projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
kvm: search_memslots: add simple LRU memslot caching
[deliverable/linux.git] / virt / kvm / kvm_main.c
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 "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
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
68
69 /*
70 * Ordering of locks:
71 *
72 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 */
74
75 DEFINE_SPINLOCK(kvm_lock);
76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
77 LIST_HEAD(vm_list);
78
79 static cpumask_var_t cpus_hardware_enabled;
80 static int kvm_usage_count = 0;
81 static atomic_t hardware_enable_failed;
82
83 struct kmem_cache *kvm_vcpu_cache;
84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
85
86 static __read_mostly struct preempt_ops kvm_preempt_ops;
87
88 struct dentry *kvm_debugfs_dir;
89
90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92 #ifdef CONFIG_COMPAT
93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 unsigned long arg);
95 #endif
96 static int hardware_enable_all(void);
97 static void hardware_disable_all(void);
98
99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
100
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
104
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
107
108 static bool largepages_enabled = true;
109
110 bool kvm_is_mmio_pfn(pfn_t pfn)
111 {
112 if (pfn_valid(pfn))
113 return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn));
114
115 return true;
116 }
117
118 /*
119 * Switches to specified vcpu, until a matching vcpu_put()
120 */
121 int vcpu_load(struct kvm_vcpu *vcpu)
122 {
123 int cpu;
124
125 if (mutex_lock_killable(&vcpu->mutex))
126 return -EINTR;
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
132 if (oldpid)
133 synchronize_rcu();
134 put_pid(oldpid);
135 }
136 cpu = get_cpu();
137 preempt_notifier_register(&vcpu->preempt_notifier);
138 kvm_arch_vcpu_load(vcpu, cpu);
139 put_cpu();
140 return 0;
141 }
142
143 void vcpu_put(struct kvm_vcpu *vcpu)
144 {
145 preempt_disable();
146 kvm_arch_vcpu_put(vcpu);
147 preempt_notifier_unregister(&vcpu->preempt_notifier);
148 preempt_enable();
149 mutex_unlock(&vcpu->mutex);
150 }
151
152 static void ack_flush(void *_completed)
153 {
154 }
155
156 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
157 {
158 int i, cpu, me;
159 cpumask_var_t cpus;
160 bool called = true;
161 struct kvm_vcpu *vcpu;
162
163 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
164
165 me = get_cpu();
166 kvm_for_each_vcpu(i, vcpu, kvm) {
167 kvm_make_request(req, vcpu);
168 cpu = vcpu->cpu;
169
170 /* Set ->requests bit before we read ->mode */
171 smp_mb();
172
173 if (cpus != NULL && cpu != -1 && cpu != me &&
174 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
175 cpumask_set_cpu(cpu, cpus);
176 }
177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
181 else
182 called = false;
183 put_cpu();
184 free_cpumask_var(cpus);
185 return called;
186 }
187
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
189 {
190 long dirty_count = kvm->tlbs_dirty;
191
192 smp_mb();
193 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194 ++kvm->stat.remote_tlb_flush;
195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
196 }
197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
198
199 void kvm_reload_remote_mmus(struct kvm *kvm)
200 {
201 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
202 }
203
204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
205 {
206 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
207 }
208
209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
210 {
211 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
212 }
213
214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215 {
216 struct page *page;
217 int r;
218
219 mutex_init(&vcpu->mutex);
220 vcpu->cpu = -1;
221 vcpu->kvm = kvm;
222 vcpu->vcpu_id = id;
223 vcpu->pid = NULL;
224 init_waitqueue_head(&vcpu->wq);
225 kvm_async_pf_vcpu_init(vcpu);
226
227 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 if (!page) {
229 r = -ENOMEM;
230 goto fail;
231 }
232 vcpu->run = page_address(page);
233
234 kvm_vcpu_set_in_spin_loop(vcpu, false);
235 kvm_vcpu_set_dy_eligible(vcpu, false);
236 vcpu->preempted = false;
237
238 r = kvm_arch_vcpu_init(vcpu);
239 if (r < 0)
240 goto fail_free_run;
241 return 0;
242
243 fail_free_run:
244 free_page((unsigned long)vcpu->run);
245 fail:
246 return r;
247 }
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 {
252 put_pid(vcpu->pid);
253 kvm_arch_vcpu_uninit(vcpu);
254 free_page((unsigned long)vcpu->run);
255 }
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 {
261 return container_of(mn, struct kvm, mmu_notifier);
262 }
263
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
267 {
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
269 int need_tlb_flush, idx;
270
271 /*
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * before returning.
280 *
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
283 *
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
288 */
289 idx = srcu_read_lock(&kvm->srcu);
290 spin_lock(&kvm->mmu_lock);
291
292 kvm->mmu_notifier_seq++;
293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294 /* we've to flush the tlb before the pages can be freed */
295 if (need_tlb_flush)
296 kvm_flush_remote_tlbs(kvm);
297
298 spin_unlock(&kvm->mmu_lock);
299
300 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
301
302 srcu_read_unlock(&kvm->srcu, idx);
303 }
304
305 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
306 struct mm_struct *mm,
307 unsigned long address,
308 pte_t pte)
309 {
310 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 int idx;
312
313 idx = srcu_read_lock(&kvm->srcu);
314 spin_lock(&kvm->mmu_lock);
315 kvm->mmu_notifier_seq++;
316 kvm_set_spte_hva(kvm, address, pte);
317 spin_unlock(&kvm->mmu_lock);
318 srcu_read_unlock(&kvm->srcu, idx);
319 }
320
321 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
322 struct mm_struct *mm,
323 unsigned long start,
324 unsigned long end)
325 {
326 struct kvm *kvm = mmu_notifier_to_kvm(mn);
327 int need_tlb_flush = 0, idx;
328
329 idx = srcu_read_lock(&kvm->srcu);
330 spin_lock(&kvm->mmu_lock);
331 /*
332 * The count increase must become visible at unlock time as no
333 * spte can be established without taking the mmu_lock and
334 * count is also read inside the mmu_lock critical section.
335 */
336 kvm->mmu_notifier_count++;
337 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
338 need_tlb_flush |= kvm->tlbs_dirty;
339 /* we've to flush the tlb before the pages can be freed */
340 if (need_tlb_flush)
341 kvm_flush_remote_tlbs(kvm);
342
343 spin_unlock(&kvm->mmu_lock);
344 srcu_read_unlock(&kvm->srcu, idx);
345 }
346
347 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
348 struct mm_struct *mm,
349 unsigned long start,
350 unsigned long end)
351 {
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
353
354 spin_lock(&kvm->mmu_lock);
355 /*
356 * This sequence increase will notify the kvm page fault that
357 * the page that is going to be mapped in the spte could have
358 * been freed.
359 */
360 kvm->mmu_notifier_seq++;
361 smp_wmb();
362 /*
363 * The above sequence increase must be visible before the
364 * below count decrease, which is ensured by the smp_wmb above
365 * in conjunction with the smp_rmb in mmu_notifier_retry().
366 */
367 kvm->mmu_notifier_count--;
368 spin_unlock(&kvm->mmu_lock);
369
370 BUG_ON(kvm->mmu_notifier_count < 0);
371 }
372
373 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
374 struct mm_struct *mm,
375 unsigned long start,
376 unsigned long end)
377 {
378 struct kvm *kvm = mmu_notifier_to_kvm(mn);
379 int young, idx;
380
381 idx = srcu_read_lock(&kvm->srcu);
382 spin_lock(&kvm->mmu_lock);
383
384 young = kvm_age_hva(kvm, start, end);
385 if (young)
386 kvm_flush_remote_tlbs(kvm);
387
388 spin_unlock(&kvm->mmu_lock);
389 srcu_read_unlock(&kvm->srcu, idx);
390
391 return young;
392 }
393
394 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
395 struct mm_struct *mm,
396 unsigned long address)
397 {
398 struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 int young, idx;
400
401 idx = srcu_read_lock(&kvm->srcu);
402 spin_lock(&kvm->mmu_lock);
403 young = kvm_test_age_hva(kvm, address);
404 spin_unlock(&kvm->mmu_lock);
405 srcu_read_unlock(&kvm->srcu, idx);
406
407 return young;
408 }
409
410 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
411 struct mm_struct *mm)
412 {
413 struct kvm *kvm = mmu_notifier_to_kvm(mn);
414 int idx;
415
416 idx = srcu_read_lock(&kvm->srcu);
417 kvm_arch_flush_shadow_all(kvm);
418 srcu_read_unlock(&kvm->srcu, idx);
419 }
420
421 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
422 .invalidate_page = kvm_mmu_notifier_invalidate_page,
423 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
424 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
425 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
426 .test_young = kvm_mmu_notifier_test_young,
427 .change_pte = kvm_mmu_notifier_change_pte,
428 .release = kvm_mmu_notifier_release,
429 };
430
431 static int kvm_init_mmu_notifier(struct kvm *kvm)
432 {
433 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
434 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
435 }
436
437 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
438
439 static int kvm_init_mmu_notifier(struct kvm *kvm)
440 {
441 return 0;
442 }
443
444 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
445
446 static void kvm_init_memslots_id(struct kvm *kvm)
447 {
448 int i;
449 struct kvm_memslots *slots = kvm->memslots;
450
451 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
452 slots->id_to_index[i] = slots->memslots[i].id = i;
453 }
454
455 static struct kvm *kvm_create_vm(unsigned long type)
456 {
457 int r, i;
458 struct kvm *kvm = kvm_arch_alloc_vm();
459
460 if (!kvm)
461 return ERR_PTR(-ENOMEM);
462
463 r = kvm_arch_init_vm(kvm, type);
464 if (r)
465 goto out_err_no_disable;
466
467 r = hardware_enable_all();
468 if (r)
469 goto out_err_no_disable;
470
471 #ifdef CONFIG_HAVE_KVM_IRQFD
472 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
473 #endif
474
475 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
476
477 r = -ENOMEM;
478 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
479 if (!kvm->memslots)
480 goto out_err_no_srcu;
481
482 /*
483 * Init kvm generation close to the maximum to easily test the
484 * code of handling generation number wrap-around.
485 */
486 kvm->memslots->generation = -150;
487
488 kvm_init_memslots_id(kvm);
489 if (init_srcu_struct(&kvm->srcu))
490 goto out_err_no_srcu;
491 if (init_srcu_struct(&kvm->irq_srcu))
492 goto out_err_no_irq_srcu;
493 for (i = 0; i < KVM_NR_BUSES; i++) {
494 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
495 GFP_KERNEL);
496 if (!kvm->buses[i])
497 goto out_err;
498 }
499
500 spin_lock_init(&kvm->mmu_lock);
501 kvm->mm = current->mm;
502 atomic_inc(&kvm->mm->mm_count);
503 kvm_eventfd_init(kvm);
504 mutex_init(&kvm->lock);
505 mutex_init(&kvm->irq_lock);
506 mutex_init(&kvm->slots_lock);
507 atomic_set(&kvm->users_count, 1);
508 INIT_LIST_HEAD(&kvm->devices);
509
510 r = kvm_init_mmu_notifier(kvm);
511 if (r)
512 goto out_err;
513
514 spin_lock(&kvm_lock);
515 list_add(&kvm->vm_list, &vm_list);
516 spin_unlock(&kvm_lock);
517
518 return kvm;
519
520 out_err:
521 cleanup_srcu_struct(&kvm->irq_srcu);
522 out_err_no_irq_srcu:
523 cleanup_srcu_struct(&kvm->srcu);
524 out_err_no_srcu:
525 hardware_disable_all();
526 out_err_no_disable:
527 for (i = 0; i < KVM_NR_BUSES; i++)
528 kfree(kvm->buses[i]);
529 kfree(kvm->memslots);
530 kvm_arch_free_vm(kvm);
531 return ERR_PTR(r);
532 }
533
534 /*
535 * Avoid using vmalloc for a small buffer.
