505 kvm_set_pfn_dirty(pfn);
1506 kvm_release_pfn_clean(pfn);
1509 void kvm_set_pfn_dirty(pfn_t pfn)
1511 if (!kvm_is_reserved_pfn(pfn)) {
1512 struct page *page = pfn_to_page(pfn);
1514 if (!PageReserved(page))
1518 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1520 void kvm_set_pfn_accessed(pfn_t pfn)
1522 if (!kvm_is_reserved_pfn(pfn))
1523 mark_page_accessed(pfn_to_page(pfn));
1525 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1527 void kvm_get_pfn(pfn_t pfn)
1529 if (!kvm_is_reserved_pfn(pfn))
1530 get_page(pfn_to_page(pfn));
1532 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1534 static int next_segment(unsigned long len, int offset)
1536 if (len > PAGE_SIZE - offset)
1537 return PAGE_SIZE - offset;
1542 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1548 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1549 if (kvm_is_error_hva(addr))
1551 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1556 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1558 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1560 gfn_t gfn = gpa >> PAGE_SHIFT;
1562 int offset = offset_in_page(gpa);
1565 while ((seg = next_segment(len, offset)) != 0) {
1566 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1576 EXPORT_SYMBOL_GPL(kvm_read_guest);
1578 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1583 gfn_t gfn = gpa >> PAGE_SHIFT;
1584 int offset = offset_in_page(gpa);
1586 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1587 if (kvm_is_error_hva(addr))
1589 pagefault_disable();
1590 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1596 EXPORT_SYMBOL(kvm_read_guest_atomic);
1598 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1599 int offset, int len)
1604 addr = gfn_to_hva(kvm, gfn);
1605 if (kvm_is_error_hva(addr))
1607 r = __copy_to_user((void __user *)addr + offset, data, len);
1610 mark_page_dirty(kvm, gfn);
1613 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1615 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1618 gfn_t gfn = gpa >> PAGE_SHIFT;
1620 int offset = offset_in_page(gpa);
1623 while ((seg = next_segment(len, offset)) != 0) {
1624 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1634 EXPORT_SYMBOL_GPL(kvm_write_guest);
1636 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1637 gpa_t gpa, unsigned long len)
1639 struct kvm_memslots *slots = kvm_memslots(kvm);
1640 int offset = offset_in_page(gpa);
1641 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1642 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1643 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1644 gfn_t nr_pages_avail;
1647 ghc->generation = slots->generation;
1649 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1650 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1651 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1655 * If the requested region crosses two memslots, we still
1656 * verify that the entire region is valid here.
1658 while (start_gfn <= end_gfn) {
1659 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1660 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1662 if (kvm_is_error_hva(ghc->hva))
1664 start_gfn += nr_pages_avail;
1666 /* Use the slow path for cross page reads and writes. */
1667 ghc->memslot = NULL;
1671 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1673 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1674 void *data, unsigned long len)
1676 struct kvm_memslots *slots = kvm_memslots(kvm);
1679 BUG_ON(len > ghc->len);
1681 if (slots->generation != ghc->generation)
1682 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1684 if (unlikely(!ghc->memslot))
1685 return kvm_write_guest(kvm, ghc->gpa, data, len);
1687 if (kvm_is_error_hva(ghc->hva))
1690 r = __copy_to_user((void __user *)ghc->hva, data, len);
1693 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1697 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1699 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1700 void *data, unsigned long len)
1702 struct kvm_memslots *slots = kvm_memslots(kvm);
1705 BUG_ON(len > ghc->len);
1707 if (slots->generation != ghc->generation)
1708 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1710 if (unlikely(!ghc->memslot))
1711 return kvm_read_guest(kvm, ghc->gpa, data, len);
1713 if (kvm_is_error_hva(ghc->hva))
1716 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1722 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1724 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1726 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1728 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1730 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1732 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1734 gfn_t gfn = gpa >> PAGE_SHIFT;
1736 int offset = offset_in_page(gpa);
1739 while ((seg = next_segment(len, offset)) != 0) {
1740 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1749 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1751 static void mark_page_dirty_in_slot(struct kvm *kvm,
1752 struct kvm_memory_slot *memslot,
1755 if (memslot && memslot->dirty_bitmap) {
1756 unsigned long rel_gfn = gfn - memslot->base_gfn;
1758 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1762 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1764 struct kvm_memory_slot *memslot;
1766 memslot = gfn_to_memslot(kvm, gfn);
1767 mark_page_dirty_in_slot(kvm, memslot, gfn);
1769 EXPORT_SYMBOL_GPL(mark_page_dirty);
