Loading...
1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
19#include <kvm/iodev.h>
20
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>
26#include <linux/mm.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>
52
53#include <asm/processor.h>
54#include <asm/io.h>
55#include <asm/ioctl.h>
56#include <asm/uaccess.h>
57#include <asm/pgtable.h>
58
59#include "coalesced_mmio.h"
60#include "async_pf.h"
61#include "vfio.h"
62
63#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h>
65
66MODULE_AUTHOR("Qumranet");
67MODULE_LICENSE("GPL");
68
69/* Architectures should define their poll value according to the halt latency */
70static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
71module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
72
73/* Default doubles per-vcpu halt_poll_ns. */
74static unsigned int halt_poll_ns_grow = 2;
75module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
76
77/* Default resets per-vcpu halt_poll_ns . */
78static unsigned int halt_poll_ns_shrink;
79module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
80
81/*
82 * Ordering of locks:
83 *
84 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
85 */
86
87DEFINE_SPINLOCK(kvm_lock);
88static DEFINE_RAW_SPINLOCK(kvm_count_lock);
89LIST_HEAD(vm_list);
90
91static cpumask_var_t cpus_hardware_enabled;
92static int kvm_usage_count;
93static atomic_t hardware_enable_failed;
94
95struct kmem_cache *kvm_vcpu_cache;
96EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
97
98static __read_mostly struct preempt_ops kvm_preempt_ops;
99
100struct dentry *kvm_debugfs_dir;
101EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
102
103static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
104 unsigned long arg);
105#ifdef CONFIG_KVM_COMPAT
106static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
107 unsigned long arg);
108#endif
109static int hardware_enable_all(void);
110static void hardware_disable_all(void);
111
112static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
113
114static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
115static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
116
117__visible bool kvm_rebooting;
118EXPORT_SYMBOL_GPL(kvm_rebooting);
119
120static bool largepages_enabled = true;
121
122bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
123{
124 if (pfn_valid(pfn))
125 return PageReserved(pfn_to_page(pfn));
126
127 return true;
128}
129
130/*
131 * Switches to specified vcpu, until a matching vcpu_put()
132 */
133int vcpu_load(struct kvm_vcpu *vcpu)
134{
135 int cpu;
136
137 if (mutex_lock_killable(&vcpu->mutex))
138 return -EINTR;
139 cpu = get_cpu();
140 preempt_notifier_register(&vcpu->preempt_notifier);
141 kvm_arch_vcpu_load(vcpu, cpu);
142 put_cpu();
143 return 0;
144}
145
146void vcpu_put(struct kvm_vcpu *vcpu)
147{
148 preempt_disable();
149 kvm_arch_vcpu_put(vcpu);
150 preempt_notifier_unregister(&vcpu->preempt_notifier);
151 preempt_enable();
152 mutex_unlock(&vcpu->mutex);
153}
154
155static void ack_flush(void *_completed)
156{
157}
158
159bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
160{
161 int i, cpu, me;
162 cpumask_var_t cpus;
163 bool called = true;
164 struct kvm_vcpu *vcpu;
165
166 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
167
168 me = get_cpu();
169 kvm_for_each_vcpu(i, vcpu, kvm) {
170 kvm_make_request(req, vcpu);
171 cpu = vcpu->cpu;
172
173 /* Set ->requests bit before we read ->mode. */
174 smp_mb__after_atomic();
175
176 if (cpus != NULL && cpu != -1 && cpu != me &&
177 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
178 cpumask_set_cpu(cpu, cpus);
179 }
180 if (unlikely(cpus == NULL))
181 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
182 else if (!cpumask_empty(cpus))
183 smp_call_function_many(cpus, ack_flush, NULL, 1);
184 else
185 called = false;
186 put_cpu();
187 free_cpumask_var(cpus);
188 return called;
189}
190
191#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
192void kvm_flush_remote_tlbs(struct kvm *kvm)
193{
194 /*
195 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
196 * kvm_make_all_cpus_request.
197 */
198 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
199
200 /*
201 * We want to publish modifications to the page tables before reading
202 * mode. Pairs with a memory barrier in arch-specific code.
203 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
204 * and smp_mb in walk_shadow_page_lockless_begin/end.
205 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
206 *
207 * There is already an smp_mb__after_atomic() before
208 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
209 * barrier here.
210 */
211 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
212 ++kvm->stat.remote_tlb_flush;
213 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
214}
215EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
216#endif
217
218void kvm_reload_remote_mmus(struct kvm *kvm)
219{
220 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
221}
222
223int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
224{
225 struct page *page;
226 int r;
227
228 mutex_init(&vcpu->mutex);
229 vcpu->cpu = -1;
230 vcpu->kvm = kvm;
231 vcpu->vcpu_id = id;
232 vcpu->pid = NULL;
233 init_swait_queue_head(&vcpu->wq);
234 kvm_async_pf_vcpu_init(vcpu);
235
236 vcpu->pre_pcpu = -1;
237 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
238
239 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
240 if (!page) {
241 r = -ENOMEM;
242 goto fail;
243 }
244 vcpu->run = page_address(page);
245
246 kvm_vcpu_set_in_spin_loop(vcpu, false);
247 kvm_vcpu_set_dy_eligible(vcpu, false);
248 vcpu->preempted = false;
249
250 r = kvm_arch_vcpu_init(vcpu);
251 if (r < 0)
252 goto fail_free_run;
253 return 0;
254
255fail_free_run:
256 free_page((unsigned long)vcpu->run);
257fail:
258 return r;
259}
260EXPORT_SYMBOL_GPL(kvm_vcpu_init);
261
262void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
263{
264 put_pid(vcpu->pid);
265 kvm_arch_vcpu_uninit(vcpu);
266 free_page((unsigned long)vcpu->run);
267}
268EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
269
270#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
271static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
272{
273 return container_of(mn, struct kvm, mmu_notifier);
274}
275
276static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
277 struct mm_struct *mm,
278 unsigned long address)
279{
280 struct kvm *kvm = mmu_notifier_to_kvm(mn);
281 int need_tlb_flush, idx;
282
283 /*
284 * When ->invalidate_page runs, the linux pte has been zapped
285 * already but the page is still allocated until
286 * ->invalidate_page returns. So if we increase the sequence
287 * here the kvm page fault will notice if the spte can't be
288 * established because the page is going to be freed. If
289 * instead the kvm page fault establishes the spte before
290 * ->invalidate_page runs, kvm_unmap_hva will release it
291 * before returning.
292 *
293 * The sequence increase only need to be seen at spin_unlock
294 * time, and not at spin_lock time.
295 *
296 * Increasing the sequence after the spin_unlock would be
297 * unsafe because the kvm page fault could then establish the
298 * pte after kvm_unmap_hva returned, without noticing the page
299 * is going to be freed.
300 */
301 idx = srcu_read_lock(&kvm->srcu);
302 spin_lock(&kvm->mmu_lock);
303
304 kvm->mmu_notifier_seq++;
305 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
306 /* we've to flush the tlb before the pages can be freed */
307 if (need_tlb_flush)
308 kvm_flush_remote_tlbs(kvm);
309
310 spin_unlock(&kvm->mmu_lock);
311
312 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
313
314 srcu_read_unlock(&kvm->srcu, idx);
315}
316
317static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
318 struct mm_struct *mm,
319 unsigned long address,
320 pte_t pte)
321{
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int idx;
324
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
327 kvm->mmu_notifier_seq++;
328 kvm_set_spte_hva(kvm, address, pte);
329 spin_unlock(&kvm->mmu_lock);
330 srcu_read_unlock(&kvm->srcu, idx);
331}
332
333static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
334 struct mm_struct *mm,
335 unsigned long start,
336 unsigned long end)
337{
338 struct kvm *kvm = mmu_notifier_to_kvm(mn);
339 int need_tlb_flush = 0, idx;
340
341 idx = srcu_read_lock(&kvm->srcu);
342 spin_lock(&kvm->mmu_lock);
343 /*
344 * The count increase must become visible at unlock time as no
345 * spte can be established without taking the mmu_lock and
346 * count is also read inside the mmu_lock critical section.
347 */
348 kvm->mmu_notifier_count++;
349 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
350 need_tlb_flush |= kvm->tlbs_dirty;
351 /* we've to flush the tlb before the pages can be freed */
352 if (need_tlb_flush)
353 kvm_flush_remote_tlbs(kvm);
354
355 spin_unlock(&kvm->mmu_lock);
356 srcu_read_unlock(&kvm->srcu, idx);
357}
358
359static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
360 struct mm_struct *mm,
361 unsigned long start,
362 unsigned long end)
363{
364 struct kvm *kvm = mmu_notifier_to_kvm(mn);
365
366 spin_lock(&kvm->mmu_lock);
367 /*
368 * This sequence increase will notify the kvm page fault that
369 * the page that is going to be mapped in the spte could have
370 * been freed.
371 */
372 kvm->mmu_notifier_seq++;
373 smp_wmb();
374 /*
375 * The above sequence increase must be visible before the
376 * below count decrease, which is ensured by the smp_wmb above
377 * in conjunction with the smp_rmb in mmu_notifier_retry().
378 */
379 kvm->mmu_notifier_count--;
380 spin_unlock(&kvm->mmu_lock);
381
382 BUG_ON(kvm->mmu_notifier_count < 0);
383}
384
385static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
386 struct mm_struct *mm,
387 unsigned long start,
388 unsigned long end)
389{
390 struct kvm *kvm = mmu_notifier_to_kvm(mn);
391 int young, idx;
392
393 idx = srcu_read_lock(&kvm->srcu);
394 spin_lock(&kvm->mmu_lock);
395
396 young = kvm_age_hva(kvm, start, end);
397 if (young)
398 kvm_flush_remote_tlbs(kvm);
399
400 spin_unlock(&kvm->mmu_lock);
401 srcu_read_unlock(&kvm->srcu, idx);
402
403 return young;
404}
405
406static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
407 struct mm_struct *mm,
408 unsigned long start,
409 unsigned long end)
410{
411 struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 int young, idx;
413
414 idx = srcu_read_lock(&kvm->srcu);
415 spin_lock(&kvm->mmu_lock);
416 /*
417 * Even though we do not flush TLB, this will still adversely
418 * affect performance on pre-Haswell Intel EPT, where there is
419 * no EPT Access Bit to clear so that we have to tear down EPT
420 * tables instead. If we find this unacceptable, we can always
421 * add a parameter to kvm_age_hva so that it effectively doesn't
422 * do anything on clear_young.
423 *
424 * Also note that currently we never issue secondary TLB flushes
425 * from clear_young, leaving this job up to the regular system
426 * cadence. If we find this inaccurate, we might come up with a
427 * more sophisticated heuristic later.
428 */
429 young = kvm_age_hva(kvm, start, end);
430 spin_unlock(&kvm->mmu_lock);
431 srcu_read_unlock(&kvm->srcu, idx);
432
433 return young;
434}
435
436static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
437 struct mm_struct *mm,
438 unsigned long address)
439{
440 struct kvm *kvm = mmu_notifier_to_kvm(mn);
441 int young, idx;
442
443 idx = srcu_read_lock(&kvm->srcu);
444 spin_lock(&kvm->mmu_lock);
445 young = kvm_test_age_hva(kvm, address);
446 spin_unlock(&kvm->mmu_lock);
447 srcu_read_unlock(&kvm->srcu, idx);
448
449 return young;
450}
451
452static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
453 struct mm_struct *mm)
454{
455 struct kvm *kvm = mmu_notifier_to_kvm(mn);
456 int idx;
457
458 idx = srcu_read_lock(&kvm->srcu);
459 kvm_arch_flush_shadow_all(kvm);
460 srcu_read_unlock(&kvm->srcu, idx);
461}
462
463static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
464 .invalidate_page = kvm_mmu_notifier_invalidate_page,
465 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
466 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
467 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
468 .clear_young = kvm_mmu_notifier_clear_young,
469 .test_young = kvm_mmu_notifier_test_young,
470 .change_pte = kvm_mmu_notifier_change_pte,
471 .release = kvm_mmu_notifier_release,
472};
473
474static int kvm_init_mmu_notifier(struct kvm *kvm)
475{
476 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
477 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
478}
479
480#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
481
482static int kvm_init_mmu_notifier(struct kvm *kvm)
483{
484 return 0;
485}
486
487#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
488
489static struct kvm_memslots *kvm_alloc_memslots(void)
490{
491 int i;
492 struct kvm_memslots *slots;
493
494 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
495 if (!slots)
496 return NULL;
497
498 /*
499 * Init kvm generation close to the maximum to easily test the
500 * code of handling generation number wrap-around.
501 */
502 slots->generation = -150;
503 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
504 slots->id_to_index[i] = slots->memslots[i].id = i;
505
506 return slots;
507}
508
509static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
510{
511 if (!memslot->dirty_bitmap)
512 return;
513
514 kvfree(memslot->dirty_bitmap);
515 memslot->dirty_bitmap = NULL;
516}
517
518/*
519 * Free any memory in @free but not in @dont.
520 */
521static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
522 struct kvm_memory_slot *dont)
523{
524 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
525 kvm_destroy_dirty_bitmap(free);
526
527 kvm_arch_free_memslot(kvm, free, dont);
528
529 free->npages = 0;
530}
531
532static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
533{
534 struct kvm_memory_slot *memslot;
535
536 if (!slots)
537 return;
538
539 kvm_for_each_memslot(memslot, slots)
540 kvm_free_memslot(kvm, memslot, NULL);
541
542 kvfree(slots);
543}
544
545static struct kvm *kvm_create_vm(unsigned long type)
546{
547 int r, i;
548 struct kvm *kvm = kvm_arch_alloc_vm();
549
550 if (!kvm)
551 return ERR_PTR(-ENOMEM);
552
553 spin_lock_init(&kvm->mmu_lock);
554 atomic_inc(¤t->mm->mm_count);
555 kvm->mm = current->mm;
556 kvm_eventfd_init(kvm);
557 mutex_init(&kvm->lock);
558 mutex_init(&kvm->irq_lock);
559 mutex_init(&kvm->slots_lock);
560 atomic_set(&kvm->users_count, 1);
561 INIT_LIST_HEAD(&kvm->devices);
562
563 r = kvm_arch_init_vm(kvm, type);
564 if (r)
565 goto out_err_no_disable;
566
567 r = hardware_enable_all();
568 if (r)
569 goto out_err_no_disable;
570
571#ifdef CONFIG_HAVE_KVM_IRQFD
572 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
573#endif
574
575 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
576
577 r = -ENOMEM;
578 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
579 kvm->memslots[i] = kvm_alloc_memslots();
580 if (!kvm->memslots[i])
581 goto out_err_no_srcu;
582 }
583
584 if (init_srcu_struct(&kvm->srcu))
585 goto out_err_no_srcu;
586 if (init_srcu_struct(&kvm->irq_srcu))
587 goto out_err_no_irq_srcu;
588 for (i = 0; i < KVM_NR_BUSES; i++) {
589 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
590 GFP_KERNEL);
591 if (!kvm->buses[i])
592 goto out_err;
593 }
594
595 r = kvm_init_mmu_notifier(kvm);
596 if (r)
597 goto out_err;
598
599 spin_lock(&kvm_lock);
600 list_add(&kvm->vm_list, &vm_list);
601 spin_unlock(&kvm_lock);
602
603 preempt_notifier_inc();
604
605 return kvm;
606
607out_err:
608 cleanup_srcu_struct(&kvm->irq_srcu);
609out_err_no_irq_srcu:
610 cleanup_srcu_struct(&kvm->srcu);
611out_err_no_srcu:
612 hardware_disable_all();
613out_err_no_disable:
614 for (i = 0; i < KVM_NR_BUSES; i++)
615 kfree(kvm->buses[i]);
616 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
617 kvm_free_memslots(kvm, kvm->memslots[i]);
618 kvm_arch_free_vm(kvm);
619 mmdrop(current->mm);
620 return ERR_PTR(r);
621}
622
623/*
624 * Avoid using vmalloc for a small buffer.
625 * Should not be used when the size is statically known.
626 */
627void *kvm_kvzalloc(unsigned long size)
628{
629 if (size > PAGE_SIZE)
630 return vzalloc(size);
631 else
632 return kzalloc(size, GFP_KERNEL);
633}
634
635static void kvm_destroy_devices(struct kvm *kvm)
636{
637 struct kvm_device *dev, *tmp;
638
639 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
640 list_del(&dev->vm_node);
641 dev->ops->destroy(dev);
642 }
643}
644
645static void kvm_destroy_vm(struct kvm *kvm)
646{
647 int i;
648 struct mm_struct *mm = kvm->mm;
649
650 kvm_arch_sync_events(kvm);
651 spin_lock(&kvm_lock);
652 list_del(&kvm->vm_list);
653 spin_unlock(&kvm_lock);
654 kvm_free_irq_routing(kvm);
655 for (i = 0; i < KVM_NR_BUSES; i++)
656 kvm_io_bus_destroy(kvm->buses[i]);
657 kvm_coalesced_mmio_free(kvm);
658#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
659 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
660#else
661 kvm_arch_flush_shadow_all(kvm);
662#endif
663 kvm_arch_destroy_vm(kvm);
664 kvm_destroy_devices(kvm);
665 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
666 kvm_free_memslots(kvm, kvm->memslots[i]);
667 cleanup_srcu_struct(&kvm->irq_srcu);
668 cleanup_srcu_struct(&kvm->srcu);
669 kvm_arch_free_vm(kvm);
670 preempt_notifier_dec();
671 hardware_disable_all();
672 mmdrop(mm);
673}
674
675void kvm_get_kvm(struct kvm *kvm)
676{
677 atomic_inc(&kvm->users_count);
678}
679EXPORT_SYMBOL_GPL(kvm_get_kvm);
680
681void kvm_put_kvm(struct kvm *kvm)
682{
683 if (atomic_dec_and_test(&kvm->users_count))
684 kvm_destroy_vm(kvm);
685}
686EXPORT_SYMBOL_GPL(kvm_put_kvm);
687
688
689static int kvm_vm_release(struct inode *inode, struct file *filp)
690{
691 struct kvm *kvm = filp->private_data;
692
693 kvm_irqfd_release(kvm);
694
695 kvm_put_kvm(kvm);
696 return 0;
697}
698
699/*
700 * Allocation size is twice as large as the actual dirty bitmap size.
701 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
702 */
703static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
704{
705 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
706
707 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
708 if (!memslot->dirty_bitmap)
709 return -ENOMEM;
710
711 return 0;
712}
713
714/*
715 * Insert memslot and re-sort memslots based on their GFN,
716 * so binary search could be used to lookup GFN.
717 * Sorting algorithm takes advantage of having initially
718 * sorted array and known changed memslot position.
719 */
720static void update_memslots(struct kvm_memslots *slots,
721 struct kvm_memory_slot *new)
722{
723 int id = new->id;
724 int i = slots->id_to_index[id];
725 struct kvm_memory_slot *mslots = slots->memslots;
726
727 WARN_ON(mslots[i].id != id);
728 if (!new->npages) {
729 WARN_ON(!mslots[i].npages);
730 if (mslots[i].npages)
731 slots->used_slots--;
732 } else {
733 if (!mslots[i].npages)
734 slots->used_slots++;
735 }
736
737 while (i < KVM_MEM_SLOTS_NUM - 1 &&
738 new->base_gfn <= mslots[i + 1].base_gfn) {
739 if (!mslots[i + 1].npages)
740 break;
741 mslots[i] = mslots[i + 1];
742 slots->id_to_index[mslots[i].id] = i;
743 i++;
744 }
745
746 /*
747 * The ">=" is needed when creating a slot with base_gfn == 0,
748 * so that it moves before all those with base_gfn == npages == 0.
749 *
750 * On the other hand, if new->npages is zero, the above loop has
751 * already left i pointing to the beginning of the empty part of
752 * mslots, and the ">=" would move the hole backwards in this
753 * case---which is wrong. So skip the loop when deleting a slot.
754 */
755 if (new->npages) {
756 while (i > 0 &&
757 new->base_gfn >= mslots[i - 1].base_gfn) {
758 mslots[i] = mslots[i - 1];
759 slots->id_to_index[mslots[i].id] = i;
760 i--;
761 }
762 } else
763 WARN_ON_ONCE(i != slots->used_slots);
764
765 mslots[i] = *new;
766 slots->id_to_index[mslots[i].id] = i;
767}
768
769static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
770{
771 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
772
773#ifdef __KVM_HAVE_READONLY_MEM
774 valid_flags |= KVM_MEM_READONLY;
775#endif
776
777 if (mem->flags & ~valid_flags)
778 return -EINVAL;
779
780 return 0;
781}
782
783static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
784 int as_id, struct kvm_memslots *slots)
785{
786 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
787
788 /*
789 * Set the low bit in the generation, which disables SPTE caching
790 * until the end of synchronize_srcu_expedited.
791 */
792 WARN_ON(old_memslots->generation & 1);
793 slots->generation = old_memslots->generation + 1;
794
795 rcu_assign_pointer(kvm->memslots[as_id], slots);
796 synchronize_srcu_expedited(&kvm->srcu);
797
798 /*
799 * Increment the new memslot generation a second time. This prevents
800 * vm exits that race with memslot updates from caching a memslot
801 * generation that will (potentially) be valid forever.
802 */
803 slots->generation++;
804
805 kvm_arch_memslots_updated(kvm, slots);
806
807 return old_memslots;
808}
809
810/*
811 * Allocate some memory and give it an address in the guest physical address
812 * space.
813 *
814 * Discontiguous memory is allowed, mostly for framebuffers.
815 *
816 * Must be called holding kvm->slots_lock for write.
817 */
818int __kvm_set_memory_region(struct kvm *kvm,
819 const struct kvm_userspace_memory_region *mem)
820{
821 int r;
822 gfn_t base_gfn;
823 unsigned long npages;
824 struct kvm_memory_slot *slot;
825 struct kvm_memory_slot old, new;
826 struct kvm_memslots *slots = NULL, *old_memslots;
827 int as_id, id;
828 enum kvm_mr_change change;
829
830 r = check_memory_region_flags(mem);
831 if (r)
832 goto out;
833
834 r = -EINVAL;
835 as_id = mem->slot >> 16;
836 id = (u16)mem->slot;
837
838 /* General sanity checks */
839 if (mem->memory_size & (PAGE_SIZE - 1))
840 goto out;
841 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
842 goto out;
843 /* We can read the guest memory with __xxx_user() later on. */
844 if ((id < KVM_USER_MEM_SLOTS) &&
845 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
846 !access_ok(VERIFY_WRITE,
847 (void __user *)(unsigned long)mem->userspace_addr,
848 mem->memory_size)))
849 goto out;
850 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
851 goto out;
852 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
853 goto out;
854
855 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
856 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
857 npages = mem->memory_size >> PAGE_SHIFT;
858
859 if (npages > KVM_MEM_MAX_NR_PAGES)
860 goto out;
861
862 new = old = *slot;
863
864 new.id = id;
865 new.base_gfn = base_gfn;
866 new.npages = npages;
867 new.flags = mem->flags;
868
869 if (npages) {
870 if (!old.npages)
871 change = KVM_MR_CREATE;
872 else { /* Modify an existing slot. */
873 if ((mem->userspace_addr != old.userspace_addr) ||
874 (npages != old.npages) ||
875 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
876 goto out;
877
878 if (base_gfn != old.base_gfn)
879 change = KVM_MR_MOVE;
880 else if (new.flags != old.flags)
881 change = KVM_MR_FLAGS_ONLY;
882 else { /* Nothing to change. */
883 r = 0;
884 goto out;
885 }
886 }
887 } else {
888 if (!old.npages)
889 goto out;
890
891 change = KVM_MR_DELETE;
892 new.base_gfn = 0;
893 new.flags = 0;
894 }
895
896 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
897 /* Check for overlaps */
898 r = -EEXIST;
899 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
900 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
901 (slot->id == id))
902 continue;
903 if (!((base_gfn + npages <= slot->base_gfn) ||
904 (base_gfn >= slot->base_gfn + slot->npages)))
905 goto out;
906 }
907 }
908
909 /* Free page dirty bitmap if unneeded */
910 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
911 new.dirty_bitmap = NULL;
912
913 r = -ENOMEM;
914 if (change == KVM_MR_CREATE) {
915 new.userspace_addr = mem->userspace_addr;
916
917 if (kvm_arch_create_memslot(kvm, &new, npages))
918 goto out_free;
919 }
920
921 /* Allocate page dirty bitmap if needed */
922 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
923 if (kvm_create_dirty_bitmap(&new) < 0)
924 goto out_free;
925 }
926
927 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
928 if (!slots)
929 goto out_free;
930 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
931
932 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
933 slot = id_to_memslot(slots, id);
934 slot->flags |= KVM_MEMSLOT_INVALID;
935
936 old_memslots = install_new_memslots(kvm, as_id, slots);
937
938 /* slot was deleted or moved, clear iommu mapping */
939 kvm_iommu_unmap_pages(kvm, &old);
940 /* From this point no new shadow pages pointing to a deleted,
941 * or moved, memslot will be created.
942 *
943 * validation of sp->gfn happens in:
944 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
945 * - kvm_is_visible_gfn (mmu_check_roots)
946 */
947 kvm_arch_flush_shadow_memslot(kvm, slot);
948
949 /*
950 * We can re-use the old_memslots from above, the only difference
951 * from the currently installed memslots is the invalid flag. This
952 * will get overwritten by update_memslots anyway.
