projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
kvm: simplify update_memslots invocation
[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_IRQCHIP
472 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
473 #endif
474 #ifdef CONFIG_HAVE_KVM_IRQFD
475 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
476 #endif
477
478 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
479
480 r = -ENOMEM;
481 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
482 if (!kvm->memslots)
483 goto out_err_no_srcu;
484
485 /*
486 * Init kvm generation close to the maximum to easily test the
487 * code of handling generation number wrap-around.
488 */
489 kvm->memslots->generation = -150;
490
491 kvm_init_memslots_id(kvm);
492 if (init_srcu_struct(&kvm->srcu))
493 goto out_err_no_srcu;
494 if (init_srcu_struct(&kvm->irq_srcu))
495 goto out_err_no_irq_srcu;
496 for (i = 0; i < KVM_NR_BUSES; i++) {
497 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
498 GFP_KERNEL);
499 if (!kvm->buses[i])
500 goto out_err;
501 }
502
503 spin_lock_init(&kvm->mmu_lock);
504 kvm->mm = current->mm;
505 atomic_inc(&kvm->mm->mm_count);
506 kvm_eventfd_init(kvm);
507 mutex_init(&kvm->lock);
508 mutex_init(&kvm->irq_lock);
509 mutex_init(&kvm->slots_lock);
510 atomic_set(&kvm->users_count, 1);
511 INIT_LIST_HEAD(&kvm->devices);
512
513 r = kvm_init_mmu_notifier(kvm);
514 if (r)
515 goto out_err;
516
517 spin_lock(&kvm_lock);
518 list_add(&kvm->vm_list, &vm_list);
519 spin_unlock(&kvm_lock);
520
521 return kvm;
522
523 out_err:
524 cleanup_srcu_struct(&kvm->irq_srcu);
525 out_err_no_irq_srcu:
526 cleanup_srcu_struct(&kvm->srcu);
527 out_err_no_srcu:
528 hardware_disable_all();
529 out_err_no_disable:
530 for (i = 0; i < KVM_NR_BUSES; i++)
531 kfree(kvm->buses[i]);
532 kfree(kvm->memslots);
533 kvm_arch_free_vm(kvm);
534 return ERR_PTR(r);
535 }
536
537 /*
538 * Avoid using vmalloc for a small buffer.
539 * Should not be used when the size is statically known.
540 */
541 void *kvm_kvzalloc(unsigned long size)
542 {
543 if (size > PAGE_SIZE)
544 return vzalloc(size);
545 else
546 return kzalloc(size, GFP_KERNEL);
547 }
548
549 void kvm_kvfree(const void *addr)
550 {
551 if (is_vmalloc_addr(addr))
552 vfree(addr);
553 else
554 kfree(addr);
555 }
556
557 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
558 {
559 if (!memslot->dirty_bitmap)
560 return;
561
562 kvm_kvfree(memslot->dirty_bitmap);
563 memslot->dirty_bitmap = NULL;
564 }
565
566 /*
567 * Free any memory in @free but not in @dont.
568 */
569 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
570 struct kvm_memory_slot *dont)
571 {
572 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
573 kvm_destroy_dirty_bitmap(free);
574
575 kvm_arch_free_memslot(kvm, free, dont);
576
577 free->npages = 0;
578 }
579
580 static void kvm_free_physmem(struct kvm *kvm)
581 {
582 struct kvm_memslots *slots = kvm->memslots;
583 struct kvm_memory_slot *memslot;
584
585 kvm_for_each_memslot(memslot, slots)
586 kvm_free_physmem_slot(kvm, memslot, NULL);
587
588 kfree(kvm->memslots);
589 }
590
591 static void kvm_destroy_devices(struct kvm *kvm)
592 {
593 struct list_head *node, *tmp;
594
595 list_for_each_safe(node, tmp, &kvm->devices) {
596 struct kvm_device *dev =
597 list_entry(node, struct kvm_device, vm_node);
598
599 list_del(node);
600 dev->ops->destroy(dev);
601 }
602 }
603
604 static void kvm_destroy_vm(struct kvm *kvm)
605 {
606 int i;
607 struct mm_struct *mm = kvm->mm;
608
609 kvm_arch_sync_events(kvm);
610 spin_lock(&kvm_lock);
611 list_del(&kvm->vm_list);
612 spin_unlock(&kvm_lock);
613 kvm_free_irq_routing(kvm);
614 for (i = 0; i < KVM_NR_BUSES; i++)
615 kvm_io_bus_destroy(kvm->buses[i]);
616 kvm_coalesced_mmio_free(kvm);
617 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
618 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
619 #else
620 kvm_arch_flush_shadow_all(kvm);
621 #endif
622 kvm_arch_destroy_vm(kvm);
623 kvm_destroy_devices(kvm);
624 kvm_free_physmem(kvm);
625 cleanup_srcu_struct(&kvm->irq_srcu);
626 cleanup_srcu_struct(&kvm->srcu);
627 kvm_arch_free_vm(kvm);
628 hardware_disable_all();
629 mmdrop(mm);
630 }
631
632 void kvm_get_kvm(struct kvm *kvm)
633 {
634 atomic_inc(&kvm->users_count);
635 }
636 EXPORT_SYMBOL_GPL(kvm_get_kvm);
637
638 void kvm_put_kvm(struct kvm *kvm)
639 {
640 if (atomic_dec_and_test(&kvm->users_count))
641 kvm_destroy_vm(kvm);
642 }
643 EXPORT_SYMBOL_GPL(kvm_put_kvm);
644
645
646 static int kvm_vm_release(struct inode *inode, struct file *filp)
647 {
648 struct kvm *kvm = filp->private_data;
649
650 kvm_irqfd_release(kvm);
651
652 kvm_put_kvm(kvm);
653 return 0;
654 }
655
656 /*
657 * Allocation size is twice as large as the actual dirty bitmap size.