1771 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1773 if (kvm_arch_vcpu_runnable(vcpu)) {
1774 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1777 if (kvm_cpu_has_pending_timer(vcpu))
1779 if (signal_pending(current))
1786 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1788 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1792 bool waited = false;
1794 start = cur = ktime_get();
1796 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1800 * This sets KVM_REQ_UNHALT if an interrupt
1803 if (kvm_vcpu_check_block(vcpu) < 0) {
1804 ++vcpu->stat.halt_successful_poll;
1808 } while (single_task_running() && ktime_before(cur, stop));
1812 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1814 if (kvm_vcpu_check_block(vcpu) < 0)
1821 finish_wait(&vcpu->wq, &wait);
1825 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1827 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1831 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1833 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1836 int cpu = vcpu->cpu;
1837 wait_queue_head_t *wqp;
1839 wqp = kvm_arch_vcpu_wq(vcpu);
1840 if (waitqueue_active(wqp)) {
1841 wake_up_interruptible(wqp);
1842 ++vcpu->stat.halt_wakeup;
1846 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1847 if (kvm_arch_vcpu_should_kick(vcpu))
1848 smp_send_reschedule(cpu);
1851 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1852 #endif /* !CONFIG_S390 */
1854 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1857 struct task_struct *task = NULL;
1861 pid = rcu_dereference(target->pid);
1863 task = get_pid_task(pid, PIDTYPE_PID);
1867 ret = yield_to(task, 1);
1868 put_task_struct(task);
1872 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1875 * Helper that checks whether a VCPU is eligible for directed yield.
1876 * Most eligible candidate to yield is decided by following heuristics:
1878 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1879 * (preempted lock holder), indicated by @in_spin_loop.
1880 * Set at the beiginning and cleared at the end of interception/PLE handler.
1882 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1883 * chance last time (mostly it has become eligible now since we have probably
1884 * yielded to lockholder in last iteration. This is done by toggling
1885 * @dy_eligible each time a VCPU checked for eligibility.)
1887 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1888 * to preempted lock-holder could result in wrong VCPU selection and CPU
1889 * burning. Giving priority for a potential lock-holder increases lock
1892 * Since algorithm is based on heuristics, accessing another VCPU data without
1893 * locking does not harm. It may result in trying to yield to same VCPU, fail
1894 * and continue with next VCPU and so on.
1896 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1898 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1901 eligible = !vcpu->spin_loop.in_spin_loop ||
1902 vcpu->spin_loop.dy_eligible;
1904 if (vcpu->spin_loop.in_spin_loop)
1905 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1913 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1915 struct kvm *kvm = me->kvm;
1916 struct kvm_vcpu *vcpu;
1917 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1923 kvm_vcpu_set_in_spin_loop(me, true);
1925 * We boost the priority of a VCPU that is runnable but not
1926 * currently running, because it got preempted by something
1927 * else and called schedule in __vcpu_run. Hopefully that
1928 * VCPU is holding the lock that we need and will release it.
1929 * We approximate round-robin by starting at the last boosted VCPU.
1931 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1932 kvm_for_each_vcpu(i, vcpu, kvm) {
1933 if (!pass && i <= last_boosted_vcpu) {
1934 i = last_boosted_vcpu;
1936 } else if (pass && i > last_boosted_vcpu)
1938 if (!ACCESS_ONCE(vcpu->preempted))
1942 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1944 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1947 yielded = kvm_vcpu_yield_to(vcpu);
1949 kvm->last_boosted_vcpu = i;
1951 } else if (yielded < 0) {
1958 kvm_vcpu_set_in_spin_loop(me, false);
1960 /* Ensure vcpu is not eligible during next spinloop */
1961 kvm_vcpu_set_dy_eligible(me, false);
1963 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1965 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1967 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1970 if (vmf->pgoff == 0)
1971 page = virt_to_page(vcpu->run);
1973 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1974 page = virt_to_page(vcpu->arch.pio_data);
1976 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1977 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1978 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1981 return kvm_arch_vcpu_fault(vcpu, vmf);
1987 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1988 .fault = kvm_vcpu_fault,
1991 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1993 vma->vm_ops = &kvm_vcpu_vm_ops;
1997 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1999 struct kvm_vcpu *vcpu = filp->private_data;
2001 kvm_put_kvm(vcpu->kvm);
2005 static struct file_operations kvm_vcpu_fops = {
2006 .release = kvm_vcpu_release,
2007 .unlocked_ioctl = kvm_vcpu_ioctl,
2008 #ifdef CONFIG_KVM_COMPAT
2009 .compat_ioctl = kvm_vcpu_compat_ioctl,
2011 .mmap = kvm_vcpu_mmap,
2012 .llseek = noop_llseek,
2016 * Allocates an inode for the vcpu.