953 */
954 slots = old_memslots;
955 }
956
957 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
958 if (r)
959 goto out_slots;
960
961 /* actual memory is freed via old in kvm_free_memslot below */
962 if (change == KVM_MR_DELETE) {
963 new.dirty_bitmap = NULL;
964 memset(&new.arch, 0, sizeof(new.arch));
965 }
966
967 update_memslots(slots, &new);
968 old_memslots = install_new_memslots(kvm, as_id, slots);
969
970 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
971
972 kvm_free_memslot(kvm, &old, &new);
973 kvfree(old_memslots);
974
975 /*
976 * IOMMU mapping: New slots need to be mapped. Old slots need to be
977 * un-mapped and re-mapped if their base changes. Since base change
978 * unmapping is handled above with slot deletion, mapping alone is
979 * needed here. Anything else the iommu might care about for existing
980 * slots (size changes, userspace addr changes and read-only flag
981 * changes) is disallowed above, so any other attribute changes getting
982 * here can be skipped.
983 */
984 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
985 r = kvm_iommu_map_pages(kvm, &new);
986 return r;
987 }
988
989 return 0;
990
991out_slots:
992 kvfree(slots);
993out_free:
994 kvm_free_memslot(kvm, &new, &old);
995out:
996 return r;
997}
998EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
999
1000int kvm_set_memory_region(struct kvm *kvm,
1001 const struct kvm_userspace_memory_region *mem)
1002{
1003 int r;
1004
1005 mutex_lock(&kvm->slots_lock);
1006 r = __kvm_set_memory_region(kvm, mem);
1007 mutex_unlock(&kvm->slots_lock);
1008 return r;
1009}
1010EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1011
1012static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1013 struct kvm_userspace_memory_region *mem)
1014{
1015 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1016 return -EINVAL;
1017
1018 return kvm_set_memory_region(kvm, mem);
1019}
1020
1021int kvm_get_dirty_log(struct kvm *kvm,
1022 struct kvm_dirty_log *log, int *is_dirty)
1023{
1024 struct kvm_memslots *slots;
1025 struct kvm_memory_slot *memslot;
1026 int r, i, as_id, id;
1027 unsigned long n;
1028 unsigned long any = 0;
1029
1030 r = -EINVAL;
1031 as_id = log->slot >> 16;
1032 id = (u16)log->slot;
1033 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1034 goto out;
1035
1036 slots = __kvm_memslots(kvm, as_id);
1037 memslot = id_to_memslot(slots, id);
1038 r = -ENOENT;
1039 if (!memslot->dirty_bitmap)
1040 goto out;
1041
1042 n = kvm_dirty_bitmap_bytes(memslot);
1043
1044 for (i = 0; !any && i < n/sizeof(long); ++i)
1045 any = memslot->dirty_bitmap[i];
1046
1047 r = -EFAULT;
1048 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1049 goto out;
1050
1051 if (any)
1052 *is_dirty = 1;
1053
1054 r = 0;
1055out:
1056 return r;
1057}
1058EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1059
1060#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1061/**
1062 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1063 * are dirty write protect them for next write.
1064 * @kvm: pointer to kvm instance
1065 * @log: slot id and address to which we copy the log
1066 * @is_dirty: flag set if any page is dirty
1067 *
1068 * We need to keep it in mind that VCPU threads can write to the bitmap
1069 * concurrently. So, to avoid losing track of dirty pages we keep the
1070 * following order:
1071 *
1072 * 1. Take a snapshot of the bit and clear it if needed.
1073 * 2. Write protect the corresponding page.
1074 * 3. Copy the snapshot to the userspace.
1075 * 4. Upon return caller flushes TLB's if needed.
1076 *
1077 * Between 2 and 4, the guest may write to the page using the remaining TLB
1078 * entry. This is not a problem because the page is reported dirty using
1079 * the snapshot taken before and step 4 ensures that writes done after
1080 * exiting to userspace will be logged for the next call.
1081 *
1082 */
1083int kvm_get_dirty_log_protect(struct kvm *kvm,
1084 struct kvm_dirty_log *log, bool *is_dirty)
1085{
1086 struct kvm_memslots *slots;
1087 struct kvm_memory_slot *memslot;
1088 int r, i, as_id, id;
1089 unsigned long n;
1090 unsigned long *dirty_bitmap;
1091 unsigned long *dirty_bitmap_buffer;
1092
1093 r = -EINVAL;
1094 as_id = log->slot >> 16;
1095 id = (u16)log->slot;
1096 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1097 goto out;
1098
1099 slots = __kvm_memslots(kvm, as_id);
1100 memslot = id_to_memslot(slots, id);
1101
1102 dirty_bitmap = memslot->dirty_bitmap;
1103 r = -ENOENT;
1104 if (!dirty_bitmap)
1105 goto out;
1106
1107 n = kvm_dirty_bitmap_bytes(memslot);
1108
1109 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1110 memset(dirty_bitmap_buffer, 0, n);
1111
1112 spin_lock(&kvm->mmu_lock);
1113 *is_dirty = false;
1114 for (i = 0; i < n / sizeof(long); i++) {
1115 unsigned long mask;
1116 gfn_t offset;
1117
1118 if (!dirty_bitmap[i])
1119 continue;
1120
1121 *is_dirty = true;
1122
1123 mask = xchg(&dirty_bitmap[i], 0);
1124 dirty_bitmap_buffer[i] = mask;
1125
1126 if (mask) {
1127 offset = i * BITS_PER_LONG;
1128 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1129 offset, mask);
1130 }
1131 }
1132
1133 spin_unlock(&kvm->mmu_lock);
1134
1135 r = -EFAULT;
1136 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1137 goto out;
1138
1139 r = 0;
1140out:
1141 return r;
1142}
1143EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1144#endif
1145
1146bool kvm_largepages_enabled(void)
1147{
1148 return largepages_enabled;
1149}
1150
1151void kvm_disable_largepages(void)
1152{
1153 largepages_enabled = false;
1154}
1155EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1156
1157struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1158{
1159 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1160}
1161EXPORT_SYMBOL_GPL(gfn_to_memslot);
1162
1163struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1164{
1165 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1166}
1167
1168bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1169{
1170 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1171
1172 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1173 memslot->flags & KVM_MEMSLOT_INVALID)
1174 return false;
1175
1176 return true;
1177}
1178EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1179
1180unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1181{
1182 struct vm_area_struct *vma;
1183 unsigned long addr, size;
1184
1185 size = PAGE_SIZE;
1186
1187 addr = gfn_to_hva(kvm, gfn);
1188 if (kvm_is_error_hva(addr))
1189 return PAGE_SIZE;
1190
1191 down_read(¤t->mm->mmap_sem);
1192 vma = find_vma(current->mm, addr);
1193 if (!vma)
1194 goto out;
1195
1196 size = vma_kernel_pagesize(vma);
1197
1198out:
1199 up_read(¤t->mm->mmap_sem);
1200
1201 return size;
1202}
1203
1204static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1205{
1206 return slot->flags & KVM_MEM_READONLY;
1207}
1208
1209static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1210 gfn_t *nr_pages, bool write)
1211{
1212 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1213 return KVM_HVA_ERR_BAD;
1214
1215 if (memslot_is_readonly(slot) && write)
1216 return KVM_HVA_ERR_RO_BAD;
1217
1218 if (nr_pages)
1219 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1220
1221 return __gfn_to_hva_memslot(slot, gfn);
1222}
1223
1224static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1225 gfn_t *nr_pages)
1226{
1227 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1228}
1229
1230unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1231 gfn_t gfn)
1232{
1233 return gfn_to_hva_many(slot, gfn, NULL);
1234}
1235EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1236
1237unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1238{
1239 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1240}
1241EXPORT_SYMBOL_GPL(gfn_to_hva);
1242
1243unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1244{
1245 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1246}
1247EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1248
1249/*
1250 * If writable is set to false, the hva returned by this function is only
1251 * allowed to be read.
1252 */
1253unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1254 gfn_t gfn, bool *writable)
1255{
1256 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1257
1258 if (!kvm_is_error_hva(hva) && writable)
1259 *writable = !memslot_is_readonly(slot);
1260
1261 return hva;
1262}
1263
1264unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1265{
1266 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1267
1268 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1269}
1270
1271unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1272{
1273 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1274
1275 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1276}
1277
1278static int get_user_page_nowait(unsigned long start, int write,
1279 struct page **page)
1280{
1281 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1282
1283 if (write)
1284 flags |= FOLL_WRITE;
1285
1286 return __get_user_pages(current, current->mm, start, 1, flags, page,
1287 NULL, NULL);
1288}
1289
1290static inline int check_user_page_hwpoison(unsigned long addr)
1291{
1292 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1293
1294 rc = __get_user_pages(current, current->mm, addr, 1,
1295 flags, NULL, NULL, NULL);
1296 return rc == -EHWPOISON;
1297}
1298
1299/*
1300 * The atomic path to get the writable pfn which will be stored in @pfn,
1301 * true indicates success, otherwise false is returned.
1302 */
1303static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1304 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1305{
1306 struct page *page[1];
1307 int npages;
1308
1309 if (!(async || atomic))
1310 return false;
1311
1312 /*
1313 * Fast pin a writable pfn only if it is a write fault request
1314 * or the caller allows to map a writable pfn for a read fault
1315 * request.
1316 */
1317 if (!(write_fault || writable))
1318 return false;
1319
1320 npages = __get_user_pages_fast(addr, 1, 1, page);
1321 if (npages == 1) {
1322 *pfn = page_to_pfn(page[0]);
1323
1324 if (writable)
1325 *writable = true;
1326 return true;
1327 }
1328
1329 return false;
1330}
1331
1332/*
1333 * The slow path to get the pfn of the specified host virtual address,
1334 * 1 indicates success, -errno is returned if error is detected.
1335 */
1336static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1337 bool *writable, kvm_pfn_t *pfn)
1338{
1339 struct page *page[1];
1340 int npages = 0;
1341
1342 might_sleep();
1343
1344 if (writable)
1345 *writable = write_fault;
1346
1347 if (async) {
1348 down_read(¤t->mm->mmap_sem);
1349 npages = get_user_page_nowait(addr, write_fault, page);
1350 up_read(¤t->mm->mmap_sem);
1351 } else
1352 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1353 write_fault, 0, page,
1354 FOLL_TOUCH|FOLL_HWPOISON);
1355 if (npages != 1)
1356 return npages;
1357
1358 /* map read fault as writable if possible */
1359 if (unlikely(!write_fault) && writable) {
1360 struct page *wpage[1];
1361
1362 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1363 if (npages == 1) {
1364 *writable = true;
1365 put_page(page[0]);
1366 page[0] = wpage[0];
1367 }
1368
1369 npages = 1;
1370 }
1371 *pfn = page_to_pfn(page[0]);
1372 return npages;
1373}
1374
1375static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1376{
1377 if (unlikely(!(vma->vm_flags & VM_READ)))
1378 return false;
1379
1380 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1381 return false;
1382
1383 return true;
1384}
1385
1386/*
1387 * Pin guest page in memory and return its pfn.
1388 * @addr: host virtual address which maps memory to the guest
1389 * @atomic: whether this function can sleep
1390 * @async: whether this function need to wait IO complete if the
1391 * host page is not in the memory
1392 * @write_fault: whether we should get a writable host page
1393 * @writable: whether it allows to map a writable host page for !@write_fault
1394 *
1395 * The function will map a writable host page for these two cases:
1396 * 1): @write_fault = true
1397 * 2): @write_fault = false && @writable, @writable will tell the caller
1398 * whether the mapping is writable.
1399 */
1400static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1401 bool write_fault, bool *writable)
1402{
1403 struct vm_area_struct *vma;
1404 kvm_pfn_t pfn = 0;
1405 int npages;
1406
1407 /* we can do it either atomically or asynchronously, not both */
1408 BUG_ON(atomic && async);
1409
1410 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1411 return pfn;
1412
1413 if (atomic)
1414 return KVM_PFN_ERR_FAULT;
1415
1416 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1417 if (npages == 1)
1418 return pfn;
1419
1420 down_read(¤t->mm->mmap_sem);
1421 if (npages == -EHWPOISON ||
1422 (!async && check_user_page_hwpoison(addr))) {
1423 pfn = KVM_PFN_ERR_HWPOISON;
1424 goto exit;
1425 }
1426
1427 vma = find_vma_intersection(current->mm, addr, addr + 1);
1428
1429 if (vma == NULL)
1430 pfn = KVM_PFN_ERR_FAULT;
1431 else if ((vma->vm_flags & VM_PFNMAP)) {
1432 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1433 vma->vm_pgoff;
1434 BUG_ON(!kvm_is_reserved_pfn(pfn));
1435 } else {
1436 if (async && vma_is_valid(vma, write_fault))
1437 *async = true;
1438 pfn = KVM_PFN_ERR_FAULT;
1439 }
1440exit:
1441 up_read(¤t->mm->mmap_sem);
1442 return pfn;
1443}
1444
1445kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1446 bool atomic, bool *async, bool write_fault,
1447 bool *writable)
1448{
1449 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1450
1451 if (addr == KVM_HVA_ERR_RO_BAD) {
1452 if (writable)
1453 *writable = false;
1454 return KVM_PFN_ERR_RO_FAULT;
1455 }
1456
1457 if (kvm_is_error_hva(addr)) {
1458 if (writable)
1459 *writable = false;
1460 return KVM_PFN_NOSLOT;
1461 }
1462
1463 /* Do not map writable pfn in the readonly memslot. */
1464 if (writable && memslot_is_readonly(slot)) {
1465 *writable = false;
1466 writable = NULL;
1467 }
1468
1469 return hva_to_pfn(addr, atomic, async, write_fault,
1470 writable);
1471}
1472EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1473
1474kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1475 bool *writable)
1476{
1477 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1478 write_fault, writable);
1479}
1480EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1481
1482kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1483{
1484 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1485}
1486EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1487
1488kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1489{
1490 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1491}
1492EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1493
1494kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1495{
1496 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1497}
1498EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1499
1500kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1501{
1502 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1503}
1504EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1505
1506kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1507{
1508 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1509}
1510EXPORT_SYMBOL_GPL(gfn_to_pfn);
1511
1512kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1513{
1514 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1515}
1516EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1517
1518int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1519 struct page **pages, int nr_pages)
1520{
1521 unsigned long addr;
1522 gfn_t entry;
1523
1524 addr = gfn_to_hva_many(slot, gfn, &entry);
1525 if (kvm_is_error_hva(addr))
1526 return -1;
1527
1528 if (entry < nr_pages)
1529 return 0;
1530
1531 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1532}
1533EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1534
1535static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1536{
1537 if (is_error_noslot_pfn(pfn))
1538 return KVM_ERR_PTR_BAD_PAGE;
1539
1540 if (kvm_is_reserved_pfn(pfn)) {
1541 WARN_ON(1);
1542 return KVM_ERR_PTR_BAD_PAGE;
1543 }
1544
1545 return pfn_to_page(pfn);
1546}
1547
1548struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1549{
1550 kvm_pfn_t pfn;
1551
1552 pfn = gfn_to_pfn(kvm, gfn);
1553
1554 return kvm_pfn_to_page(pfn);
1555}
1556EXPORT_SYMBOL_GPL(gfn_to_page);
1557
1558struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1559{
1560 kvm_pfn_t pfn;
1561
1562 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1563
1564 return kvm_pfn_to_page(pfn);
1565}
1566EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1567
1568void kvm_release_page_clean(struct page *page)
1569{
1570 WARN_ON(is_error_page(page));
1571
1572 kvm_release_pfn_clean(page_to_pfn(page));
1573}
1574EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1575
1576void kvm_release_pfn_clean(kvm_pfn_t pfn)
1577{
1578 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1579 put_page(pfn_to_page(pfn));
1580}
1581EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1582
1583void kvm_release_page_dirty(struct page *page)
1584{
1585 WARN_ON(is_error_page(page));
1586
1587 kvm_release_pfn_dirty(page_to_pfn(page));
1588}
1589EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1590
1591static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1592{
1593 kvm_set_pfn_dirty(pfn);
1594 kvm_release_pfn_clean(pfn);
1595}
1596
1597void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1598{
1599 if (!kvm_is_reserved_pfn(pfn)) {
1600 struct page *page = pfn_to_page(pfn);
1601
1602 if (!PageReserved(page))
1603 SetPageDirty(page);
1604 }
1605}
1606EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1607
1608void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1609{
1610 if (!kvm_is_reserved_pfn(pfn))
1611 mark_page_accessed(pfn_to_page(pfn));
1612}
1613EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1614
1615void kvm_get_pfn(kvm_pfn_t pfn)
1616{
1617 if (!kvm_is_reserved_pfn(pfn))
1618 get_page(pfn_to_page(pfn));
1619}
1620EXPORT_SYMBOL_GPL(kvm_get_pfn);
1621
1622static int next_segment(unsigned long len, int offset)
1623{
1624 if (len > PAGE_SIZE - offset)
1625 return PAGE_SIZE - offset;
1626 else
1627 return len;
1628}
1629
1630static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1631 void *data, int offset, int len)
1632{
1633 int r;
1634 unsigned long addr;
1635
1636 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1637 if (kvm_is_error_hva(addr))
1638 return -EFAULT;
1639 r = __copy_from_user(data, (void __user *)addr + offset, len);
1640 if (r)
1641 return -EFAULT;
1642 return 0;
1643}
1644
1645int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1646 int len)
1647{
1648 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1649
1650 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1651}
1652EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1653
1654int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1655 int offset, int len)
1656{
1657 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1658
1659 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1660}
1661EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1662
1663int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1664{
1665 gfn_t gfn = gpa >> PAGE_SHIFT;
1666 int seg;
1667 int offset = offset_in_page(gpa);
1668 int ret;
1669
1670 while ((seg = next_segment(len, offset)) != 0) {
1671 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1672 if (ret < 0)
1673 return ret;
1674 offset = 0;
1675 len -= seg;
1676 data += seg;
1677 ++gfn;
1678 }
1679 return 0;
1680}
1681EXPORT_SYMBOL_GPL(kvm_read_guest);
1682
1683int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1684{
1685 gfn_t gfn = gpa >> PAGE_SHIFT;
1686 int seg;
1687 int offset = offset_in_page(gpa);
1688 int ret;
1689
1690 while ((seg = next_segment(len, offset)) != 0) {
1691 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1692 if (ret < 0)
1693 return ret;
1694 offset = 0;
1695 len -= seg;
1696 data += seg;
1697 ++gfn;
1698 }
1699 return 0;
1700}
1701EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1702
1703static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1704 void *data, int offset, unsigned long len)
1705{
1706 int r;
1707 unsigned long addr;
1708
1709 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1710 if (kvm_is_error_hva(addr))
1711 return -EFAULT;
1712 pagefault_disable();
1713 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1714 pagefault_enable();
1715 if (r)
1716 return -EFAULT;
1717 return 0;
1718}
1719
1720int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1721 unsigned long len)
1722{
1723 gfn_t gfn = gpa >> PAGE_SHIFT;
1724 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1725 int offset = offset_in_page(gpa);
1726
1727 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1728}
1729EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1730
1731int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1732 void *data, unsigned long len)
1733{
1734 gfn_t gfn = gpa >> PAGE_SHIFT;
1735 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1736 int offset = offset_in_page(gpa);
1737
1738 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1739}
1740EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1741
1742static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1743 const void *data, int offset, int len)
1744{
1745 int r;
1746 unsigned long addr;
1747
1748 addr = gfn_to_hva_memslot(memslot, gfn);
1749 if (kvm_is_error_hva(addr))
1750 return -EFAULT;
1751 r = __copy_to_user((void __user *)addr + offset, data, len);
1752 if (r)
1753 return -EFAULT;
1754 mark_page_dirty_in_slot(memslot, gfn);
1755 return 0;
1756}
1757
1758int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1759 const void *data, int offset, int len)
1760{
1761 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1762
1763 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1764}
1765EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1766
1767int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1768 const void *data, int offset, int len)
1769{
1770 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1771
1772 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1773}
1774EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1775
1776int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1777 unsigned long len)
1778{
1779 gfn_t gfn = gpa >> PAGE_SHIFT;
1780 int seg;
1781 int offset = offset_in_page(gpa);
1782 int ret;
1783
1784 while ((seg = next_segment(len, offset)) != 0) {
1785 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1786 if (ret < 0)
1787 return ret;
1788 offset = 0;
1789 len -= seg;
1790 data += seg;
1791 ++gfn;
1792 }
1793 return 0;
1794}
1795EXPORT_SYMBOL_GPL(kvm_write_guest);
1796
1797int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1798 unsigned long len)
1799{
1800 gfn_t gfn = gpa >> PAGE_SHIFT;
1801 int seg;
1802 int offset = offset_in_page(gpa);
1803 int ret;
1804
1805 while ((seg = next_segment(len, offset)) != 0) {
1806 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1807 if (ret < 0)
1808 return ret;
1809 offset = 0;
1810 len -= seg;
1811 data += seg;
1812 ++gfn;
1813 }
1814 return 0;
1815}
1816EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1817
1818int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1819 gpa_t gpa, unsigned long len)
1820{
1821 struct kvm_memslots *slots = kvm_memslots(kvm);
1822 int offset = offset_in_page(gpa);
1823 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1824 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1825 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1826 gfn_t nr_pages_avail;
1827
1828 ghc->gpa = gpa;
1829 ghc->generation = slots->generation;
1830 ghc->len = len;
1831 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1832 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1833 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1834 ghc->hva += offset;
1835 } else {
1836 /*
1837 * If the requested region crosses two memslots, we still
1838 * verify that the entire region is valid here.
1839 */
1840 while (start_gfn <= end_gfn) {
1841 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1842 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1843 &nr_pages_avail);
1844 if (kvm_is_error_hva(ghc->hva))
1845 return -EFAULT;
1846 start_gfn += nr_pages_avail;
1847 }
1848 /* Use the slow path for cross page reads and writes. */
1849 ghc->memslot = NULL;
1850 }
1851 return 0;
1852}
1853EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1854
1855int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1856 void *data, unsigned long len)
1857{
1858 struct kvm_memslots *slots = kvm_memslots(kvm);
1859 int r;
1860
1861 BUG_ON(len > ghc->len);
1862
1863 if (slots->generation != ghc->generation)
1864 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1865
1866 if (unlikely(!ghc->memslot))
1867 return kvm_write_guest(kvm, ghc->gpa, data, len);
1868
1869 if (kvm_is_error_hva(ghc->hva))
1870 return -EFAULT;
1871
1872 r = __copy_to_user((void __user *)ghc->hva, data, len);
1873 if (r)
1874 return -EFAULT;
1875 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1876
1877 return 0;
1878}
1879EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1880
1881int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1882 void *data, unsigned long len)
1883{
1884 struct kvm_memslots *slots = kvm_memslots(kvm);
1885 int r;
1886
1887 BUG_ON(len > ghc->len);
1888
1889 if (slots->generation != ghc->generation)
1890 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1891
1892 if (unlikely(!ghc->memslot))
1893 return kvm_read_guest(kvm, ghc->gpa, data, len);
1894
1895 if (kvm_is_error_hva(ghc->hva))
1896 return -EFAULT;
1897
1898 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1899 if (r)
1900 return -EFAULT;
1901
1902 return 0;
1903}
1904EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1905
1906int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1907{
1908 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1909
1910 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1911}
1912EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1913
1914int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1915{
1916 gfn_t gfn = gpa >> PAGE_SHIFT;
1917 int seg;
1918 int offset = offset_in_page(gpa);
1919 int ret;
1920
1921 while ((seg = next_segment(len, offset)) != 0) {
1922 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1923 if (ret < 0)
1924 return ret;
1925 offset = 0;
1926 len -= seg;
1927 ++gfn;
1928 }
1929 return 0;
1930}
1931EXPORT_SYMBOL_GPL(kvm_clear_guest);
1932
1933static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1934 gfn_t gfn)
1935{
1936 if (memslot && memslot->dirty_bitmap) {
1937 unsigned long rel_gfn = gfn - memslot->base_gfn;
1938
1939 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1940 }
1941}
1942
1943void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1944{
1945 struct kvm_memory_slot *memslot;
1946
1947 memslot = gfn_to_memslot(kvm, gfn);
1948 mark_page_dirty_in_slot(memslot, gfn);
1949}
1950EXPORT_SYMBOL_GPL(mark_page_dirty);
1951
1952void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1953{
1954 struct kvm_memory_slot *memslot;
1955
1956 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1957 mark_page_dirty_in_slot(memslot, gfn);
1958}
1959EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1960
1961static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1962{
1963 unsigned int old, val, grow;
1964
1965 old = val = vcpu->halt_poll_ns;
1966 grow = READ_ONCE(halt_poll_ns_grow);
1967 /* 10us base */
1968 if (val == 0 && grow)
1969 val = 10000;
1970 else
1971 val *= grow;
1972
1973 if (val > halt_poll_ns)
1974 val = halt_poll_ns;
1975
1976 vcpu->halt_poll_ns = val;
1977 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1978}
1979
1980static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1981{
1982 unsigned int old, val, shrink;
1983
1984 old = val = vcpu->halt_poll_ns;
1985 shrink = READ_ONCE(halt_poll_ns_shrink);
1986 if (shrink == 0)
1987 val = 0;
1988 else
1989 val /= shrink;
1990
1991 vcpu->halt_poll_ns = val;
1992 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1993}
1994
1995static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1996{
1997 if (kvm_arch_vcpu_runnable(vcpu)) {
1998 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1999 return -EINTR;
2000 }
2001 if (kvm_cpu_has_pending_timer(vcpu))
2002 return -EINTR;
2003 if (signal_pending(current))
2004 return -EINTR;
2005
2006 return 0;
2007}
2008
2009/*
2010 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2011 */
2012void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2013{
2014 ktime_t start, cur;
2015 DECLARE_SWAITQUEUE(wait);
2016 bool waited = false;
2017 u64 block_ns;
2018
2019 start = cur = ktime_get();
2020 if (vcpu->halt_poll_ns) {
2021 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2022
2023 ++vcpu->stat.halt_attempted_poll;
2024 do {
2025 /*
2026 * This sets KVM_REQ_UNHALT if an interrupt
2027 * arrives.