536 * Should not be used when the size is statically known.
537 */
538 void *kvm_kvzalloc(unsigned long size)
539 {
540 if (size > PAGE_SIZE)
541 return vzalloc(size);
542 else
543 return kzalloc(size, GFP_KERNEL);
544 }
545
546 void kvm_kvfree(const void *addr)
547 {
548 if (is_vmalloc_addr(addr))
549 vfree(addr);
550 else
551 kfree(addr);
552 }
553
554 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
555 {
556 if (!memslot->dirty_bitmap)
557 return;
558
559 kvm_kvfree(memslot->dirty_bitmap);
560 memslot->dirty_bitmap = NULL;
561 }
562
563 /*
564 * Free any memory in @free but not in @dont.
565 */
566 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
567 struct kvm_memory_slot *dont)
568 {
569 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
570 kvm_destroy_dirty_bitmap(free);
571
572 kvm_arch_free_memslot(kvm, free, dont);
573
574 free->npages = 0;
575 }
576
577 static void kvm_free_physmem(struct kvm *kvm)
578 {
579 struct kvm_memslots *slots = kvm->memslots;
580 struct kvm_memory_slot *memslot;
581
582 kvm_for_each_memslot(memslot, slots)
583 kvm_free_physmem_slot(kvm, memslot, NULL);
584
585 kfree(kvm->memslots);
586 }
587
588 static void kvm_destroy_devices(struct kvm *kvm)
589 {
590 struct list_head *node, *tmp;
591
592 list_for_each_safe(node, tmp, &kvm->devices) {
593 struct kvm_device *dev =
594 list_entry(node, struct kvm_device, vm_node);
595
596 list_del(node);
597 dev->ops->destroy(dev);
598 }
599 }
600
601 static void kvm_destroy_vm(struct kvm *kvm)
602 {
603 int i;
604 struct mm_struct *mm = kvm->mm;
605
606 kvm_arch_sync_events(kvm);
607 spin_lock(&kvm_lock);
608 list_del(&kvm->vm_list);
609 spin_unlock(&kvm_lock);
610 kvm_free_irq_routing(kvm);
611 for (i = 0; i < KVM_NR_BUSES; i++)
612 kvm_io_bus_destroy(kvm->buses[i]);
613 kvm_coalesced_mmio_free(kvm);
614 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
615 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
616 #else
617 kvm_arch_flush_shadow_all(kvm);
618 #endif
619 kvm_arch_destroy_vm(kvm);
620 kvm_destroy_devices(kvm);
621 kvm_free_physmem(kvm);
622 cleanup_srcu_struct(&kvm->irq_srcu);
623 cleanup_srcu_struct(&kvm->srcu);
624 kvm_arch_free_vm(kvm);
625 hardware_disable_all();
626 mmdrop(mm);
627 }
628
629 void kvm_get_kvm(struct kvm *kvm)
630 {
631 atomic_inc(&kvm->users_count);
632 }
633 EXPORT_SYMBOL_GPL(kvm_get_kvm);
634
635 void kvm_put_kvm(struct kvm *kvm)
636 {
637 if (atomic_dec_and_test(&kvm->users_count))
638 kvm_destroy_vm(kvm);
639 }
640 EXPORT_SYMBOL_GPL(kvm_put_kvm);
641
642
643 static int kvm_vm_release(struct inode *inode, struct file *filp)
644 {
645 struct kvm *kvm = filp->private_data;
646
647 kvm_irqfd_release(kvm);
648
649 kvm_put_kvm(kvm);
650 return 0;
651 }
652
653 /*
654 * Allocation size is twice as large as the actual dirty bitmap size.
655 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
656 */
657 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
658 {
659 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
660
661 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
662 if (!memslot->dirty_bitmap)
663 return -ENOMEM;
664
665 return 0;
666 }
667
668 /*
669 * Insert memslot and re-sort memslots based on their size,
670 * so the larger slots will get better fit. Sorting algorithm
671 * takes advantage of having initially sorted array and
672 * known changed memslot position.
673 */
674 static void update_memslots(struct kvm_memslots *slots,
675 struct kvm_memory_slot *new)
676 {
677 int id = new->id;
678 int i = slots->id_to_index[id];
679 struct kvm_memory_slot *mslots = slots->memslots;
680
681 WARN_ON(mslots[i].id != id);
682 while (i < KVM_MEM_SLOTS_NUM - 1 &&
683 new->npages < mslots[i + 1].npages) {
684 mslots[i] = mslots[i + 1];
685 slots->id_to_index[mslots[i].id] = i;
686 i++;
687 }
688 while (i > 0 &&
689 new->npages > mslots[i - 1].npages) {
690 mslots[i] = mslots[i - 1];
691 slots->id_to_index[mslots[i].id] = i;
692 i--;
693 }
694
695 mslots[i] = *new;
696 slots->id_to_index[mslots[i].id] = i;
697 }
698
699 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
700 {
701 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
702
703 #ifdef __KVM_HAVE_READONLY_MEM
704 valid_flags |= KVM_MEM_READONLY;
705 #endif
706
707 if (mem->flags & ~valid_flags)
708 return -EINVAL;
709
710 return 0;
711 }
712
713 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
714 struct kvm_memslots *slots)
715 {
716 struct kvm_memslots *old_memslots = kvm->memslots;
717
718 /*
719 * Set the low bit in the generation, which disables SPTE caching
720 * until the end of synchronize_srcu_expedited.
721 */
722 WARN_ON(old_memslots->generation & 1);
723 slots->generation = old_memslots->generation + 1;
724
725 rcu_assign_pointer(kvm->memslots, slots);
726 synchronize_srcu_expedited(&kvm->srcu);
727
728 /*
729 * Increment the new memslot generation a second time. This prevents
730 * vm exits that race with memslot updates from caching a memslot
731 * generation that will (potentially) be valid forever.
732 */
733 slots->generation++;
734
735 kvm_arch_memslots_updated(kvm);
736
737 return old_memslots;
738 }
739
740 /*
741 * Allocate some memory and give it an address in the guest physical address
742 * space.
743 *
744 * Discontiguous memory is allowed, mostly for framebuffers.
745 *
746 * Must be called holding kvm->slots_lock for write.
747 */
748 int __kvm_set_memory_region(struct kvm *kvm,
749 struct kvm_userspace_memory_region *mem)
750 {
751 int r;
752 gfn_t base_gfn;
753 unsigned long npages;
754 struct kvm_memory_slot *slot;
755 struct kvm_memory_slot old, new;
756 struct kvm_memslots *slots = NULL, *old_memslots;
757 enum kvm_mr_change change;
758
759 r = check_memory_region_flags(mem);
760 if (r)
761 goto out;
762
763 r = -EINVAL;
764 /* General sanity checks */
765 if (mem->memory_size & (PAGE_SIZE - 1))
766 goto out;
767 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
768 goto out;
769 /* We can read the guest memory with __xxx_user() later on. */
770 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
771 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
772 !access_ok(VERIFY_WRITE,
773 (void __user *)(unsigned long)mem->userspace_addr,
774 mem->memory_size)))
775 goto out;
776 if (mem->slot >= KVM_MEM_SLOTS_NUM)
777 goto out;
778 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
779 goto out;
780
781 slot = id_to_memslot(kvm->memslots, mem->slot);
782 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
783 npages = mem->memory_size >> PAGE_SHIFT;
784
785 if (npages > KVM_MEM_MAX_NR_PAGES)
786 goto out;
787
788 if (!npages)
789 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
790
791 new = old = *slot;
792
793 new.id = mem->slot;
794 new.base_gfn = base_gfn;
795 new.npages = npages;
796 new.flags = mem->flags;
797
798 if (npages) {
799 if (!old.npages)
800 change = KVM_MR_CREATE;
801 else { /* Modify an existing slot. */
802 if ((mem->userspace_addr != old.userspace_addr) ||
803 (npages != old.npages) ||
804 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
805 goto out;
806
807 if (base_gfn != old.base_gfn)
808 change = KVM_MR_MOVE;
809 else if (new.flags != old.flags)
810 change = KVM_MR_FLAGS_ONLY;
811 else { /* Nothing to change. */
812 r = 0;
813 goto out;
814 }
815 }
816 } else if (old.npages) {
817 change = KVM_MR_DELETE;
818 } else /* Modify a non-existent slot: disallowed. */
819 goto out;
820
821 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
822 /* Check for overlaps */
823 r = -EEXIST;
824 kvm_for_each_memslot(slot, kvm->memslots) {
825 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
826 (slot->id == mem->slot))
827 continue;
828 if (!((base_gfn + npages <= slot->base_gfn) ||
829 (base_gfn >= slot->base_gfn + slot->npages)))
830 goto out;
831 }
832 }
833
834 /* Free page dirty bitmap if unneeded */
835 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
836 new.dirty_bitmap = NULL;
837
838 r = -ENOMEM;
839 if (change == KVM_MR_CREATE) {
840 new.userspace_addr = mem->userspace_addr;
841
842 if (kvm_arch_create_memslot(kvm, &new, npages))
843 goto out_free;
844 }
845
846 /* Allocate page dirty bitmap if needed */
847 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
848 if (kvm_create_dirty_bitmap(&new) < 0)
849 goto out_free;
850 }
851
852 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
853 GFP_KERNEL);
854 if (!slots)
855 goto out_free;
856
857 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
858 slot = id_to_memslot(slots, mem->slot);
859 slot->flags |= KVM_MEMSLOT_INVALID;
860
861 old_memslots = install_new_memslots(kvm, slots);
862
863 /* slot was deleted or moved, clear iommu mapping */
864 kvm_iommu_unmap_pages(kvm, &old);
865 /* From this point no new shadow pages pointing to a deleted,
866 * or moved, memslot will be created.
867 *
868 * validation of sp->gfn happens in:
869 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
870 * - kvm_is_visible_gfn (mmu_check_roots)
871 */
872 kvm_arch_flush_shadow_memslot(kvm, slot);
873
874 /*
875 * We can re-use the old_memslots from above, the only difference
876 * from the currently installed memslots is the invalid flag. This
877 * will get overwritten by update_memslots anyway.