658 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
659 */
660 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
661 {
662 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
663
664 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
665 if (!memslot->dirty_bitmap)
666 return -ENOMEM;
667
668 return 0;
669 }
670
671 /*
672 * Insert memslot and re-sort memslots based on their size,
673 * so the larger slots will get better fit. Sorting algorithm
674 * takes advantage of having initially sorted array and
675 * known changed memslot position.
676 */
677 static void update_memslots(struct kvm_memslots *slots,
678 struct kvm_memory_slot *new)
679 {
680 int id = new->id;
681 int i = slots->id_to_index[id];
682 struct kvm_memory_slot *mslots = slots->memslots;
683
684 WARN_ON(mslots[i].id != id);
685 if (new->npages != mslots[i].npages) {
686 if (new->npages < mslots[i].npages) {
687 while (i < KVM_MEM_SLOTS_NUM - 1 &&
688 new->npages < mslots[i + 1].npages) {
689 mslots[i] = mslots[i + 1];
690 slots->id_to_index[mslots[i].id] = i;
691 i++;
692 }
693 } else {
694 while (i > 0 &&
695 new->npages > mslots[i - 1].npages) {
696 mslots[i] = mslots[i - 1];
697 slots->id_to_index[mslots[i].id] = i;
698 i--;
699 }
700 }
701 }
702
703 mslots[i] = *new;
704 slots->id_to_index[mslots[i].id] = i;
705 }
706
707 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
708 {
709 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
710
711 #ifdef __KVM_HAVE_READONLY_MEM
712 valid_flags |= KVM_MEM_READONLY;
713 #endif
714
715 if (mem->flags & ~valid_flags)
716 return -EINVAL;
717
718 return 0;
719 }
720
721 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
722 struct kvm_memslots *slots)
723 {
724 struct kvm_memslots *old_memslots = kvm->memslots;
725
726 /*
727 * Set the low bit in the generation, which disables SPTE caching
728 * until the end of synchronize_srcu_expedited.
729 */
730 WARN_ON(old_memslots->generation & 1);
731 slots->generation = old_memslots->generation + 1;
732
733 rcu_assign_pointer(kvm->memslots, slots);
734 synchronize_srcu_expedited(&kvm->srcu);
735
736 /*
737 * Increment the new memslot generation a second time. This prevents
738 * vm exits that race with memslot updates from caching a memslot
739 * generation that will (potentially) be valid forever.
740 */
741 slots->generation++;
742
743 kvm_arch_memslots_updated(kvm);
744
745 return old_memslots;
746 }
747
748 /*
749 * Allocate some memory and give it an address in the guest physical address
750 * space.
751 *
752 * Discontiguous memory is allowed, mostly for framebuffers.
753 *
754 * Must be called holding kvm->slots_lock for write.
755 */
756 int __kvm_set_memory_region(struct kvm *kvm,
757 struct kvm_userspace_memory_region *mem)
758 {
759 int r;
760 gfn_t base_gfn;
761 unsigned long npages;
762 struct kvm_memory_slot *slot;
763 struct kvm_memory_slot old, new;
764 struct kvm_memslots *slots = NULL, *old_memslots;
765 enum kvm_mr_change change;
766
767 r = check_memory_region_flags(mem);
768 if (r)
769 goto out;
770
771 r = -EINVAL;
772 /* General sanity checks */
773 if (mem->memory_size & (PAGE_SIZE - 1))
774 goto out;
775 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
776 goto out;
777 /* We can read the guest memory with __xxx_user() later on. */
778 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
779 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
780 !access_ok(VERIFY_WRITE,
781 (void __user *)(unsigned long)mem->userspace_addr,
782 mem->memory_size)))
783 goto out;
784 if (mem->slot >= KVM_MEM_SLOTS_NUM)
785 goto out;
786 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
787 goto out;
788
789 slot = id_to_memslot(kvm->memslots, mem->slot);
790 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
791 npages = mem->memory_size >> PAGE_SHIFT;
792
793 if (npages > KVM_MEM_MAX_NR_PAGES)
794 goto out;
795
796 if (!npages)
797 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
798
799 new = old = *slot;
800
801 new.id = mem->slot;
802 new.base_gfn = base_gfn;
803 new.npages = npages;
804 new.flags = mem->flags;
805
806 if (npages) {
807 if (!old.npages)
808 change = KVM_MR_CREATE;
809 else { /* Modify an existing slot. */
810 if ((mem->userspace_addr != old.userspace_addr) ||
811 (npages != old.npages) ||
812 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
813 goto out;
814
815 if (base_gfn != old.base_gfn)
816 change = KVM_MR_MOVE;
817 else if (new.flags != old.flags)
818 change = KVM_MR_FLAGS_ONLY;
819 else { /* Nothing to change. */
820 r = 0;
821 goto out;
822 }
823 }
824 } else if (old.npages) {
825 change = KVM_MR_DELETE;
826 } else /* Modify a non-existent slot: disallowed. */
827 goto out;
828
829 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
830 /* Check for overlaps */
831 r = -EEXIST;
832 kvm_for_each_memslot(slot, kvm->memslots) {
833 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
834 (slot->id == mem->slot))
835 continue;
836 if (!((base_gfn + npages <= slot->base_gfn) ||
837 (base_gfn >= slot->base_gfn + slot->npages)))
838 goto out;
839 }
840 }
841
842 /* Free page dirty bitmap if unneeded */
843 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
844 new.dirty_bitmap = NULL;
845
846 r = -ENOMEM;
847 if (change == KVM_MR_CREATE) {
848 new.userspace_addr = mem->userspace_addr;
849
850 if (kvm_arch_create_memslot(kvm, &new, npages))
851 goto out_free;
852 }
853
854 /* Allocate page dirty bitmap if needed */
855 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
856 if (kvm_create_dirty_bitmap(&new) < 0)
857 goto out_free;
858 }
859
860 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
861 GFP_KERNEL);
862 if (!slots)
863 goto out_free;
864
865 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
866 slot = id_to_memslot(slots, mem->slot);
867 slot->flags |= KVM_MEMSLOT_INVALID;
868
869 old_memslots = install_new_memslots(kvm, slots);
870
871 /* slot was deleted or moved, clear iommu mapping */
872 kvm_iommu_unmap_pages(kvm, &old);
873 /* From this point no new shadow pages pointing to a deleted,
874 * or moved, memslot will be created.