2028 */
2029 if (kvm_vcpu_check_block(vcpu) < 0) {
2030 ++vcpu->stat.halt_successful_poll;
2031 goto out;
2032 }
2033 cur = ktime_get();
2034 } while (single_task_running() && ktime_before(cur, stop));
2035 }
2036
2037 kvm_arch_vcpu_blocking(vcpu);
2038
2039 for (;;) {
2040 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2041
2042 if (kvm_vcpu_check_block(vcpu) < 0)
2043 break;
2044
2045 waited = true;
2046 schedule();
2047 }
2048
2049 finish_swait(&vcpu->wq, &wait);
2050 cur = ktime_get();
2051
2052 kvm_arch_vcpu_unblocking(vcpu);
2053out:
2054 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2055
2056 if (halt_poll_ns) {
2057 if (block_ns <= vcpu->halt_poll_ns)
2058 ;
2059 /* we had a long block, shrink polling */
2060 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2061 shrink_halt_poll_ns(vcpu);
2062 /* we had a short halt and our poll time is too small */
2063 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2064 block_ns < halt_poll_ns)
2065 grow_halt_poll_ns(vcpu);
2066 } else
2067 vcpu->halt_poll_ns = 0;
2068
2069 trace_kvm_vcpu_wakeup(block_ns, waited);
2070}
2071EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2072
2073#ifndef CONFIG_S390
2074/*
2075 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2076 */
2077void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2078{
2079 int me;
2080 int cpu = vcpu->cpu;
2081 struct swait_queue_head *wqp;
2082
2083 wqp = kvm_arch_vcpu_wq(vcpu);
2084 if (swait_active(wqp)) {
2085 swake_up(wqp);
2086 ++vcpu->stat.halt_wakeup;
2087 }
2088
2089 me = get_cpu();
2090 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2091 if (kvm_arch_vcpu_should_kick(vcpu))
2092 smp_send_reschedule(cpu);
2093 put_cpu();
2094}
2095EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2096#endif /* !CONFIG_S390 */
2097
2098int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2099{
2100 struct pid *pid;
2101 struct task_struct *task = NULL;
2102 int ret = 0;
2103
2104 rcu_read_lock();
2105 pid = rcu_dereference(target->pid);
2106 if (pid)
2107 task = get_pid_task(pid, PIDTYPE_PID);
2108 rcu_read_unlock();
2109 if (!task)
2110 return ret;
2111 ret = yield_to(task, 1);
2112 put_task_struct(task);
2113
2114 return ret;
2115}
2116EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2117
2118/*
2119 * Helper that checks whether a VCPU is eligible for directed yield.
2120 * Most eligible candidate to yield is decided by following heuristics:
2121 *
2122 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2123 * (preempted lock holder), indicated by @in_spin_loop.
2124 * Set at the beiginning and cleared at the end of interception/PLE handler.
2125 *
2126 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2127 * chance last time (mostly it has become eligible now since we have probably
2128 * yielded to lockholder in last iteration. This is done by toggling
2129 * @dy_eligible each time a VCPU checked for eligibility.)
2130 *
2131 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2132 * to preempted lock-holder could result in wrong VCPU selection and CPU
2133 * burning. Giving priority for a potential lock-holder increases lock
2134 * progress.
2135 *
2136 * Since algorithm is based on heuristics, accessing another VCPU data without
2137 * locking does not harm. It may result in trying to yield to same VCPU, fail
2138 * and continue with next VCPU and so on.
2139 */
2140static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2141{
2142#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2143 bool eligible;
2144
2145 eligible = !vcpu->spin_loop.in_spin_loop ||
2146 vcpu->spin_loop.dy_eligible;
2147
2148 if (vcpu->spin_loop.in_spin_loop)
2149 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2150
2151 return eligible;
2152#else
2153 return true;
2154#endif
2155}
2156
2157void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2158{
2159 struct kvm *kvm = me->kvm;
2160 struct kvm_vcpu *vcpu;
2161 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2162 int yielded = 0;
2163 int try = 3;
2164 int pass;
2165 int i;
2166
2167 kvm_vcpu_set_in_spin_loop(me, true);
2168 /*
2169 * We boost the priority of a VCPU that is runnable but not
2170 * currently running, because it got preempted by something
2171 * else and called schedule in __vcpu_run. Hopefully that
2172 * VCPU is holding the lock that we need and will release it.
2173 * We approximate round-robin by starting at the last boosted VCPU.
2174 */
2175 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2176 kvm_for_each_vcpu(i, vcpu, kvm) {
2177 if (!pass && i <= last_boosted_vcpu) {
2178 i = last_boosted_vcpu;
2179 continue;
2180 } else if (pass && i > last_boosted_vcpu)
2181 break;
2182 if (!ACCESS_ONCE(vcpu->preempted))
2183 continue;
2184 if (vcpu == me)
2185 continue;
2186 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2187 continue;
2188 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2189 continue;
2190
2191 yielded = kvm_vcpu_yield_to(vcpu);
2192 if (yielded > 0) {
2193 kvm->last_boosted_vcpu = i;
2194 break;
2195 } else if (yielded < 0) {
2196 try--;
2197 if (!try)
2198 break;
2199 }
2200 }
2201 }
2202 kvm_vcpu_set_in_spin_loop(me, false);
2203
2204 /* Ensure vcpu is not eligible during next spinloop */
2205 kvm_vcpu_set_dy_eligible(me, false);
2206}
2207EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2208
2209static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2210{
2211 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2212 struct page *page;
2213
2214 if (vmf->pgoff == 0)
2215 page = virt_to_page(vcpu->run);
2216#ifdef CONFIG_X86
2217 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2218 page = virt_to_page(vcpu->arch.pio_data);
2219#endif
2220#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2221 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2222 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2223#endif
2224 else
2225 return kvm_arch_vcpu_fault(vcpu, vmf);
2226 get_page(page);
2227 vmf->page = page;
2228 return 0;
2229}
2230
2231static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2232 .fault = kvm_vcpu_fault,
2233};
2234
2235static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2236{
2237 vma->vm_ops = &kvm_vcpu_vm_ops;
2238 return 0;
2239}
2240
2241static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2242{
2243 struct kvm_vcpu *vcpu = filp->private_data;
2244
2245 kvm_put_kvm(vcpu->kvm);
2246 return 0;
2247}
2248
2249static struct file_operations kvm_vcpu_fops = {
2250 .release = kvm_vcpu_release,
2251 .unlocked_ioctl = kvm_vcpu_ioctl,
2252#ifdef CONFIG_KVM_COMPAT
2253 .compat_ioctl = kvm_vcpu_compat_ioctl,
2254#endif
2255 .mmap = kvm_vcpu_mmap,
2256 .llseek = noop_llseek,
2257};
2258
2259/*
2260 * Allocates an inode for the vcpu.
2261 */
2262static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2263{
2264 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2265}
2266
2267/*
2268 * Creates some virtual cpus. Good luck creating more than one.
2269 */
2270static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2271{
2272 int r;
2273 struct kvm_vcpu *vcpu;
2274
2275 if (id >= KVM_MAX_VCPUS)
2276 return -EINVAL;
2277
2278 vcpu = kvm_arch_vcpu_create(kvm, id);
2279 if (IS_ERR(vcpu))
2280 return PTR_ERR(vcpu);
2281
2282 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2283
2284 r = kvm_arch_vcpu_setup(vcpu);
2285 if (r)
2286 goto vcpu_destroy;
2287
2288 mutex_lock(&kvm->lock);
2289 if (!kvm_vcpu_compatible(vcpu)) {
2290 r = -EINVAL;
2291 goto unlock_vcpu_destroy;
2292 }
2293 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2294 r = -EINVAL;
2295 goto unlock_vcpu_destroy;
2296 }
2297 if (kvm_get_vcpu_by_id(kvm, id)) {
2298 r = -EEXIST;
2299 goto unlock_vcpu_destroy;
2300 }
2301
2302 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2303
2304 /* Now it's all set up, let userspace reach it */
2305 kvm_get_kvm(kvm);
2306 r = create_vcpu_fd(vcpu);
2307 if (r < 0) {
2308 kvm_put_kvm(kvm);
2309 goto unlock_vcpu_destroy;
2310 }
2311
2312 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2313
2314 /*
2315 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2316 * before kvm->online_vcpu's incremented value.
2317 */
2318 smp_wmb();
2319 atomic_inc(&kvm->online_vcpus);
2320
2321 mutex_unlock(&kvm->lock);
2322 kvm_arch_vcpu_postcreate(vcpu);
2323 return r;
2324
2325unlock_vcpu_destroy:
2326 mutex_unlock(&kvm->lock);
2327vcpu_destroy:
2328 kvm_arch_vcpu_destroy(vcpu);
2329 return r;
2330}
2331
2332static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2333{
2334 if (sigset) {
2335 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2336 vcpu->sigset_active = 1;
2337 vcpu->sigset = *sigset;
2338 } else
2339 vcpu->sigset_active = 0;
2340 return 0;
2341}
2342
2343static long kvm_vcpu_ioctl(struct file *filp,
2344 unsigned int ioctl, unsigned long arg)
2345{
2346 struct kvm_vcpu *vcpu = filp->private_data;
2347 void __user *argp = (void __user *)arg;
2348 int r;
2349 struct kvm_fpu *fpu = NULL;
2350 struct kvm_sregs *kvm_sregs = NULL;
2351
2352 if (vcpu->kvm->mm != current->mm)
2353 return -EIO;
2354
2355 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2356 return -EINVAL;
2357
2358#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2359 /*
2360 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2361 * so vcpu_load() would break it.
2362 */
2363 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2364 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2365#endif
2366
2367
2368 r = vcpu_load(vcpu);
2369 if (r)
2370 return r;
2371 switch (ioctl) {
2372 case KVM_RUN:
2373 r = -EINVAL;
2374 if (arg)
2375 goto out;
2376 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2377 /* The thread running this VCPU changed. */
2378 struct pid *oldpid = vcpu->pid;
2379 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2380
2381 rcu_assign_pointer(vcpu->pid, newpid);
2382 if (oldpid)
2383 synchronize_rcu();
2384 put_pid(oldpid);
2385 }
2386 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2387 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2388 break;
2389 case KVM_GET_REGS: {
2390 struct kvm_regs *kvm_regs;
2391
2392 r = -ENOMEM;
2393 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2394 if (!kvm_regs)
2395 goto out;
2396 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2397 if (r)
2398 goto out_free1;
2399 r = -EFAULT;
2400 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2401 goto out_free1;
2402 r = 0;
2403out_free1:
2404 kfree(kvm_regs);
2405 break;
2406 }
2407 case KVM_SET_REGS: {
2408 struct kvm_regs *kvm_regs;
2409
2410 r = -ENOMEM;
2411 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2412 if (IS_ERR(kvm_regs)) {
2413 r = PTR_ERR(kvm_regs);
2414 goto out;
2415 }
2416 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2417 kfree(kvm_regs);
2418 break;
2419 }
2420 case KVM_GET_SREGS: {
2421 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2422 r = -ENOMEM;
2423 if (!kvm_sregs)
2424 goto out;
2425 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2426 if (r)
2427 goto out;
2428 r = -EFAULT;
2429 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2430 goto out;
2431 r = 0;
2432 break;
2433 }
2434 case KVM_SET_SREGS: {
2435 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2436 if (IS_ERR(kvm_sregs)) {
2437 r = PTR_ERR(kvm_sregs);
2438 kvm_sregs = NULL;
2439 goto out;
2440 }
2441 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2442 break;
2443 }
2444 case KVM_GET_MP_STATE: {
2445 struct kvm_mp_state mp_state;
2446
2447 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2448 if (r)
2449 goto out;
2450 r = -EFAULT;
2451 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2452 goto out;
2453 r = 0;
2454 break;
2455 }
2456 case KVM_SET_MP_STATE: {
2457 struct kvm_mp_state mp_state;
2458
2459 r = -EFAULT;
2460 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2461 goto out;
2462 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2463 break;
2464 }
2465 case KVM_TRANSLATE: {
2466 struct kvm_translation tr;
2467
2468 r = -EFAULT;
2469 if (copy_from_user(&tr, argp, sizeof(tr)))
2470 goto out;
2471 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2472 if (r)
2473 goto out;
2474 r = -EFAULT;
2475 if (copy_to_user(argp, &tr, sizeof(tr)))
2476 goto out;
2477 r = 0;
2478 break;
2479 }
2480 case KVM_SET_GUEST_DEBUG: {
2481 struct kvm_guest_debug dbg;
2482
2483 r = -EFAULT;
2484 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2485 goto out;
2486 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2487 break;
2488 }
2489 case KVM_SET_SIGNAL_MASK: {
2490 struct kvm_signal_mask __user *sigmask_arg = argp;
2491 struct kvm_signal_mask kvm_sigmask;
2492 sigset_t sigset, *p;
2493
2494 p = NULL;
2495 if (argp) {
2496 r = -EFAULT;
2497 if (copy_from_user(&kvm_sigmask, argp,
2498 sizeof(kvm_sigmask)))
2499 goto out;
2500 r = -EINVAL;
2501 if (kvm_sigmask.len != sizeof(sigset))
2502 goto out;
2503 r = -EFAULT;
2504 if (copy_from_user(&sigset, sigmask_arg->sigset,
2505 sizeof(sigset)))
2506 goto out;
2507 p = &sigset;
2508 }
2509 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2510 break;
2511 }
2512 case KVM_GET_FPU: {
2513 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2514 r = -ENOMEM;
2515 if (!fpu)
2516 goto out;
2517 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2518 if (r)
2519 goto out;
2520 r = -EFAULT;
2521 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2522 goto out;
2523 r = 0;
2524 break;
2525 }
2526 case KVM_SET_FPU: {
2527 fpu = memdup_user(argp, sizeof(*fpu));
2528 if (IS_ERR(fpu)) {
2529 r = PTR_ERR(fpu);
2530 fpu = NULL;
2531 goto out;
2532 }
2533 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2534 break;
2535 }
2536 default:
2537 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2538 }
2539out:
2540 vcpu_put(vcpu);
2541 kfree(fpu);
2542 kfree(kvm_sregs);
2543 return r;
2544}
2545
2546#ifdef CONFIG_KVM_COMPAT
2547static long kvm_vcpu_compat_ioctl(struct file *filp,
2548 unsigned int ioctl, unsigned long arg)
2549{
2550 struct kvm_vcpu *vcpu = filp->private_data;
2551 void __user *argp = compat_ptr(arg);
2552 int r;
2553
2554 if (vcpu->kvm->mm != current->mm)
2555 return -EIO;
2556
2557 switch (ioctl) {
2558 case KVM_SET_SIGNAL_MASK: {
2559 struct kvm_signal_mask __user *sigmask_arg = argp;
2560 struct kvm_signal_mask kvm_sigmask;
2561 compat_sigset_t csigset;
2562 sigset_t sigset;
2563
2564 if (argp) {
2565 r = -EFAULT;
2566 if (copy_from_user(&kvm_sigmask, argp,
2567 sizeof(kvm_sigmask)))
2568 goto out;
2569 r = -EINVAL;
2570 if (kvm_sigmask.len != sizeof(csigset))
2571 goto out;
2572 r = -EFAULT;
2573 if (copy_from_user(&csigset, sigmask_arg->sigset,
2574 sizeof(csigset)))
2575 goto out;
2576 sigset_from_compat(&sigset, &csigset);
2577 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2578 } else
2579 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2580 break;
2581 }
2582 default:
2583 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2584 }
2585
2586out:
2587 return r;
2588}
2589#endif
2590
2591static int kvm_device_ioctl_attr(struct kvm_device *dev,
2592 int (*accessor)(struct kvm_device *dev,
2593 struct kvm_device_attr *attr),
2594 unsigned long arg)
2595{
2596 struct kvm_device_attr attr;
2597
2598 if (!accessor)
2599 return -EPERM;
2600
2601 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2602 return -EFAULT;
2603
2604 return accessor(dev, &attr);
2605}
2606
2607static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2608 unsigned long arg)
2609{
2610 struct kvm_device *dev = filp->private_data;
2611
2612 switch (ioctl) {
2613 case KVM_SET_DEVICE_ATTR:
2614 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2615 case KVM_GET_DEVICE_ATTR:
2616 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2617 case KVM_HAS_DEVICE_ATTR:
2618 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2619 default:
2620 if (dev->ops->ioctl)
2621 return dev->ops->ioctl(dev, ioctl, arg);
2622
2623 return -ENOTTY;
2624 }
2625}
2626
2627static int kvm_device_release(struct inode *inode, struct file *filp)
2628{
2629 struct kvm_device *dev = filp->private_data;
2630 struct kvm *kvm = dev->kvm;
2631
2632 kvm_put_kvm(kvm);
2633 return 0;
2634}
2635
2636static const struct file_operations kvm_device_fops = {
2637 .unlocked_ioctl = kvm_device_ioctl,
2638#ifdef CONFIG_KVM_COMPAT
2639 .compat_ioctl = kvm_device_ioctl,
2640#endif
2641 .release = kvm_device_release,
2642};
2643
2644struct kvm_device *kvm_device_from_filp(struct file *filp)
2645{
2646 if (filp->f_op != &kvm_device_fops)
2647 return NULL;
2648
2649 return filp->private_data;
2650}
2651
2652static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2653#ifdef CONFIG_KVM_MPIC
2654 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2655 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2656#endif
2657
2658#ifdef CONFIG_KVM_XICS
2659 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2660#endif
2661};
2662
2663int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2664{
2665 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2666 return -ENOSPC;
2667
2668 if (kvm_device_ops_table[type] != NULL)
2669 return -EEXIST;
2670
2671 kvm_device_ops_table[type] = ops;
2672 return 0;
2673}
2674
2675void kvm_unregister_device_ops(u32 type)
2676{
2677 if (kvm_device_ops_table[type] != NULL)
2678 kvm_device_ops_table[type] = NULL;
2679}
2680
2681static int kvm_ioctl_create_device(struct kvm *kvm,
2682 struct kvm_create_device *cd)
2683{
2684 struct kvm_device_ops *ops = NULL;
2685 struct kvm_device *dev;
2686 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2687 int ret;
2688
2689 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2690 return -ENODEV;
2691
2692 ops = kvm_device_ops_table[cd->type];
2693 if (ops == NULL)
2694 return -ENODEV;
2695
2696 if (test)
2697 return 0;
2698
2699 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2700 if (!dev)
2701 return -ENOMEM;
2702
2703 dev->ops = ops;
2704 dev->kvm = kvm;
2705
2706 ret = ops->create(dev, cd->type);
2707 if (ret < 0) {
2708 kfree(dev);
2709 return ret;
2710 }
2711
2712 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2713 if (ret < 0) {
2714 ops->destroy(dev);
2715 return ret;
2716 }
2717
2718 list_add(&dev->vm_node, &kvm->devices);
2719 kvm_get_kvm(kvm);
2720 cd->fd = ret;
2721 return 0;
2722}
2723
2724static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2725{
2726 switch (arg) {
2727 case KVM_CAP_USER_MEMORY:
2728 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2729 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2730 case KVM_CAP_INTERNAL_ERROR_DATA:
2731#ifdef CONFIG_HAVE_KVM_MSI
2732 case KVM_CAP_SIGNAL_MSI:
2733#endif
2734#ifdef CONFIG_HAVE_KVM_IRQFD
2735 case KVM_CAP_IRQFD:
2736 case KVM_CAP_IRQFD_RESAMPLE:
2737#endif
2738 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2739 case KVM_CAP_CHECK_EXTENSION_VM:
2740 return 1;
2741#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2742 case KVM_CAP_IRQ_ROUTING:
2743 return KVM_MAX_IRQ_ROUTES;
2744#endif
2745#if KVM_ADDRESS_SPACE_NUM > 1
2746 case KVM_CAP_MULTI_ADDRESS_SPACE:
2747 return KVM_ADDRESS_SPACE_NUM;
2748#endif
2749 default:
2750 break;
2751 }
2752 return kvm_vm_ioctl_check_extension(kvm, arg);
2753}
2754
2755static long kvm_vm_ioctl(struct file *filp,
2756 unsigned int ioctl, unsigned long arg)
2757{
2758 struct kvm *kvm = filp->private_data;
2759 void __user *argp = (void __user *)arg;
2760 int r;
2761
2762 if (kvm->mm != current->mm)
2763 return -EIO;
2764 switch (ioctl) {
2765 case KVM_CREATE_VCPU:
2766 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2767 break;
2768 case KVM_SET_USER_MEMORY_REGION: {
2769 struct kvm_userspace_memory_region kvm_userspace_mem;
2770
2771 r = -EFAULT;
2772 if (copy_from_user(&kvm_userspace_mem, argp,
2773 sizeof(kvm_userspace_mem)))
2774 goto out;
2775
2776 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2777 break;
2778 }
2779 case KVM_GET_DIRTY_LOG: {
2780 struct kvm_dirty_log log;
2781
2782 r = -EFAULT;
2783 if (copy_from_user(&log, argp, sizeof(log)))
2784 goto out;
2785 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2786 break;
2787 }
2788#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2789 case KVM_REGISTER_COALESCED_MMIO: {
2790 struct kvm_coalesced_mmio_zone zone;
2791
2792 r = -EFAULT;
2793 if (copy_from_user(&zone, argp, sizeof(zone)))
2794 goto out;
2795 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2796 break;
2797 }
2798 case KVM_UNREGISTER_COALESCED_MMIO: {
2799 struct kvm_coalesced_mmio_zone zone;
2800
2801 r = -EFAULT;
2802 if (copy_from_user(&zone, argp, sizeof(zone)))
2803 goto out;
2804 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2805 break;
2806 }
2807#endif
2808 case KVM_IRQFD: {
2809 struct kvm_irqfd data;
2810
2811 r = -EFAULT;
2812 if (copy_from_user(&data, argp, sizeof(data)))
2813 goto out;
2814 r = kvm_irqfd(kvm, &data);
2815 break;
2816 }
2817 case KVM_IOEVENTFD: {
2818 struct kvm_ioeventfd data;
2819
2820 r = -EFAULT;
2821 if (copy_from_user(&data, argp, sizeof(data)))
2822 goto out;
2823 r = kvm_ioeventfd(kvm, &data);
2824 break;
2825 }
2826#ifdef CONFIG_HAVE_KVM_MSI
2827 case KVM_SIGNAL_MSI: {
2828 struct kvm_msi msi;
2829
2830 r = -EFAULT;
2831 if (copy_from_user(&msi, argp, sizeof(msi)))
2832 goto out;
2833 r = kvm_send_userspace_msi(kvm, &msi);
2834 break;
2835 }
2836#endif
2837#ifdef __KVM_HAVE_IRQ_LINE
2838 case KVM_IRQ_LINE_STATUS:
2839 case KVM_IRQ_LINE: {
2840 struct kvm_irq_level irq_event;
2841
2842 r = -EFAULT;
2843 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2844 goto out;
2845
2846 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2847 ioctl == KVM_IRQ_LINE_STATUS);
2848 if (r)
2849 goto out;
2850
2851 r = -EFAULT;
2852 if (ioctl == KVM_IRQ_LINE_STATUS) {
2853 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2854 goto out;
2855 }
2856
2857 r = 0;
2858 break;
2859 }
2860#endif
2861#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2862 case KVM_SET_GSI_ROUTING: {
2863 struct kvm_irq_routing routing;
2864 struct kvm_irq_routing __user *urouting;
2865 struct kvm_irq_routing_entry *entries;
2866
2867 r = -EFAULT;
2868 if (copy_from_user(&routing, argp, sizeof(routing)))
2869 goto out;
2870 r = -EINVAL;
2871 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2872 goto out;
2873 if (routing.flags)
2874 goto out;
2875 r = -ENOMEM;
2876 entries = vmalloc(routing.nr * sizeof(*entries));
2877 if (!entries)
2878 goto out;
2879 r = -EFAULT;
2880 urouting = argp;
2881 if (copy_from_user(entries, urouting->entries,
2882 routing.nr * sizeof(*entries)))
2883 goto out_free_irq_routing;
2884 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2885 routing.flags);
2886out_free_irq_routing:
2887 vfree(entries);
2888 break;
2889 }
2890#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2891 case KVM_CREATE_DEVICE: {
2892 struct kvm_create_device cd;
2893
2894 r = -EFAULT;
2895 if (copy_from_user(&cd, argp, sizeof(cd)))
2896 goto out;
2897
2898 r = kvm_ioctl_create_device(kvm, &cd);
2899 if (r)
2900 goto out;
2901
2902 r = -EFAULT;
2903 if (copy_to_user(argp, &cd, sizeof(cd)))
2904 goto out;
2905
2906 r = 0;
2907 break;
2908 }
2909 case KVM_CHECK_EXTENSION:
2910 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2911 break;
2912 default:
2913 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2914 }
2915out:
2916 return r;
2917}
2918
2919#ifdef CONFIG_KVM_COMPAT
2920struct compat_kvm_dirty_log {
2921 __u32 slot;
2922 __u32 padding1;
2923 union {
2924 compat_uptr_t dirty_bitmap; /* one bit per page */
2925 __u64 padding2;
2926 };
2927};
2928
2929static long kvm_vm_compat_ioctl(struct file *filp,
2930 unsigned int ioctl, unsigned long arg)
2931{
2932 struct kvm *kvm = filp->private_data;
2933 int r;
2934
2935 if (kvm->mm != current->mm)
2936 return -EIO;
2937 switch (ioctl) {
2938 case KVM_GET_DIRTY_LOG: {
2939 struct compat_kvm_dirty_log compat_log;
2940 struct kvm_dirty_log log;
2941
2942 r = -EFAULT;
2943 if (copy_from_user(&compat_log, (void __user *)arg,
2944 sizeof(compat_log)))
2945 goto out;
2946 log.slot = compat_log.slot;
2947 log.padding1 = compat_log.padding1;
2948 log.padding2 = compat_log.padding2;
2949 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2950
2951 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2952 break;
2953 }
2954 default:
2955 r = kvm_vm_ioctl(filp, ioctl, arg);
2956 }
2957
2958out:
2959 return r;
2960}
2961#endif
2962
2963static struct file_operations kvm_vm_fops = {
2964 .release = kvm_vm_release,
2965 .unlocked_ioctl = kvm_vm_ioctl,
2966#ifdef CONFIG_KVM_COMPAT
2967 .compat_ioctl = kvm_vm_compat_ioctl,
2968#endif
2969 .llseek = noop_llseek,
2970};
2971
2972static int kvm_dev_ioctl_create_vm(unsigned long type)
2973{
2974 int r;
2975 struct kvm *kvm;
2976
2977 kvm = kvm_create_vm(type);
2978 if (IS_ERR(kvm))
2979 return PTR_ERR(kvm);
2980#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2981 r = kvm_coalesced_mmio_init(kvm);
2982 if (r < 0) {
2983 kvm_put_kvm(kvm);
2984 return r;
2985 }
2986#endif
2987 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2988 if (r < 0)
2989 kvm_put_kvm(kvm);
2990
2991 return r;
2992}
2993
2994static long kvm_dev_ioctl(struct file *filp,
2995 unsigned int ioctl, unsigned long arg)
2996{
2997 long r = -EINVAL;
2998
2999 switch (ioctl) {
3000 case KVM_GET_API_VERSION:
3001 if (arg)
3002 goto out;
3003 r = KVM_API_VERSION;
3004 break;
3005 case KVM_CREATE_VM:
3006 r = kvm_dev_ioctl_create_vm(arg);
3007 break;
3008 case KVM_CHECK_EXTENSION:
3009 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3010 break;
3011 case KVM_GET_VCPU_MMAP_SIZE:
3012 if (arg)
3013 goto out;
3014 r = PAGE_SIZE; /* struct kvm_run */
3015#ifdef CONFIG_X86
3016 r += PAGE_SIZE; /* pio data page */
3017#endif
3018#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3019 r += PAGE_SIZE; /* coalesced mmio ring page */
3020#endif
3021 break;
3022 case KVM_TRACE_ENABLE:
3023 case KVM_TRACE_PAUSE:
3024 case KVM_TRACE_DISABLE:
3025 r = -EOPNOTSUPP;
3026 break;
3027 default:
3028 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3029 }
3030out:
3031 return r;
3032}
3033
3034static struct file_operations kvm_chardev_ops = {
3035 .unlocked_ioctl = kvm_dev_ioctl,
3036 .compat_ioctl = kvm_dev_ioctl,
3037 .llseek = noop_llseek,
3038};
3039
3040static struct miscdevice kvm_dev = {
3041 KVM_MINOR,
3042 "kvm",
3043 &kvm_chardev_ops,
3044};
3045
3046static void hardware_enable_nolock(void *junk)
3047{
3048 int cpu = raw_smp_processor_id();
3049 int r;
3050
3051 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3052 return;
3053
3054 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3055
3056 r = kvm_arch_hardware_enable();
3057
3058 if (r) {
3059 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3060 atomic_inc(&hardware_enable_failed);
3061 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3062 }
3063}
3064
3065static void hardware_enable(void)
3066{
3067 raw_spin_lock(&kvm_count_lock);
3068 if (kvm_usage_count)
3069 hardware_enable_nolock(NULL);
3070 raw_spin_unlock(&kvm_count_lock);
3071}
3072
3073static void hardware_disable_nolock(void *junk)
3074{
3075 int cpu = raw_smp_processor_id();
3076
3077 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3078 return;
3079 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3080 kvm_arch_hardware_disable();
3081}
3082
3083static void hardware_disable(void)
3084{
3085 raw_spin_lock(&kvm_count_lock);
3086 if (kvm_usage_count)
3087 hardware_disable_nolock(NULL);
3088 raw_spin_unlock(&kvm_count_lock);
3089}
3090
3091static void hardware_disable_all_nolock(void)
3092{
3093 BUG_ON(!kvm_usage_count);
3094
3095 kvm_usage_count--;
3096 if (!kvm_usage_count)
3097 on_each_cpu(hardware_disable_nolock, NULL, 1);
3098}
3099
3100static void hardware_disable_all(void)
3101{
3102 raw_spin_lock(&kvm_count_lock);
3103 hardware_disable_all_nolock();
3104 raw_spin_unlock(&kvm_count_lock);
3105}
3106
3107static int hardware_enable_all(void)
3108{
3109 int r = 0;
3110
3111 raw_spin_lock(&kvm_count_lock);
3112
3113 kvm_usage_count++;
3114 if (kvm_usage_count == 1) {
3115 atomic_set(&hardware_enable_failed, 0);
3116 on_each_cpu(hardware_enable_nolock, NULL, 1);
3117
3118 if (atomic_read(&hardware_enable_failed)) {
3119 hardware_disable_all_nolock();
3120 r = -EBUSY;
3121 }
3122 }
3123
3124 raw_spin_unlock(&kvm_count_lock);
3125
3126 return r;
3127}
3128
3129static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3130 void *v)
3131{
3132 val &= ~CPU_TASKS_FROZEN;
3133 switch (val) {
3134 case CPU_DYING:
3135 hardware_disable();
3136 break;
3137 case CPU_STARTING:
3138 hardware_enable();
3139 break;
3140 }
3141 return NOTIFY_OK;
3142}
3143
3144static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3145 void *v)
3146{
3147 /*
3148 * Some (well, at least mine) BIOSes hang on reboot if
3149 * in vmx root mode.