878 */
879 slots = old_memslots;
880 }
881
882 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
883 if (r)
884 goto out_slots;
885
886 /* actual memory is freed via old in kvm_free_physmem_slot below */
887 if (change == KVM_MR_DELETE) {
888 new.dirty_bitmap = NULL;
889 memset(&new.arch, 0, sizeof(new.arch));
890 }
891
892 update_memslots(slots, &new);
893 old_memslots = install_new_memslots(kvm, slots);
894
895 kvm_arch_commit_memory_region(kvm, mem, &old, change);
896
897 kvm_free_physmem_slot(kvm, &old, &new);
898 kfree(old_memslots);
899
900 /*
901 * IOMMU mapping: New slots need to be mapped. Old slots need to be
902 * un-mapped and re-mapped if their base changes. Since base change
903 * unmapping is handled above with slot deletion, mapping alone is
904 * needed here. Anything else the iommu might care about for existing
905 * slots (size changes, userspace addr changes and read-only flag
906 * changes) is disallowed above, so any other attribute changes getting
907 * here can be skipped.
908 */
909 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
910 r = kvm_iommu_map_pages(kvm, &new);
911 return r;
912 }
913
914 return 0;
915
916 out_slots:
917 kfree(slots);
918 out_free:
919 kvm_free_physmem_slot(kvm, &new, &old);
920 out:
921 return r;
922 }
923 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
924
925 int kvm_set_memory_region(struct kvm *kvm,
926 struct kvm_userspace_memory_region *mem)
927 {
928 int r;
929
930 mutex_lock(&kvm->slots_lock);
931 r = __kvm_set_memory_region(kvm, mem);
932 mutex_unlock(&kvm->slots_lock);
933 return r;
934 }
935 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
936
937 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
938 struct kvm_userspace_memory_region *mem)
939 {
940 if (mem->slot >= KVM_USER_MEM_SLOTS)
941 return -EINVAL;
942 return kvm_set_memory_region(kvm, mem);
943 }
944
945 int kvm_get_dirty_log(struct kvm *kvm,
946 struct kvm_dirty_log *log, int *is_dirty)
947 {
948 struct kvm_memory_slot *memslot;
949 int r, i;
950 unsigned long n;
951 unsigned long any = 0;
952
953 r = -EINVAL;
954 if (log->slot >= KVM_USER_MEM_SLOTS)
955 goto out;
956
957 memslot = id_to_memslot(kvm->memslots, log->slot);
958 r = -ENOENT;
959 if (!memslot->dirty_bitmap)
960 goto out;
961
962 n = kvm_dirty_bitmap_bytes(memslot);
963
964 for (i = 0; !any && i < n/sizeof(long); ++i)
965 any = memslot->dirty_bitmap[i];
966
967 r = -EFAULT;
968 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
969 goto out;
970
971 if (any)
972 *is_dirty = 1;
973
974 r = 0;
975 out:
976 return r;
977 }
978 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
979
980 bool kvm_largepages_enabled(void)
981 {
982 return largepages_enabled;
983 }
984
985 void kvm_disable_largepages(void)
986 {
987 largepages_enabled = false;
988 }
989 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
990
991 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
992 {
993 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
994 }
995 EXPORT_SYMBOL_GPL(gfn_to_memslot);
996
997 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
998 {
999 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1000
1001 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1002 memslot->flags & KVM_MEMSLOT_INVALID)
1003 return 0;
1004
1005 return 1;
1006 }
1007 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1008
1009 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1010 {
1011 struct vm_area_struct *vma;
1012 unsigned long addr, size;
1013
1014 size = PAGE_SIZE;
1015
1016 addr = gfn_to_hva(kvm, gfn);
1017 if (kvm_is_error_hva(addr))
1018 return PAGE_SIZE;
1019
1020 down_read(&current->mm->mmap_sem);
1021 vma = find_vma(current->mm, addr);
1022 if (!vma)
1023 goto out;
1024
1025 size = vma_kernel_pagesize(vma);
1026
1027 out:
1028 up_read(&current->mm->mmap_sem);
1029
1030 return size;
1031 }
1032
1033 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1034 {
1035 return slot->flags & KVM_MEM_READONLY;
1036 }
1037
1038 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1039 gfn_t *nr_pages, bool write)
1040 {
1041 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1042 return KVM_HVA_ERR_BAD;
1043
1044 if (memslot_is_readonly(slot) && write)
1045 return KVM_HVA_ERR_RO_BAD;
1046
1047 if (nr_pages)
1048 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1049
1050 return __gfn_to_hva_memslot(slot, gfn);
1051 }
1052
1053 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1054 gfn_t *nr_pages)
1055 {
1056 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1057 }
1058
1059 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1060 gfn_t gfn)
1061 {
1062 return gfn_to_hva_many(slot, gfn, NULL);
1063 }
1064 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1065
1066 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1067 {
1068 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1069 }
1070 EXPORT_SYMBOL_GPL(gfn_to_hva);
1071
1072 /*
1073 * If writable is set to false, the hva returned by this function is only
1074 * allowed to be read.
1075 */
1076 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1077 gfn_t gfn, bool *writable)
1078 {
1079 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1080
1081 if (!kvm_is_error_hva(hva) && writable)
1082 *writable = !memslot_is_readonly(slot);
1083
1084 return hva;
1085 }
1086
1087 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1088 {
1089 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1090
1091 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1092 }
1093
1094 static int kvm_read_hva(void *data, void __user *hva, int len)
1095 {
1096 return __copy_from_user(data, hva, len);
1097 }
1098
1099 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1100 {
1101 return __copy_from_user_inatomic(data, hva, len);
1102 }
1103
1104 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1105 unsigned long start, int write, struct page **page)
1106 {
1107 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1108
1109 if (write)
1110 flags |= FOLL_WRITE;
1111
1112 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1113 }
1114
1115 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1116 unsigned long addr, bool write_fault,
1117 struct page **pagep)
1118 {
1119 int npages;
1120 int locked = 1;
1121 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1122 (pagep ? FOLL_GET : 0) |
1123 (write_fault ? FOLL_WRITE : 0);
1124
1125 /*
1126 * If retrying the fault, we get here *not* having allowed the filemap
1127 * to wait on the page lock. We should now allow waiting on the IO with
1128 * the mmap semaphore released.
1129 */
1130 down_read(&mm->mmap_sem);
1131 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1132 &locked);
1133 if (!locked) {
1134 VM_BUG_ON(npages);
1135
1136 if (!pagep)
1137 return 0;
1138
1139 /*
1140 * The previous call has now waited on the IO. Now we can
1141 * retry and complete. Pass TRIED to ensure we do not re
1142 * schedule async IO (see e.g. filemap_fault).
1143 */
1144 down_read(&mm->mmap_sem);
1145 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1146 pagep, NULL, NULL);
1147 }
1148 up_read(&mm->mmap_sem);
1149 return npages;
1150 }
1151
1152 static inline int check_user_page_hwpoison(unsigned long addr)
1153 {
1154 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1155
1156 rc = __get_user_pages(current, current->mm, addr, 1,
1157 flags, NULL, NULL, NULL);
1158 return rc == -EHWPOISON;
1159 }
1160
1161 /*
1162 * The atomic path to get the writable pfn which will be stored in @pfn,
1163 * true indicates success, otherwise false is returned.
1164 */
1165 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1166 bool write_fault, bool *writable, pfn_t *pfn)
1167 {
1168 struct page *page[1];
1169 int npages;
1170
1171 if (!(async || atomic))
1172 return false;
1173
1174 /*
1175 * Fast pin a writable pfn only if it is a write fault request
1176 * or the caller allows to map a writable pfn for a read fault
1177 * request.
1178 */
1179 if (!(write_fault || writable))
1180 return false;
1181
1182 npages = __get_user_pages_fast(addr, 1, 1, page);
1183 if (npages == 1) {
1184 *pfn = page_to_pfn(page[0]);
1185
1186 if (writable)
1187 *writable = true;
1188 return true;
1189 }
1190
1191 return false;
1192 }
1193
1194 /*
1195 * The slow path to get the pfn of the specified host virtual address,
1196 * 1 indicates success, -errno is returned if error is detected.
1197 */
1198 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1199 bool *writable, pfn_t *pfn)
1200 {
1201 struct page *page[1];
1202 int npages = 0;
1203
1204 might_sleep();
1205
1206 if (writable)
1207 *writable = write_fault;
1208
1209 if (async) {
1210 down_read(&current->mm->mmap_sem);
1211 npages = get_user_page_nowait(current, current->mm,
1212 addr, write_fault, page);
1213 up_read(&current->mm->mmap_sem);
1214 } else {
1215 /*
1216 * By now we have tried gup_fast, and possibly async_pf, and we
1217 * are certainly not atomic. Time to retry the gup, allowing
1218 * mmap semaphore to be relinquished in the case of IO.
1219 */
1220 npages = kvm_get_user_page_io(current, current->mm, addr,
1221 write_fault, page);
1222 }
1223 if (npages != 1)
1224 return npages;
1225
1226 /* map read fault as writable if possible */
1227 if (unlikely(!write_fault) && writable) {
1228 struct page *wpage[1];
1229
1230 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1231 if (npages == 1) {
1232 *writable = true;
1233 put_page(page[0]);
1234 page[0] = wpage[0];
1235 }
1236
1237 npages = 1;
1238 }
1239 *pfn = page_to_pfn(page[0]);
1240 return npages;
1241 }
1242
1243 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1244 {
1245 if (unlikely(!(vma->vm_flags & VM_READ)))
1246 return false;
1247
1248 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1249 return false;
1250
1251 return true;
1252 }
1253
1254 /*
1255 * Pin guest page in memory and return its pfn.
1256 * @addr: host virtual address which maps memory to the guest
1257 * @atomic: whether this function can sleep
1258 * @async: whether this function need to wait IO complete if the
1259 * host page is not in the memory
1260 * @write_fault: whether we should get a writable host page
1261 * @writable: whether it allows to map a writable host page for !@write_fault
1262 *
1263 * The function will map a writable host page for these two cases:
1264 * 1): @write_fault = true
1265 * 2): @write_fault = false && @writable, @writable will tell the caller
1266 * whether the mapping is writable.