2018 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2020 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2024 * Creates some virtual cpus. Good luck creating more than one.
2026 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2029 struct kvm_vcpu *vcpu, *v;
2031 if (id >= KVM_MAX_VCPUS)
2034 vcpu = kvm_arch_vcpu_create(kvm, id);
2036 return PTR_ERR(vcpu);
2038 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2040 r = kvm_arch_vcpu_setup(vcpu);
2044 mutex_lock(&kvm->lock);
2045 if (!kvm_vcpu_compatible(vcpu)) {
2047 goto unlock_vcpu_destroy;
2049 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2051 goto unlock_vcpu_destroy;
2054 kvm_for_each_vcpu(r, v, kvm)
2055 if (v->vcpu_id == id) {
2057 goto unlock_vcpu_destroy;
2060 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2062 /* Now it's all set up, let userspace reach it */
2064 r = create_vcpu_fd(vcpu);
2067 goto unlock_vcpu_destroy;
2070 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2072 atomic_inc(&kvm->online_vcpus);
2074 mutex_unlock(&kvm->lock);
2075 kvm_arch_vcpu_postcreate(vcpu);
2078 unlock_vcpu_destroy:
2079 mutex_unlock(&kvm->lock);
2081 kvm_arch_vcpu_destroy(vcpu);
2085 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2088 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2089 vcpu->sigset_active = 1;
2090 vcpu->sigset = *sigset;
2092 vcpu->sigset_active = 0;
2096 static long kvm_vcpu_ioctl(struct file *filp,
2097 unsigned int ioctl, unsigned long arg)
2099 struct kvm_vcpu *vcpu = filp->private_data;
2100 void __user *argp = (void __user *)arg;
2102 struct kvm_fpu *fpu = NULL;
2103 struct kvm_sregs *kvm_sregs = NULL;
2105 if (vcpu->kvm->mm != current->mm)
2108 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2111 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2113 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2114 * so vcpu_load() would break it.
2116 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2117 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2121 r = vcpu_load(vcpu);
2129 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2130 /* The thread running this VCPU changed. */
2131 struct pid *oldpid = vcpu->pid;
2132 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2134 rcu_assign_pointer(vcpu->pid, newpid);
2139 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2140 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2142 case KVM_GET_REGS: {
2143 struct kvm_regs *kvm_regs;
2146 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2149 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2153 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2160 case KVM_SET_REGS: {
2161 struct kvm_regs *kvm_regs;
2164 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2165 if (IS_ERR(kvm_regs)) {
2166 r = PTR_ERR(kvm_regs);
2169 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2173 case KVM_GET_SREGS: {
2174 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2178 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2182 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2187 case KVM_SET_SREGS: {
2188 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2189 if (IS_ERR(kvm_sregs)) {
2190 r = PTR_ERR(kvm_sregs);
2194 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2197 case KVM_GET_MP_STATE: {
2198 struct kvm_mp_state mp_state;
2200 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2204 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2209 case KVM_SET_MP_STATE: {
2210 struct kvm_mp_state mp_state;
2213 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2215 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2218 case KVM_TRANSLATE: {
2219 struct kvm_translation tr;
2222 if (copy_from_user(&tr, argp, sizeof(tr)))
2224 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2228 if (copy_to_user(argp, &tr, sizeof(tr)))
2233 case KVM_SET_GUEST_DEBUG: {
2234 struct kvm_guest_debug dbg;