3150 *
3151 * And Intel TXT required VMX off for all cpu when system shutdown.
3152 */
3153 pr_info("kvm: exiting hardware virtualization\n");
3154 kvm_rebooting = true;
3155 on_each_cpu(hardware_disable_nolock, NULL, 1);
3156 return NOTIFY_OK;
3157}
3158
3159static struct notifier_block kvm_reboot_notifier = {
3160 .notifier_call = kvm_reboot,
3161 .priority = 0,
3162};
3163
3164static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3165{
3166 int i;
3167
3168 for (i = 0; i < bus->dev_count; i++) {
3169 struct kvm_io_device *pos = bus->range[i].dev;
3170
3171 kvm_iodevice_destructor(pos);
3172 }
3173 kfree(bus);
3174}
3175
3176static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3177 const struct kvm_io_range *r2)
3178{
3179 gpa_t addr1 = r1->addr;
3180 gpa_t addr2 = r2->addr;
3181
3182 if (addr1 < addr2)
3183 return -1;
3184
3185 /* If r2->len == 0, match the exact address. If r2->len != 0,
3186 * accept any overlapping write. Any order is acceptable for
3187 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3188 * we process all of them.
3189 */
3190 if (r2->len) {
3191 addr1 += r1->len;
3192 addr2 += r2->len;
3193 }
3194
3195 if (addr1 > addr2)
3196 return 1;
3197
3198 return 0;
3199}
3200
3201static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3202{
3203 return kvm_io_bus_cmp(p1, p2);
3204}
3205
3206static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3207 gpa_t addr, int len)
3208{
3209 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3210 .addr = addr,
3211 .len = len,
3212 .dev = dev,
3213 };
3214
3215 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3216 kvm_io_bus_sort_cmp, NULL);
3217
3218 return 0;
3219}
3220
3221static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3222 gpa_t addr, int len)
3223{
3224 struct kvm_io_range *range, key;
3225 int off;
3226
3227 key = (struct kvm_io_range) {
3228 .addr = addr,
3229 .len = len,
3230 };
3231
3232 range = bsearch(&key, bus->range, bus->dev_count,
3233 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3234 if (range == NULL)
3235 return -ENOENT;
3236
3237 off = range - bus->range;
3238
3239 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3240 off--;
3241
3242 return off;
3243}
3244
3245static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3246 struct kvm_io_range *range, const void *val)
3247{
3248 int idx;
3249
3250 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3251 if (idx < 0)
3252 return -EOPNOTSUPP;
3253
3254 while (idx < bus->dev_count &&
3255 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3256 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3257 range->len, val))
3258 return idx;
3259 idx++;
3260 }
3261
3262 return -EOPNOTSUPP;
3263}
3264
3265/* kvm_io_bus_write - called under kvm->slots_lock */
3266int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3267 int len, const void *val)
3268{
3269 struct kvm_io_bus *bus;
3270 struct kvm_io_range range;
3271 int r;
3272
3273 range = (struct kvm_io_range) {
3274 .addr = addr,
3275 .len = len,
3276 };
3277
3278 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3279 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3280 return r < 0 ? r : 0;
3281}
3282
3283/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3284int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3285 gpa_t addr, int len, const void *val, long cookie)
3286{
3287 struct kvm_io_bus *bus;
3288 struct kvm_io_range range;
3289
3290 range = (struct kvm_io_range) {
3291 .addr = addr,
3292 .len = len,
3293 };
3294
3295 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3296
3297 /* First try the device referenced by cookie. */
3298 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3299 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3300 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3301 val))
3302 return cookie;
3303
3304 /*
3305 * cookie contained garbage; fall back to search and return the
3306 * correct cookie value.
3307 */
3308 return __kvm_io_bus_write(vcpu, bus, &range, val);
3309}
3310
3311static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3312 struct kvm_io_range *range, void *val)
3313{
3314 int idx;
3315
3316 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3317 if (idx < 0)
3318 return -EOPNOTSUPP;
3319
3320 while (idx < bus->dev_count &&
3321 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3322 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3323 range->len, val))
3324 return idx;
3325 idx++;
3326 }
3327
3328 return -EOPNOTSUPP;
3329}
3330EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3331
3332/* kvm_io_bus_read - called under kvm->slots_lock */
3333int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3334 int len, void *val)
3335{
3336 struct kvm_io_bus *bus;
3337 struct kvm_io_range range;
3338 int r;
3339
3340 range = (struct kvm_io_range) {
3341 .addr = addr,
3342 .len = len,
3343 };
3344
3345 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3346 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3347 return r < 0 ? r : 0;
3348}
3349
3350
3351/* Caller must hold slots_lock. */
3352int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3353 int len, struct kvm_io_device *dev)
3354{
3355 struct kvm_io_bus *new_bus, *bus;
3356
3357 bus = kvm->buses[bus_idx];
3358 /* exclude ioeventfd which is limited by maximum fd */
3359 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3360 return -ENOSPC;
3361
3362 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3363 sizeof(struct kvm_io_range)), GFP_KERNEL);
3364 if (!new_bus)
3365 return -ENOMEM;
3366 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3367 sizeof(struct kvm_io_range)));
3368 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3369 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3370 synchronize_srcu_expedited(&kvm->srcu);
3371 kfree(bus);
3372
3373 return 0;
3374}
3375
3376/* Caller must hold slots_lock. */
3377int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3378 struct kvm_io_device *dev)
3379{
3380 int i, r;
3381 struct kvm_io_bus *new_bus, *bus;
3382
3383 bus = kvm->buses[bus_idx];
3384 r = -ENOENT;
3385 for (i = 0; i < bus->dev_count; i++)
3386 if (bus->range[i].dev == dev) {
3387 r = 0;
3388 break;
3389 }
3390
3391 if (r)
3392 return r;
3393
3394 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3395 sizeof(struct kvm_io_range)), GFP_KERNEL);
3396 if (!new_bus)
3397 return -ENOMEM;
3398
3399 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3400 new_bus->dev_count--;
3401 memcpy(new_bus->range + i, bus->range + i + 1,
3402 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3403
3404 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3405 synchronize_srcu_expedited(&kvm->srcu);
3406 kfree(bus);
3407 return r;
3408}
3409
3410static struct notifier_block kvm_cpu_notifier = {
3411 .notifier_call = kvm_cpu_hotplug,
3412};
3413
3414static int vm_stat_get(void *_offset, u64 *val)
3415{
3416 unsigned offset = (long)_offset;
3417 struct kvm *kvm;
3418
3419 *val = 0;
3420 spin_lock(&kvm_lock);
3421 list_for_each_entry(kvm, &vm_list, vm_list)
3422 *val += *(u32 *)((void *)kvm + offset);
3423 spin_unlock(&kvm_lock);
3424 return 0;
3425}
3426
3427DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3428
3429static int vcpu_stat_get(void *_offset, u64 *val)
3430{
3431 unsigned offset = (long)_offset;
3432 struct kvm *kvm;
3433 struct kvm_vcpu *vcpu;
3434 int i;
3435
3436 *val = 0;
3437 spin_lock(&kvm_lock);
3438 list_for_each_entry(kvm, &vm_list, vm_list)
3439 kvm_for_each_vcpu(i, vcpu, kvm)
3440 *val += *(u32 *)((void *)vcpu + offset);
3441
3442 spin_unlock(&kvm_lock);
3443 return 0;
3444}
3445
3446DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3447
3448static const struct file_operations *stat_fops[] = {
3449 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3450 [KVM_STAT_VM] = &vm_stat_fops,
3451};
3452
3453static int kvm_init_debug(void)
3454{
3455 int r = -EEXIST;
3456 struct kvm_stats_debugfs_item *p;
3457
3458 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3459 if (kvm_debugfs_dir == NULL)
3460 goto out;
3461
3462 for (p = debugfs_entries; p->name; ++p) {
3463 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3464 (void *)(long)p->offset,
3465 stat_fops[p->kind]))
3466 goto out_dir;
3467 }
3468
3469 return 0;
3470
3471out_dir:
3472 debugfs_remove_recursive(kvm_debugfs_dir);
3473out:
3474 return r;
3475}
3476
3477static int kvm_suspend(void)
3478{
3479 if (kvm_usage_count)
3480 hardware_disable_nolock(NULL);
3481 return 0;
3482}
3483
3484static void kvm_resume(void)
3485{
3486 if (kvm_usage_count) {
3487 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3488 hardware_enable_nolock(NULL);
3489 }
3490}
3491
3492static struct syscore_ops kvm_syscore_ops = {
3493 .suspend = kvm_suspend,
3494 .resume = kvm_resume,
3495};
3496
3497static inline
3498struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3499{
3500 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3501}
3502
3503static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3504{
3505 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3506
3507 if (vcpu->preempted)
3508 vcpu->preempted = false;
3509
3510 kvm_arch_sched_in(vcpu, cpu);
3511
3512 kvm_arch_vcpu_load(vcpu, cpu);
3513}
3514
3515static void kvm_sched_out(struct preempt_notifier *pn,
3516 struct task_struct *next)
3517{
3518 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3519
3520 if (current->state == TASK_RUNNING)
3521 vcpu->preempted = true;
3522 kvm_arch_vcpu_put(vcpu);
3523}
3524
3525int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3526 struct module *module)
3527{
3528 int r;
3529 int cpu;
3530
3531 r = kvm_arch_init(opaque);
3532 if (r)
3533 goto out_fail;
3534
3535 /*
3536 * kvm_arch_init makes sure there's at most one caller
3537 * for architectures that support multiple implementations,
3538 * like intel and amd on x86.
3539 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3540 * conflicts in case kvm is already setup for another implementation.
3541 */
3542 r = kvm_irqfd_init();
3543 if (r)
3544 goto out_irqfd;
3545
3546 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3547 r = -ENOMEM;
3548 goto out_free_0;
3549 }
3550
3551 r = kvm_arch_hardware_setup();
3552 if (r < 0)
3553 goto out_free_0a;
3554
3555 for_each_online_cpu(cpu) {
3556 smp_call_function_single(cpu,
3557 kvm_arch_check_processor_compat,
3558 &r, 1);
3559 if (r < 0)
3560 goto out_free_1;
3561 }
3562
3563 r = register_cpu_notifier(&kvm_cpu_notifier);
3564 if (r)
3565 goto out_free_2;
3566 register_reboot_notifier(&kvm_reboot_notifier);
3567
3568 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3569 if (!vcpu_align)
3570 vcpu_align = __alignof__(struct kvm_vcpu);
3571 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3572 0, NULL);
3573 if (!kvm_vcpu_cache) {
3574 r = -ENOMEM;
3575 goto out_free_3;
3576 }
3577
3578 r = kvm_async_pf_init();
3579 if (r)
3580 goto out_free;
3581
3582 kvm_chardev_ops.owner = module;
3583 kvm_vm_fops.owner = module;
3584 kvm_vcpu_fops.owner = module;
3585
3586 r = misc_register(&kvm_dev);
3587 if (r) {
3588 pr_err("kvm: misc device register failed\n");
3589 goto out_unreg;
3590 }
3591
3592 register_syscore_ops(&kvm_syscore_ops);
3593
3594 kvm_preempt_ops.sched_in = kvm_sched_in;
3595 kvm_preempt_ops.sched_out = kvm_sched_out;
3596
3597 r = kvm_init_debug();
3598 if (r) {
3599 pr_err("kvm: create debugfs files failed\n");
3600 goto out_undebugfs;
3601 }
3602
3603 r = kvm_vfio_ops_init();
3604 WARN_ON(r);
3605
3606 return 0;
3607
3608out_undebugfs:
3609 unregister_syscore_ops(&kvm_syscore_ops);
3610 misc_deregister(&kvm_dev);
3611out_unreg:
3612 kvm_async_pf_deinit();
3613out_free:
3614 kmem_cache_destroy(kvm_vcpu_cache);
3615out_free_3:
3616 unregister_reboot_notifier(&kvm_reboot_notifier);
3617 unregister_cpu_notifier(&kvm_cpu_notifier);
3618out_free_2:
3619out_free_1:
3620 kvm_arch_hardware_unsetup();
3621out_free_0a:
3622 free_cpumask_var(cpus_hardware_enabled);
3623out_free_0:
3624 kvm_irqfd_exit();
3625out_irqfd:
3626 kvm_arch_exit();
3627out_fail:
3628 return r;
3629}
3630EXPORT_SYMBOL_GPL(kvm_init);
3631
3632void kvm_exit(void)
3633{
3634 debugfs_remove_recursive(kvm_debugfs_dir);
3635 misc_deregister(&kvm_dev);
3636 kmem_cache_destroy(kvm_vcpu_cache);
3637 kvm_async_pf_deinit();
3638 unregister_syscore_ops(&kvm_syscore_ops);
3639 unregister_reboot_notifier(&kvm_reboot_notifier);
3640 unregister_cpu_notifier(&kvm_cpu_notifier);
3641 on_each_cpu(hardware_disable_nolock, NULL, 1);
3642 kvm_arch_hardware_unsetup();
3643 kvm_arch_exit();
3644 kvm_irqfd_exit();
3645 free_cpumask_var(cpus_hardware_enabled);
3646 kvm_vfio_ops_exit();
3647}
3648EXPORT_SYMBOL_GPL(kvm_exit);
1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
19#include <kvm/iodev.h>
20
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>
26#include <linux/mm.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>
52
53#include <asm/processor.h>
54#include <asm/io.h>
55#include <asm/ioctl.h>
56#include <linux/uaccess.h>
57#include <asm/pgtable.h>
58
59#include "coalesced_mmio.h"
60#include "async_pf.h"
61#include "vfio.h"
62
63#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h>
65
66/* Worst case buffer size needed for holding an integer. */
67#define ITOA_MAX_LEN 12
68
69MODULE_AUTHOR("Qumranet");
70MODULE_LICENSE("GPL");
71
72/* Architectures should define their poll value according to the halt latency */
73unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75EXPORT_SYMBOL_GPL(halt_poll_ns);
76
77/* Default doubles per-vcpu halt_poll_ns. */
78unsigned int halt_poll_ns_grow = 2;
79module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
80EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
81
82/* Default resets per-vcpu halt_poll_ns . */
83unsigned int halt_poll_ns_shrink;
84module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
85EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
86
87/*
88 * Ordering of locks:
89 *
90 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
91 */
92
93DEFINE_SPINLOCK(kvm_lock);
94static DEFINE_RAW_SPINLOCK(kvm_count_lock);
95LIST_HEAD(vm_list);
96
97static cpumask_var_t cpus_hardware_enabled;
98static int kvm_usage_count;
99static atomic_t hardware_enable_failed;
100
101struct kmem_cache *kvm_vcpu_cache;
102EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
103
104static __read_mostly struct preempt_ops kvm_preempt_ops;
105
106struct dentry *kvm_debugfs_dir;
107EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
108
109static int kvm_debugfs_num_entries;
110static const struct file_operations *stat_fops_per_vm[];
111
112static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
113 unsigned long arg);
114#ifdef CONFIG_KVM_COMPAT
115static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
116 unsigned long arg);
117#endif
118static int hardware_enable_all(void);
119static void hardware_disable_all(void);
120
121static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
122
123static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
124static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
125
126__visible bool kvm_rebooting;
127EXPORT_SYMBOL_GPL(kvm_rebooting);
128
129static bool largepages_enabled = true;
130
131bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
132{
133 if (pfn_valid(pfn))
134 return PageReserved(pfn_to_page(pfn));
135
136 return true;
137}
138
139/*
140 * Switches to specified vcpu, until a matching vcpu_put()
141 */
142int vcpu_load(struct kvm_vcpu *vcpu)
143{
144 int cpu;
145
146 if (mutex_lock_killable(&vcpu->mutex))
147 return -EINTR;
148 cpu = get_cpu();
149 preempt_notifier_register(&vcpu->preempt_notifier);
150 kvm_arch_vcpu_load(vcpu, cpu);
151 put_cpu();
152 return 0;
153}
154EXPORT_SYMBOL_GPL(vcpu_load);
155
156void vcpu_put(struct kvm_vcpu *vcpu)
157{
158 preempt_disable();
159 kvm_arch_vcpu_put(vcpu);
160 preempt_notifier_unregister(&vcpu->preempt_notifier);
161 preempt_enable();
162 mutex_unlock(&vcpu->mutex);
163}
164EXPORT_SYMBOL_GPL(vcpu_put);
165
166static void ack_flush(void *_completed)
167{
168}
169
170bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
171{
172 int i, cpu, me;
173 cpumask_var_t cpus;
174 bool called = true;
175 struct kvm_vcpu *vcpu;
176
177 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
178
179 me = get_cpu();
180 kvm_for_each_vcpu(i, vcpu, kvm) {
181 kvm_make_request(req, vcpu);
182 cpu = vcpu->cpu;
183
184 /* Set ->requests bit before we read ->mode. */
185 smp_mb__after_atomic();
186
187 if (cpus != NULL && cpu != -1 && cpu != me &&
188 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
189 cpumask_set_cpu(cpu, cpus);
190 }
191 if (unlikely(cpus == NULL))
192 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
193 else if (!cpumask_empty(cpus))
194 smp_call_function_many(cpus, ack_flush, NULL, 1);
195 else
196 called = false;
197 put_cpu();
198 free_cpumask_var(cpus);
199 return called;
200}
201
202#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
203void kvm_flush_remote_tlbs(struct kvm *kvm)
204{
205 /*
206 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
207 * kvm_make_all_cpus_request.
208 */
209 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
210
211 /*
212 * We want to publish modifications to the page tables before reading
213 * mode. Pairs with a memory barrier in arch-specific code.
214 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
215 * and smp_mb in walk_shadow_page_lockless_begin/end.
216 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
217 *
218 * There is already an smp_mb__after_atomic() before
219 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
220 * barrier here.