875 *
876 * validation of sp->gfn happens in:
877 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
878 * - kvm_is_visible_gfn (mmu_check_roots)
879 */
880 kvm_arch_flush_shadow_memslot(kvm, slot);
881
882 /*
883 * We can re-use the old_memslots from above, the only difference
884 * from the currently installed memslots is the invalid flag. This
885 * will get overwritten by update_memslots anyway.
886 */
887 slots = old_memslots;
888 }
889
890 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
891 if (r)
892 goto out_slots;
893
894 /* actual memory is freed via old in kvm_free_physmem_slot below */
895 if (change == KVM_MR_DELETE) {
896 new.dirty_bitmap = NULL;
897 memset(&new.arch, 0, sizeof(new.arch));
898 }
899
900 update_memslots(slots, &new);
901 old_memslots = install_new_memslots(kvm, slots);
902
903 kvm_arch_commit_memory_region(kvm, mem, &old, change);
904
905 kvm_free_physmem_slot(kvm, &old, &new);
906 kfree(old_memslots);
907
908 /*
909 * IOMMU mapping: New slots need to be mapped. Old slots need to be
910 * un-mapped and re-mapped if their base changes. Since base change
911 * unmapping is handled above with slot deletion, mapping alone is
912 * needed here. Anything else the iommu might care about for existing
913 * slots (size changes, userspace addr changes and read-only flag
914 * changes) is disallowed above, so any other attribute changes getting
915 * here can be skipped.
916 */
917 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
918 r = kvm_iommu_map_pages(kvm, &new);
919 return r;
920 }
921
922 return 0;
923
924 out_slots:
925 kfree(slots);
926 out_free:
927 kvm_free_physmem_slot(kvm, &new, &old);
928 out:
929 return r;
930 }
931 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
932
933 int kvm_set_memory_region(struct kvm *kvm,
934 struct kvm_userspace_memory_region *mem)
935 {
936 int r;
937
938 mutex_lock(&kvm->slots_lock);
939 r = __kvm_set_memory_region(kvm, mem);
940 mutex_unlock(&kvm->slots_lock);
941 return r;
942 }
943 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
944
945 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
946 struct kvm_userspace_memory_region *mem)
947 {
948 if (mem->slot >= KVM_USER_MEM_SLOTS)
949 return -EINVAL;
950 return kvm_set_memory_region(kvm, mem);
951 }
952
953 int kvm_get_dirty_log(struct kvm *kvm,
954 struct kvm_dirty_log *log, int *is_dirty)
955 {
956 struct kvm_memory_slot *memslot;
957 int r, i;
958 unsigned long n;
959 unsigned long any = 0;
960
961 r = -EINVAL;
962 if (log->slot >= KVM_USER_MEM_SLOTS)
963 goto out;
964
965 memslot = id_to_memslot(kvm->memslots, log->slot);
966 r = -ENOENT;
967 if (!memslot->dirty_bitmap)
968 goto out;
969
970 n = kvm_dirty_bitmap_bytes(memslot);
971
972 for (i = 0; !any && i < n/sizeof(long); ++i)
973 any = memslot->dirty_bitmap[i];
974
975 r = -EFAULT;
976 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
977 goto out;
978
979 if (any)
980 *is_dirty = 1;
981
982 r = 0;
983 out:
984 return r;
985 }
986 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
987
988 bool kvm_largepages_enabled(void)
989 {
990 return largepages_enabled;
991 }
992
993 void kvm_disable_largepages(void)
994 {
995 largepages_enabled = false;
996 }
997 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
998
999 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1000 {
1001 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1002 }
1003 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1004
1005 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1006 {
1007 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1008
1009 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1010 memslot->flags & KVM_MEMSLOT_INVALID)
1011 return 0;
1012
1013 return 1;
1014 }
1015 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1016
1017 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1018 {
1019 struct vm_area_struct *vma;
1020 unsigned long addr, size;
1021
1022 size = PAGE_SIZE;
1023
1024 addr = gfn_to_hva(kvm, gfn);
1025 if (kvm_is_error_hva(addr))
1026 return PAGE_SIZE;
1027
1028 down_read(&current->mm->mmap_sem);
1029 vma = find_vma(current->mm, addr);
1030 if (!vma)
1031 goto out;
1032
1033 size = vma_kernel_pagesize(vma);
1034
1035 out:
1036 up_read(&current->mm->mmap_sem);
1037
1038 return size;
1039 }
1040
1041 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1042 {
1043 return slot->flags & KVM_MEM_READONLY;
1044 }
1045
1046 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1047 gfn_t *nr_pages, bool write)
1048 {
1049 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1050 return KVM_HVA_ERR_BAD;
1051
1052 if (memslot_is_readonly(slot) && write)
1053 return KVM_HVA_ERR_RO_BAD;
1054
1055 if (nr_pages)
1056 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1057
1058 return __gfn_to_hva_memslot(slot, gfn);
1059 }
1060
1061 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1062 gfn_t *nr_pages)
1063 {
1064 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1065 }
1066
1067 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1068 gfn_t gfn)
1069 {
1070 return gfn_to_hva_many(slot, gfn, NULL);
1071 }
1072 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1073
1074 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1075 {
1076 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1077 }
1078 EXPORT_SYMBOL_GPL(gfn_to_hva);
1079
1080 /*
1081 * If writable is set to false, the hva returned by this function is only
1082 * allowed to be read.