2237 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2239 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2242 case KVM_SET_SIGNAL_MASK: {
2243 struct kvm_signal_mask __user *sigmask_arg = argp;
2244 struct kvm_signal_mask kvm_sigmask;
2245 sigset_t sigset, *p;
2250 if (copy_from_user(&kvm_sigmask, argp,
2251 sizeof(kvm_sigmask)))
2254 if (kvm_sigmask.len != sizeof(sigset))
2257 if (copy_from_user(&sigset, sigmask_arg->sigset,
2262 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2266 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2270 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2274 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2280 fpu = memdup_user(argp, sizeof(*fpu));
2286 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2290 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2299 #ifdef CONFIG_KVM_COMPAT
2300 static long kvm_vcpu_compat_ioctl(struct file *filp,
2301 unsigned int ioctl, unsigned long arg)
2303 struct kvm_vcpu *vcpu = filp->private_data;
2304 void __user *argp = compat_ptr(arg);
2307 if (vcpu->kvm->mm != current->mm)
2311 case KVM_SET_SIGNAL_MASK: {
2312 struct kvm_signal_mask __user *sigmask_arg = argp;
2313 struct kvm_signal_mask kvm_sigmask;
2314 compat_sigset_t csigset;
2319 if (copy_from_user(&kvm_sigmask, argp,
2320 sizeof(kvm_sigmask)))
2323 if (kvm_sigmask.len != sizeof(csigset))
2326 if (copy_from_user(&csigset, sigmask_arg->sigset,
2329 sigset_from_compat(&sigset, &csigset);
2330 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2332 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2336 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2344 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2345 int (*accessor)(struct kvm_device *dev,
2346 struct kvm_device_attr *attr),
2349 struct kvm_device_attr attr;
2354 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2357 return accessor(dev, &attr);
2360 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2363 struct kvm_device *dev = filp->private_data;
2366 case KVM_SET_DEVICE_ATTR:
2367 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2368 case KVM_GET_DEVICE_ATTR:
2369 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2370 case KVM_HAS_DEVICE_ATTR:
2371 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2373 if (dev->ops->ioctl)
2374 return dev->ops->ioctl(dev, ioctl, arg);
2380 static int kvm_device_release(struct inode *inode, struct file *filp)
2382 struct kvm_device *dev = filp->private_data;
2383 struct kvm *kvm = dev->kvm;
2389 static const struct file_operations kvm_device_fops = {
2390 .unlocked_ioctl = kvm_device_ioctl,
2391 #ifdef CONFIG_KVM_COMPAT
2392 .compat_ioctl = kvm_device_ioctl,
2394 .release = kvm_device_release,
2397 struct kvm_device *kvm_device_from_filp(struct file *filp)
2399 if (filp->f_op != &kvm_device_fops)
2402 return filp->private_data;
2405 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2406 #ifdef CONFIG_KVM_MPIC
2407 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2408 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2411 #ifdef CONFIG_KVM_XICS
2412 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2416 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2418 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2421 if (kvm_device_ops_table[type] != NULL)
2424 kvm_device_ops_table[type] = ops;
2428 void kvm_unregister_device_ops(u32 type)
2430 if (kvm_device_ops_table[type] != NULL)
2431 kvm_device_ops_table[type] = NULL;
2434 static int kvm_ioctl_create_device(struct kvm *kvm,
2435 struct kvm_create_device *cd)
2437 struct kvm_device_ops *ops = NULL;
2438 struct kvm_device *dev;
2439 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2442 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2445 ops = kvm_device_ops_table[cd->type];
2452 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2459 ret = ops->create(dev, cd->type);
2465 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2471 list_add(&dev->vm_node, &kvm->devices);