1267 */
1268 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1269 bool write_fault, bool *writable)
1270 {
1271 struct vm_area_struct *vma;
1272 pfn_t pfn = 0;
1273 int npages;
1274
1275 /* we can do it either atomically or asynchronously, not both */
1276 BUG_ON(atomic && async);
1277
1278 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1279 return pfn;
1280
1281 if (atomic)
1282 return KVM_PFN_ERR_FAULT;
1283
1284 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1285 if (npages == 1)
1286 return pfn;
1287
1288 down_read(&current->mm->mmap_sem);
1289 if (npages == -EHWPOISON ||
1290 (!async && check_user_page_hwpoison(addr))) {
1291 pfn = KVM_PFN_ERR_HWPOISON;
1292 goto exit;
1293 }
1294
1295 vma = find_vma_intersection(current->mm, addr, addr + 1);
1296
1297 if (vma == NULL)
1298 pfn = KVM_PFN_ERR_FAULT;
1299 else if ((vma->vm_flags & VM_PFNMAP)) {
1300 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1301 vma->vm_pgoff;
1302 BUG_ON(!kvm_is_mmio_pfn(pfn));
1303 } else {
1304 if (async && vma_is_valid(vma, write_fault))
1305 *async = true;
1306 pfn = KVM_PFN_ERR_FAULT;
1307 }
1308 exit:
1309 up_read(&current->mm->mmap_sem);
1310 return pfn;
1311 }
1312
1313 static pfn_t
1314 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1315 bool *async, bool write_fault, bool *writable)
1316 {
1317 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1318
1319 if (addr == KVM_HVA_ERR_RO_BAD)
1320 return KVM_PFN_ERR_RO_FAULT;
1321
1322 if (kvm_is_error_hva(addr))
1323 return KVM_PFN_NOSLOT;
1324
1325 /* Do not map writable pfn in the readonly memslot. */
1326 if (writable && memslot_is_readonly(slot)) {
1327 *writable = false;
1328 writable = NULL;
1329 }
1330
1331 return hva_to_pfn(addr, atomic, async, write_fault,
1332 writable);
1333 }
1334
1335 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1336 bool write_fault, bool *writable)
1337 {
1338 struct kvm_memory_slot *slot;
1339
1340 if (async)
1341 *async = false;
1342
1343 slot = gfn_to_memslot(kvm, gfn);
1344
1345 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1346 writable);
1347 }
1348
1349 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1350 {
1351 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1352 }
1353 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1354
1355 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1356 bool write_fault, bool *writable)
1357 {
1358 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1359 }
1360 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1361
1362 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1363 {
1364 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1365 }
1366 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1367
1368 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1369 bool *writable)
1370 {
1371 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1372 }
1373 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1374
1375 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1376 {
1377 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1378 }
1379
1380 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1381 {
1382 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1383 }
1384 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1385
1386 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1387 int nr_pages)
1388 {
1389 unsigned long addr;
1390 gfn_t entry;
1391
1392 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1393 if (kvm_is_error_hva(addr))
1394 return -1;
1395
1396 if (entry < nr_pages)
1397 return 0;
1398
1399 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1400 }
1401 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1402
1403 static struct page *kvm_pfn_to_page(pfn_t pfn)
1404 {
1405 if (is_error_noslot_pfn(pfn))
1406 return KVM_ERR_PTR_BAD_PAGE;
1407
1408 if (kvm_is_mmio_pfn(pfn)) {
1409 WARN_ON(1);
1410 return KVM_ERR_PTR_BAD_PAGE;
1411 }
1412
1413 return pfn_to_page(pfn);
1414 }
1415
1416 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1417 {
1418 pfn_t pfn;
1419
1420 pfn = gfn_to_pfn(kvm, gfn);
1421
1422 return kvm_pfn_to_page(pfn);
1423 }
1424
1425 EXPORT_SYMBOL_GPL(gfn_to_page);
1426
1427 void kvm_release_page_clean(struct page *page)
1428 {
1429 WARN_ON(is_error_page(page));
1430
1431 kvm_release_pfn_clean(page_to_pfn(page));
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1434
1435 void kvm_release_pfn_clean(pfn_t pfn)
1436 {
1437 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1438 put_page(pfn_to_page(pfn));
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1441
1442 void kvm_release_page_dirty(struct page *page)
1443 {
1444 WARN_ON(is_error_page(page));
1445
1446 kvm_release_pfn_dirty(page_to_pfn(page));
1447 }
1448 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1449
1450 static void kvm_release_pfn_dirty(pfn_t pfn)
1451 {
1452 kvm_set_pfn_dirty(pfn);
1453 kvm_release_pfn_clean(pfn);
1454 }
1455
1456 void kvm_set_pfn_dirty(pfn_t pfn)
1457 {
1458 if (!kvm_is_mmio_pfn(pfn)) {
1459 struct page *page = pfn_to_page(pfn);
1460 if (!PageReserved(page))
1461 SetPageDirty(page);
1462 }
1463 }
1464 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1465
1466 void kvm_set_pfn_accessed(pfn_t pfn)
1467 {
1468 if (!kvm_is_mmio_pfn(pfn))
1469 mark_page_accessed(pfn_to_page(pfn));
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1472
1473 void kvm_get_pfn(pfn_t pfn)
1474 {
1475 if (!kvm_is_mmio_pfn(pfn))
1476 get_page(pfn_to_page(pfn));
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1479
1480 static int next_segment(unsigned long len, int offset)
1481 {
1482 if (len > PAGE_SIZE - offset)
1483 return PAGE_SIZE - offset;
1484 else
1485 return len;
1486 }
1487
1488 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1489 int len)
1490 {
1491 int r;
1492 unsigned long addr;
1493
1494 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1495 if (kvm_is_error_hva(addr))
1496 return -EFAULT;
1497 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1498 if (r)
1499 return -EFAULT;
1500 return 0;
1501 }
1502 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1503
1504 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1505 {
1506 gfn_t gfn = gpa >> PAGE_SHIFT;
1507 int seg;
1508 int offset = offset_in_page(gpa);
1509 int ret;
1510
1511 while ((seg = next_segment(len, offset)) != 0) {
1512 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1513 if (ret < 0)
1514 return ret;
1515 offset = 0;
1516 len -= seg;
1517 data += seg;
1518 ++gfn;
1519 }
1520 return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_read_guest);
1523
1524 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1525 unsigned long len)
1526 {
1527 int r;
1528 unsigned long addr;
1529 gfn_t gfn = gpa >> PAGE_SHIFT;
1530 int offset = offset_in_page(gpa);
1531
1532 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1533 if (kvm_is_error_hva(addr))
1534 return -EFAULT;
1535 pagefault_disable();
1536 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1537 pagefault_enable();
1538 if (r)
1539 return -EFAULT;
1540 return 0;
1541 }
1542 EXPORT_SYMBOL(kvm_read_guest_atomic);
1543
1544 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1545 int offset, int len)
1546 {
1547 int r;
1548 unsigned long addr;
1549
1550 addr = gfn_to_hva(kvm, gfn);
1551 if (kvm_is_error_hva(addr))
1552 return -EFAULT;
1553 r = __copy_to_user((void __user *)addr + offset, data, len);
1554 if (r)
1555 return -EFAULT;
1556 mark_page_dirty(kvm, gfn);
1557 return 0;
1558 }
1559 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1560
1561 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1562 unsigned long len)
1563 {
1564 gfn_t gfn = gpa >> PAGE_SHIFT;
1565 int seg;
1566 int offset = offset_in_page(gpa);
1567 int ret;
1568
1569 while ((seg = next_segment(len, offset)) != 0) {
1570 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1571 if (ret < 0)
1572 return ret;
1573 offset = 0;
1574 len -= seg;
1575 data += seg;
1576 ++gfn;
1577 }
1578 return 0;
1579 }
1580
1581 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1582 gpa_t gpa, unsigned long len)
1583 {
1584 struct kvm_memslots *slots = kvm_memslots(kvm);
1585 int offset = offset_in_page(gpa);
1586 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1587 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1588 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1589 gfn_t nr_pages_avail;
1590
1591 ghc->gpa = gpa;
1592 ghc->generation = slots->generation;
1593 ghc->len = len;
1594 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1595 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1596 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1597 ghc->hva += offset;
1598 } else {
1599 /*
1600 * If the requested region crosses two memslots, we still
1601 * verify that the entire region is valid here.
1602 */
1603 while (start_gfn <= end_gfn) {
1604 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1605 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1606 &nr_pages_avail);
1607 if (kvm_is_error_hva(ghc->hva))
1608 return -EFAULT;
1609 start_gfn += nr_pages_avail;
1610 }
1611 /* Use the slow path for cross page reads and writes. */
1612 ghc->memslot = NULL;
1613 }
1614 return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1617
1618 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1619 void *data, unsigned long len)
1620 {
1621 struct kvm_memslots *slots = kvm_memslots(kvm);
1622 int r;
1623
1624 BUG_ON(len > ghc->len);
1625
1626 if (slots->generation != ghc->generation)
1627 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1628
1629 if (unlikely(!ghc->memslot))
1630 return kvm_write_guest(kvm, ghc->gpa, data, len);
1631
1632 if (kvm_is_error_hva(ghc->hva))
1633 return -EFAULT;
1634
1635 r = __copy_to_user((void __user *)ghc->hva, data, len);
1636 if (r)
1637 return -EFAULT;
1638 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1639
1640 return 0;
1641 }
1642 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1643
1644 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1645 void *data, unsigned long len)
1646 {
1647 struct kvm_memslots *slots = kvm_memslots(kvm);
1648 int r;
1649
1650 BUG_ON(len > ghc->len);
1651
1652 if (slots->generation != ghc->generation)
1653 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1654
1655 if (unlikely(!ghc->memslot))
1656 return kvm_read_guest(kvm, ghc->gpa, data, len);
1657
1658 if (kvm_is_error_hva(ghc->hva))
1659 return -EFAULT;
1660
1661 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1662 if (r)
1663 return -EFAULT;
1664
1665 return 0;
1666 }
1667 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1668
1669 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1670 {
1671 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1672
1673 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1674 }
1675 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1676
1677 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1678 {
1679 gfn_t gfn = gpa >> PAGE_SHIFT;
1680 int seg;
1681 int offset = offset_in_page(gpa);
1682 int ret;
1683
1684 while ((seg = next_segment(len, offset)) != 0) {
1685 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1686 if (ret < 0)
1687 return ret;
1688 offset = 0;
1689 len -= seg;
1690 ++gfn;
1691 }
1692 return 0;
1693 }
1694 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1695
1696 static void mark_page_dirty_in_slot(struct kvm *kvm,
1697 struct kvm_memory_slot *memslot,
1698 gfn_t gfn)
1699 {
1700 if (memslot && memslot->dirty_bitmap) {
1701 unsigned long rel_gfn = gfn - memslot->base_gfn;
1702
1703 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1704 }
1705 }
1706
1707 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1708 {
1709 struct kvm_memory_slot *memslot;
1710
1711 memslot = gfn_to_memslot(kvm, gfn);
1712 mark_page_dirty_in_slot(kvm, memslot, gfn);
1713 }
1714 EXPORT_SYMBOL_GPL(mark_page_dirty);
1715
1716 /*
1717 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1718 */
1719 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1720 {
1721 DEFINE_WAIT(wait);
1722
1723 for (;;) {
1724 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1725
1726 if (kvm_arch_vcpu_runnable(vcpu)) {
1727 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1728 break;
1729 }
1730 if (kvm_cpu_has_pending_timer(vcpu))
1731 break;
1732 if (signal_pending(current))
1733 break;
1734
1735 schedule();
1736 }
1737
1738 finish_wait(&vcpu->wq, &wait);
1739 }
1740 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1741
1742 #ifndef CONFIG_S390
1743 /*
1744 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1745 */
1746 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1747 {
1748 int me;
1749 int cpu = vcpu->cpu;
1750 wait_queue_head_t *wqp;
1751
1752 wqp = kvm_arch_vcpu_wq(vcpu);
1753 if (waitqueue_active(wqp)) {
1754 wake_up_interruptible(wqp);
1755 ++vcpu->stat.halt_wakeup;
1756 }
1757
1758 me = get_cpu();
1759 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1760 if (kvm_arch_vcpu_should_kick(vcpu))
1761 smp_send_reschedule(cpu);
1762 put_cpu();
1763 }
1764 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1765 #endif /* !CONFIG_S390 */
1766
1767 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1768 {
1769 struct pid *pid;
1770 struct task_struct *task = NULL;
1771 int ret = 0;
1772
1773 rcu_read_lock();
1774 pid = rcu_dereference(target->pid);
1775 if (pid)
1776 task = get_pid_task(pid, PIDTYPE_PID);
1777 rcu_read_unlock();
1778 if (!task)
1779 return ret;
1780 if (task->flags & PF_VCPU) {
1781 put_task_struct(task);
1782 return ret;
1783 }
1784 ret = yield_to(task, 1);
1785 put_task_struct(task);
1786
1787 return ret;
1788 }
1789 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1790
1791 /*
1792 * Helper that checks whether a VCPU is eligible for directed yield.