1083 */
1084 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1085 gfn_t gfn, bool *writable)
1086 {
1087 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1088
1089 if (!kvm_is_error_hva(hva) && writable)
1090 *writable = !memslot_is_readonly(slot);
1091
1092 return hva;
1093 }
1094
1095 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1096 {
1097 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1098
1099 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1100 }
1101
1102 static int kvm_read_hva(void *data, void __user *hva, int len)
1103 {
1104 return __copy_from_user(data, hva, len);
1105 }
1106
1107 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1108 {
1109 return __copy_from_user_inatomic(data, hva, len);
1110 }
1111
1112 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1113 unsigned long start, int write, struct page **page)
1114 {
1115 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1116
1117 if (write)
1118 flags |= FOLL_WRITE;
1119
1120 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1121 }
1122
1123 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1124 unsigned long addr, bool write_fault,
1125 struct page **pagep)
1126 {
1127 int npages;
1128 int locked = 1;
1129 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1130 (pagep ? FOLL_GET : 0) |
1131 (write_fault ? FOLL_WRITE : 0);
1132
1133 /*
1134 * If retrying the fault, we get here *not* having allowed the filemap
1135 * to wait on the page lock. We should now allow waiting on the IO with
1136 * the mmap semaphore released.
1137 */
1138 down_read(&mm->mmap_sem);
1139 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1140 &locked);
1141 if (!locked) {
1142 VM_BUG_ON(npages);
1143
1144 if (!pagep)
1145 return 0;
1146
1147 /*
1148 * The previous call has now waited on the IO. Now we can
1149 * retry and complete. Pass TRIED to ensure we do not re
1150 * schedule async IO (see e.g. filemap_fault).
1151 */
1152 down_read(&mm->mmap_sem);
1153 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1154 pagep, NULL, NULL);
1155 }
1156 up_read(&mm->mmap_sem);
1157 return npages;
1158 }
1159
1160 static inline int check_user_page_hwpoison(unsigned long addr)
1161 {
1162 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1163
1164 rc = __get_user_pages(current, current->mm, addr, 1,
1165 flags, NULL, NULL, NULL);
1166 return rc == -EHWPOISON;
1167 }
1168
1169 /*
1170 * The atomic path to get the writable pfn which will be stored in @pfn,
1171 * true indicates success, otherwise false is returned.
1172 */
1173 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1174 bool write_fault, bool *writable, pfn_t *pfn)
1175 {
1176 struct page *page[1];
1177 int npages;
1178
1179 if (!(async || atomic))
1180 return false;
1181
1182 /*
1183 * Fast pin a writable pfn only if it is a write fault request
1184 * or the caller allows to map a writable pfn for a read fault
1185 * request.
1186 */
1187 if (!(write_fault || writable))
1188 return false;
1189
1190 npages = __get_user_pages_fast(addr, 1, 1, page);
1191 if (npages == 1) {
1192 *pfn = page_to_pfn(page[0]);
1193
1194 if (writable)
1195 *writable = true;
1196 return true;
1197 }
1198
1199 return false;
1200 }
1201
1202 /*
1203 * The slow path to get the pfn of the specified host virtual address,
1204 * 1 indicates success, -errno is returned if error is detected.
1205 */
1206 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1207 bool *writable, pfn_t *pfn)
1208 {
1209 struct page *page[1];
1210 int npages = 0;
1211
1212 might_sleep();
1213
1214 if (writable)
1215 *writable = write_fault;
1216
1217 if (async) {
1218 down_read(&current->mm->mmap_sem);
1219 npages = get_user_page_nowait(current, current->mm,
1220 addr, write_fault, page);
1221 up_read(&current->mm->mmap_sem);
1222 } else {
1223 /*
1224 * By now we have tried gup_fast, and possibly async_pf, and we
1225 * are certainly not atomic. Time to retry the gup, allowing
1226 * mmap semaphore to be relinquished in the case of IO.
1227 */
1228 npages = kvm_get_user_page_io(current, current->mm, addr,
1229 write_fault, page);
1230 }
1231 if (npages != 1)
1232 return npages;
1233
1234 /* map read fault as writable if possible */
1235 if (unlikely(!write_fault) && writable) {
1236 struct page *wpage[1];
1237
1238 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1239 if (npages == 1) {
1240 *writable = true;
1241 put_page(page[0]);
1242 page[0] = wpage[0];
1243 }
1244
1245 npages = 1;
1246 }
1247 *pfn = page_to_pfn(page[0]);
1248 return npages;
1249 }
1250
1251 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1252 {
1253 if (unlikely(!(vma->vm_flags & VM_READ)))
1254 return false;
1255
1256 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1257 return false;
1258
1259 return true;
1260 }
1261
1262 /*
1263 * Pin guest page in memory and return its pfn.
1264 * @addr: host virtual address which maps memory to the guest
1265 * @atomic: whether this function can sleep
1266 * @async: whether this function need to wait IO complete if the
1267 * host page is not in the memory
1268 * @write_fault: whether we should get a writable host page
1269 * @writable: whether it allows to map a writable host page for !@write_fault
1270 *
1271 * The function will map a writable host page for these two cases:
1272 * 1): @write_fault = true
1273 * 2): @write_fault = false && @writable, @writable will tell the caller
1274 * whether the mapping is writable.