2477 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2480 case KVM_CAP_USER_MEMORY:
2481 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2482 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2483 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2484 case KVM_CAP_SET_BOOT_CPU_ID:
2486 case KVM_CAP_INTERNAL_ERROR_DATA:
2487 #ifdef CONFIG_HAVE_KVM_MSI
2488 case KVM_CAP_SIGNAL_MSI:
2490 #ifdef CONFIG_HAVE_KVM_IRQFD
2491 case KVM_CAP_IRQFD_RESAMPLE:
2493 case KVM_CAP_CHECK_EXTENSION_VM:
2495 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2496 case KVM_CAP_IRQ_ROUTING:
2497 return KVM_MAX_IRQ_ROUTES;
2502 return kvm_vm_ioctl_check_extension(kvm, arg);
2505 static long kvm_vm_ioctl(struct file *filp,
2506 unsigned int ioctl, unsigned long arg)
2508 struct kvm *kvm = filp->private_data;
2509 void __user *argp = (void __user *)arg;
2512 if (kvm->mm != current->mm)
2515 case KVM_CREATE_VCPU:
2516 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2518 case KVM_SET_USER_MEMORY_REGION: {
2519 struct kvm_userspace_memory_region kvm_userspace_mem;
2522 if (copy_from_user(&kvm_userspace_mem, argp,
2523 sizeof(kvm_userspace_mem)))
2526 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2529 case KVM_GET_DIRTY_LOG: {
2530 struct kvm_dirty_log log;
2533 if (copy_from_user(&log, argp, sizeof(log)))
2535 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2538 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2539 case KVM_REGISTER_COALESCED_MMIO: {
2540 struct kvm_coalesced_mmio_zone zone;
2543 if (copy_from_user(&zone, argp, sizeof(zone)))
2545 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2548 case KVM_UNREGISTER_COALESCED_MMIO: {
2549 struct kvm_coalesced_mmio_zone zone;
2552 if (copy_from_user(&zone, argp, sizeof(zone)))
2554 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2559 struct kvm_irqfd data;
2562 if (copy_from_user(&data, argp, sizeof(data)))
2564 r = kvm_irqfd(kvm, &data);
2567 case KVM_IOEVENTFD: {
2568 struct kvm_ioeventfd data;
2571 if (copy_from_user(&data, argp, sizeof(data)))
2573 r = kvm_ioeventfd(kvm, &data);
2576 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2577 case KVM_SET_BOOT_CPU_ID:
2579 mutex_lock(&kvm->lock);
2580 if (atomic_read(&kvm->online_vcpus) != 0)
2583 kvm->bsp_vcpu_id = arg;
2584 mutex_unlock(&kvm->lock);
2587 #ifdef CONFIG_HAVE_KVM_MSI
2588 case KVM_SIGNAL_MSI: {
2592 if (copy_from_user(&msi, argp, sizeof(msi)))
2594 r = kvm_send_userspace_msi(kvm, &msi);
2598 #ifdef __KVM_HAVE_IRQ_LINE
2599 case KVM_IRQ_LINE_STATUS:
2600 case KVM_IRQ_LINE: {
2601 struct kvm_irq_level irq_event;
2604 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2607 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2608 ioctl == KVM_IRQ_LINE_STATUS);
2613 if (ioctl == KVM_IRQ_LINE_STATUS) {
2614 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2622 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2623 case KVM_SET_GSI_ROUTING: {
2624 struct kvm_irq_routing routing;
2625 struct kvm_irq_routing __user *urouting;
2626 struct kvm_irq_routing_entry *entries;
2629 if (copy_from_user(&routing, argp, sizeof(routing)))
2632 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2637 entries = vmalloc(routing.nr * sizeof(*entries));
2642 if (copy_from_user(entries, urouting->entries,
2643 routing.nr * sizeof(*entries)))
2644 goto out_free_irq_routing;
2645 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2647 out_free_irq_routing:
2651 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2652 case KVM_CREATE_DEVICE: {
2653 struct kvm_create_device cd;
2656 if (copy_from_user(&cd, argp, sizeof(cd)))
2659 r = kvm_ioctl_create_device(kvm, &cd);
2664 if (copy_to_user(argp, &cd, sizeof(cd)))
2670 case KVM_CHECK_EXTENSION:
2671 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2674 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2680 #ifdef CONFIG_KVM_COMPAT
2681 struct compat_kvm_dirty_log {
2685 compat_uptr_t dirty_bitmap; /* one bit per page */
2690 static long kvm_vm_compat_ioctl(struct file *filp,