221 */
222 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
223 ++kvm->stat.remote_tlb_flush;
224 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
225}
226EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
227#endif
228
229void kvm_reload_remote_mmus(struct kvm *kvm)
230{
231 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
232}
233
234int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
235{
236 struct page *page;
237 int r;
238
239 mutex_init(&vcpu->mutex);
240 vcpu->cpu = -1;
241 vcpu->kvm = kvm;
242 vcpu->vcpu_id = id;
243 vcpu->pid = NULL;
244 init_swait_queue_head(&vcpu->wq);
245 kvm_async_pf_vcpu_init(vcpu);
246
247 vcpu->pre_pcpu = -1;
248 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
249
250 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
251 if (!page) {
252 r = -ENOMEM;
253 goto fail;
254 }
255 vcpu->run = page_address(page);
256
257 kvm_vcpu_set_in_spin_loop(vcpu, false);
258 kvm_vcpu_set_dy_eligible(vcpu, false);
259 vcpu->preempted = false;
260
261 r = kvm_arch_vcpu_init(vcpu);
262 if (r < 0)
263 goto fail_free_run;
264 return 0;
265
266fail_free_run:
267 free_page((unsigned long)vcpu->run);
268fail:
269 return r;
270}
271EXPORT_SYMBOL_GPL(kvm_vcpu_init);
272
273void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
274{
275 put_pid(vcpu->pid);
276 kvm_arch_vcpu_uninit(vcpu);
277 free_page((unsigned long)vcpu->run);
278}
279EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
280
281#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
282static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
283{
284 return container_of(mn, struct kvm, mmu_notifier);
285}
286
287static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
288 struct mm_struct *mm,
289 unsigned long address)
290{
291 struct kvm *kvm = mmu_notifier_to_kvm(mn);
292 int need_tlb_flush, idx;
293
294 /*
295 * When ->invalidate_page runs, the linux pte has been zapped
296 * already but the page is still allocated until
297 * ->invalidate_page returns. So if we increase the sequence
298 * here the kvm page fault will notice if the spte can't be
299 * established because the page is going to be freed. If
300 * instead the kvm page fault establishes the spte before
301 * ->invalidate_page runs, kvm_unmap_hva will release it
302 * before returning.
303 *
304 * The sequence increase only need to be seen at spin_unlock
305 * time, and not at spin_lock time.
306 *
307 * Increasing the sequence after the spin_unlock would be
308 * unsafe because the kvm page fault could then establish the
309 * pte after kvm_unmap_hva returned, without noticing the page
310 * is going to be freed.
311 */
312 idx = srcu_read_lock(&kvm->srcu);
313 spin_lock(&kvm->mmu_lock);
314
315 kvm->mmu_notifier_seq++;
316 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
317 /* we've to flush the tlb before the pages can be freed */
318 if (need_tlb_flush)
319 kvm_flush_remote_tlbs(kvm);
320
321 spin_unlock(&kvm->mmu_lock);
322
323 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
324
325 srcu_read_unlock(&kvm->srcu, idx);
326}
327
328static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
329 struct mm_struct *mm,
330 unsigned long address,
331 pte_t pte)
332{
333 struct kvm *kvm = mmu_notifier_to_kvm(mn);
334 int idx;
335
336 idx = srcu_read_lock(&kvm->srcu);
337 spin_lock(&kvm->mmu_lock);
338 kvm->mmu_notifier_seq++;
339 kvm_set_spte_hva(kvm, address, pte);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
342}
343
344static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
345 struct mm_struct *mm,
346 unsigned long start,
347 unsigned long end)
348{
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 int need_tlb_flush = 0, idx;
351
352 idx = srcu_read_lock(&kvm->srcu);
353 spin_lock(&kvm->mmu_lock);
354 /*
355 * The count increase must become visible at unlock time as no
356 * spte can be established without taking the mmu_lock and
357 * count is also read inside the mmu_lock critical section.
358 */
359 kvm->mmu_notifier_count++;
360 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
361 need_tlb_flush |= kvm->tlbs_dirty;
362 /* we've to flush the tlb before the pages can be freed */
363 if (need_tlb_flush)
364 kvm_flush_remote_tlbs(kvm);
365
366 spin_unlock(&kvm->mmu_lock);
367 srcu_read_unlock(&kvm->srcu, idx);
368}
369
370static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long start,
373 unsigned long end)
374{
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
376
377 spin_lock(&kvm->mmu_lock);
378 /*
379 * This sequence increase will notify the kvm page fault that
380 * the page that is going to be mapped in the spte could have
381 * been freed.
382 */
383 kvm->mmu_notifier_seq++;
384 smp_wmb();
385 /*
386 * The above sequence increase must be visible before the
387 * below count decrease, which is ensured by the smp_wmb above
388 * in conjunction with the smp_rmb in mmu_notifier_retry().
389 */
390 kvm->mmu_notifier_count--;
391 spin_unlock(&kvm->mmu_lock);
392
393 BUG_ON(kvm->mmu_notifier_count < 0);
394}
395
396static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
397 struct mm_struct *mm,
398 unsigned long start,
399 unsigned long end)
400{
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
402 int young, idx;
403
404 idx = srcu_read_lock(&kvm->srcu);
405 spin_lock(&kvm->mmu_lock);
406
407 young = kvm_age_hva(kvm, start, end);
408 if (young)
409 kvm_flush_remote_tlbs(kvm);
410
411 spin_unlock(&kvm->mmu_lock);
412 srcu_read_unlock(&kvm->srcu, idx);
413
414 return young;
415}
416
417static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
418 struct mm_struct *mm,
419 unsigned long start,
420 unsigned long end)
421{
422 struct kvm *kvm = mmu_notifier_to_kvm(mn);
423 int young, idx;
424
425 idx = srcu_read_lock(&kvm->srcu);
426 spin_lock(&kvm->mmu_lock);
427 /*
428 * Even though we do not flush TLB, this will still adversely
429 * affect performance on pre-Haswell Intel EPT, where there is
430 * no EPT Access Bit to clear so that we have to tear down EPT
431 * tables instead. If we find this unacceptable, we can always
432 * add a parameter to kvm_age_hva so that it effectively doesn't
433 * do anything on clear_young.
434 *
435 * Also note that currently we never issue secondary TLB flushes
436 * from clear_young, leaving this job up to the regular system
437 * cadence. If we find this inaccurate, we might come up with a
438 * more sophisticated heuristic later.
439 */
440 young = kvm_age_hva(kvm, start, end);
441 spin_unlock(&kvm->mmu_lock);
442 srcu_read_unlock(&kvm->srcu, idx);
443
444 return young;
445}
446
447static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
448 struct mm_struct *mm,
449 unsigned long address)
450{
451 struct kvm *kvm = mmu_notifier_to_kvm(mn);
452 int young, idx;
453
454 idx = srcu_read_lock(&kvm->srcu);
455 spin_lock(&kvm->mmu_lock);
456 young = kvm_test_age_hva(kvm, address);
457 spin_unlock(&kvm->mmu_lock);
458 srcu_read_unlock(&kvm->srcu, idx);
459
460 return young;
461}
462
463static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
464 struct mm_struct *mm)
465{
466 struct kvm *kvm = mmu_notifier_to_kvm(mn);
467 int idx;
468
469 idx = srcu_read_lock(&kvm->srcu);
470 kvm_arch_flush_shadow_all(kvm);
471 srcu_read_unlock(&kvm->srcu, idx);
472}
473
474static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
475 .invalidate_page = kvm_mmu_notifier_invalidate_page,
476 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
477 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
478 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
479 .clear_young = kvm_mmu_notifier_clear_young,
480 .test_young = kvm_mmu_notifier_test_young,
481 .change_pte = kvm_mmu_notifier_change_pte,
482 .release = kvm_mmu_notifier_release,
483};
484
485static int kvm_init_mmu_notifier(struct kvm *kvm)
486{
487 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
488 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
489}
490
491#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
492
493static int kvm_init_mmu_notifier(struct kvm *kvm)
494{
495 return 0;
496}
497
498#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
499
500static struct kvm_memslots *kvm_alloc_memslots(void)
501{
502 int i;
503 struct kvm_memslots *slots;
504
505 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
506 if (!slots)
507 return NULL;
508
509 /*
510 * Init kvm generation close to the maximum to easily test the
511 * code of handling generation number wrap-around.
512 */
513 slots->generation = -150;
514 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
515 slots->id_to_index[i] = slots->memslots[i].id = i;
516
517 return slots;
518}
519
520static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
521{
522 if (!memslot->dirty_bitmap)
523 return;
524
525 kvfree(memslot->dirty_bitmap);
526 memslot->dirty_bitmap = NULL;
527}
528
529/*
530 * Free any memory in @free but not in @dont.
531 */
532static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
533 struct kvm_memory_slot *dont)
534{
535 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
536 kvm_destroy_dirty_bitmap(free);
537
538 kvm_arch_free_memslot(kvm, free, dont);
539
540 free->npages = 0;
541}
542
543static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
544{
545 struct kvm_memory_slot *memslot;
546
547 if (!slots)
548 return;
549
550 kvm_for_each_memslot(memslot, slots)
551 kvm_free_memslot(kvm, memslot, NULL);
552
553 kvfree(slots);
554}
555
556static void kvm_destroy_vm_debugfs(struct kvm *kvm)
557{
558 int i;
559
560 if (!kvm->debugfs_dentry)
561 return;
562
563 debugfs_remove_recursive(kvm->debugfs_dentry);
564
565 if (kvm->debugfs_stat_data) {
566 for (i = 0; i < kvm_debugfs_num_entries; i++)
567 kfree(kvm->debugfs_stat_data[i]);
568 kfree(kvm->debugfs_stat_data);
569 }
570}
571
572static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
573{
574 char dir_name[ITOA_MAX_LEN * 2];
575 struct kvm_stat_data *stat_data;
576 struct kvm_stats_debugfs_item *p;
577
578 if (!debugfs_initialized())
579 return 0;
580
581 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
582 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
583 kvm_debugfs_dir);
584 if (!kvm->debugfs_dentry)
585 return -ENOMEM;
586
587 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
588 sizeof(*kvm->debugfs_stat_data),
589 GFP_KERNEL);
590 if (!kvm->debugfs_stat_data)
591 return -ENOMEM;
592
593 for (p = debugfs_entries; p->name; p++) {
594 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
595 if (!stat_data)
596 return -ENOMEM;
597
598 stat_data->kvm = kvm;
599 stat_data->offset = p->offset;
600 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
601 if (!debugfs_create_file(p->name, 0644,
602 kvm->debugfs_dentry,
603 stat_data,
604 stat_fops_per_vm[p->kind]))
605 return -ENOMEM;
606 }
607 return 0;
608}
609
610static struct kvm *kvm_create_vm(unsigned long type)
611{
612 int r, i;
613 struct kvm *kvm = kvm_arch_alloc_vm();
614
615 if (!kvm)
616 return ERR_PTR(-ENOMEM);
617
618 spin_lock_init(&kvm->mmu_lock);
619 atomic_inc(¤t->mm->mm_count);
620 kvm->mm = current->mm;
621 kvm_eventfd_init(kvm);
622 mutex_init(&kvm->lock);
623 mutex_init(&kvm->irq_lock);
624 mutex_init(&kvm->slots_lock);
625 atomic_set(&kvm->users_count, 1);
626 INIT_LIST_HEAD(&kvm->devices);
627
628 r = kvm_arch_init_vm(kvm, type);
629 if (r)
630 goto out_err_no_disable;
631
632 r = hardware_enable_all();
633 if (r)
634 goto out_err_no_disable;
635
636#ifdef CONFIG_HAVE_KVM_IRQFD
637 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
638#endif
639
640 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
641
642 r = -ENOMEM;
643 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
644 kvm->memslots[i] = kvm_alloc_memslots();
645 if (!kvm->memslots[i])
646 goto out_err_no_srcu;
647 }
648
649 if (init_srcu_struct(&kvm->srcu))
650 goto out_err_no_srcu;
651 if (init_srcu_struct(&kvm->irq_srcu))
652 goto out_err_no_irq_srcu;
653 for (i = 0; i < KVM_NR_BUSES; i++) {
654 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
655 GFP_KERNEL);
656 if (!kvm->buses[i])
657 goto out_err;
658 }
659
660 r = kvm_init_mmu_notifier(kvm);
661 if (r)
662 goto out_err;
663
664 spin_lock(&kvm_lock);
665 list_add(&kvm->vm_list, &vm_list);
666 spin_unlock(&kvm_lock);
667
668 preempt_notifier_inc();
669
670 return kvm;
671
672out_err:
673 cleanup_srcu_struct(&kvm->irq_srcu);
674out_err_no_irq_srcu:
675 cleanup_srcu_struct(&kvm->srcu);
676out_err_no_srcu:
677 hardware_disable_all();
678out_err_no_disable:
679 for (i = 0; i < KVM_NR_BUSES; i++)
680 kfree(kvm->buses[i]);
681 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
682 kvm_free_memslots(kvm, kvm->memslots[i]);
683 kvm_arch_free_vm(kvm);
684 mmdrop(current->mm);
685 return ERR_PTR(r);
686}
687
688/*
689 * Avoid using vmalloc for a small buffer.
690 * Should not be used when the size is statically known.
691 */
692void *kvm_kvzalloc(unsigned long size)
693{
694 if (size > PAGE_SIZE)
695 return vzalloc(size);
696 else
697 return kzalloc(size, GFP_KERNEL);
698}
699
700static void kvm_destroy_devices(struct kvm *kvm)
701{
702 struct kvm_device *dev, *tmp;
703
704 /*
705 * We do not need to take the kvm->lock here, because nobody else
706 * has a reference to the struct kvm at this point and therefore
707 * cannot access the devices list anyhow.
708 */
709 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
710 list_del(&dev->vm_node);
711 dev->ops->destroy(dev);
712 }
713}
714
715static void kvm_destroy_vm(struct kvm *kvm)
716{
717 int i;
718 struct mm_struct *mm = kvm->mm;
719
720 kvm_destroy_vm_debugfs(kvm);
721 kvm_arch_sync_events(kvm);
722 spin_lock(&kvm_lock);
723 list_del(&kvm->vm_list);
724 spin_unlock(&kvm_lock);
725 kvm_free_irq_routing(kvm);
726 for (i = 0; i < KVM_NR_BUSES; i++) {
727 if (kvm->buses[i])
728 kvm_io_bus_destroy(kvm->buses[i]);
729 kvm->buses[i] = NULL;
730 }
731 kvm_coalesced_mmio_free(kvm);
732#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
733 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
734#else
735 kvm_arch_flush_shadow_all(kvm);
736#endif
737 kvm_arch_destroy_vm(kvm);
738 kvm_destroy_devices(kvm);
739 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
740 kvm_free_memslots(kvm, kvm->memslots[i]);
741 cleanup_srcu_struct(&kvm->irq_srcu);
742 cleanup_srcu_struct(&kvm->srcu);
743 kvm_arch_free_vm(kvm);
744 preempt_notifier_dec();
745 hardware_disable_all();
746 mmdrop(mm);
747}
748
749void kvm_get_kvm(struct kvm *kvm)
750{
751 atomic_inc(&kvm->users_count);
752}
753EXPORT_SYMBOL_GPL(kvm_get_kvm);
754
755void kvm_put_kvm(struct kvm *kvm)
756{
757 if (atomic_dec_and_test(&kvm->users_count))
758 kvm_destroy_vm(kvm);
759}
760EXPORT_SYMBOL_GPL(kvm_put_kvm);
761
762
763static int kvm_vm_release(struct inode *inode, struct file *filp)
764{
765 struct kvm *kvm = filp->private_data;
766
767 kvm_irqfd_release(kvm);
768
769 kvm_put_kvm(kvm);
770 return 0;
771}
772
773/*
774 * Allocation size is twice as large as the actual dirty bitmap size.
775 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
776 */
777static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
778{
779 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
780
781 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
782 if (!memslot->dirty_bitmap)
783 return -ENOMEM;
784
785 return 0;
786}
787
788/*
789 * Insert memslot and re-sort memslots based on their GFN,
790 * so binary search could be used to lookup GFN.
791 * Sorting algorithm takes advantage of having initially
792 * sorted array and known changed memslot position.
793 */
794static void update_memslots(struct kvm_memslots *slots,
795 struct kvm_memory_slot *new)
796{
797 int id = new->id;
798 int i = slots->id_to_index[id];
799 struct kvm_memory_slot *mslots = slots->memslots;
800
801 WARN_ON(mslots[i].id != id);
802 if (!new->npages) {
803 WARN_ON(!mslots[i].npages);
804 if (mslots[i].npages)
805 slots->used_slots--;
806 } else {
807 if (!mslots[i].npages)
808 slots->used_slots++;
809 }
810
811 while (i < KVM_MEM_SLOTS_NUM - 1 &&
812 new->base_gfn <= mslots[i + 1].base_gfn) {
813 if (!mslots[i + 1].npages)
814 break;
815 mslots[i] = mslots[i + 1];
816 slots->id_to_index[mslots[i].id] = i;
817 i++;
818 }
819
820 /*
821 * The ">=" is needed when creating a slot with base_gfn == 0,
822 * so that it moves before all those with base_gfn == npages == 0.
823 *
824 * On the other hand, if new->npages is zero, the above loop has
825 * already left i pointing to the beginning of the empty part of
826 * mslots, and the ">=" would move the hole backwards in this
827 * case---which is wrong. So skip the loop when deleting a slot.
828 */
829 if (new->npages) {
830 while (i > 0 &&
831 new->base_gfn >= mslots[i - 1].base_gfn) {
832 mslots[i] = mslots[i - 1];
833 slots->id_to_index[mslots[i].id] = i;
834 i--;
835 }
836 } else
837 WARN_ON_ONCE(i != slots->used_slots);
838
839 mslots[i] = *new;
840 slots->id_to_index[mslots[i].id] = i;
841}
842
843static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
844{
845 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
846
847#ifdef __KVM_HAVE_READONLY_MEM
848 valid_flags |= KVM_MEM_READONLY;
849#endif
850
851 if (mem->flags & ~valid_flags)
852 return -EINVAL;
853
854 return 0;
855}
856
857static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
858 int as_id, struct kvm_memslots *slots)
859{
860 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
861
862 /*
863 * Set the low bit in the generation, which disables SPTE caching
864 * until the end of synchronize_srcu_expedited.
865 */
866 WARN_ON(old_memslots->generation & 1);
867 slots->generation = old_memslots->generation + 1;
868
869 rcu_assign_pointer(kvm->memslots[as_id], slots);
870 synchronize_srcu_expedited(&kvm->srcu);
871
872 /*
873 * Increment the new memslot generation a second time. This prevents
874 * vm exits that race with memslot updates from caching a memslot
875 * generation that will (potentially) be valid forever.
876 */
877 slots->generation++;
878
879 kvm_arch_memslots_updated(kvm, slots);
880
881 return old_memslots;
882}
883
884/*
885 * Allocate some memory and give it an address in the guest physical address
886 * space.
887 *
888 * Discontiguous memory is allowed, mostly for framebuffers.
889 *
890 * Must be called holding kvm->slots_lock for write.
891 */
892int __kvm_set_memory_region(struct kvm *kvm,
893 const struct kvm_userspace_memory_region *mem)
894{
895 int r;
896 gfn_t base_gfn;
897 unsigned long npages;
898 struct kvm_memory_slot *slot;
899 struct kvm_memory_slot old, new;
900 struct kvm_memslots *slots = NULL, *old_memslots;
901 int as_id, id;
902 enum kvm_mr_change change;
903
904 r = check_memory_region_flags(mem);
905 if (r)
906 goto out;
907
908 r = -EINVAL;
909 as_id = mem->slot >> 16;
910 id = (u16)mem->slot;
911
912 /* General sanity checks */
913 if (mem->memory_size & (PAGE_SIZE - 1))
914 goto out;
915 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
916 goto out;
917 /* We can read the guest memory with __xxx_user() later on. */
918 if ((id < KVM_USER_MEM_SLOTS) &&
919 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
920 !access_ok(VERIFY_WRITE,
921 (void __user *)(unsigned long)mem->userspace_addr,
922 mem->memory_size)))
923 goto out;
924 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
925 goto out;
926 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
927 goto out;
928
929 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
930 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
931 npages = mem->memory_size >> PAGE_SHIFT;
932
933 if (npages > KVM_MEM_MAX_NR_PAGES)
934 goto out;
935
936 new = old = *slot;
937
938 new.id = id;
939 new.base_gfn = base_gfn;
940 new.npages = npages;
941 new.flags = mem->flags;
942
943 if (npages) {
944 if (!old.npages)
945 change = KVM_MR_CREATE;
946 else { /* Modify an existing slot. */
947 if ((mem->userspace_addr != old.userspace_addr) ||
948 (npages != old.npages) ||
949 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
950 goto out;
951
952 if (base_gfn != old.base_gfn)
953 change = KVM_MR_MOVE;
954 else if (new.flags != old.flags)
955 change = KVM_MR_FLAGS_ONLY;
956 else { /* Nothing to change. */
957 r = 0;
958 goto out;
959 }
960 }
961 } else {
962 if (!old.npages)
963 goto out;
964
965 change = KVM_MR_DELETE;
966 new.base_gfn = 0;
967 new.flags = 0;
968 }
969
970 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
971 /* Check for overlaps */
972 r = -EEXIST;
973 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
974 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
975 (slot->id == id))
976 continue;
977 if (!((base_gfn + npages <= slot->base_gfn) ||
978 (base_gfn >= slot->base_gfn + slot->npages)))
979 goto out;
980 }
981 }
982
983 /* Free page dirty bitmap if unneeded */
984 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
985 new.dirty_bitmap = NULL;
986
987 r = -ENOMEM;
988 if (change == KVM_MR_CREATE) {
989 new.userspace_addr = mem->userspace_addr;
990
991 if (kvm_arch_create_memslot(kvm, &new, npages))
992 goto out_free;
993 }
994
995 /* Allocate page dirty bitmap if needed */
996 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
997 if (kvm_create_dirty_bitmap(&new) < 0)
998 goto out_free;
999 }
1000
1001 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
1002 if (!slots)
1003 goto out_free;
1004 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1005
1006 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1007 slot = id_to_memslot(slots, id);
1008 slot->flags |= KVM_MEMSLOT_INVALID;
1009
1010 old_memslots = install_new_memslots(kvm, as_id, slots);
1011
1012 /* slot was deleted or moved, clear iommu mapping */
1013 kvm_iommu_unmap_pages(kvm, &old);
1014 /* From this point no new shadow pages pointing to a deleted,
1015 * or moved, memslot will be created.
1016 *
1017 * validation of sp->gfn happens in:
1018 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1019 * - kvm_is_visible_gfn (mmu_check_roots)
1020 */
1021 kvm_arch_flush_shadow_memslot(kvm, slot);
1022
1023 /*
1024 * We can re-use the old_memslots from above, the only difference
1025 * from the currently installed memslots is the invalid flag. This
1026 * will get overwritten by update_memslots anyway.
1027 */
1028 slots = old_memslots;
1029 }
1030
1031 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1032 if (r)
1033 goto out_slots;
1034
1035 /* actual memory is freed via old in kvm_free_memslot below */
1036 if (change == KVM_MR_DELETE) {
1037 new.dirty_bitmap = NULL;
1038 memset(&new.arch, 0, sizeof(new.arch));
1039 }
1040
1041 update_memslots(slots, &new);
1042 old_memslots = install_new_memslots(kvm, as_id, slots);
1043
1044 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1045
1046 kvm_free_memslot(kvm, &old, &new);
1047 kvfree(old_memslots);
1048
1049 /*
1050 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1051 * un-mapped and re-mapped if their base changes. Since base change
1052 * unmapping is handled above with slot deletion, mapping alone is
1053 * needed here. Anything else the iommu might care about for existing
1054 * slots (size changes, userspace addr changes and read-only flag
1055 * changes) is disallowed above, so any other attribute changes getting
1056 * here can be skipped.
1057 */
1058 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1059 r = kvm_iommu_map_pages(kvm, &new);
1060 return r;
1061 }
1062
1063 return 0;
1064
1065out_slots:
1066 kvfree(slots);
1067out_free:
1068 kvm_free_memslot(kvm, &new, &old);
1069out:
1070 return r;
1071}
1072EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1073
1074int kvm_set_memory_region(struct kvm *kvm,
1075 const struct kvm_userspace_memory_region *mem)
1076{
1077 int r;
1078
1079 mutex_lock(&kvm->slots_lock);
1080 r = __kvm_set_memory_region(kvm, mem);
1081 mutex_unlock(&kvm->slots_lock);
1082 return r;
1083}
1084EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1085
1086static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1087 struct kvm_userspace_memory_region *mem)
1088{
1089 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1090 return -EINVAL;
1091
1092 return kvm_set_memory_region(kvm, mem);
1093}
1094
1095int kvm_get_dirty_log(struct kvm *kvm,
1096 struct kvm_dirty_log *log, int *is_dirty)
1097{
1098 struct kvm_memslots *slots;
1099 struct kvm_memory_slot *memslot;
1100 int r, i, as_id, id;
1101 unsigned long n;
1102 unsigned long any = 0;
1103
1104 r = -EINVAL;
1105 as_id = log->slot >> 16;
1106 id = (u16)log->slot;
1107 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1108 goto out;
1109
1110 slots = __kvm_memslots(kvm, as_id);
1111 memslot = id_to_memslot(slots, id);
1112 r = -ENOENT;
1113 if (!memslot->dirty_bitmap)
1114 goto out;
1115
1116 n = kvm_dirty_bitmap_bytes(memslot);
1117
1118 for (i = 0; !any && i < n/sizeof(long); ++i)
1119 any = memslot->dirty_bitmap[i];
1120
1121 r = -EFAULT;
1122 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1123 goto out;
1124
1125 if (any)
1126 *is_dirty = 1;
1127
1128 r = 0;
1129out:
1130 return r;
1131}
1132EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1133
1134#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1135/**
1136 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1137 * are dirty write protect them for next write.