1275 */
1276 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1277 bool write_fault, bool *writable)
1278 {
1279 struct vm_area_struct *vma;
1280 pfn_t pfn = 0;
1281 int npages;
1282
1283 /* we can do it either atomically or asynchronously, not both */
1284 BUG_ON(atomic && async);
1285
1286 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1287 return pfn;
1288
1289 if (atomic)
1290 return KVM_PFN_ERR_FAULT;
1291
1292 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1293 if (npages == 1)
1294 return pfn;
1295
1296 down_read(&current->mm->mmap_sem);
1297 if (npages == -EHWPOISON ||
1298 (!async && check_user_page_hwpoison(addr))) {
1299 pfn = KVM_PFN_ERR_HWPOISON;
1300 goto exit;
1301 }
1302
1303 vma = find_vma_intersection(current->mm, addr, addr + 1);
1304
1305 if (vma == NULL)
1306 pfn = KVM_PFN_ERR_FAULT;
1307 else if ((vma->vm_flags & VM_PFNMAP)) {
1308 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1309 vma->vm_pgoff;
1310 BUG_ON(!kvm_is_mmio_pfn(pfn));
1311 } else {
1312 if (async && vma_is_valid(vma, write_fault))
1313 *async = true;
1314 pfn = KVM_PFN_ERR_FAULT;
1315 }
1316 exit:
1317 up_read(&current->mm->mmap_sem);
1318 return pfn;
1319 }
1320
1321 static pfn_t
1322 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1323 bool *async, bool write_fault, bool *writable)
1324 {
1325 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1326
1327 if (addr == KVM_HVA_ERR_RO_BAD)
1328 return KVM_PFN_ERR_RO_FAULT;
1329
1330 if (kvm_is_error_hva(addr))
1331 return KVM_PFN_NOSLOT;
1332
1333 /* Do not map writable pfn in the readonly memslot. */
1334 if (writable && memslot_is_readonly(slot)) {
1335 *writable = false;
1336 writable = NULL;
1337 }
1338
1339 return hva_to_pfn(addr, atomic, async, write_fault,
1340 writable);
1341 }
1342
1343 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1344 bool write_fault, bool *writable)
1345 {
1346 struct kvm_memory_slot *slot;
1347
1348 if (async)
1349 *async = false;
1350
1351 slot = gfn_to_memslot(kvm, gfn);
1352
1353 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1354 writable);
1355 }
1356
1357 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1358 {
1359 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1360 }
1361 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1362
1363 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1364 bool write_fault, bool *writable)
1365 {
1366 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1367 }
1368 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1369
1370 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1371 {
1372 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1373 }
1374 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1375
1376 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1377 bool *writable)
1378 {
1379 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1380 }
1381 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1382
1383 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1384 {
1385 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1386 }
1387
1388 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1389 {
1390 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1391 }
1392 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1393
1394 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1395 int nr_pages)
1396 {
1397 unsigned long addr;
1398 gfn_t entry;
1399
1400 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1401 if (kvm_is_error_hva(addr))
1402 return -1;
1403
1404 if (entry < nr_pages)
1405 return 0;
1406
1407 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1408 }
1409 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1410
1411 static struct page *kvm_pfn_to_page(pfn_t pfn)
1412 {
1413 if (is_error_noslot_pfn(pfn))
1414 return KVM_ERR_PTR_BAD_PAGE;
1415
1416 if (kvm_is_mmio_pfn(pfn)) {
1417 WARN_ON(1);
1418 return KVM_ERR_PTR_BAD_PAGE;
1419 }
1420
1421 return pfn_to_page(pfn);
1422 }
1423
1424 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1425 {
1426 pfn_t pfn;
1427
1428 pfn = gfn_to_pfn(kvm, gfn);
1429
1430 return kvm_pfn_to_page(pfn);
1431 }
1432
1433 EXPORT_SYMBOL_GPL(gfn_to_page);
1434
1435 void kvm_release_page_clean(struct page *page)
1436 {
1437 WARN_ON(is_error_page(page));
1438
1439 kvm_release_pfn_clean(page_to_pfn(page));
1440 }
1441 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1442
1443 void kvm_release_pfn_clean(pfn_t pfn)
1444 {
1445 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1446 put_page(pfn_to_page(pfn));
1447 }
1448 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1449
1450 void kvm_release_page_dirty(struct page *page)
1451 {
1452 WARN_ON(is_error_page(page));
1453
1454 kvm_release_pfn_dirty(page_to_pfn(page));
1455 }
1456 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1457
1458 static void kvm_release_pfn_dirty(pfn_t pfn)
1459 {
1460 kvm_set_pfn_dirty(pfn);
1461 kvm_release_pfn_clean(pfn);
1462 }
1463
1464 void kvm_set_pfn_dirty(pfn_t pfn)
1465 {
1466 if (!kvm_is_mmio_pfn(pfn)) {
1467 struct page *page = pfn_to_page(pfn);
1468 if (!PageReserved(page))
1469 SetPageDirty(page);
1470 }
1471 }
1472 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1473
1474 void kvm_set_pfn_accessed(pfn_t pfn)
1475 {
1476 if (!kvm_is_mmio_pfn(pfn))
1477 mark_page_accessed(pfn_to_page(pfn));
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1480
1481 void kvm_get_pfn(pfn_t pfn)
1482 {
1483 if (!kvm_is_mmio_pfn(pfn))
1484 get_page(pfn_to_page(pfn));
1485 }
1486 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1487
1488 static int next_segment(unsigned long len, int offset)
1489 {
1490 if (len > PAGE_SIZE - offset)
1491 return PAGE_SIZE - offset;
1492 else
1493 return len;
1494 }
1495
1496 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1497 int len)
1498 {
1499 int r;
1500 unsigned long addr;
1501
1502 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1503 if (kvm_is_error_hva(addr))
1504 return -EFAULT;
1505 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1506 if (r)
1507 return -EFAULT;
1508 return 0;
1509 }
1510 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1511
1512 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1513 {
1514 gfn_t gfn = gpa >> PAGE_SHIFT;
1515 int seg;
1516 int offset = offset_in_page(gpa);
1517 int ret;
1518
1519 while ((seg = next_segment(len, offset)) != 0) {
1520 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1521 if (ret < 0)
1522 return ret;
1523 offset = 0;
1524 len -= seg;
1525 data += seg;
1526 ++gfn;
1527 }
1528 return 0;
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_read_guest);
1531
1532 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1533 unsigned long len)
1534 {
1535 int r;
1536 unsigned long addr;
1537 gfn_t gfn = gpa >> PAGE_SHIFT;
1538 int offset = offset_in_page(gpa);
1539
1540 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1541 if (kvm_is_error_hva(addr))