2691 unsigned int ioctl, unsigned long arg)
2693 struct kvm *kvm = filp->private_data;
2696 if (kvm->mm != current->mm)
2699 case KVM_GET_DIRTY_LOG: {
2700 struct compat_kvm_dirty_log compat_log;
2701 struct kvm_dirty_log log;
2704 if (copy_from_user(&compat_log, (void __user *)arg,
2705 sizeof(compat_log)))
2707 log.slot = compat_log.slot;
2708 log.padding1 = compat_log.padding1;
2709 log.padding2 = compat_log.padding2;
2710 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2712 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2716 r = kvm_vm_ioctl(filp, ioctl, arg);
2724 static struct file_operations kvm_vm_fops = {
2725 .release = kvm_vm_release,
2726 .unlocked_ioctl = kvm_vm_ioctl,
2727 #ifdef CONFIG_KVM_COMPAT
2728 .compat_ioctl = kvm_vm_compat_ioctl,
2730 .llseek = noop_llseek,
2733 static int kvm_dev_ioctl_create_vm(unsigned long type)
2738 kvm = kvm_create_vm(type);
2740 return PTR_ERR(kvm);
2741 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2742 r = kvm_coalesced_mmio_init(kvm);
2748 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2755 static long kvm_dev_ioctl(struct file *filp,
2756 unsigned int ioctl, unsigned long arg)
2761 case KVM_GET_API_VERSION:
2764 r = KVM_API_VERSION;
2767 r = kvm_dev_ioctl_create_vm(arg);
2769 case KVM_CHECK_EXTENSION:
2770 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2772 case KVM_GET_VCPU_MMAP_SIZE:
2775 r = PAGE_SIZE; /* struct kvm_run */
2777 r += PAGE_SIZE; /* pio data page */
2779 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2780 r += PAGE_SIZE; /* coalesced mmio ring page */
2783 case KVM_TRACE_ENABLE:
2784 case KVM_TRACE_PAUSE:
2785 case KVM_TRACE_DISABLE:
2789 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2795 static struct file_operations kvm_chardev_ops = {
2796 .unlocked_ioctl = kvm_dev_ioctl,
2797 .compat_ioctl = kvm_dev_ioctl,
2798 .llseek = noop_llseek,
2801 static struct miscdevice kvm_dev = {
2807 static void hardware_enable_nolock(void *junk)
2809 int cpu = raw_smp_processor_id();
2812 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2815 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2817 r = kvm_arch_hardware_enable();
2820 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2821 atomic_inc(&hardware_enable_failed);
2822 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2826 static void hardware_enable(void)
2828 raw_spin_lock(&kvm_count_lock);
2829 if (kvm_usage_count)
2830 hardware_enable_nolock(NULL);
2831 raw_spin_unlock(&kvm_count_lock);
2834 static void hardware_disable_nolock(void *junk)
2836 int cpu = raw_smp_processor_id();
2838 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2840 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2841 kvm_arch_hardware_disable();
2844 static void hardware_disable(void)
2846 raw_spin_lock(&kvm_count_lock);
2847 if (kvm_usage_count)
2848 hardware_disable_nolock(NULL);
2849 raw_spin_unlock(&kvm_count_lock);
2852 static void hardware_disable_all_nolock(void)
2854 BUG_ON(!kvm_usage_count);
2857 if (!kvm_usage_count)
2858 on_each_cpu(hardware_disable_nolock, NULL, 1);
2861 static void hardware_disable_all(void)
2863 raw_spin_lock(&kvm_count_lock);
2864 hardware_disable_all_nolock();
2865 raw_spin_unlock(&kvm_count_lock);
2868 static int hardware_enable_all(void)
2872 raw_spin_lock(&kvm_count_lock);
2875 if (kvm_usage_count == 1) {
2876 atomic_set(&hardware_enable_failed, 0);
2877 on_each_cpu(hardware_enable_nolock, NULL, 1);
2879 if (atomic_read(&hardware_enable_failed)) {
2880 hardware_disable_all_nolock();
2885 raw_spin_unlock(&kvm_count_lock);
2890 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2895 val &= ~CPU_TASKS_FROZEN;
2898 pr_info("kvm: disabling virtualization on CPU%d\n",
2903 pr_info("kvm: enabling virtualization on CPU%d\n",
2911 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2915 * Some (well, at least mine) BIOSes hang on reboot if
2918 * And Intel TXT required VMX off for all cpu when system shutdown.