1793 * Most eligible candidate to yield is decided by following heuristics:
1794 *
1795 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1796 * (preempted lock holder), indicated by @in_spin_loop.
1797 * Set at the beiginning and cleared at the end of interception/PLE handler.
1798 *
1799 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1800 * chance last time (mostly it has become eligible now since we have probably
1801 * yielded to lockholder in last iteration. This is done by toggling
1802 * @dy_eligible each time a VCPU checked for eligibility.)
1803 *
1804 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1805 * to preempted lock-holder could result in wrong VCPU selection and CPU
1806 * burning. Giving priority for a potential lock-holder increases lock
1807 * progress.
1808 *
1809 * Since algorithm is based on heuristics, accessing another VCPU data without
1810 * locking does not harm. It may result in trying to yield to same VCPU, fail
1811 * and continue with next VCPU and so on.
1812 */
1813 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1814 {
1815 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1816 bool eligible;
1817
1818 eligible = !vcpu->spin_loop.in_spin_loop ||
1819 vcpu->spin_loop.dy_eligible;
1820
1821 if (vcpu->spin_loop.in_spin_loop)
1822 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1823
1824 return eligible;
1825 #else
1826 return true;
1827 #endif
1828 }
1829
1830 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1831 {
1832 struct kvm *kvm = me->kvm;
1833 struct kvm_vcpu *vcpu;
1834 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1835 int yielded = 0;
1836 int try = 3;
1837 int pass;
1838 int i;
1839
1840 kvm_vcpu_set_in_spin_loop(me, true);
1841 /*
1842 * We boost the priority of a VCPU that is runnable but not
1843 * currently running, because it got preempted by something
1844 * else and called schedule in __vcpu_run. Hopefully that
1845 * VCPU is holding the lock that we need and will release it.
1846 * We approximate round-robin by starting at the last boosted VCPU.
1847 */
1848 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1849 kvm_for_each_vcpu(i, vcpu, kvm) {
1850 if (!pass && i <= last_boosted_vcpu) {
1851 i = last_boosted_vcpu;
1852 continue;
1853 } else if (pass && i > last_boosted_vcpu)
1854 break;
1855 if (!ACCESS_ONCE(vcpu->preempted))
1856 continue;
1857 if (vcpu == me)
1858 continue;
1859 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1860 continue;
1861 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1862 continue;
1863
1864 yielded = kvm_vcpu_yield_to(vcpu);
1865 if (yielded > 0) {
1866 kvm->last_boosted_vcpu = i;
1867 break;
1868 } else if (yielded < 0) {
1869 try--;
1870 if (!try)
1871 break;
1872 }
1873 }
1874 }
1875 kvm_vcpu_set_in_spin_loop(me, false);
1876
1877 /* Ensure vcpu is not eligible during next spinloop */
1878 kvm_vcpu_set_dy_eligible(me, false);
1879 }
1880 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1881
1882 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1883 {
1884 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1885 struct page *page;
1886
1887 if (vmf->pgoff == 0)
1888 page = virt_to_page(vcpu->run);
1889 #ifdef CONFIG_X86
1890 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1891 page = virt_to_page(vcpu->arch.pio_data);
1892 #endif
1893 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1894 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1895 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1896 #endif
1897 else
1898 return kvm_arch_vcpu_fault(vcpu, vmf);
1899 get_page(page);
1900 vmf->page = page;
1901 return 0;
1902 }
1903
1904 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1905 .fault = kvm_vcpu_fault,
1906 };
1907
1908 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1909 {
1910 vma->vm_ops = &kvm_vcpu_vm_ops;
1911 return 0;
1912 }
1913
1914 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1915 {
1916 struct kvm_vcpu *vcpu = filp->private_data;
1917
1918 kvm_put_kvm(vcpu->kvm);
1919 return 0;
1920 }
1921
1922 static struct file_operations kvm_vcpu_fops = {
1923 .release = kvm_vcpu_release,
1924 .unlocked_ioctl = kvm_vcpu_ioctl,
1925 #ifdef CONFIG_COMPAT
1926 .compat_ioctl = kvm_vcpu_compat_ioctl,
1927 #endif
1928 .mmap = kvm_vcpu_mmap,
1929 .llseek = noop_llseek,
1930 };
1931
1932 /*
1933 * Allocates an inode for the vcpu.
1934 */
1935 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1936 {
1937 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1938 }
1939
1940 /*
1941 * Creates some virtual cpus. Good luck creating more than one.
1942 */
1943 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1944 {
1945 int r;
1946 struct kvm_vcpu *vcpu, *v;
1947
1948 if (id >= KVM_MAX_VCPUS)
1949 return -EINVAL;
1950
1951 vcpu = kvm_arch_vcpu_create(kvm, id);
1952 if (IS_ERR(vcpu))
1953 return PTR_ERR(vcpu);
1954
1955 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1956
1957 r = kvm_arch_vcpu_setup(vcpu);
1958 if (r)
1959 goto vcpu_destroy;
1960
1961 mutex_lock(&kvm->lock);
1962 if (!kvm_vcpu_compatible(vcpu)) {
1963 r = -EINVAL;
1964 goto unlock_vcpu_destroy;
1965 }
1966 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1967 r = -EINVAL;
1968 goto unlock_vcpu_destroy;
1969 }
1970
1971 kvm_for_each_vcpu(r, v, kvm)
1972 if (v->vcpu_id == id) {
1973 r = -EEXIST;
1974 goto unlock_vcpu_destroy;
1975 }
1976
1977 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1978
1979 /* Now it's all set up, let userspace reach it */
1980 kvm_get_kvm(kvm);
1981 r = create_vcpu_fd(vcpu);
1982 if (r < 0) {
1983 kvm_put_kvm(kvm);
1984 goto unlock_vcpu_destroy;
1985 }
1986
1987 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1988 smp_wmb();
1989 atomic_inc(&kvm->online_vcpus);
1990
1991 mutex_unlock(&kvm->lock);
1992 kvm_arch_vcpu_postcreate(vcpu);
1993 return r;
1994
1995 unlock_vcpu_destroy:
1996 mutex_unlock(&kvm->lock);
1997 vcpu_destroy:
1998 kvm_arch_vcpu_destroy(vcpu);
1999 return r;
2000 }
2001
2002 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2003 {
2004 if (sigset) {
2005 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2006 vcpu->sigset_active = 1;
2007 vcpu->sigset = *sigset;
2008 } else
2009 vcpu->sigset_active = 0;
2010 return 0;
2011 }
2012
2013 static long kvm_vcpu_ioctl(struct file *filp,
2014 unsigned int ioctl, unsigned long arg)
2015 {
2016 struct kvm_vcpu *vcpu = filp->private_data;
2017 void __user *argp = (void __user *)arg;
2018 int r;
2019 struct kvm_fpu *fpu = NULL;
2020 struct kvm_sregs *kvm_sregs = NULL;
2021
2022 if (vcpu->kvm->mm != current->mm)
2023 return -EIO;
2024
2025 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2026 return -EINVAL;
2027
2028 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2029 /*
2030 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2031 * so vcpu_load() would break it.