1542 return -EFAULT;
1543 pagefault_disable();
1544 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1545 pagefault_enable();
1546 if (r)
1547 return -EFAULT;
1548 return 0;
1549 }
1550 EXPORT_SYMBOL(kvm_read_guest_atomic);
1551
1552 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1553 int offset, int len)
1554 {
1555 int r;
1556 unsigned long addr;
1557
1558 addr = gfn_to_hva(kvm, gfn);
1559 if (kvm_is_error_hva(addr))
1560 return -EFAULT;
1561 r = __copy_to_user((void __user *)addr + offset, data, len);
1562 if (r)
1563 return -EFAULT;
1564 mark_page_dirty(kvm, gfn);
1565 return 0;
1566 }
1567 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1568
1569 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1570 unsigned long len)
1571 {
1572 gfn_t gfn = gpa >> PAGE_SHIFT;
1573 int seg;
1574 int offset = offset_in_page(gpa);
1575 int ret;
1576
1577 while ((seg = next_segment(len, offset)) != 0) {
1578 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1579 if (ret < 0)
1580 return ret;
1581 offset = 0;
1582 len -= seg;
1583 data += seg;
1584 ++gfn;
1585 }
1586 return 0;
1587 }
1588
1589 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1590 gpa_t gpa, unsigned long len)
1591 {
1592 struct kvm_memslots *slots = kvm_memslots(kvm);
1593 int offset = offset_in_page(gpa);
1594 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1595 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1596 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1597 gfn_t nr_pages_avail;
1598
1599 ghc->gpa = gpa;
1600 ghc->generation = slots->generation;
1601 ghc->len = len;
1602 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1603 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1604 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1605 ghc->hva += offset;
1606 } else {
1607 /*
1608 * If the requested region crosses two memslots, we still
1609 * verify that the entire region is valid here.
1610 */
1611 while (start_gfn <= end_gfn) {
1612 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1613 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1614 &nr_pages_avail);
1615 if (kvm_is_error_hva(ghc->hva))
1616 return -EFAULT;
1617 start_gfn += nr_pages_avail;
1618 }
1619 /* Use the slow path for cross page reads and writes. */
1620 ghc->memslot = NULL;
1621 }
1622 return 0;
1623 }
1624 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1625
1626 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1627 void *data, unsigned long len)
1628 {
1629 struct kvm_memslots *slots = kvm_memslots(kvm);
1630 int r;
1631
1632 BUG_ON(len > ghc->len);
1633
1634 if (slots->generation != ghc->generation)
1635 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1636
1637 if (unlikely(!ghc->memslot))
1638 return kvm_write_guest(kvm, ghc->gpa, data, len);
1639
1640 if (kvm_is_error_hva(ghc->hva))
1641 return -EFAULT;
1642
1643 r = __copy_to_user((void __user *)ghc->hva, data, len);
1644 if (r)
1645 return -EFAULT;
1646 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1647
1648 return 0;
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1651
1652 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1653 void *data, unsigned long len)
1654 {
1655 struct kvm_memslots *slots = kvm_memslots(kvm);
1656 int r;
1657
1658 BUG_ON(len > ghc->len);
1659
1660 if (slots->generation != ghc->generation)
1661 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1662
1663 if (unlikely(!ghc->memslot))
1664 return kvm_read_guest(kvm, ghc->gpa, data, len);
1665
1666 if (kvm_is_error_hva(ghc->hva))
1667 return -EFAULT;
1668
1669 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1670 if (r)
1671 return -EFAULT;
1672
1673 return 0;
1674 }
1675 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1676
1677 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1678 {
1679 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1680
1681 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1682 }
1683 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1684
1685 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1686 {
1687 gfn_t gfn = gpa >> PAGE_SHIFT;
1688 int seg;
1689 int offset = offset_in_page(gpa);
1690 int ret;
1691
1692 while ((seg = next_segment(len, offset)) != 0) {
1693 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1694 if (ret < 0)
1695 return ret;
1696 offset = 0;
1697 len -= seg;
1698 ++gfn;
1699 }
1700 return 0;
1701 }
1702 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1703
1704 static void mark_page_dirty_in_slot(struct kvm *kvm,
1705 struct kvm_memory_slot *memslot,
1706 gfn_t gfn)
1707 {
1708 if (memslot && memslot->dirty_bitmap) {
1709 unsigned long rel_gfn = gfn - memslot->base_gfn;
1710
1711 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1712 }
1713 }
1714
1715 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1716 {
1717 struct kvm_memory_slot *memslot;
1718
1719 memslot = gfn_to_memslot(kvm, gfn);
1720 mark_page_dirty_in_slot(kvm, memslot, gfn);
1721 }
1722 EXPORT_SYMBOL_GPL(mark_page_dirty);
1723
1724 /*
1725 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1726 */
1727 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1728 {
1729 DEFINE_WAIT(wait);
1730
1731 for (;;) {
1732 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1733
1734 if (kvm_arch_vcpu_runnable(vcpu)) {
1735 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1736 break;
1737 }
1738 if (kvm_cpu_has_pending_timer(vcpu))
1739 break;
1740 if (signal_pending(current))
1741 break;
1742
1743 schedule();
1744 }
1745
1746 finish_wait(&vcpu->wq, &wait);
1747 }
1748 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1749
1750 #ifndef CONFIG_S390
1751 /*
1752 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1753 */
1754 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1755 {
1756 int me;
1757 int cpu = vcpu->cpu;
1758 wait_queue_head_t *wqp;
1759
1760 wqp = kvm_arch_vcpu_wq(vcpu);
1761 if (waitqueue_active(wqp)) {
1762 wake_up_interruptible(wqp);
1763 ++vcpu->stat.halt_wakeup;
1764 }
1765
1766 me = get_cpu();
1767 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1768 if (kvm_arch_vcpu_should_kick(vcpu))
1769 smp_send_reschedule(cpu);
1770 put_cpu();
1771 }
1772 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1773 #endif /* !CONFIG_S390 */
1774
1775 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1776 {
1777 struct pid *pid;
1778 struct task_struct *task = NULL;
1779 int ret = 0;
1780
1781 rcu_read_lock();
1782 pid = rcu_dereference(target->pid);
1783 if (pid)
1784 task = get_pid_task(pid, PIDTYPE_PID);
1785 rcu_read_unlock();
1786 if (!task)
1787 return ret;
1788 if (task->flags & PF_VCPU) {
1789 put_task_struct(task);
1790 return ret;
1791 }
1792 ret = yield_to(task, 1);
1793 put_task_struct(task);
1794
1795 return ret;
1796 }
1797 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1798
1799 /*
1800 * Helper that checks whether a VCPU is eligible for directed yield.