2920 pr_info("kvm: exiting hardware virtualization\n");
2921 kvm_rebooting = true;
2922 on_each_cpu(hardware_disable_nolock, NULL, 1);
2926 static struct notifier_block kvm_reboot_notifier = {
2927 .notifier_call = kvm_reboot,
2931 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2935 for (i = 0; i < bus->dev_count; i++) {
2936 struct kvm_io_device *pos = bus->range[i].dev;
2938 kvm_iodevice_destructor(pos);
2943 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2944 const struct kvm_io_range *r2)
2946 if (r1->addr < r2->addr)
2948 if (r1->addr + r1->len > r2->addr + r2->len)
2953 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2955 return kvm_io_bus_cmp(p1, p2);
2958 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2959 gpa_t addr, int len)
2961 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2967 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2968 kvm_io_bus_sort_cmp, NULL);
2973 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2974 gpa_t addr, int len)
2976 struct kvm_io_range *range, key;
2979 key = (struct kvm_io_range) {
2984 range = bsearch(&key, bus->range, bus->dev_count,
2985 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2989 off = range - bus->range;
2991 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2997 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2998 struct kvm_io_range *range, const void *val)
3002 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3006 while (idx < bus->dev_count &&
3007 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3008 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
3017 /* kvm_io_bus_write - called under kvm->slots_lock */
3018 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3019 int len, const void *val)
3021 struct kvm_io_bus *bus;
3022 struct kvm_io_range range;
3025 range = (struct kvm_io_range) {
3030 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3031 r = __kvm_io_bus_write(bus, &range, val);
3032 return r < 0 ? r : 0;
3035 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3036 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3037 int len, const void *val, long cookie)
3039 struct kvm_io_bus *bus;
3040 struct kvm_io_range range;
3042 range = (struct kvm_io_range) {
3047 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3049 /* First try the device referenced by cookie. */
3050 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3051 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3052 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
3057 * cookie contained garbage; fall back to search and return the
3058 * correct cookie value.
3060 return __kvm_io_bus_write(bus, &range, val);
3063 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
3068 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3072 while (idx < bus->dev_count &&
3073 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3074 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3082 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3084 /* kvm_io_bus_read - called under kvm->slots_lock */
3085 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3088 struct kvm_io_bus *bus;
3089 struct kvm_io_range range;
3092 range = (struct kvm_io_range) {
3097 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3098 r = __kvm_io_bus_read(bus, &range, val);
3099 return r < 0 ? r : 0;
3103 /* Caller must hold slots_lock. */
3104 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3105 int len, struct kvm_io_device *dev)
3107 struct kvm_io_bus *new_bus, *bus;
3109 bus = kvm->buses[bus_idx];
3110 /* exclude ioeventfd which is limited by maximum fd */
3111 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3114 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3115 sizeof(struct kvm_io_range)), GFP_KERNEL);
3118 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3119 sizeof(struct kvm_io_range)));
3120 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3121 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3122 synchronize_srcu_expedited(&kvm->srcu);
3128 /* Caller must hold slots_lock. */
3129 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3130 struct kvm_io_device *dev)
3133 struct kvm_io_bus *new_bus, *bus;
3135 bus = kvm->buses[bus_idx];
3137 for (i = 0; i < bus->dev_count; i++)
3138 if (bus->range[i].dev == dev) {
3146 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3147 sizeof(struct kvm_io_range)), GFP_KERNEL);
3151 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3152 new_bus->dev_count--;
3153 memcpy(new_bus->range + i, bus->range + i + 1,
3154 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3156 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3157 synchronize_srcu_expedited(&kvm->srcu);