1138 * @kvm: pointer to kvm instance
1139 * @log: slot id and address to which we copy the log
1140 * @is_dirty: flag set if any page is dirty
1141 *
1142 * We need to keep it in mind that VCPU threads can write to the bitmap
1143 * concurrently. So, to avoid losing track of dirty pages we keep the
1144 * following order:
1145 *
1146 * 1. Take a snapshot of the bit and clear it if needed.
1147 * 2. Write protect the corresponding page.
1148 * 3. Copy the snapshot to the userspace.
1149 * 4. Upon return caller flushes TLB's if needed.
1150 *
1151 * Between 2 and 4, the guest may write to the page using the remaining TLB
1152 * entry. This is not a problem because the page is reported dirty using
1153 * the snapshot taken before and step 4 ensures that writes done after
1154 * exiting to userspace will be logged for the next call.
1155 *
1156 */
1157int kvm_get_dirty_log_protect(struct kvm *kvm,
1158 struct kvm_dirty_log *log, bool *is_dirty)
1159{
1160 struct kvm_memslots *slots;
1161 struct kvm_memory_slot *memslot;
1162 int r, i, as_id, id;
1163 unsigned long n;
1164 unsigned long *dirty_bitmap;
1165 unsigned long *dirty_bitmap_buffer;
1166
1167 r = -EINVAL;
1168 as_id = log->slot >> 16;
1169 id = (u16)log->slot;
1170 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1171 goto out;
1172
1173 slots = __kvm_memslots(kvm, as_id);
1174 memslot = id_to_memslot(slots, id);
1175
1176 dirty_bitmap = memslot->dirty_bitmap;
1177 r = -ENOENT;
1178 if (!dirty_bitmap)
1179 goto out;
1180
1181 n = kvm_dirty_bitmap_bytes(memslot);
1182
1183 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1184 memset(dirty_bitmap_buffer, 0, n);
1185
1186 spin_lock(&kvm->mmu_lock);
1187 *is_dirty = false;
1188 for (i = 0; i < n / sizeof(long); i++) {
1189 unsigned long mask;
1190 gfn_t offset;
1191
1192 if (!dirty_bitmap[i])
1193 continue;
1194
1195 *is_dirty = true;
1196
1197 mask = xchg(&dirty_bitmap[i], 0);
1198 dirty_bitmap_buffer[i] = mask;
1199
1200 if (mask) {
1201 offset = i * BITS_PER_LONG;
1202 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1203 offset, mask);
1204 }
1205 }
1206
1207 spin_unlock(&kvm->mmu_lock);
1208
1209 r = -EFAULT;
1210 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1211 goto out;
1212
1213 r = 0;
1214out:
1215 return r;
1216}
1217EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1218#endif
1219
1220bool kvm_largepages_enabled(void)
1221{
1222 return largepages_enabled;
1223}
1224
1225void kvm_disable_largepages(void)
1226{
1227 largepages_enabled = false;
1228}
1229EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1230
1231struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1232{
1233 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1234}
1235EXPORT_SYMBOL_GPL(gfn_to_memslot);
1236
1237struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1238{
1239 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1240}
1241
1242bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1243{
1244 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1245
1246 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1247 memslot->flags & KVM_MEMSLOT_INVALID)
1248 return false;
1249
1250 return true;
1251}
1252EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1253
1254unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1255{
1256 struct vm_area_struct *vma;
1257 unsigned long addr, size;
1258
1259 size = PAGE_SIZE;
1260
1261 addr = gfn_to_hva(kvm, gfn);
1262 if (kvm_is_error_hva(addr))
1263 return PAGE_SIZE;
1264
1265 down_read(¤t->mm->mmap_sem);
1266 vma = find_vma(current->mm, addr);
1267 if (!vma)
1268 goto out;
1269
1270 size = vma_kernel_pagesize(vma);
1271
1272out:
1273 up_read(¤t->mm->mmap_sem);
1274
1275 return size;
1276}
1277
1278static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1279{
1280 return slot->flags & KVM_MEM_READONLY;
1281}
1282
1283static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1284 gfn_t *nr_pages, bool write)
1285{
1286 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1287 return KVM_HVA_ERR_BAD;
1288
1289 if (memslot_is_readonly(slot) && write)
1290 return KVM_HVA_ERR_RO_BAD;
1291
1292 if (nr_pages)
1293 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1294
1295 return __gfn_to_hva_memslot(slot, gfn);
1296}
1297
1298static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1299 gfn_t *nr_pages)
1300{
1301 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1302}
1303
1304unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1305 gfn_t gfn)
1306{
1307 return gfn_to_hva_many(slot, gfn, NULL);
1308}
1309EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1310
1311unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1312{
1313 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1314}
1315EXPORT_SYMBOL_GPL(gfn_to_hva);
1316
1317unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1318{
1319 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1320}
1321EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1322
1323/*
1324 * If writable is set to false, the hva returned by this function is only
1325 * allowed to be read.
1326 */
1327unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1328 gfn_t gfn, bool *writable)
1329{
1330 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1331
1332 if (!kvm_is_error_hva(hva) && writable)
1333 *writable = !memslot_is_readonly(slot);
1334
1335 return hva;
1336}
1337
1338unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1339{
1340 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1341
1342 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1343}
1344
1345unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1346{
1347 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1348
1349 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1350}
1351
1352static int get_user_page_nowait(unsigned long start, int write,
1353 struct page **page)
1354{
1355 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1356
1357 if (write)
1358 flags |= FOLL_WRITE;
1359
1360 return get_user_pages(start, 1, flags, page, NULL);
1361}
1362
1363static inline int check_user_page_hwpoison(unsigned long addr)
1364{
1365 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1366
1367 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1368 return rc == -EHWPOISON;
1369}
1370
1371/*
1372 * The atomic path to get the writable pfn which will be stored in @pfn,
1373 * true indicates success, otherwise false is returned.
1374 */
1375static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1376 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1377{
1378 struct page *page[1];
1379 int npages;
1380
1381 if (!(async || atomic))
1382 return false;
1383
1384 /*
1385 * Fast pin a writable pfn only if it is a write fault request
1386 * or the caller allows to map a writable pfn for a read fault
1387 * request.
1388 */
1389 if (!(write_fault || writable))
1390 return false;
1391
1392 npages = __get_user_pages_fast(addr, 1, 1, page);
1393 if (npages == 1) {
1394 *pfn = page_to_pfn(page[0]);
1395
1396 if (writable)
1397 *writable = true;
1398 return true;
1399 }
1400
1401 return false;
1402}
1403
1404/*
1405 * The slow path to get the pfn of the specified host virtual address,
1406 * 1 indicates success, -errno is returned if error is detected.
1407 */
1408static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1409 bool *writable, kvm_pfn_t *pfn)
1410{
1411 struct page *page[1];
1412 int npages = 0;
1413
1414 might_sleep();
1415
1416 if (writable)
1417 *writable = write_fault;
1418
1419 if (async) {
1420 down_read(¤t->mm->mmap_sem);
1421 npages = get_user_page_nowait(addr, write_fault, page);
1422 up_read(¤t->mm->mmap_sem);
1423 } else {
1424 unsigned int flags = FOLL_HWPOISON;
1425
1426 if (write_fault)
1427 flags |= FOLL_WRITE;
1428
1429 npages = get_user_pages_unlocked(addr, 1, page, flags);
1430 }
1431 if (npages != 1)
1432 return npages;
1433
1434 /* map read fault as writable if possible */
1435 if (unlikely(!write_fault) && writable) {
1436 struct page *wpage[1];
1437
1438 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1439 if (npages == 1) {
1440 *writable = true;
1441 put_page(page[0]);
1442 page[0] = wpage[0];
1443 }
1444
1445 npages = 1;
1446 }
1447 *pfn = page_to_pfn(page[0]);
1448 return npages;
1449}
1450
1451static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1452{
1453 if (unlikely(!(vma->vm_flags & VM_READ)))
1454 return false;
1455
1456 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1457 return false;
1458
1459 return true;
1460}
1461
1462static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1463 unsigned long addr, bool *async,
1464 bool write_fault, kvm_pfn_t *p_pfn)
1465{
1466 unsigned long pfn;
1467 int r;
1468
1469 r = follow_pfn(vma, addr, &pfn);
1470 if (r) {
1471 /*
1472 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1473 * not call the fault handler, so do it here.
1474 */
1475 bool unlocked = false;
1476 r = fixup_user_fault(current, current->mm, addr,
1477 (write_fault ? FAULT_FLAG_WRITE : 0),
1478 &unlocked);
1479 if (unlocked)
1480 return -EAGAIN;
1481 if (r)
1482 return r;
1483
1484 r = follow_pfn(vma, addr, &pfn);
1485 if (r)
1486 return r;
1487
1488 }
1489
1490
1491 /*
1492 * Get a reference here because callers of *hva_to_pfn* and
1493 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1494 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1495 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1496 * simply do nothing for reserved pfns.
1497 *
1498 * Whoever called remap_pfn_range is also going to call e.g.
1499 * unmap_mapping_range before the underlying pages are freed,
1500 * causing a call to our MMU notifier.
1501 */
1502 kvm_get_pfn(pfn);
1503
1504 *p_pfn = pfn;
1505 return 0;
1506}
1507
1508/*
1509 * Pin guest page in memory and return its pfn.
1510 * @addr: host virtual address which maps memory to the guest
1511 * @atomic: whether this function can sleep
1512 * @async: whether this function need to wait IO complete if the
1513 * host page is not in the memory
1514 * @write_fault: whether we should get a writable host page
1515 * @writable: whether it allows to map a writable host page for !@write_fault
1516 *
1517 * The function will map a writable host page for these two cases:
1518 * 1): @write_fault = true
1519 * 2): @write_fault = false && @writable, @writable will tell the caller
1520 * whether the mapping is writable.
1521 */
1522static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1523 bool write_fault, bool *writable)
1524{
1525 struct vm_area_struct *vma;
1526 kvm_pfn_t pfn = 0;
1527 int npages, r;
1528
1529 /* we can do it either atomically or asynchronously, not both */
1530 BUG_ON(atomic && async);
1531
1532 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1533 return pfn;
1534
1535 if (atomic)
1536 return KVM_PFN_ERR_FAULT;
1537
1538 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1539 if (npages == 1)
1540 return pfn;
1541
1542 down_read(¤t->mm->mmap_sem);
1543 if (npages == -EHWPOISON ||
1544 (!async && check_user_page_hwpoison(addr))) {
1545 pfn = KVM_PFN_ERR_HWPOISON;
1546 goto exit;
1547 }
1548
1549retry:
1550 vma = find_vma_intersection(current->mm, addr, addr + 1);
1551
1552 if (vma == NULL)
1553 pfn = KVM_PFN_ERR_FAULT;
1554 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1555 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1556 if (r == -EAGAIN)
1557 goto retry;
1558 if (r < 0)
1559 pfn = KVM_PFN_ERR_FAULT;
1560 } else {
1561 if (async && vma_is_valid(vma, write_fault))
1562 *async = true;
1563 pfn = KVM_PFN_ERR_FAULT;
1564 }
1565exit:
1566 up_read(¤t->mm->mmap_sem);
1567 return pfn;
1568}
1569
1570kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1571 bool atomic, bool *async, bool write_fault,
1572 bool *writable)
1573{
1574 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1575
1576 if (addr == KVM_HVA_ERR_RO_BAD) {
1577 if (writable)
1578 *writable = false;
1579 return KVM_PFN_ERR_RO_FAULT;
1580 }
1581
1582 if (kvm_is_error_hva(addr)) {
1583 if (writable)
1584 *writable = false;
1585 return KVM_PFN_NOSLOT;
1586 }
1587
1588 /* Do not map writable pfn in the readonly memslot. */
1589 if (writable && memslot_is_readonly(slot)) {
1590 *writable = false;
1591 writable = NULL;
1592 }
1593
1594 return hva_to_pfn(addr, atomic, async, write_fault,
1595 writable);
1596}
1597EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1598
1599kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1600 bool *writable)
1601{
1602 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1603 write_fault, writable);
1604}
1605EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1606
1607kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1608{
1609 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1610}
1611EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1612
1613kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1614{
1615 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1616}
1617EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1618
1619kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1620{
1621 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1622}
1623EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1624
1625kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1626{
1627 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1628}
1629EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1630
1631kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1632{
1633 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1634}
1635EXPORT_SYMBOL_GPL(gfn_to_pfn);
1636
1637kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1638{
1639 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1640}
1641EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1642
1643int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1644 struct page **pages, int nr_pages)
1645{
1646 unsigned long addr;
1647 gfn_t entry;
1648
1649 addr = gfn_to_hva_many(slot, gfn, &entry);
1650 if (kvm_is_error_hva(addr))
1651 return -1;
1652
1653 if (entry < nr_pages)
1654 return 0;
1655
1656 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1657}
1658EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1659
1660static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1661{
1662 if (is_error_noslot_pfn(pfn))
1663 return KVM_ERR_PTR_BAD_PAGE;
1664
1665 if (kvm_is_reserved_pfn(pfn)) {
1666 WARN_ON(1);
1667 return KVM_ERR_PTR_BAD_PAGE;
1668 }
1669
1670 return pfn_to_page(pfn);
1671}
1672
1673struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1674{
1675 kvm_pfn_t pfn;
1676
1677 pfn = gfn_to_pfn(kvm, gfn);
1678
1679 return kvm_pfn_to_page(pfn);
1680}
1681EXPORT_SYMBOL_GPL(gfn_to_page);
1682
1683struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1684{
1685 kvm_pfn_t pfn;
1686
1687 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1688
1689 return kvm_pfn_to_page(pfn);
1690}
1691EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1692
1693void kvm_release_page_clean(struct page *page)
1694{
1695 WARN_ON(is_error_page(page));
1696
1697 kvm_release_pfn_clean(page_to_pfn(page));
1698}
1699EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1700
1701void kvm_release_pfn_clean(kvm_pfn_t pfn)
1702{
1703 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1704 put_page(pfn_to_page(pfn));
1705}
1706EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1707
1708void kvm_release_page_dirty(struct page *page)
1709{
1710 WARN_ON(is_error_page(page));
1711
1712 kvm_release_pfn_dirty(page_to_pfn(page));
1713}
1714EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1715
1716static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1717{
1718 kvm_set_pfn_dirty(pfn);
1719 kvm_release_pfn_clean(pfn);
1720}
1721
1722void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1723{
1724 if (!kvm_is_reserved_pfn(pfn)) {
1725 struct page *page = pfn_to_page(pfn);
1726
1727 if (!PageReserved(page))
1728 SetPageDirty(page);
1729 }
1730}
1731EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1732
1733void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1734{
1735 if (!kvm_is_reserved_pfn(pfn))
1736 mark_page_accessed(pfn_to_page(pfn));
1737}
1738EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1739
1740void kvm_get_pfn(kvm_pfn_t pfn)
1741{
1742 if (!kvm_is_reserved_pfn(pfn))
1743 get_page(pfn_to_page(pfn));
1744}
1745EXPORT_SYMBOL_GPL(kvm_get_pfn);
1746
1747static int next_segment(unsigned long len, int offset)
1748{
1749 if (len > PAGE_SIZE - offset)
1750 return PAGE_SIZE - offset;
1751 else
1752 return len;
1753}
1754
1755static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1756 void *data, int offset, int len)
1757{
1758 int r;
1759 unsigned long addr;
1760
1761 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1762 if (kvm_is_error_hva(addr))
1763 return -EFAULT;
1764 r = __copy_from_user(data, (void __user *)addr + offset, len);
1765 if (r)
1766 return -EFAULT;
1767 return 0;
1768}
1769
1770int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1771 int len)
1772{
1773 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1774
1775 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1776}
1777EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1778
1779int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1780 int offset, int len)
1781{
1782 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1783
1784 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1785}
1786EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1787
1788int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1789{
1790 gfn_t gfn = gpa >> PAGE_SHIFT;
1791 int seg;
1792 int offset = offset_in_page(gpa);
1793 int ret;
1794
1795 while ((seg = next_segment(len, offset)) != 0) {
1796 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1797 if (ret < 0)
1798 return ret;
1799 offset = 0;
1800 len -= seg;
1801 data += seg;
1802 ++gfn;
1803 }
1804 return 0;
1805}
1806EXPORT_SYMBOL_GPL(kvm_read_guest);
1807
1808int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1809{
1810 gfn_t gfn = gpa >> PAGE_SHIFT;
1811 int seg;
1812 int offset = offset_in_page(gpa);
1813 int ret;
1814
1815 while ((seg = next_segment(len, offset)) != 0) {
1816 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1817 if (ret < 0)
1818 return ret;
1819 offset = 0;
1820 len -= seg;
1821 data += seg;
1822 ++gfn;
1823 }
1824 return 0;
1825}
1826EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1827
1828static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1829 void *data, int offset, unsigned long len)
1830{
1831 int r;
1832 unsigned long addr;
1833
1834 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1835 if (kvm_is_error_hva(addr))
1836 return -EFAULT;
1837 pagefault_disable();
1838 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1839 pagefault_enable();
1840 if (r)
1841 return -EFAULT;
1842 return 0;
1843}
1844
1845int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1846 unsigned long len)
1847{
1848 gfn_t gfn = gpa >> PAGE_SHIFT;
1849 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1850 int offset = offset_in_page(gpa);
1851
1852 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1853}
1854EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1855
1856int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1857 void *data, unsigned long len)
1858{
1859 gfn_t gfn = gpa >> PAGE_SHIFT;
1860 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1861 int offset = offset_in_page(gpa);
1862
1863 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1864}
1865EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1866
1867static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1868 const void *data, int offset, int len)
1869{
1870 int r;
1871 unsigned long addr;
1872
1873 addr = gfn_to_hva_memslot(memslot, gfn);
1874 if (kvm_is_error_hva(addr))
1875 return -EFAULT;
1876 r = __copy_to_user((void __user *)addr + offset, data, len);
1877 if (r)
1878 return -EFAULT;
1879 mark_page_dirty_in_slot(memslot, gfn);
1880 return 0;
1881}
1882
1883int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1884 const void *data, int offset, int len)
1885{
1886 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1887
1888 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1889}
1890EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1891
1892int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1893 const void *data, int offset, int len)
1894{
1895 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1896
1897 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1898}
1899EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1900
1901int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1902 unsigned long len)
1903{
1904 gfn_t gfn = gpa >> PAGE_SHIFT;
1905 int seg;
1906 int offset = offset_in_page(gpa);
1907 int ret;
1908
1909 while ((seg = next_segment(len, offset)) != 0) {
1910 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1911 if (ret < 0)
1912 return ret;
1913 offset = 0;
1914 len -= seg;
1915 data += seg;
1916 ++gfn;
1917 }
1918 return 0;
1919}
1920EXPORT_SYMBOL_GPL(kvm_write_guest);
1921
1922int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1923 unsigned long len)
1924{
1925 gfn_t gfn = gpa >> PAGE_SHIFT;
1926 int seg;
1927 int offset = offset_in_page(gpa);
1928 int ret;
1929
1930 while ((seg = next_segment(len, offset)) != 0) {
1931 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1932 if (ret < 0)
1933 return ret;
1934 offset = 0;
1935 len -= seg;
1936 data += seg;
1937 ++gfn;
1938 }
1939 return 0;
1940}
1941EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1942
1943int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1944 gpa_t gpa, unsigned long len)
1945{
1946 struct kvm_memslots *slots = kvm_memslots(kvm);
1947 int offset = offset_in_page(gpa);
1948 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1949 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1950 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1951 gfn_t nr_pages_avail;
1952
1953 ghc->gpa = gpa;
1954 ghc->generation = slots->generation;
1955 ghc->len = len;
1956 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1957 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1958 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1959 ghc->hva += offset;
1960 } else {
1961 /*
1962 * If the requested region crosses two memslots, we still
1963 * verify that the entire region is valid here.
1964 */
1965 while (start_gfn <= end_gfn) {
1966 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1967 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1968 &nr_pages_avail);
1969 if (kvm_is_error_hva(ghc->hva))
1970 return -EFAULT;
1971 start_gfn += nr_pages_avail;
1972 }
1973 /* Use the slow path for cross page reads and writes. */
1974 ghc->memslot = NULL;
1975 }
1976 return 0;
1977}
1978EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1979
1980int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1981 void *data, int offset, unsigned long len)
1982{
1983 struct kvm_memslots *slots = kvm_memslots(kvm);
1984 int r;
1985 gpa_t gpa = ghc->gpa + offset;
1986
1987 BUG_ON(len + offset > ghc->len);
1988
1989 if (slots->generation != ghc->generation)
1990 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1991
1992 if (unlikely(!ghc->memslot))
1993 return kvm_write_guest(kvm, gpa, data, len);
1994
1995 if (kvm_is_error_hva(ghc->hva))
1996 return -EFAULT;
1997
1998 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1999 if (r)
2000 return -EFAULT;
2001 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2002
2003 return 0;
2004}
2005EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2006
2007int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2008 void *data, unsigned long len)
2009{
2010 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2011}
2012EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2013
2014int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2015 void *data, unsigned long len)
2016{
2017 struct kvm_memslots *slots = kvm_memslots(kvm);
2018 int r;
2019
2020 BUG_ON(len > ghc->len);
2021
2022 if (slots->generation != ghc->generation)
2023 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
2024
2025 if (unlikely(!ghc->memslot))
2026 return kvm_read_guest(kvm, ghc->gpa, data, len);
2027
2028 if (kvm_is_error_hva(ghc->hva))
2029 return -EFAULT;
2030
2031 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2032 if (r)
2033 return -EFAULT;
2034
2035 return 0;
2036}
2037EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2038
2039int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2040{
2041 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2042
2043 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2044}
2045EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2046
2047int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2048{
2049 gfn_t gfn = gpa >> PAGE_SHIFT;
2050 int seg;
2051 int offset = offset_in_page(gpa);
2052 int ret;
2053
2054 while ((seg = next_segment(len, offset)) != 0) {
2055 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2056 if (ret < 0)
2057 return ret;
2058 offset = 0;
2059 len -= seg;
2060 ++gfn;
2061 }
2062 return 0;
2063}
2064EXPORT_SYMBOL_GPL(kvm_clear_guest);
2065
2066static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2067 gfn_t gfn)
2068{
2069 if (memslot && memslot->dirty_bitmap) {
2070 unsigned long rel_gfn = gfn - memslot->base_gfn;
2071
2072 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2073 }
2074}
2075
2076void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2077{
2078 struct kvm_memory_slot *memslot;
2079
2080 memslot = gfn_to_memslot(kvm, gfn);
2081 mark_page_dirty_in_slot(memslot, gfn);
2082}
2083EXPORT_SYMBOL_GPL(mark_page_dirty);
2084
2085void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2086{
2087 struct kvm_memory_slot *memslot;
2088
2089 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2090 mark_page_dirty_in_slot(memslot, gfn);
2091}
2092EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2093
2094static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2095{
2096 unsigned int old, val, grow;
2097
2098 old = val = vcpu->halt_poll_ns;
2099 grow = READ_ONCE(halt_poll_ns_grow);
2100 /* 10us base */
2101 if (val == 0 && grow)
2102 val = 10000;
2103 else
2104 val *= grow;
2105
2106 if (val > halt_poll_ns)
2107 val = halt_poll_ns;
2108
2109 vcpu->halt_poll_ns = val;
2110 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2111}
2112
2113static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2114{
2115 unsigned int old, val, shrink;
2116
2117 old = val = vcpu->halt_poll_ns;
2118 shrink = READ_ONCE(halt_poll_ns_shrink);
2119 if (shrink == 0)
2120 val = 0;
2121 else
2122 val /= shrink;
2123
2124 vcpu->halt_poll_ns = val;
2125 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2126}
2127
2128static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2129{
2130 if (kvm_arch_vcpu_runnable(vcpu)) {
2131 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2132 return -EINTR;
2133 }
2134 if (kvm_cpu_has_pending_timer(vcpu))
2135 return -EINTR;
2136 if (signal_pending(current))
2137 return -EINTR;
2138
2139 return 0;
2140}
2141
2142/*
2143 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2144 */
2145void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2146{
2147 ktime_t start, cur;
2148 DECLARE_SWAITQUEUE(wait);
2149 bool waited = false;
2150 u64 block_ns;
2151
2152 start = cur = ktime_get();
2153 if (vcpu->halt_poll_ns) {
2154 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2155
2156 ++vcpu->stat.halt_attempted_poll;
2157 do {
2158 /*
2159 * This sets KVM_REQ_UNHALT if an interrupt
2160 * arrives.