2032 */
2033 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2034 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2035 #endif
2036
2037
2038 r = vcpu_load(vcpu);
2039 if (r)
2040 return r;
2041 switch (ioctl) {
2042 case KVM_RUN:
2043 r = -EINVAL;
2044 if (arg)
2045 goto out;
2046 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2047 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2048 break;
2049 case KVM_GET_REGS: {
2050 struct kvm_regs *kvm_regs;
2051
2052 r = -ENOMEM;
2053 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2054 if (!kvm_regs)
2055 goto out;
2056 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2057 if (r)
2058 goto out_free1;
2059 r = -EFAULT;
2060 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2061 goto out_free1;
2062 r = 0;
2063 out_free1:
2064 kfree(kvm_regs);
2065 break;
2066 }
2067 case KVM_SET_REGS: {
2068 struct kvm_regs *kvm_regs;
2069
2070 r = -ENOMEM;
2071 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2072 if (IS_ERR(kvm_regs)) {
2073 r = PTR_ERR(kvm_regs);
2074 goto out;
2075 }
2076 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2077 kfree(kvm_regs);
2078 break;
2079 }
2080 case KVM_GET_SREGS: {
2081 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2082 r = -ENOMEM;
2083 if (!kvm_sregs)
2084 goto out;
2085 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2086 if (r)
2087 goto out;
2088 r = -EFAULT;
2089 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2090 goto out;
2091 r = 0;
2092 break;
2093 }
2094 case KVM_SET_SREGS: {
2095 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2096 if (IS_ERR(kvm_sregs)) {
2097 r = PTR_ERR(kvm_sregs);
2098 kvm_sregs = NULL;
2099 goto out;
2100 }
2101 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2102 break;
2103 }
2104 case KVM_GET_MP_STATE: {
2105 struct kvm_mp_state mp_state;
2106
2107 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2108 if (r)
2109 goto out;
2110 r = -EFAULT;
2111 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2112 goto out;
2113 r = 0;
2114 break;
2115 }
2116 case KVM_SET_MP_STATE: {
2117 struct kvm_mp_state mp_state;
2118
2119 r = -EFAULT;
2120 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2121 goto out;
2122 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2123 break;
2124 }
2125 case KVM_TRANSLATE: {
2126 struct kvm_translation tr;
2127
2128 r = -EFAULT;
2129 if (copy_from_user(&tr, argp, sizeof tr))
2130 goto out;
2131 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2132 if (r)
2133 goto out;
2134 r = -EFAULT;
2135 if (copy_to_user(argp, &tr, sizeof tr))
2136 goto out;
2137 r = 0;
2138 break;
2139 }
2140 case KVM_SET_GUEST_DEBUG: {
2141 struct kvm_guest_debug dbg;
2142
2143 r = -EFAULT;
2144 if (copy_from_user(&dbg, argp, sizeof dbg))
2145 goto out;
2146 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2147 break;
2148 }
2149 case KVM_SET_SIGNAL_MASK: {
2150 struct kvm_signal_mask __user *sigmask_arg = argp;
2151 struct kvm_signal_mask kvm_sigmask;
2152 sigset_t sigset, *p;
2153
2154 p = NULL;
2155 if (argp) {
2156 r = -EFAULT;
2157 if (copy_from_user(&kvm_sigmask, argp,
2158 sizeof kvm_sigmask))
2159 goto out;
2160 r = -EINVAL;
2161 if (kvm_sigmask.len != sizeof sigset)
2162 goto out;
2163 r = -EFAULT;
2164 if (copy_from_user(&sigset, sigmask_arg->sigset,
2165 sizeof sigset))
2166 goto out;
2167 p = &sigset;
2168 }
2169 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2170 break;
2171 }
2172 case KVM_GET_FPU: {
2173 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2174 r = -ENOMEM;
2175 if (!fpu)
2176 goto out;
2177 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2178 if (r)
2179 goto out;
2180 r = -EFAULT;
2181 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2182 goto out;
2183 r = 0;
2184 break;
2185 }
2186 case KVM_SET_FPU: {
2187 fpu = memdup_user(argp, sizeof(*fpu));
2188 if (IS_ERR(fpu)) {
2189 r = PTR_ERR(fpu);
2190 fpu = NULL;
2191 goto out;
2192 }
2193 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2194 break;
2195 }
2196 default:
2197 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2198 }
2199 out:
2200 vcpu_put(vcpu);
2201 kfree(fpu);
2202 kfree(kvm_sregs);
2203 return r;
2204 }
2205
2206 #ifdef CONFIG_COMPAT
2207 static long kvm_vcpu_compat_ioctl(struct file *filp,
2208 unsigned int ioctl, unsigned long arg)
2209 {
2210 struct kvm_vcpu *vcpu = filp->private_data;
2211 void __user *argp = compat_ptr(arg);
2212 int r;
2213
2214 if (vcpu->kvm->mm != current->mm)
2215 return -EIO;
2216
2217 switch (ioctl) {
2218 case KVM_SET_SIGNAL_MASK: {
2219 struct kvm_signal_mask __user *sigmask_arg = argp;
2220 struct kvm_signal_mask kvm_sigmask;
2221 compat_sigset_t csigset;
2222 sigset_t sigset;
2223
2224 if (argp) {
2225 r = -EFAULT;
2226 if (copy_from_user(&kvm_sigmask, argp,
2227 sizeof kvm_sigmask))
2228 goto out;
2229 r = -EINVAL;
2230 if (kvm_sigmask.len != sizeof csigset)
2231 goto out;
2232 r = -EFAULT;
2233 if (copy_from_user(&csigset, sigmask_arg->sigset,
2234 sizeof csigset))
2235 goto out;
2236 sigset_from_compat(&sigset, &csigset);
2237 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2238 } else
2239 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2240 break;
2241 }
2242 default:
2243 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2244 }
2245
2246 out:
2247 return r;
2248 }
2249 #endif
2250
2251 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2252 int (*accessor)(struct kvm_device *dev,
2253 struct kvm_device_attr *attr),
2254 unsigned long arg)
2255 {
2256 struct kvm_device_attr attr;
2257
2258 if (!accessor)
2259 return -EPERM;
2260
2261 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2262 return -EFAULT;
2263
2264 return accessor(dev, &attr);
2265 }
2266
2267 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2268 unsigned long arg)
2269 {
2270 struct kvm_device *dev = filp->private_data;
2271
2272 switch (ioctl) {
2273 case KVM_SET_DEVICE_ATTR:
2274 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2275 case KVM_GET_DEVICE_ATTR:
2276 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2277 case KVM_HAS_DEVICE_ATTR:
2278 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2279 default:
2280 if (dev->ops->ioctl)
2281 return dev->ops->ioctl(dev, ioctl, arg);
2282
2283 return -ENOTTY;
2284 }
2285 }
2286
2287 static int kvm_device_release(struct inode *inode, struct file *filp)
2288 {
2289 struct kvm_device *dev = filp->private_data;
2290 struct kvm *kvm = dev->kvm;
2291
2292 kvm_put_kvm(kvm);
2293 return 0;
2294 }
2295
2296 static const struct file_operations kvm_device_fops = {
2297 .unlocked_ioctl = kvm_device_ioctl,
2298 #ifdef CONFIG_COMPAT
2299 .compat_ioctl = kvm_device_ioctl,
2300 #endif
2301 .release = kvm_device_release,
2302 };
2303
2304 struct kvm_device *kvm_device_from_filp(struct file *filp)
2305 {
2306 if (filp->f_op != &kvm_device_fops)
2307 return NULL;
2308
2309 return filp->private_data;
2310 }
2311
2312 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2313 #ifdef CONFIG_KVM_MPIC
2314 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2315 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2316 #endif
2317
2318 #ifdef CONFIG_KVM_XICS
2319 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2320 #endif
2321 };
2322
2323 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2324 {
2325 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2326 return -ENOSPC;
2327
2328 if (kvm_device_ops_table[type] != NULL)
2329 return -EEXIST;
2330
2331 kvm_device_ops_table[type] = ops;
2332 return 0;
2333 }
2334
2335 void kvm_unregister_device_ops(u32 type)
2336 {
2337 if (kvm_device_ops_table[type] != NULL)
2338 kvm_device_ops_table[type] = NULL;
2339 }
2340
2341 static int kvm_ioctl_create_device(struct kvm *kvm,
2342 struct kvm_create_device *cd)
2343 {
2344 struct kvm_device_ops *ops = NULL;
2345 struct kvm_device *dev;
2346 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2347 int ret;
2348
2349 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2350 return -ENODEV;
2351
2352 ops = kvm_device_ops_table[cd->type];
2353 if (ops == NULL)
2354 return -ENODEV;
2355
2356 if (test)
2357 return 0;
2358
2359 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2360 if (!dev)
2361 return -ENOMEM;
2362
2363 dev->ops = ops;
2364 dev->kvm = kvm;
2365
2366 ret = ops->create(dev, cd->type);
2367 if (ret < 0) {
2368 kfree(dev);
2369 return ret;
2370 }
2371
2372 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2373 if (ret < 0) {
2374 ops->destroy(dev);
2375 return ret;
2376 }
2377
2378 list_add(&dev->vm_node, &kvm->devices);
2379 kvm_get_kvm(kvm);
2380 cd->fd = ret;
2381 return 0;
2382 }
2383
2384 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2385 {
2386 switch (arg) {
2387 case KVM_CAP_USER_MEMORY:
2388 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2389 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2390 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2391 case KVM_CAP_SET_BOOT_CPU_ID:
2392 #endif
2393 case KVM_CAP_INTERNAL_ERROR_DATA:
2394 #ifdef CONFIG_HAVE_KVM_MSI
2395 case KVM_CAP_SIGNAL_MSI:
2396 #endif
2397 #ifdef CONFIG_HAVE_KVM_IRQFD
2398 case KVM_CAP_IRQFD_RESAMPLE:
2399 #endif
2400 case KVM_CAP_CHECK_EXTENSION_VM:
2401 return 1;
2402 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2403 case KVM_CAP_IRQ_ROUTING:
2404 return KVM_MAX_IRQ_ROUTES;
2405 #endif
2406 default:
2407 break;
2408 }
2409 return kvm_vm_ioctl_check_extension(kvm, arg);
2410 }
2411
2412 static long kvm_vm_ioctl(struct file *filp,
2413 unsigned int ioctl, unsigned long arg)
2414 {
2415 struct kvm *kvm = filp->private_data;
2416 void __user *argp = (void __user *)arg;
2417 int r;
2418
2419 if (kvm->mm != current->mm)
2420 return -EIO;
2421 switch (ioctl) {
2422 case KVM_CREATE_VCPU:
2423 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2424 break;
2425 case KVM_SET_USER_MEMORY_REGION: {
2426 struct kvm_userspace_memory_region kvm_userspace_mem;
2427
2428 r = -EFAULT;
2429 if (copy_from_user(&kvm_userspace_mem, argp,
2430 sizeof kvm_userspace_mem))
2431 goto out;
2432
2433 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2434 break;
2435 }
2436 case KVM_GET_DIRTY_LOG: {
2437 struct kvm_dirty_log log;
2438
2439 r = -EFAULT;
2440 if (copy_from_user(&log, argp, sizeof log))
2441 goto out;
2442 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2443 break;
2444 }
2445 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2446 case KVM_REGISTER_COALESCED_MMIO: {
2447 struct kvm_coalesced_mmio_zone zone;
2448 r = -EFAULT;
2449 if (copy_from_user(&zone, argp, sizeof zone))
2450 goto out;
2451 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2452 break;
2453 }
2454 case KVM_UNREGISTER_COALESCED_MMIO: {
2455 struct kvm_coalesced_mmio_zone zone;
2456 r = -EFAULT;
2457 if (copy_from_user(&zone, argp, sizeof zone))
2458 goto out;
2459 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2460 break;
2461 }
2462 #endif
2463 case KVM_IRQFD: {
2464 struct kvm_irqfd data;
2465
2466 r = -EFAULT;
2467 if (copy_from_user(&data, argp, sizeof data))
2468 goto out;
2469 r = kvm_irqfd(kvm, &data);
2470 break;
2471 }
2472 case KVM_IOEVENTFD: {
2473 struct kvm_ioeventfd data;
2474
2475 r = -EFAULT;
2476 if (copy_from_user(&data, argp, sizeof data))
2477 goto out;
2478 r = kvm_ioeventfd(kvm, &data);
2479 break;
2480 }
2481 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2482 case KVM_SET_BOOT_CPU_ID:
2483 r = 0;
2484 mutex_lock(&kvm->lock);
2485 if (atomic_read(&kvm->online_vcpus) != 0)
2486 r = -EBUSY;
2487 else
2488 kvm->bsp_vcpu_id = arg;
2489 mutex_unlock(&kvm->lock);