1801 * Most eligible candidate to yield is decided by following heuristics:
1802 *
1803 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1804 * (preempted lock holder), indicated by @in_spin_loop.
1805 * Set at the beiginning and cleared at the end of interception/PLE handler.
1806 *
1807 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1808 * chance last time (mostly it has become eligible now since we have probably
1809 * yielded to lockholder in last iteration. This is done by toggling
1810 * @dy_eligible each time a VCPU checked for eligibility.)
1811 *
1812 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1813 * to preempted lock-holder could result in wrong VCPU selection and CPU
1814 * burning. Giving priority for a potential lock-holder increases lock
1815 * progress.
1816 *
1817 * Since algorithm is based on heuristics, accessing another VCPU data without
1818 * locking does not harm. It may result in trying to yield to same VCPU, fail
1819 * and continue with next VCPU and so on.
1820 */
1821 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1822 {
1823 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1824 bool eligible;
1825
1826 eligible = !vcpu->spin_loop.in_spin_loop ||
1827 vcpu->spin_loop.dy_eligible;
1828
1829 if (vcpu->spin_loop.in_spin_loop)
1830 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1831
1832 return eligible;
1833 #else
1834 return true;
1835 #endif
1836 }
1837
1838 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1839 {
1840 struct kvm *kvm = me->kvm;
1841 struct kvm_vcpu *vcpu;
1842 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1843 int yielded = 0;
1844 int try = 3;
1845 int pass;
1846 int i;
1847
1848 kvm_vcpu_set_in_spin_loop(me, true);
1849 /*
1850 * We boost the priority of a VCPU that is runnable but not
1851 * currently running, because it got preempted by something
1852 * else and called schedule in __vcpu_run. Hopefully that
1853 * VCPU is holding the lock that we need and will release it.
1854 * We approximate round-robin by starting at the last boosted VCPU.
1855 */
1856 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1857 kvm_for_each_vcpu(i, vcpu, kvm) {
1858 if (!pass && i <= last_boosted_vcpu) {
1859 i = last_boosted_vcpu;
1860 continue;
1861 } else if (pass && i > last_boosted_vcpu)
1862 break;
1863 if (!ACCESS_ONCE(vcpu->preempted))
1864 continue;
1865 if (vcpu == me)
1866 continue;
1867 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1868 continue;
1869 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1870 continue;
1871
1872 yielded = kvm_vcpu_yield_to(vcpu);
1873 if (yielded > 0) {
1874 kvm->last_boosted_vcpu = i;
1875 break;
1876 } else if (yielded < 0) {
1877 try--;
1878 if (!try)
1879 break;
1880 }
1881 }
1882 }
1883 kvm_vcpu_set_in_spin_loop(me, false);
1884
1885 /* Ensure vcpu is not eligible during next spinloop */
1886 kvm_vcpu_set_dy_eligible(me, false);
1887 }
1888 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1889
1890 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1891 {
1892 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1893 struct page *page;
1894
1895 if (vmf->pgoff == 0)
1896 page = virt_to_page(vcpu->run);
1897 #ifdef CONFIG_X86
1898 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1899 page = virt_to_page(vcpu->arch.pio_data);
1900 #endif
1901 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1902 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1903 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1904 #endif
1905 else
1906 return kvm_arch_vcpu_fault(vcpu, vmf);
1907 get_page(page);
1908 vmf->page = page;
1909 return 0;
1910 }
1911
1912 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1913 .fault = kvm_vcpu_fault,
1914 };
1915
1916 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1917 {
1918 vma->vm_ops = &kvm_vcpu_vm_ops;
1919 return 0;
1920 }
1921
1922 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1923 {
1924 struct kvm_vcpu *vcpu = filp->private_data;
1925
1926 kvm_put_kvm(vcpu->kvm);
1927 return 0;
1928 }
1929
1930 static struct file_operations kvm_vcpu_fops = {
1931 .release = kvm_vcpu_release,
1932 .unlocked_ioctl = kvm_vcpu_ioctl,
1933 #ifdef CONFIG_COMPAT
1934 .compat_ioctl = kvm_vcpu_compat_ioctl,
1935 #endif
1936 .mmap = kvm_vcpu_mmap,
1937 .llseek = noop_llseek,
1938 };
1939
1940 /*
1941 * Allocates an inode for the vcpu.
1942 */
1943 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1944 {
1945 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1946 }
1947
1948 /*
1949 * Creates some virtual cpus. Good luck creating more than one.