2161 */
2162 if (kvm_vcpu_check_block(vcpu) < 0) {
2163 ++vcpu->stat.halt_successful_poll;
2164 if (!vcpu_valid_wakeup(vcpu))
2165 ++vcpu->stat.halt_poll_invalid;
2166 goto out;
2167 }
2168 cur = ktime_get();
2169 } while (single_task_running() && ktime_before(cur, stop));
2170 }
2171
2172 kvm_arch_vcpu_blocking(vcpu);
2173
2174 for (;;) {
2175 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2176
2177 if (kvm_vcpu_check_block(vcpu) < 0)
2178 break;
2179
2180 waited = true;
2181 schedule();
2182 }
2183
2184 finish_swait(&vcpu->wq, &wait);
2185 cur = ktime_get();
2186
2187 kvm_arch_vcpu_unblocking(vcpu);
2188out:
2189 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2190
2191 if (!vcpu_valid_wakeup(vcpu))
2192 shrink_halt_poll_ns(vcpu);
2193 else if (halt_poll_ns) {
2194 if (block_ns <= vcpu->halt_poll_ns)
2195 ;
2196 /* we had a long block, shrink polling */
2197 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2198 shrink_halt_poll_ns(vcpu);
2199 /* we had a short halt and our poll time is too small */
2200 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2201 block_ns < halt_poll_ns)
2202 grow_halt_poll_ns(vcpu);
2203 } else
2204 vcpu->halt_poll_ns = 0;
2205
2206 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2207 kvm_arch_vcpu_block_finish(vcpu);
2208}
2209EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2210
2211#ifndef CONFIG_S390
2212void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2213{
2214 struct swait_queue_head *wqp;
2215
2216 wqp = kvm_arch_vcpu_wq(vcpu);
2217 if (swait_active(wqp)) {
2218 swake_up(wqp);
2219 ++vcpu->stat.halt_wakeup;
2220 }
2221
2222}
2223EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2224
2225/*
2226 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2227 */
2228void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2229{
2230 int me;
2231 int cpu = vcpu->cpu;
2232
2233 kvm_vcpu_wake_up(vcpu);
2234 me = get_cpu();
2235 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2236 if (kvm_arch_vcpu_should_kick(vcpu))
2237 smp_send_reschedule(cpu);
2238 put_cpu();
2239}
2240EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2241#endif /* !CONFIG_S390 */
2242
2243int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2244{
2245 struct pid *pid;
2246 struct task_struct *task = NULL;
2247 int ret = 0;
2248
2249 rcu_read_lock();
2250 pid = rcu_dereference(target->pid);
2251 if (pid)
2252 task = get_pid_task(pid, PIDTYPE_PID);
2253 rcu_read_unlock();
2254 if (!task)
2255 return ret;
2256 ret = yield_to(task, 1);
2257 put_task_struct(task);
2258
2259 return ret;
2260}
2261EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2262
2263/*
2264 * Helper that checks whether a VCPU is eligible for directed yield.
2265 * Most eligible candidate to yield is decided by following heuristics:
2266 *
2267 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2268 * (preempted lock holder), indicated by @in_spin_loop.
2269 * Set at the beiginning and cleared at the end of interception/PLE handler.
2270 *
2271 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2272 * chance last time (mostly it has become eligible now since we have probably
2273 * yielded to lockholder in last iteration. This is done by toggling
2274 * @dy_eligible each time a VCPU checked for eligibility.)
2275 *
2276 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2277 * to preempted lock-holder could result in wrong VCPU selection and CPU
2278 * burning. Giving priority for a potential lock-holder increases lock
2279 * progress.
2280 *
2281 * Since algorithm is based on heuristics, accessing another VCPU data without
2282 * locking does not harm. It may result in trying to yield to same VCPU, fail
2283 * and continue with next VCPU and so on.
2284 */
2285static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2286{
2287#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2288 bool eligible;
2289
2290 eligible = !vcpu->spin_loop.in_spin_loop ||
2291 vcpu->spin_loop.dy_eligible;
2292
2293 if (vcpu->spin_loop.in_spin_loop)
2294 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2295
2296 return eligible;
2297#else
2298 return true;
2299#endif
2300}
2301
2302void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2303{
2304 struct kvm *kvm = me->kvm;
2305 struct kvm_vcpu *vcpu;
2306 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2307 int yielded = 0;
2308 int try = 3;
2309 int pass;
2310 int i;
2311
2312 kvm_vcpu_set_in_spin_loop(me, true);
2313 /*
2314 * We boost the priority of a VCPU that is runnable but not
2315 * currently running, because it got preempted by something
2316 * else and called schedule in __vcpu_run. Hopefully that
2317 * VCPU is holding the lock that we need and will release it.
2318 * We approximate round-robin by starting at the last boosted VCPU.
2319 */
2320 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2321 kvm_for_each_vcpu(i, vcpu, kvm) {
2322 if (!pass && i <= last_boosted_vcpu) {
2323 i = last_boosted_vcpu;
2324 continue;
2325 } else if (pass && i > last_boosted_vcpu)
2326 break;
2327 if (!ACCESS_ONCE(vcpu->preempted))
2328 continue;
2329 if (vcpu == me)
2330 continue;
2331 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2332 continue;
2333 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2334 continue;
2335
2336 yielded = kvm_vcpu_yield_to(vcpu);
2337 if (yielded > 0) {
2338 kvm->last_boosted_vcpu = i;
2339 break;
2340 } else if (yielded < 0) {
2341 try--;
2342 if (!try)
2343 break;
2344 }
2345 }
2346 }
2347 kvm_vcpu_set_in_spin_loop(me, false);
2348
2349 /* Ensure vcpu is not eligible during next spinloop */
2350 kvm_vcpu_set_dy_eligible(me, false);
2351}
2352EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2353
2354static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2355{
2356 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2357 struct page *page;
2358
2359 if (vmf->pgoff == 0)
2360 page = virt_to_page(vcpu->run);
2361#ifdef CONFIG_X86
2362 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2363 page = virt_to_page(vcpu->arch.pio_data);
2364#endif
2365#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2366 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2367 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2368#endif
2369 else
2370 return kvm_arch_vcpu_fault(vcpu, vmf);
2371 get_page(page);
2372 vmf->page = page;
2373 return 0;
2374}
2375
2376static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2377 .fault = kvm_vcpu_fault,
2378};
2379
2380static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2381{
2382 vma->vm_ops = &kvm_vcpu_vm_ops;
2383 return 0;
2384}
2385
2386static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2387{
2388 struct kvm_vcpu *vcpu = filp->private_data;
2389
2390 debugfs_remove_recursive(vcpu->debugfs_dentry);
2391 kvm_put_kvm(vcpu->kvm);
2392 return 0;
2393}
2394
2395static struct file_operations kvm_vcpu_fops = {
2396 .release = kvm_vcpu_release,
2397 .unlocked_ioctl = kvm_vcpu_ioctl,
2398#ifdef CONFIG_KVM_COMPAT
2399 .compat_ioctl = kvm_vcpu_compat_ioctl,
2400#endif
2401 .mmap = kvm_vcpu_mmap,
2402 .llseek = noop_llseek,
2403};
2404
2405/*
2406 * Allocates an inode for the vcpu.
2407 */
2408static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2409{
2410 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2411}
2412
2413static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2414{
2415 char dir_name[ITOA_MAX_LEN * 2];
2416 int ret;
2417
2418 if (!kvm_arch_has_vcpu_debugfs())
2419 return 0;
2420
2421 if (!debugfs_initialized())
2422 return 0;
2423
2424 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2425 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2426 vcpu->kvm->debugfs_dentry);
2427 if (!vcpu->debugfs_dentry)
2428 return -ENOMEM;
2429
2430 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2431 if (ret < 0) {
2432 debugfs_remove_recursive(vcpu->debugfs_dentry);
2433 return ret;
2434 }
2435
2436 return 0;
2437}
2438
2439/*
2440 * Creates some virtual cpus. Good luck creating more than one.
2441 */
2442static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2443{
2444 int r;
2445 struct kvm_vcpu *vcpu;
2446
2447 if (id >= KVM_MAX_VCPU_ID)
2448 return -EINVAL;
2449
2450 mutex_lock(&kvm->lock);
2451 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2452 mutex_unlock(&kvm->lock);
2453 return -EINVAL;
2454 }
2455
2456 kvm->created_vcpus++;
2457 mutex_unlock(&kvm->lock);
2458
2459 vcpu = kvm_arch_vcpu_create(kvm, id);
2460 if (IS_ERR(vcpu)) {
2461 r = PTR_ERR(vcpu);
2462 goto vcpu_decrement;
2463 }
2464
2465 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2466
2467 r = kvm_arch_vcpu_setup(vcpu);
2468 if (r)
2469 goto vcpu_destroy;
2470
2471 r = kvm_create_vcpu_debugfs(vcpu);
2472 if (r)
2473 goto vcpu_destroy;
2474
2475 mutex_lock(&kvm->lock);
2476 if (kvm_get_vcpu_by_id(kvm, id)) {
2477 r = -EEXIST;
2478 goto unlock_vcpu_destroy;
2479 }
2480
2481 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2482
2483 /* Now it's all set up, let userspace reach it */
2484 kvm_get_kvm(kvm);
2485 r = create_vcpu_fd(vcpu);
2486 if (r < 0) {
2487 kvm_put_kvm(kvm);
2488 goto unlock_vcpu_destroy;
2489 }
2490
2491 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2492
2493 /*
2494 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2495 * before kvm->online_vcpu's incremented value.
2496 */
2497 smp_wmb();
2498 atomic_inc(&kvm->online_vcpus);
2499
2500 mutex_unlock(&kvm->lock);
2501 kvm_arch_vcpu_postcreate(vcpu);
2502 return r;
2503
2504unlock_vcpu_destroy:
2505 mutex_unlock(&kvm->lock);
2506 debugfs_remove_recursive(vcpu->debugfs_dentry);
2507vcpu_destroy:
2508 kvm_arch_vcpu_destroy(vcpu);
2509vcpu_decrement:
2510 mutex_lock(&kvm->lock);
2511 kvm->created_vcpus--;
2512 mutex_unlock(&kvm->lock);
2513 return r;
2514}
2515
2516static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2517{
2518 if (sigset) {
2519 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2520 vcpu->sigset_active = 1;
2521 vcpu->sigset = *sigset;
2522 } else
2523 vcpu->sigset_active = 0;
2524 return 0;
2525}
2526
2527static long kvm_vcpu_ioctl(struct file *filp,
2528 unsigned int ioctl, unsigned long arg)
2529{
2530 struct kvm_vcpu *vcpu = filp->private_data;
2531 void __user *argp = (void __user *)arg;
2532 int r;
2533 struct kvm_fpu *fpu = NULL;
2534 struct kvm_sregs *kvm_sregs = NULL;
2535
2536 if (vcpu->kvm->mm != current->mm)
2537 return -EIO;
2538
2539 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2540 return -EINVAL;
2541
2542#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2543 /*
2544 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2545 * so vcpu_load() would break it.
2546 */
2547 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2548 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2549#endif
2550
2551
2552 r = vcpu_load(vcpu);
2553 if (r)
2554 return r;
2555 switch (ioctl) {
2556 case KVM_RUN:
2557 r = -EINVAL;
2558 if (arg)
2559 goto out;
2560 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2561 /* The thread running this VCPU changed. */
2562 struct pid *oldpid = vcpu->pid;
2563 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2564
2565 rcu_assign_pointer(vcpu->pid, newpid);
2566 if (oldpid)
2567 synchronize_rcu();
2568 put_pid(oldpid);
2569 }
2570 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2571 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2572 break;
2573 case KVM_GET_REGS: {
2574 struct kvm_regs *kvm_regs;
2575
2576 r = -ENOMEM;
2577 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2578 if (!kvm_regs)
2579 goto out;
2580 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2581 if (r)
2582 goto out_free1;
2583 r = -EFAULT;
2584 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2585 goto out_free1;
2586 r = 0;
2587out_free1:
2588 kfree(kvm_regs);
2589 break;
2590 }
2591 case KVM_SET_REGS: {
2592 struct kvm_regs *kvm_regs;
2593
2594 r = -ENOMEM;
2595 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2596 if (IS_ERR(kvm_regs)) {
2597 r = PTR_ERR(kvm_regs);
2598 goto out;
2599 }
2600 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2601 kfree(kvm_regs);
2602 break;
2603 }
2604 case KVM_GET_SREGS: {
2605 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2606 r = -ENOMEM;
2607 if (!kvm_sregs)
2608 goto out;
2609 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2610 if (r)
2611 goto out;
2612 r = -EFAULT;
2613 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2614 goto out;
2615 r = 0;
2616 break;
2617 }
2618 case KVM_SET_SREGS: {
2619 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2620 if (IS_ERR(kvm_sregs)) {
2621 r = PTR_ERR(kvm_sregs);
2622 kvm_sregs = NULL;
2623 goto out;
2624 }
2625 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2626 break;
2627 }
2628 case KVM_GET_MP_STATE: {
2629 struct kvm_mp_state mp_state;
2630
2631 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2632 if (r)
2633 goto out;
2634 r = -EFAULT;
2635 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2636 goto out;
2637 r = 0;
2638 break;
2639 }
2640 case KVM_SET_MP_STATE: {
2641 struct kvm_mp_state mp_state;
2642
2643 r = -EFAULT;
2644 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2645 goto out;
2646 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2647 break;
2648 }
2649 case KVM_TRANSLATE: {
2650 struct kvm_translation tr;
2651
2652 r = -EFAULT;
2653 if (copy_from_user(&tr, argp, sizeof(tr)))
2654 goto out;
2655 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2656 if (r)
2657 goto out;
2658 r = -EFAULT;
2659 if (copy_to_user(argp, &tr, sizeof(tr)))
2660 goto out;
2661 r = 0;
2662 break;
2663 }
2664 case KVM_SET_GUEST_DEBUG: {
2665 struct kvm_guest_debug dbg;
2666
2667 r = -EFAULT;
2668 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2669 goto out;
2670 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2671 break;
2672 }
2673 case KVM_SET_SIGNAL_MASK: {
2674 struct kvm_signal_mask __user *sigmask_arg = argp;
2675 struct kvm_signal_mask kvm_sigmask;
2676 sigset_t sigset, *p;
2677
2678 p = NULL;
2679 if (argp) {
2680 r = -EFAULT;
2681 if (copy_from_user(&kvm_sigmask, argp,
2682 sizeof(kvm_sigmask)))
2683 goto out;
2684 r = -EINVAL;
2685 if (kvm_sigmask.len != sizeof(sigset))
2686 goto out;
2687 r = -EFAULT;
2688 if (copy_from_user(&sigset, sigmask_arg->sigset,
2689 sizeof(sigset)))
2690 goto out;
2691 p = &sigset;
2692 }
2693 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2694 break;
2695 }
2696 case KVM_GET_FPU: {
2697 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2698 r = -ENOMEM;
2699 if (!fpu)
2700 goto out;
2701 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2702 if (r)
2703 goto out;
2704 r = -EFAULT;
2705 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2706 goto out;
2707 r = 0;
2708 break;
2709 }
2710 case KVM_SET_FPU: {
2711 fpu = memdup_user(argp, sizeof(*fpu));
2712 if (IS_ERR(fpu)) {
2713 r = PTR_ERR(fpu);
2714 fpu = NULL;
2715 goto out;
2716 }
2717 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2718 break;
2719 }
2720 default:
2721 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2722 }
2723out:
2724 vcpu_put(vcpu);
2725 kfree(fpu);
2726 kfree(kvm_sregs);
2727 return r;
2728}
2729
2730#ifdef CONFIG_KVM_COMPAT
2731static long kvm_vcpu_compat_ioctl(struct file *filp,
2732 unsigned int ioctl, unsigned long arg)
2733{
2734 struct kvm_vcpu *vcpu = filp->private_data;
2735 void __user *argp = compat_ptr(arg);
2736 int r;
2737
2738 if (vcpu->kvm->mm != current->mm)
2739 return -EIO;
2740
2741 switch (ioctl) {
2742 case KVM_SET_SIGNAL_MASK: {
2743 struct kvm_signal_mask __user *sigmask_arg = argp;
2744 struct kvm_signal_mask kvm_sigmask;
2745 compat_sigset_t csigset;
2746 sigset_t sigset;
2747
2748 if (argp) {
2749 r = -EFAULT;
2750 if (copy_from_user(&kvm_sigmask, argp,
2751 sizeof(kvm_sigmask)))
2752 goto out;
2753 r = -EINVAL;
2754 if (kvm_sigmask.len != sizeof(csigset))
2755 goto out;
2756 r = -EFAULT;
2757 if (copy_from_user(&csigset, sigmask_arg->sigset,
2758 sizeof(csigset)))
2759 goto out;
2760 sigset_from_compat(&sigset, &csigset);
2761 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2762 } else
2763 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2764 break;
2765 }
2766 default:
2767 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2768 }
2769
2770out:
2771 return r;
2772}
2773#endif
2774
2775static int kvm_device_ioctl_attr(struct kvm_device *dev,
2776 int (*accessor)(struct kvm_device *dev,
2777 struct kvm_device_attr *attr),
2778 unsigned long arg)
2779{
2780 struct kvm_device_attr attr;
2781
2782 if (!accessor)
2783 return -EPERM;
2784
2785 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2786 return -EFAULT;
2787
2788 return accessor(dev, &attr);
2789}
2790
2791static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2792 unsigned long arg)
2793{
2794 struct kvm_device *dev = filp->private_data;
2795
2796 switch (ioctl) {
2797 case KVM_SET_DEVICE_ATTR:
2798 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2799 case KVM_GET_DEVICE_ATTR:
2800 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2801 case KVM_HAS_DEVICE_ATTR:
2802 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2803 default:
2804 if (dev->ops->ioctl)
2805 return dev->ops->ioctl(dev, ioctl, arg);
2806
2807 return -ENOTTY;
2808 }
2809}
2810
2811static int kvm_device_release(struct inode *inode, struct file *filp)
2812{
2813 struct kvm_device *dev = filp->private_data;
2814 struct kvm *kvm = dev->kvm;
2815
2816 kvm_put_kvm(kvm);
2817 return 0;
2818}
2819
2820static const struct file_operations kvm_device_fops = {
2821 .unlocked_ioctl = kvm_device_ioctl,
2822#ifdef CONFIG_KVM_COMPAT
2823 .compat_ioctl = kvm_device_ioctl,
2824#endif
2825 .release = kvm_device_release,
2826};
2827
2828struct kvm_device *kvm_device_from_filp(struct file *filp)
2829{
2830 if (filp->f_op != &kvm_device_fops)
2831 return NULL;
2832
2833 return filp->private_data;
2834}
2835
2836static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2837#ifdef CONFIG_KVM_MPIC
2838 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2839 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2840#endif
2841
2842#ifdef CONFIG_KVM_XICS
2843 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2844#endif
2845};
2846
2847int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2848{
2849 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2850 return -ENOSPC;
2851
2852 if (kvm_device_ops_table[type] != NULL)
2853 return -EEXIST;
2854
2855 kvm_device_ops_table[type] = ops;
2856 return 0;
2857}
2858
2859void kvm_unregister_device_ops(u32 type)
2860{
2861 if (kvm_device_ops_table[type] != NULL)
2862 kvm_device_ops_table[type] = NULL;
2863}
2864
2865static int kvm_ioctl_create_device(struct kvm *kvm,
2866 struct kvm_create_device *cd)
2867{
2868 struct kvm_device_ops *ops = NULL;
2869 struct kvm_device *dev;
2870 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2871 int ret;
2872
2873 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2874 return -ENODEV;
2875
2876 ops = kvm_device_ops_table[cd->type];
2877 if (ops == NULL)
2878 return -ENODEV;
2879
2880 if (test)
2881 return 0;
2882
2883 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2884 if (!dev)
2885 return -ENOMEM;
2886
2887 dev->ops = ops;
2888 dev->kvm = kvm;
2889
2890 mutex_lock(&kvm->lock);
2891 ret = ops->create(dev, cd->type);
2892 if (ret < 0) {
2893 mutex_unlock(&kvm->lock);
2894 kfree(dev);
2895 return ret;
2896 }
2897 list_add(&dev->vm_node, &kvm->devices);
2898 mutex_unlock(&kvm->lock);
2899
2900 if (ops->init)
2901 ops->init(dev);
2902
2903 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2904 if (ret < 0) {
2905 mutex_lock(&kvm->lock);
2906 list_del(&dev->vm_node);
2907 mutex_unlock(&kvm->lock);
2908 ops->destroy(dev);
2909 return ret;
2910 }
2911
2912 kvm_get_kvm(kvm);
2913 cd->fd = ret;
2914 return 0;
2915}
2916
2917static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2918{
2919 switch (arg) {
2920 case KVM_CAP_USER_MEMORY:
2921 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2922 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2923 case KVM_CAP_INTERNAL_ERROR_DATA:
2924#ifdef CONFIG_HAVE_KVM_MSI
2925 case KVM_CAP_SIGNAL_MSI:
2926#endif
2927#ifdef CONFIG_HAVE_KVM_IRQFD
2928 case KVM_CAP_IRQFD:
2929 case KVM_CAP_IRQFD_RESAMPLE:
2930#endif
2931 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2932 case KVM_CAP_CHECK_EXTENSION_VM:
2933 return 1;
2934#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2935 case KVM_CAP_IRQ_ROUTING:
2936 return KVM_MAX_IRQ_ROUTES;
2937#endif
2938#if KVM_ADDRESS_SPACE_NUM > 1
2939 case KVM_CAP_MULTI_ADDRESS_SPACE:
2940 return KVM_ADDRESS_SPACE_NUM;
2941#endif
2942 case KVM_CAP_MAX_VCPU_ID:
2943 return KVM_MAX_VCPU_ID;
2944 default:
2945 break;
2946 }
2947 return kvm_vm_ioctl_check_extension(kvm, arg);
2948}
2949
2950static long kvm_vm_ioctl(struct file *filp,
2951 unsigned int ioctl, unsigned long arg)
2952{
2953 struct kvm *kvm = filp->private_data;
2954 void __user *argp = (void __user *)arg;
2955 int r;
2956
2957 if (kvm->mm != current->mm)
2958 return -EIO;
2959 switch (ioctl) {
2960 case KVM_CREATE_VCPU:
2961 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2962 break;
2963 case KVM_SET_USER_MEMORY_REGION: {
2964 struct kvm_userspace_memory_region kvm_userspace_mem;
2965
2966 r = -EFAULT;
2967 if (copy_from_user(&kvm_userspace_mem, argp,
2968 sizeof(kvm_userspace_mem)))
2969 goto out;
2970
2971 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2972 break;
2973 }
2974 case KVM_GET_DIRTY_LOG: {
2975 struct kvm_dirty_log log;
2976
2977 r = -EFAULT;
2978 if (copy_from_user(&log, argp, sizeof(log)))
2979 goto out;
2980 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2981 break;
2982 }
2983#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2984 case KVM_REGISTER_COALESCED_MMIO: {
2985 struct kvm_coalesced_mmio_zone zone;
2986
2987 r = -EFAULT;
2988 if (copy_from_user(&zone, argp, sizeof(zone)))
2989 goto out;
2990 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2991 break;
2992 }
2993 case KVM_UNREGISTER_COALESCED_MMIO: {
2994 struct kvm_coalesced_mmio_zone zone;
2995
2996 r = -EFAULT;
2997 if (copy_from_user(&zone, argp, sizeof(zone)))
2998 goto out;
2999 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3000 break;
3001 }
3002#endif
3003 case KVM_IRQFD: {
3004 struct kvm_irqfd data;
3005
3006 r = -EFAULT;
3007 if (copy_from_user(&data, argp, sizeof(data)))
3008 goto out;
3009 r = kvm_irqfd(kvm, &data);
3010 break;
3011 }
3012 case KVM_IOEVENTFD: {
3013 struct kvm_ioeventfd data;
3014
3015 r = -EFAULT;
3016 if (copy_from_user(&data, argp, sizeof(data)))
3017 goto out;
3018 r = kvm_ioeventfd(kvm, &data);
3019 break;
3020 }
3021#ifdef CONFIG_HAVE_KVM_MSI
3022 case KVM_SIGNAL_MSI: {
3023 struct kvm_msi msi;
3024
3025 r = -EFAULT;
3026 if (copy_from_user(&msi, argp, sizeof(msi)))
3027 goto out;
3028 r = kvm_send_userspace_msi(kvm, &msi);
3029 break;
3030 }
3031#endif
3032#ifdef __KVM_HAVE_IRQ_LINE
3033 case KVM_IRQ_LINE_STATUS:
3034 case KVM_IRQ_LINE: {
3035 struct kvm_irq_level irq_event;
3036
3037 r = -EFAULT;
3038 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3039 goto out;
3040
3041 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3042 ioctl == KVM_IRQ_LINE_STATUS);
3043 if (r)
3044 goto out;
3045
3046 r = -EFAULT;
3047 if (ioctl == KVM_IRQ_LINE_STATUS) {
3048 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3049 goto out;
3050 }
3051
3052 r = 0;
3053 break;
3054 }
3055#endif
3056#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3057 case KVM_SET_GSI_ROUTING: {
3058 struct kvm_irq_routing routing;
3059 struct kvm_irq_routing __user *urouting;
3060 struct kvm_irq_routing_entry *entries = NULL;
3061
3062 r = -EFAULT;
3063 if (copy_from_user(&routing, argp, sizeof(routing)))
3064 goto out;
3065 r = -EINVAL;
3066 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3067 goto out;
3068 if (routing.flags)
3069 goto out;
3070 if (routing.nr) {
3071 r = -ENOMEM;
3072 entries = vmalloc(routing.nr * sizeof(*entries));
3073 if (!entries)
3074 goto out;
3075 r = -EFAULT;
3076 urouting = argp;
3077 if (copy_from_user(entries, urouting->entries,
3078 routing.nr * sizeof(*entries)))
3079 goto out_free_irq_routing;
3080 }
3081 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3082 routing.flags);
3083out_free_irq_routing:
3084 vfree(entries);
3085 break;
3086 }
3087#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3088 case KVM_CREATE_DEVICE: {
3089 struct kvm_create_device cd;
3090
3091 r = -EFAULT;
3092 if (copy_from_user(&cd, argp, sizeof(cd)))
3093 goto out;
3094
3095 r = kvm_ioctl_create_device(kvm, &cd);
3096 if (r)
3097 goto out;
3098
3099 r = -EFAULT;
3100 if (copy_to_user(argp, &cd, sizeof(cd)))
3101 goto out;
3102
3103 r = 0;
3104 break;
3105 }
3106 case KVM_CHECK_EXTENSION:
3107 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3108 break;
3109 default:
3110 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3111 }
3112out:
3113 return r;
3114}
3115
3116#ifdef CONFIG_KVM_COMPAT
3117struct compat_kvm_dirty_log {
3118 __u32 slot;
3119 __u32 padding1;
3120 union {
3121 compat_uptr_t dirty_bitmap; /* one bit per page */
3122 __u64 padding2;
3123 };
3124};
3125
3126static long kvm_vm_compat_ioctl(struct file *filp,
3127 unsigned int ioctl, unsigned long arg)
3128{
3129 struct kvm *kvm = filp->private_data;
3130 int r;
3131
3132 if (kvm->mm != current->mm)
3133 return -EIO;
3134 switch (ioctl) {
3135 case KVM_GET_DIRTY_LOG: {
3136 struct compat_kvm_dirty_log compat_log;
3137 struct kvm_dirty_log log;
3138
3139 r = -EFAULT;
3140 if (copy_from_user(&compat_log, (void __user *)arg,
3141 sizeof(compat_log)))
3142 goto out;
3143 log.slot = compat_log.slot;
3144 log.padding1 = compat_log.padding1;
3145 log.padding2 = compat_log.padding2;
3146 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3147
3148 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3149 break;
3150 }
3151 default:
3152 r = kvm_vm_ioctl(filp, ioctl, arg);
3153 }
3154
3155out:
3156 return r;
3157}
3158#endif
3159
3160static struct file_operations kvm_vm_fops = {
3161 .release = kvm_vm_release,
3162 .unlocked_ioctl = kvm_vm_ioctl,
3163#ifdef CONFIG_KVM_COMPAT
3164 .compat_ioctl = kvm_vm_compat_ioctl,
3165#endif
3166 .llseek = noop_llseek,
3167};
3168
3169static int kvm_dev_ioctl_create_vm(unsigned long type)
3170{
3171 int r;
3172 struct kvm *kvm;
3173 struct file *file;
3174
3175 kvm = kvm_create_vm(type);
3176 if (IS_ERR(kvm))
3177 return PTR_ERR(kvm);
3178#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3179 r = kvm_coalesced_mmio_init(kvm);
3180 if (r < 0) {
3181 kvm_put_kvm(kvm);
3182 return r;
3183 }
3184#endif
3185 r = get_unused_fd_flags(O_CLOEXEC);
3186 if (r < 0) {
3187 kvm_put_kvm(kvm);
3188 return r;
3189 }
3190 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3191 if (IS_ERR(file)) {
3192 put_unused_fd(r);
3193 kvm_put_kvm(kvm);
3194 return PTR_ERR(file);
3195 }
3196
3197 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3198 put_unused_fd(r);
3199 fput(file);
3200 return -ENOMEM;
3201 }
3202
3203 fd_install(r, file);
3204 return r;
3205}
3206
3207static long kvm_dev_ioctl(struct file *filp,
3208 unsigned int ioctl, unsigned long arg)
3209{
3210 long r = -EINVAL;
3211
3212 switch (ioctl) {
3213 case KVM_GET_API_VERSION:
3214 if (arg)
3215 goto out;
3216 r = KVM_API_VERSION;
3217 break;
3218 case KVM_CREATE_VM:
3219 r = kvm_dev_ioctl_create_vm(arg);
3220 break;
3221 case KVM_CHECK_EXTENSION:
3222 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3223 break;
3224 case KVM_GET_VCPU_MMAP_SIZE:
3225 if (arg)
3226 goto out;
3227 r = PAGE_SIZE; /* struct kvm_run */
3228#ifdef CONFIG_X86
3229 r += PAGE_SIZE; /* pio data page */
3230#endif
3231#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3232 r += PAGE_SIZE; /* coalesced mmio ring page */
3233#endif
3234 break;
3235 case KVM_TRACE_ENABLE:
3236 case KVM_TRACE_PAUSE:
3237 case KVM_TRACE_DISABLE:
3238 r = -EOPNOTSUPP;
3239 break;
3240 default:
3241 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3242 }
3243out:
3244 return r;
3245}
3246
3247static struct file_operations kvm_chardev_ops = {
3248 .unlocked_ioctl = kvm_dev_ioctl,
3249 .compat_ioctl = kvm_dev_ioctl,
3250 .llseek = noop_llseek,
3251};
3252
3253static struct miscdevice kvm_dev = {
3254 KVM_MINOR,
3255 "kvm",
3256 &kvm_chardev_ops,
3257};
3258
3259static void hardware_enable_nolock(void *junk)
3260{
3261 int cpu = raw_smp_processor_id();
3262 int r;
3263
3264 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3265 return;
3266
3267 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3268
3269 r = kvm_arch_hardware_enable();
3270
3271 if (r) {
3272 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3273 atomic_inc(&hardware_enable_failed);
3274 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3275 }
3276}
3277
3278static int kvm_starting_cpu(unsigned int cpu)
3279{
3280 raw_spin_lock(&kvm_count_lock);
3281 if (kvm_usage_count)
3282 hardware_enable_nolock(NULL);
3283 raw_spin_unlock(&kvm_count_lock);
3284 return 0;
3285}
3286
3287static void hardware_disable_nolock(void *junk)
3288{
3289 int cpu = raw_smp_processor_id();
3290
3291 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3292 return;
3293 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3294 kvm_arch_hardware_disable();
3295}
3296
3297static int kvm_dying_cpu(unsigned int cpu)
3298{
3299 raw_spin_lock(&kvm_count_lock);
3300 if (kvm_usage_count)
3301 hardware_disable_nolock(NULL);
3302 raw_spin_unlock(&kvm_count_lock);
3303 return 0;
3304}
3305
3306static void hardware_disable_all_nolock(void)
3307{
3308 BUG_ON(!kvm_usage_count);
3309
3310 kvm_usage_count--;
3311 if (!kvm_usage_count)
3312 on_each_cpu(hardware_disable_nolock, NULL, 1);
3313}
3314
3315static void hardware_disable_all(void)
3316{
3317 raw_spin_lock(&kvm_count_lock);
3318 hardware_disable_all_nolock();
3319 raw_spin_unlock(&kvm_count_lock);
3320}
3321
3322static int hardware_enable_all(void)
3323{
3324 int r = 0;
3325
3326 raw_spin_lock(&kvm_count_lock);
3327
3328 kvm_usage_count++;
3329 if (kvm_usage_count == 1) {
3330 atomic_set(&hardware_enable_failed, 0);
3331 on_each_cpu(hardware_enable_nolock, NULL, 1);
3332
3333 if (atomic_read(&hardware_enable_failed)) {
3334 hardware_disable_all_nolock();
3335 r = -EBUSY;
3336 }
3337 }
3338
3339 raw_spin_unlock(&kvm_count_lock);
3340
3341 return r;
3342}
3343
3344static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3345 void *v)
3346{
3347 /*
3348 * Some (well, at least mine) BIOSes hang on reboot if
3349 * in vmx root mode.