2490 break;
2491 #endif
2492 #ifdef CONFIG_HAVE_KVM_MSI
2493 case KVM_SIGNAL_MSI: {
2494 struct kvm_msi msi;
2495
2496 r = -EFAULT;
2497 if (copy_from_user(&msi, argp, sizeof msi))
2498 goto out;
2499 r = kvm_send_userspace_msi(kvm, &msi);
2500 break;
2501 }
2502 #endif
2503 #ifdef __KVM_HAVE_IRQ_LINE
2504 case KVM_IRQ_LINE_STATUS:
2505 case KVM_IRQ_LINE: {
2506 struct kvm_irq_level irq_event;
2507
2508 r = -EFAULT;
2509 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2510 goto out;
2511
2512 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2513 ioctl == KVM_IRQ_LINE_STATUS);
2514 if (r)
2515 goto out;
2516
2517 r = -EFAULT;
2518 if (ioctl == KVM_IRQ_LINE_STATUS) {
2519 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2520 goto out;
2521 }
2522
2523 r = 0;
2524 break;
2525 }
2526 #endif
2527 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2528 case KVM_SET_GSI_ROUTING: {
2529 struct kvm_irq_routing routing;
2530 struct kvm_irq_routing __user *urouting;
2531 struct kvm_irq_routing_entry *entries;
2532
2533 r = -EFAULT;
2534 if (copy_from_user(&routing, argp, sizeof(routing)))
2535 goto out;
2536 r = -EINVAL;
2537 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2538 goto out;
2539 if (routing.flags)
2540 goto out;
2541 r = -ENOMEM;
2542 entries = vmalloc(routing.nr * sizeof(*entries));
2543 if (!entries)
2544 goto out;
2545 r = -EFAULT;
2546 urouting = argp;
2547 if (copy_from_user(entries, urouting->entries,
2548 routing.nr * sizeof(*entries)))
2549 goto out_free_irq_routing;
2550 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2551 routing.flags);
2552 out_free_irq_routing:
2553 vfree(entries);
2554 break;
2555 }
2556 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2557 case KVM_CREATE_DEVICE: {
2558 struct kvm_create_device cd;
2559
2560 r = -EFAULT;
2561 if (copy_from_user(&cd, argp, sizeof(cd)))
2562 goto out;
2563
2564 r = kvm_ioctl_create_device(kvm, &cd);
2565 if (r)
2566 goto out;
2567
2568 r = -EFAULT;
2569 if (copy_to_user(argp, &cd, sizeof(cd)))
2570 goto out;
2571
2572 r = 0;
2573 break;
2574 }
2575 case KVM_CHECK_EXTENSION:
2576 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2577 break;
2578 default:
2579 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2580 }
2581 out:
2582 return r;
2583 }
2584
2585 #ifdef CONFIG_COMPAT
2586 struct compat_kvm_dirty_log {
2587 __u32 slot;
2588 __u32 padding1;
2589 union {
2590 compat_uptr_t dirty_bitmap; /* one bit per page */
2591 __u64 padding2;
2592 };
2593 };
2594
2595 static long kvm_vm_compat_ioctl(struct file *filp,
2596 unsigned int ioctl, unsigned long arg)
2597 {
2598 struct kvm *kvm = filp->private_data;
2599 int r;
2600
2601 if (kvm->mm != current->mm)
2602 return -EIO;
2603 switch (ioctl) {
2604 case KVM_GET_DIRTY_LOG: {
2605 struct compat_kvm_dirty_log compat_log;
2606 struct kvm_dirty_log log;
2607
2608 r = -EFAULT;
2609 if (copy_from_user(&compat_log, (void __user *)arg,
2610 sizeof(compat_log)))
2611 goto out;
2612 log.slot = compat_log.slot;
2613 log.padding1 = compat_log.padding1;
2614 log.padding2 = compat_log.padding2;
2615 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2616
2617 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2618 break;
2619 }
2620 default:
2621 r = kvm_vm_ioctl(filp, ioctl, arg);
2622 }
2623
2624 out:
2625 return r;
2626 }
2627 #endif
2628
2629 static struct file_operations kvm_vm_fops = {
2630 .release = kvm_vm_release,
2631 .unlocked_ioctl = kvm_vm_ioctl,
2632 #ifdef CONFIG_COMPAT
2633 .compat_ioctl = kvm_vm_compat_ioctl,
2634 #endif
2635 .llseek = noop_llseek,
2636 };
2637
2638 static int kvm_dev_ioctl_create_vm(unsigned long type)
2639 {
2640 int r;
2641 struct kvm *kvm;
2642
2643 kvm = kvm_create_vm(type);
2644 if (IS_ERR(kvm))
2645 return PTR_ERR(kvm);
2646 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2647 r = kvm_coalesced_mmio_init(kvm);
2648 if (r < 0) {
2649 kvm_put_kvm(kvm);
2650 return r;
2651 }
2652 #endif
2653 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2654 if (r < 0)
2655 kvm_put_kvm(kvm);
2656
2657 return r;
2658 }
2659
2660 static long kvm_dev_ioctl(struct file *filp,
2661 unsigned int ioctl, unsigned long arg)
2662 {
2663 long r = -EINVAL;
2664
2665 switch (ioctl) {
2666 case KVM_GET_API_VERSION:
2667 if (arg)
2668 goto out;
2669 r = KVM_API_VERSION;
2670 break;
2671 case KVM_CREATE_VM:
2672 r = kvm_dev_ioctl_create_vm(arg);
2673 break;
2674 case KVM_CHECK_EXTENSION:
2675 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2676 break;
2677 case KVM_GET_VCPU_MMAP_SIZE:
2678 if (arg)
2679 goto out;
2680 r = PAGE_SIZE; /* struct kvm_run */
2681 #ifdef CONFIG_X86
2682 r += PAGE_SIZE; /* pio data page */
2683 #endif
2684 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2685 r += PAGE_SIZE; /* coalesced mmio ring page */
2686 #endif
2687 break;
2688 case KVM_TRACE_ENABLE:
2689 case KVM_TRACE_PAUSE:
2690 case KVM_TRACE_DISABLE:
2691 r = -EOPNOTSUPP;
2692 break;
2693 default:
2694 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2695 }
2696 out:
2697 return r;
2698 }
2699
2700 static struct file_operations kvm_chardev_ops = {
2701 .unlocked_ioctl = kvm_dev_ioctl,
2702 .compat_ioctl = kvm_dev_ioctl,
2703 .llseek = noop_llseek,
2704 };
2705
2706 static struct miscdevice kvm_dev = {
2707 KVM_MINOR,
2708 "kvm",
2709 &kvm_chardev_ops,
2710 };
2711
2712 static void hardware_enable_nolock(void *junk)
2713 {
2714 int cpu = raw_smp_processor_id();
2715 int r;
2716
2717 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2718 return;
2719
2720 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2721
2722 r = kvm_arch_hardware_enable();
2723
2724 if (r) {
2725 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2726 atomic_inc(&hardware_enable_failed);
2727 printk(KERN_INFO "kvm: enabling virtualization on "
2728 "CPU%d failed\n", cpu);
2729 }
2730 }
2731
2732 static void hardware_enable(void)
2733 {
2734 raw_spin_lock(&kvm_count_lock);
2735 if (kvm_usage_count)
2736 hardware_enable_nolock(NULL);
2737 raw_spin_unlock(&kvm_count_lock);
2738 }
2739
2740 static void hardware_disable_nolock(void *junk)
2741 {
2742 int cpu = raw_smp_processor_id();
2743
2744 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2745 return;
2746 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2747 kvm_arch_hardware_disable();
2748 }
2749
2750 static void hardware_disable(void)
2751 {
2752 raw_spin_lock(&kvm_count_lock);
2753 if (kvm_usage_count)
2754 hardware_disable_nolock(NULL);
2755 raw_spin_unlock(&kvm_count_lock);
2756 }
2757
2758 static void hardware_disable_all_nolock(void)
2759 {
2760 BUG_ON(!kvm_usage_count);
2761
2762 kvm_usage_count--;
2763 if (!kvm_usage_count)
2764 on_each_cpu(hardware_disable_nolock, NULL, 1);
2765 }
2766
2767 static void hardware_disable_all(void)
2768 {
2769 raw_spin_lock(&kvm_count_lock);
2770 hardware_disable_all_nolock();
2771 raw_spin_unlock(&kvm_count_lock);
2772 }
2773
2774 static int hardware_enable_all(void)
2775 {
2776 int r = 0;
2777
2778 raw_spin_lock(&kvm_count_lock);
2779
2780 kvm_usage_count++;
2781 if (kvm_usage_count == 1) {
2782 atomic_set(&hardware_enable_failed, 0);
2783 on_each_cpu(hardware_enable_nolock, NULL, 1);
2784
2785 if (atomic_read(&hardware_enable_failed)) {
2786 hardware_disable_all_nolock();
2787 r = -EBUSY;
2788 }
2789 }
2790
2791 raw_spin_unlock(&kvm_count_lock);
2792
2793 return r;
2794 }
2795
2796 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2797 void *v)
2798 {
2799 int cpu = (long)v;
2800
2801 val &= ~CPU_TASKS_FROZEN;
2802 switch (val) {
2803 case CPU_DYING:
2804 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2805 cpu);
2806 hardware_disable();
2807 break;
2808 case CPU_STARTING:
2809 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2810 cpu);
2811 hardware_enable();
2812 break;
2813 }
2814 return NOTIFY_OK;
2815 }
2816
2817 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2818 void *v)
2819 {
2820 /*
2821 * Some (well, at least mine) BIOSes hang on reboot if
2822 * in vmx root mode.
2823 *
2824 * And Intel TXT required VMX off for all cpu when system shutdown.
2825 */
2826 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2827 kvm_rebooting = true;
2828 on_each_cpu(hardware_disable_nolock, NULL, 1);
2829 return NOTIFY_OK;
2830 }
2831
2832 static struct notifier_block kvm_reboot_notifier = {
2833 .notifier_call = kvm_reboot,
2834 .priority = 0,
2835 };
2836
2837 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2838 {
2839 int i;
2840
2841 for (i = 0; i < bus->dev_count; i++) {
2842 struct kvm_io_device *pos = bus->range[i].dev;
2843
2844 kvm_iodevice_destructor(pos);
2845 }
2846 kfree(bus);
2847 }
2848
2849 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2850 const struct kvm_io_range *r2)
2851 {
2852 if (r1->addr < r2->addr)
2853 return -1;
2854 if (r1->addr + r1->len > r2->addr + r2->len)
2855 return 1;
2856 return 0;
2857 }
2858
2859 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2860 {
2861 return kvm_io_bus_cmp(p1, p2);
2862 }
2863
2864 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2865 gpa_t addr, int len)
2866 {
2867 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2868 .addr = addr,
2869 .len = len,
2870 .dev = dev,
2871 };
2872
2873 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2874 kvm_io_bus_sort_cmp, NULL);
2875
2876 return 0;
2877 }
2878
2879 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2880 gpa_t addr, int len)
2881 {
2882 struct kvm_io_range *range, key;
2883 int off;
2884
2885 key = (struct kvm_io_range) {
2886 .addr = addr,
2887 .len = len,
2888 };
2889
2890 range = bsearch(&key, bus->range, bus->dev_count,
2891 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2892 if (range == NULL)
2893 return -ENOENT;
2894
2895 off = range - bus->range;
2896
2897 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2898 off--;
2899
2900 return off;
2901 }
2902
2903 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2904 struct kvm_io_range *range, const void *val)
2905 {
2906 int idx;
2907
2908 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2909 if (idx < 0)
2910 return -EOPNOTSUPP;
2911
2912 while (idx < bus->dev_count &&
2913 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2914 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2915 range->len, val))
2916 return idx;
2917 idx++;
2918 }
2919
2920 return -EOPNOTSUPP;
2921 }
2922
2923 /* kvm_io_bus_write - called under kvm->slots_lock */
2924 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2925 int len, const void *val)
2926 {
2927 struct kvm_io_bus *bus;
2928 struct kvm_io_range range;
2929 int r;
2930
2931 range = (struct kvm_io_range) {
2932 .addr = addr,
2933 .len = len,
2934 };
2935
2936 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2937 r = __kvm_io_bus_write(bus, &range, val);
2938 return r < 0 ? r : 0;
2939 }
2940
2941 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2942 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2943 int len, const void *val, long cookie)
2944 {
2945 struct kvm_io_bus *bus;
2946 struct kvm_io_range range;
2947
2948 range = (struct kvm_io_range) {
2949 .addr = addr,
2950 .len = len,
2951 };
2952
2953 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2954
2955 /* First try the device referenced by cookie. */
2956 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2957 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2958 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2959 val))
2960 return cookie;
2961
2962 /*
2963 * cookie contained garbage; fall back to search and return the
2964 * correct cookie value.