3162 static struct notifier_block kvm_cpu_notifier = {
3163 .notifier_call = kvm_cpu_hotplug,
3166 static int vm_stat_get(void *_offset, u64 *val)
3168 unsigned offset = (long)_offset;
3172 spin_lock(&kvm_lock);
3173 list_for_each_entry(kvm, &vm_list, vm_list)
3174 *val += *(u32 *)((void *)kvm + offset);
3175 spin_unlock(&kvm_lock);
3179 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3181 static int vcpu_stat_get(void *_offset, u64 *val)
3183 unsigned offset = (long)_offset;
3185 struct kvm_vcpu *vcpu;
3189 spin_lock(&kvm_lock);
3190 list_for_each_entry(kvm, &vm_list, vm_list)
3191 kvm_for_each_vcpu(i, vcpu, kvm)
3192 *val += *(u32 *)((void *)vcpu + offset);
3194 spin_unlock(&kvm_lock);
3198 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3200 static const struct file_operations *stat_fops[] = {
3201 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3202 [KVM_STAT_VM] = &vm_stat_fops,
3205 static int kvm_init_debug(void)
3208 struct kvm_stats_debugfs_item *p;
3210 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3211 if (kvm_debugfs_dir == NULL)
3214 for (p = debugfs_entries; p->name; ++p) {
3215 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3216 (void *)(long)p->offset,
3217 stat_fops[p->kind]);
3218 if (p->dentry == NULL)
3225 debugfs_remove_recursive(kvm_debugfs_dir);
3230 static void kvm_exit_debug(void)
3232 struct kvm_stats_debugfs_item *p;
3234 for (p = debugfs_entries; p->name; ++p)
3235 debugfs_remove(p->dentry);
3236 debugfs_remove(kvm_debugfs_dir);
3239 static int kvm_suspend(void)
3241 if (kvm_usage_count)
3242 hardware_disable_nolock(NULL);
3246 static void kvm_resume(void)
3248 if (kvm_usage_count) {
3249 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3250 hardware_enable_nolock(NULL);
3254 static struct syscore_ops kvm_syscore_ops = {
3255 .suspend = kvm_suspend,
3256 .resume = kvm_resume,
3260 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3262 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3265 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3267 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3269 if (vcpu->preempted)
3270 vcpu->preempted = false;
3272 kvm_arch_sched_in(vcpu, cpu);
3274 kvm_arch_vcpu_load(vcpu, cpu);
3277 static void kvm_sched_out(struct preempt_notifier *pn,
3278 struct task_struct *next)
3280 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3282 if (current->state == TASK_RUNNING)
3283 vcpu->preempted = true;
3284 kvm_arch_vcpu_put(vcpu);
3287 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3288 struct module *module)
3293 r = kvm_arch_init(opaque);
3298 * kvm_arch_init makes sure there's at most one caller
3299 * for architectures that support multiple implementations,
3300 * like intel and amd on x86.
3301 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3302 * conflicts in case kvm is already setup for another implementation.
3304 r = kvm_irqfd_init();
3308 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3313 r = kvm_arch_hardware_setup();
3317 for_each_online_cpu(cpu) {
3318 smp_call_function_single(cpu,
3319 kvm_arch_check_processor_compat,
3325 r = register_cpu_notifier(&kvm_cpu_notifier);
3328 register_reboot_notifier(&kvm_reboot_notifier);
3330 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3332 vcpu_align = __alignof__(struct kvm_vcpu);
3333 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3335 if (!kvm_vcpu_cache) {
3340 r = kvm_async_pf_init();
3344 kvm_chardev_ops.owner = module;
3345 kvm_vm_fops.owner = module;
3346 kvm_vcpu_fops.owner = module;
3348 r = misc_register(&kvm_dev);
3350 pr_err("kvm: misc device register failed\n");
3354 register_syscore_ops(&kvm_syscore_ops);
3356 kvm_preempt_ops.sched_in = kvm_sched_in;
3357 kvm_preempt_ops.sched_out = kvm_sched_out;
3359 r = kvm_init_debug();
3361 pr_err("kvm: create debugfs files failed\n");
3365 r = kvm_vfio_ops_init();
3371 unregister_syscore_ops(&kvm_syscore_ops);
3372 misc_deregister(&kvm_dev);
3374 kvm_async_pf_deinit();
3376 kmem_cache_destroy(kvm_vcpu_cache);
3378 unregister_reboot_notifier(&kvm_reboot_notifier);
3379 unregister_cpu_notifier(&kvm_cpu_notifier);
3382 kvm_arch_hardware_unsetup();
3384 free_cpumask_var(cpus_hardware_enabled);
3392 EXPORT_SYMBOL_GPL(kvm_init);
3397 misc_deregister(&kvm_dev);
3398 kmem_cache_destroy(kvm_vcpu_cache);
3399 kvm_async_pf_deinit();
3400 unregister_syscore_ops(&kvm_syscore_ops);
3401 unregister_reboot_notifier(&kvm_reboot_notifier);
3402 unregister_cpu_notifier(&kvm_cpu_notifier);
3403 on_each_cpu(hardware_disable_nolock, NULL, 1);
3404 kvm_arch_hardware_unsetup();
3407 free_cpumask_var(cpus_hardware_enabled);
3408 kvm_vfio_ops_exit();
3410 EXPORT_SYMBOL_GPL(kvm_exit);