3350 *
3351 * And Intel TXT required VMX off for all cpu when system shutdown.
3352 */
3353 pr_info("kvm: exiting hardware virtualization\n");
3354 kvm_rebooting = true;
3355 on_each_cpu(hardware_disable_nolock, NULL, 1);
3356 return NOTIFY_OK;
3357}
3358
3359static struct notifier_block kvm_reboot_notifier = {
3360 .notifier_call = kvm_reboot,
3361 .priority = 0,
3362};
3363
3364static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3365{
3366 int i;
3367
3368 for (i = 0; i < bus->dev_count; i++) {
3369 struct kvm_io_device *pos = bus->range[i].dev;
3370
3371 kvm_iodevice_destructor(pos);
3372 }
3373 kfree(bus);
3374}
3375
3376static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3377 const struct kvm_io_range *r2)
3378{
3379 gpa_t addr1 = r1->addr;
3380 gpa_t addr2 = r2->addr;
3381
3382 if (addr1 < addr2)
3383 return -1;
3384
3385 /* If r2->len == 0, match the exact address. If r2->len != 0,
3386 * accept any overlapping write. Any order is acceptable for
3387 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3388 * we process all of them.
3389 */
3390 if (r2->len) {
3391 addr1 += r1->len;
3392 addr2 += r2->len;
3393 }
3394
3395 if (addr1 > addr2)
3396 return 1;
3397
3398 return 0;
3399}
3400
3401static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3402{
3403 return kvm_io_bus_cmp(p1, p2);
3404}
3405
3406static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3407 gpa_t addr, int len)
3408{
3409 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3410 .addr = addr,
3411 .len = len,
3412 .dev = dev,
3413 };
3414
3415 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3416 kvm_io_bus_sort_cmp, NULL);
3417
3418 return 0;
3419}
3420
3421static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3422 gpa_t addr, int len)
3423{
3424 struct kvm_io_range *range, key;
3425 int off;
3426
3427 key = (struct kvm_io_range) {
3428 .addr = addr,
3429 .len = len,
3430 };
3431
3432 range = bsearch(&key, bus->range, bus->dev_count,
3433 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3434 if (range == NULL)
3435 return -ENOENT;
3436
3437 off = range - bus->range;
3438
3439 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3440 off--;
3441
3442 return off;
3443}
3444
3445static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3446 struct kvm_io_range *range, const void *val)
3447{
3448 int idx;
3449
3450 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3451 if (idx < 0)
3452 return -EOPNOTSUPP;
3453
3454 while (idx < bus->dev_count &&
3455 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3456 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3457 range->len, val))
3458 return idx;
3459 idx++;
3460 }
3461
3462 return -EOPNOTSUPP;
3463}
3464
3465/* kvm_io_bus_write - called under kvm->slots_lock */
3466int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3467 int len, const void *val)
3468{
3469 struct kvm_io_bus *bus;
3470 struct kvm_io_range range;
3471 int r;
3472
3473 range = (struct kvm_io_range) {
3474 .addr = addr,
3475 .len = len,
3476 };
3477
3478 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3479 if (!bus)
3480 return -ENOMEM;
3481 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3482 return r < 0 ? r : 0;
3483}
3484
3485/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3486int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3487 gpa_t addr, int len, const void *val, long cookie)
3488{
3489 struct kvm_io_bus *bus;
3490 struct kvm_io_range range;
3491
3492 range = (struct kvm_io_range) {
3493 .addr = addr,
3494 .len = len,
3495 };
3496
3497 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3498 if (!bus)
3499 return -ENOMEM;
3500
3501 /* First try the device referenced by cookie. */
3502 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3503 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3504 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3505 val))
3506 return cookie;
3507
3508 /*
3509 * cookie contained garbage; fall back to search and return the
3510 * correct cookie value.
3511 */
3512 return __kvm_io_bus_write(vcpu, bus, &range, val);
3513}
3514
3515static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3516 struct kvm_io_range *range, void *val)
3517{
3518 int idx;
3519
3520 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3521 if (idx < 0)
3522 return -EOPNOTSUPP;
3523
3524 while (idx < bus->dev_count &&
3525 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3526 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3527 range->len, val))
3528 return idx;
3529 idx++;
3530 }
3531
3532 return -EOPNOTSUPP;
3533}
3534EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3535
3536/* kvm_io_bus_read - called under kvm->slots_lock */
3537int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3538 int len, void *val)
3539{
3540 struct kvm_io_bus *bus;
3541 struct kvm_io_range range;
3542 int r;
3543
3544 range = (struct kvm_io_range) {
3545 .addr = addr,
3546 .len = len,
3547 };
3548
3549 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3550 if (!bus)
3551 return -ENOMEM;
3552 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3553 return r < 0 ? r : 0;
3554}
3555
3556
3557/* Caller must hold slots_lock. */
3558int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3559 int len, struct kvm_io_device *dev)
3560{
3561 struct kvm_io_bus *new_bus, *bus;
3562
3563 bus = kvm->buses[bus_idx];
3564 if (!bus)
3565 return -ENOMEM;
3566
3567 /* exclude ioeventfd which is limited by maximum fd */
3568 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3569 return -ENOSPC;
3570
3571 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3572 sizeof(struct kvm_io_range)), GFP_KERNEL);
3573 if (!new_bus)
3574 return -ENOMEM;
3575 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3576 sizeof(struct kvm_io_range)));
3577 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3578 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3579 synchronize_srcu_expedited(&kvm->srcu);
3580 kfree(bus);
3581
3582 return 0;
3583}
3584
3585/* Caller must hold slots_lock. */
3586void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3587 struct kvm_io_device *dev)
3588{
3589 int i;
3590 struct kvm_io_bus *new_bus, *bus;
3591
3592 bus = kvm->buses[bus_idx];
3593 if (!bus)
3594 return;
3595
3596 for (i = 0; i < bus->dev_count; i++)
3597 if (bus->range[i].dev == dev) {
3598 break;
3599 }
3600
3601 if (i == bus->dev_count)
3602 return;
3603
3604 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3605 sizeof(struct kvm_io_range)), GFP_KERNEL);
3606 if (!new_bus) {
3607 pr_err("kvm: failed to shrink bus, removing it completely\n");
3608 goto broken;
3609 }
3610
3611 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3612 new_bus->dev_count--;
3613 memcpy(new_bus->range + i, bus->range + i + 1,
3614 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3615
3616broken:
3617 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3618 synchronize_srcu_expedited(&kvm->srcu);
3619 kfree(bus);
3620 return;
3621}
3622
3623struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3624 gpa_t addr)
3625{
3626 struct kvm_io_bus *bus;
3627 int dev_idx, srcu_idx;
3628 struct kvm_io_device *iodev = NULL;
3629
3630 srcu_idx = srcu_read_lock(&kvm->srcu);
3631
3632 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3633 if (!bus)
3634 goto out_unlock;
3635
3636 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3637 if (dev_idx < 0)
3638 goto out_unlock;
3639
3640 iodev = bus->range[dev_idx].dev;
3641
3642out_unlock:
3643 srcu_read_unlock(&kvm->srcu, srcu_idx);
3644
3645 return iodev;
3646}
3647EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3648
3649static int kvm_debugfs_open(struct inode *inode, struct file *file,
3650 int (*get)(void *, u64 *), int (*set)(void *, u64),
3651 const char *fmt)
3652{
3653 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3654 inode->i_private;
3655
3656 /* The debugfs files are a reference to the kvm struct which
3657 * is still valid when kvm_destroy_vm is called.
3658 * To avoid the race between open and the removal of the debugfs
3659 * directory we test against the users count.
3660 */
3661 if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3662 return -ENOENT;
3663
3664 if (simple_attr_open(inode, file, get, set, fmt)) {
3665 kvm_put_kvm(stat_data->kvm);
3666 return -ENOMEM;
3667 }
3668
3669 return 0;
3670}
3671
3672static int kvm_debugfs_release(struct inode *inode, struct file *file)
3673{
3674 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3675 inode->i_private;
3676
3677 simple_attr_release(inode, file);
3678 kvm_put_kvm(stat_data->kvm);
3679
3680 return 0;
3681}
3682
3683static int vm_stat_get_per_vm(void *data, u64 *val)
3684{
3685 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3686
3687 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3688
3689 return 0;
3690}
3691
3692static int vm_stat_clear_per_vm(void *data, u64 val)
3693{
3694 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3695
3696 if (val)
3697 return -EINVAL;
3698
3699 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3700
3701 return 0;
3702}
3703
3704static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3705{
3706 __simple_attr_check_format("%llu\n", 0ull);
3707 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3708 vm_stat_clear_per_vm, "%llu\n");
3709}
3710
3711static const struct file_operations vm_stat_get_per_vm_fops = {
3712 .owner = THIS_MODULE,
3713 .open = vm_stat_get_per_vm_open,
3714 .release = kvm_debugfs_release,
3715 .read = simple_attr_read,
3716 .write = simple_attr_write,
3717 .llseek = generic_file_llseek,
3718};
3719
3720static int vcpu_stat_get_per_vm(void *data, u64 *val)
3721{
3722 int i;
3723 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3724 struct kvm_vcpu *vcpu;
3725
3726 *val = 0;
3727
3728 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3729 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3730
3731 return 0;
3732}
3733
3734static int vcpu_stat_clear_per_vm(void *data, u64 val)
3735{
3736 int i;
3737 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3738 struct kvm_vcpu *vcpu;
3739
3740 if (val)
3741 return -EINVAL;
3742
3743 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3744 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3745
3746 return 0;
3747}
3748
3749static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3750{
3751 __simple_attr_check_format("%llu\n", 0ull);
3752 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3753 vcpu_stat_clear_per_vm, "%llu\n");
3754}
3755
3756static const struct file_operations vcpu_stat_get_per_vm_fops = {
3757 .owner = THIS_MODULE,
3758 .open = vcpu_stat_get_per_vm_open,
3759 .release = kvm_debugfs_release,
3760 .read = simple_attr_read,
3761 .write = simple_attr_write,
3762 .llseek = generic_file_llseek,
3763};
3764
3765static const struct file_operations *stat_fops_per_vm[] = {
3766 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3767 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3768};
3769
3770static int vm_stat_get(void *_offset, u64 *val)
3771{
3772 unsigned offset = (long)_offset;
3773 struct kvm *kvm;
3774 struct kvm_stat_data stat_tmp = {.offset = offset};
3775 u64 tmp_val;
3776
3777 *val = 0;
3778 spin_lock(&kvm_lock);
3779 list_for_each_entry(kvm, &vm_list, vm_list) {
3780 stat_tmp.kvm = kvm;
3781 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3782 *val += tmp_val;
3783 }
3784 spin_unlock(&kvm_lock);
3785 return 0;
3786}
3787
3788static int vm_stat_clear(void *_offset, u64 val)
3789{
3790 unsigned offset = (long)_offset;
3791 struct kvm *kvm;
3792 struct kvm_stat_data stat_tmp = {.offset = offset};
3793
3794 if (val)
3795 return -EINVAL;
3796
3797 spin_lock(&kvm_lock);
3798 list_for_each_entry(kvm, &vm_list, vm_list) {
3799 stat_tmp.kvm = kvm;
3800 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3801 }
3802 spin_unlock(&kvm_lock);
3803
3804 return 0;
3805}
3806
3807DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3808
3809static int vcpu_stat_get(void *_offset, u64 *val)
3810{
3811 unsigned offset = (long)_offset;
3812 struct kvm *kvm;
3813 struct kvm_stat_data stat_tmp = {.offset = offset};
3814 u64 tmp_val;
3815
3816 *val = 0;
3817 spin_lock(&kvm_lock);
3818 list_for_each_entry(kvm, &vm_list, vm_list) {
3819 stat_tmp.kvm = kvm;
3820 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3821 *val += tmp_val;
3822 }
3823 spin_unlock(&kvm_lock);
3824 return 0;
3825}
3826
3827static int vcpu_stat_clear(void *_offset, u64 val)
3828{
3829 unsigned offset = (long)_offset;
3830 struct kvm *kvm;
3831 struct kvm_stat_data stat_tmp = {.offset = offset};
3832
3833 if (val)
3834 return -EINVAL;
3835
3836 spin_lock(&kvm_lock);
3837 list_for_each_entry(kvm, &vm_list, vm_list) {
3838 stat_tmp.kvm = kvm;
3839 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3840 }
3841 spin_unlock(&kvm_lock);
3842
3843 return 0;
3844}
3845
3846DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3847 "%llu\n");
3848
3849static const struct file_operations *stat_fops[] = {
3850 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3851 [KVM_STAT_VM] = &vm_stat_fops,
3852};
3853
3854static int kvm_init_debug(void)
3855{
3856 int r = -EEXIST;
3857 struct kvm_stats_debugfs_item *p;
3858
3859 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3860 if (kvm_debugfs_dir == NULL)
3861 goto out;
3862
3863 kvm_debugfs_num_entries = 0;
3864 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3865 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3866 (void *)(long)p->offset,
3867 stat_fops[p->kind]))
3868 goto out_dir;
3869 }
3870
3871 return 0;
3872
3873out_dir:
3874 debugfs_remove_recursive(kvm_debugfs_dir);
3875out:
3876 return r;
3877}
3878
3879static int kvm_suspend(void)
3880{
3881 if (kvm_usage_count)
3882 hardware_disable_nolock(NULL);
3883 return 0;
3884}
3885
3886static void kvm_resume(void)
3887{
3888 if (kvm_usage_count) {
3889 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3890 hardware_enable_nolock(NULL);
3891 }
3892}
3893
3894static struct syscore_ops kvm_syscore_ops = {
3895 .suspend = kvm_suspend,
3896 .resume = kvm_resume,
3897};
3898
3899static inline
3900struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3901{
3902 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3903}
3904
3905static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3906{
3907 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3908
3909 if (vcpu->preempted)
3910 vcpu->preempted = false;
3911
3912 kvm_arch_sched_in(vcpu, cpu);
3913
3914 kvm_arch_vcpu_load(vcpu, cpu);
3915}
3916
3917static void kvm_sched_out(struct preempt_notifier *pn,
3918 struct task_struct *next)
3919{
3920 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3921
3922 if (current->state == TASK_RUNNING)
3923 vcpu->preempted = true;
3924 kvm_arch_vcpu_put(vcpu);
3925}
3926
3927int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3928 struct module *module)
3929{
3930 int r;
3931 int cpu;
3932
3933 r = kvm_arch_init(opaque);
3934 if (r)
3935 goto out_fail;
3936
3937 /*
3938 * kvm_arch_init makes sure there's at most one caller
3939 * for architectures that support multiple implementations,
3940 * like intel and amd on x86.
3941 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3942 * conflicts in case kvm is already setup for another implementation.
3943 */
3944 r = kvm_irqfd_init();
3945 if (r)
3946 goto out_irqfd;
3947
3948 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3949 r = -ENOMEM;
3950 goto out_free_0;
3951 }
3952
3953 r = kvm_arch_hardware_setup();
3954 if (r < 0)
3955 goto out_free_0a;
3956
3957 for_each_online_cpu(cpu) {
3958 smp_call_function_single(cpu,
3959 kvm_arch_check_processor_compat,
3960 &r, 1);
3961 if (r < 0)
3962 goto out_free_1;
3963 }
3964
3965 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
3966 kvm_starting_cpu, kvm_dying_cpu);
3967 if (r)
3968 goto out_free_2;
3969 register_reboot_notifier(&kvm_reboot_notifier);
3970
3971 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3972 if (!vcpu_align)
3973 vcpu_align = __alignof__(struct kvm_vcpu);
3974 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3975 0, NULL);
3976 if (!kvm_vcpu_cache) {
3977 r = -ENOMEM;
3978 goto out_free_3;
3979 }
3980
3981 r = kvm_async_pf_init();
3982 if (r)
3983 goto out_free;
3984
3985 kvm_chardev_ops.owner = module;
3986 kvm_vm_fops.owner = module;
3987 kvm_vcpu_fops.owner = module;
3988
3989 r = misc_register(&kvm_dev);
3990 if (r) {
3991 pr_err("kvm: misc device register failed\n");
3992 goto out_unreg;
3993 }
3994
3995 register_syscore_ops(&kvm_syscore_ops);
3996
3997 kvm_preempt_ops.sched_in = kvm_sched_in;
3998 kvm_preempt_ops.sched_out = kvm_sched_out;
3999
4000 r = kvm_init_debug();
4001 if (r) {
4002 pr_err("kvm: create debugfs files failed\n");
4003 goto out_undebugfs;
4004 }
4005
4006 r = kvm_vfio_ops_init();
4007 WARN_ON(r);
4008
4009 return 0;
4010
4011out_undebugfs:
4012 unregister_syscore_ops(&kvm_syscore_ops);
4013 misc_deregister(&kvm_dev);
4014out_unreg:
4015 kvm_async_pf_deinit();
4016out_free:
4017 kmem_cache_destroy(kvm_vcpu_cache);
4018out_free_3:
4019 unregister_reboot_notifier(&kvm_reboot_notifier);
4020 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4021out_free_2:
4022out_free_1:
4023 kvm_arch_hardware_unsetup();
4024out_free_0a:
4025 free_cpumask_var(cpus_hardware_enabled);
4026out_free_0:
4027 kvm_irqfd_exit();
4028out_irqfd:
4029 kvm_arch_exit();
4030out_fail:
4031 return r;
4032}
4033EXPORT_SYMBOL_GPL(kvm_init);
4034
4035void kvm_exit(void)
4036{
4037 debugfs_remove_recursive(kvm_debugfs_dir);
4038 misc_deregister(&kvm_dev);
4039 kmem_cache_destroy(kvm_vcpu_cache);
4040 kvm_async_pf_deinit();
4041 unregister_syscore_ops(&kvm_syscore_ops);
4042 unregister_reboot_notifier(&kvm_reboot_notifier);
4043 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4044 on_each_cpu(hardware_disable_nolock, NULL, 1);
4045 kvm_arch_hardware_unsetup();
4046 kvm_arch_exit();
4047 kvm_irqfd_exit();
4048 free_cpumask_var(cpus_hardware_enabled);
4049 kvm_vfio_ops_exit();
4050}
4051EXPORT_SYMBOL_GPL(kvm_exit);