2965 */
2966 return __kvm_io_bus_write(bus, &range, val);
2967 }
2968
2969 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2970 void *val)
2971 {
2972 int idx;
2973
2974 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2975 if (idx < 0)
2976 return -EOPNOTSUPP;
2977
2978 while (idx < bus->dev_count &&
2979 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2980 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2981 range->len, val))
2982 return idx;
2983 idx++;
2984 }
2985
2986 return -EOPNOTSUPP;
2987 }
2988 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2989
2990 /* kvm_io_bus_read - called under kvm->slots_lock */
2991 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2992 int len, void *val)
2993 {
2994 struct kvm_io_bus *bus;
2995 struct kvm_io_range range;
2996 int r;
2997
2998 range = (struct kvm_io_range) {
2999 .addr = addr,
3000 .len = len,
3001 };
3002
3003 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3004 r = __kvm_io_bus_read(bus, &range, val);
3005 return r < 0 ? r : 0;
3006 }
3007
3008
3009 /* Caller must hold slots_lock. */
3010 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3011 int len, struct kvm_io_device *dev)
3012 {
3013 struct kvm_io_bus *new_bus, *bus;
3014
3015 bus = kvm->buses[bus_idx];
3016 /* exclude ioeventfd which is limited by maximum fd */
3017 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3018 return -ENOSPC;
3019
3020 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3021 sizeof(struct kvm_io_range)), GFP_KERNEL);
3022 if (!new_bus)
3023 return -ENOMEM;
3024 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3025 sizeof(struct kvm_io_range)));
3026 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3027 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3028 synchronize_srcu_expedited(&kvm->srcu);
3029 kfree(bus);
3030
3031 return 0;
3032 }
3033
3034 /* Caller must hold slots_lock. */
3035 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3036 struct kvm_io_device *dev)
3037 {
3038 int i, r;
3039 struct kvm_io_bus *new_bus, *bus;
3040
3041 bus = kvm->buses[bus_idx];
3042 r = -ENOENT;
3043 for (i = 0; i < bus->dev_count; i++)
3044 if (bus->range[i].dev == dev) {
3045 r = 0;
3046 break;
3047 }
3048
3049 if (r)
3050 return r;
3051
3052 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3053 sizeof(struct kvm_io_range)), GFP_KERNEL);
3054 if (!new_bus)
3055 return -ENOMEM;
3056
3057 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3058 new_bus->dev_count--;
3059 memcpy(new_bus->range + i, bus->range + i + 1,
3060 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3061
3062 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3063 synchronize_srcu_expedited(&kvm->srcu);
3064 kfree(bus);
3065 return r;
3066 }
3067
3068 static struct notifier_block kvm_cpu_notifier = {
3069 .notifier_call = kvm_cpu_hotplug,
3070 };
3071
3072 static int vm_stat_get(void *_offset, u64 *val)
3073 {
3074 unsigned offset = (long)_offset;
3075 struct kvm *kvm;
3076
3077 *val = 0;
3078 spin_lock(&kvm_lock);
3079 list_for_each_entry(kvm, &vm_list, vm_list)
3080 *val += *(u32 *)((void *)kvm + offset);
3081 spin_unlock(&kvm_lock);
3082 return 0;
3083 }
3084
3085 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3086
3087 static int vcpu_stat_get(void *_offset, u64 *val)
3088 {
3089 unsigned offset = (long)_offset;
3090 struct kvm *kvm;
3091 struct kvm_vcpu *vcpu;
3092 int i;
3093
3094 *val = 0;
3095 spin_lock(&kvm_lock);
3096 list_for_each_entry(kvm, &vm_list, vm_list)
3097 kvm_for_each_vcpu(i, vcpu, kvm)
3098 *val += *(u32 *)((void *)vcpu + offset);
3099
3100 spin_unlock(&kvm_lock);
3101 return 0;
3102 }
3103
3104 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3105
3106 static const struct file_operations *stat_fops[] = {
3107 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3108 [KVM_STAT_VM] = &vm_stat_fops,
3109 };
3110
3111 static int kvm_init_debug(void)
3112 {
3113 int r = -EEXIST;
3114 struct kvm_stats_debugfs_item *p;
3115
3116 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3117 if (kvm_debugfs_dir == NULL)
3118 goto out;
3119
3120 for (p = debugfs_entries; p->name; ++p) {
3121 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3122 (void *)(long)p->offset,
3123 stat_fops[p->kind]);
3124 if (p->dentry == NULL)
3125 goto out_dir;
3126 }
3127
3128 return 0;
3129
3130 out_dir:
3131 debugfs_remove_recursive(kvm_debugfs_dir);
3132 out:
3133 return r;
3134 }
3135
3136 static void kvm_exit_debug(void)
3137 {
3138 struct kvm_stats_debugfs_item *p;
3139
3140 for (p = debugfs_entries; p->name; ++p)
3141 debugfs_remove(p->dentry);
3142 debugfs_remove(kvm_debugfs_dir);
3143 }
3144
3145 static int kvm_suspend(void)
3146 {
3147 if (kvm_usage_count)
3148 hardware_disable_nolock(NULL);
3149 return 0;
3150 }
3151
3152 static void kvm_resume(void)
3153 {
3154 if (kvm_usage_count) {
3155 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3156 hardware_enable_nolock(NULL);
3157 }
3158 }
3159
3160 static struct syscore_ops kvm_syscore_ops = {
3161 .suspend = kvm_suspend,
3162 .resume = kvm_resume,
3163 };
3164
3165 static inline
3166 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3167 {
3168 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3169 }
3170
3171 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3172 {
3173 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3174 if (vcpu->preempted)
3175 vcpu->preempted = false;
3176
3177 kvm_arch_sched_in(vcpu, cpu);
3178
3179 kvm_arch_vcpu_load(vcpu, cpu);
3180 }
3181
3182 static void kvm_sched_out(struct preempt_notifier *pn,
3183 struct task_struct *next)
3184 {
3185 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3186
3187 if (current->state == TASK_RUNNING)
3188 vcpu->preempted = true;
3189 kvm_arch_vcpu_put(vcpu);
3190 }
3191
3192 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3193 struct module *module)
3194 {
3195 int r;
3196 int cpu;
3197
3198 r = kvm_arch_init(opaque);
3199 if (r)
3200 goto out_fail;
3201
3202 /*
3203 * kvm_arch_init makes sure there's at most one caller
3204 * for architectures that support multiple implementations,
3205 * like intel and amd on x86.
3206 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3207 * conflicts in case kvm is already setup for another implementation.
3208 */
3209 r = kvm_irqfd_init();
3210 if (r)
3211 goto out_irqfd;
3212
3213 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3214 r = -ENOMEM;
3215 goto out_free_0;
3216 }
3217
3218 r = kvm_arch_hardware_setup();
3219 if (r < 0)
3220 goto out_free_0a;
3221
3222 for_each_online_cpu(cpu) {
3223 smp_call_function_single(cpu,
3224 kvm_arch_check_processor_compat,
3225 &r, 1);
3226 if (r < 0)
3227 goto out_free_1;
3228 }
3229
3230 r = register_cpu_notifier(&kvm_cpu_notifier);
3231 if (r)
3232 goto out_free_2;
3233 register_reboot_notifier(&kvm_reboot_notifier);
3234
3235 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3236 if (!vcpu_align)
3237 vcpu_align = __alignof__(struct kvm_vcpu);
3238 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3239 0, NULL);
3240 if (!kvm_vcpu_cache) {
3241 r = -ENOMEM;
3242 goto out_free_3;
3243 }
3244
3245 r = kvm_async_pf_init();
3246 if (r)
3247 goto out_free;
3248
3249 kvm_chardev_ops.owner = module;
3250 kvm_vm_fops.owner = module;
3251 kvm_vcpu_fops.owner = module;
3252
3253 r = misc_register(&kvm_dev);
3254 if (r) {
3255 printk(KERN_ERR "kvm: misc device register failed\n");
3256 goto out_unreg;
3257 }
3258
3259 register_syscore_ops(&kvm_syscore_ops);
3260
3261 kvm_preempt_ops.sched_in = kvm_sched_in;
3262 kvm_preempt_ops.sched_out = kvm_sched_out;
3263
3264 r = kvm_init_debug();
3265 if (r) {
3266 printk(KERN_ERR "kvm: create debugfs files failed\n");
3267 goto out_undebugfs;
3268 }
3269
3270 r = kvm_vfio_ops_init();
3271 WARN_ON(r);
3272
3273 return 0;
3274
3275 out_undebugfs:
3276 unregister_syscore_ops(&kvm_syscore_ops);
3277 misc_deregister(&kvm_dev);
3278 out_unreg:
3279 kvm_async_pf_deinit();
3280 out_free:
3281 kmem_cache_destroy(kvm_vcpu_cache);
3282 out_free_3:
3283 unregister_reboot_notifier(&kvm_reboot_notifier);
3284 unregister_cpu_notifier(&kvm_cpu_notifier);
3285 out_free_2:
3286 out_free_1:
3287 kvm_arch_hardware_unsetup();
3288 out_free_0a:
3289 free_cpumask_var(cpus_hardware_enabled);
3290 out_free_0:
3291 kvm_irqfd_exit();
3292 out_irqfd:
3293 kvm_arch_exit();
3294 out_fail:
3295 return r;
3296 }
3297 EXPORT_SYMBOL_GPL(kvm_init);
3298
3299 void kvm_exit(void)
3300 {
3301 kvm_exit_debug();
3302 misc_deregister(&kvm_dev);
3303 kmem_cache_destroy(kvm_vcpu_cache);
3304 kvm_async_pf_deinit();
3305 unregister_syscore_ops(&kvm_syscore_ops);
3306 unregister_reboot_notifier(&kvm_reboot_notifier);
3307 unregister_cpu_notifier(&kvm_cpu_notifier);
3308 on_each_cpu(hardware_disable_nolock, NULL, 1);
3309 kvm_arch_hardware_unsetup();
3310 kvm_arch_exit();
3311 kvm_irqfd_exit();
3312 free_cpumask_var(cpus_hardware_enabled);
3313 kvm_vfio_ops_exit();
3314 }
3315 EXPORT_SYMBOL_GPL(kvm_exit);
Linux kernel - Deliverables
This page took 0.097116 seconds and 6 git commands to generate.