1950 */
1951 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1952 {
1953 int r;
1954 struct kvm_vcpu *vcpu, *v;
1955
1956 if (id >= KVM_MAX_VCPUS)
1957 return -EINVAL;
1958
1959 vcpu = kvm_arch_vcpu_create(kvm, id);
1960 if (IS_ERR(vcpu))
1961 return PTR_ERR(vcpu);
1962
1963 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1964
1965 r = kvm_arch_vcpu_setup(vcpu);
1966 if (r)
1967 goto vcpu_destroy;
1968
1969 mutex_lock(&kvm->lock);
1970 if (!kvm_vcpu_compatible(vcpu)) {
1971 r = -EINVAL;
1972 goto unlock_vcpu_destroy;
1973 }
1974 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1975 r = -EINVAL;
1976 goto unlock_vcpu_destroy;
1977 }
1978
1979 kvm_for_each_vcpu(r, v, kvm)
1980 if (v->vcpu_id == id) {
1981 r = -EEXIST;
1982 goto unlock_vcpu_destroy;
1983 }
1984
1985 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1986
1987 /* Now it's all set up, let userspace reach it */
1988 kvm_get_kvm(kvm);
1989 r = create_vcpu_fd(vcpu);
1990 if (r < 0) {
1991 kvm_put_kvm(kvm);
1992 goto unlock_vcpu_destroy;
1993 }
1994
1995 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1996 smp_wmb();
1997 atomic_inc(&kvm->online_vcpus);
1998
1999 mutex_unlock(&kvm->lock);
2000 kvm_arch_vcpu_postcreate(vcpu);
2001 return r;
2002
2003 unlock_vcpu_destroy:
2004 mutex_unlock(&kvm->lock);
2005 vcpu_destroy:
2006 kvm_arch_vcpu_destroy(vcpu);
2007 return r;
2008 }
2009
2010 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2011 {
2012 if (sigset) {
2013 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2014 vcpu->sigset_active = 1;
2015 vcpu->sigset = *sigset;
2016 } else
2017 vcpu->sigset_active = 0;
2018 return 0;
2019 }
2020
2021 static long kvm_vcpu_ioctl(struct file *filp,
2022 unsigned int ioctl, unsigned long arg)
2023 {
2024 struct kvm_vcpu *vcpu = filp->private_data;
2025 void __user *argp = (void __user *)arg;
2026 int r;
2027 struct kvm_fpu *fpu = NULL;
2028 struct kvm_sregs *kvm_sregs = NULL;
2029
2030 if (vcpu->kvm->mm != current->mm)
2031 return -EIO;
2032
2033 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2034 return -EINVAL;
2035
2036 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2037 /*
2038 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2039 * so vcpu_load() would break it.
2040 */
2041 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2042 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2043 #endif
2044
2045
2046 r = vcpu_load(vcpu);
2047 if (r)
2048 return r;
2049 switch (ioctl) {
2050 case KVM_RUN:
2051 r = -EINVAL;
2052 if (arg)
2053 goto out;
2054 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2055 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2056 break;
2057 case KVM_GET_REGS: {
2058 struct kvm_regs *kvm_regs;
2059
2060 r = -ENOMEM;
2061 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2062 if (!kvm_regs)
2063 goto out;
2064 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2065 if (r)
2066 goto out_free1;
2067 r = -EFAULT;
2068 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2069 goto out_free1;
2070 r = 0;
2071 out_free1:
2072 kfree(kvm_regs);
2073 break;
2074 }
2075 case KVM_SET_REGS: {
2076 struct kvm_regs *kvm_regs;
2077
2078 r = -ENOMEM;
2079 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2080 if (IS_ERR(kvm_regs)) {
2081 r = PTR_ERR(kvm_regs);
2082 goto out;
2083 }
2084 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2085 kfree(kvm_regs);
2086 break;
2087 }
2088 case KVM_GET_SREGS: {
2089 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2090 r = -ENOMEM;
2091 if (!kvm_sregs)
2092 goto out;
2093 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2094 if (r)
2095 goto out;
2096 r = -EFAULT;
2097 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2098 goto out;
2099 r = 0;
2100 break;
2101 }
2102 case KVM_SET_SREGS: {
2103 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2104 if (IS_ERR(kvm_sregs)) {
2105 r = PTR_ERR(kvm_sregs);
2106 kvm_sregs = NULL;
2107 goto out;
2108 }
2109 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2110 break;
2111 }
2112 case KVM_GET_MP_STATE: {
2113 struct kvm_mp_state mp_state;
2114
2115 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2116 if (r)
2117 goto out;
2118 r = -EFAULT;
2119 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2120 goto out;
2121 r = 0;
2122 break;
2123 }
2124 case KVM_SET_MP_STATE: {
2125 struct kvm_mp_state mp_state;
2126
2127 r = -EFAULT;
2128 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2129 goto out;
2130 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2131 break;
2132 }
2133 case KVM_TRANSLATE: {
2134 struct kvm_translation tr;
2135
2136 r = -EFAULT;
2137 if (copy_from_user(&tr, argp, sizeof tr))
2138 goto out;
2139 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2140 if (r)
2141 goto out;
2142 r = -EFAULT;
2143 if (copy_to_user(argp, &tr, sizeof tr))
2144 goto out;
2145 r = 0;
2146 break;
2147 }
2148 case KVM_SET_GUEST_DEBUG: {
2149 struct kvm_guest_debug dbg;
2150
2151 r = -EFAULT;
2152 if (copy_from_user(&dbg, argp, sizeof dbg))
2153 goto out;
2154 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2155 break;
2156 }
2157 case KVM_SET_SIGNAL_MASK: {
2158 struct kvm_signal_mask __user *sigmask_arg = argp;
2159 struct kvm_signal_mask kvm_sigmask;
2160 sigset_t sigset, *p;
2161
2162 p = NULL;
2163 if (argp) {
2164 r = -EFAULT;
2165 if (copy_from_user(&kvm_sigmask, argp,
2166 sizeof kvm_sigmask))
2167 goto out;
2168 r = -EINVAL;
2169 if (kvm_sigmask.len != sizeof sigset)
2170 goto out;
2171 r = -EFAULT;
2172 if (copy_from_user(&sigset, sigmask_arg->sigset,
2173 sizeof sigset))
2174 goto out;
2175 p = &sigset;
2176 }
2177 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2178 break;
2179 }
2180 case KVM_GET_FPU: {
2181 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2182 r = -ENOMEM;
2183 if (!fpu)
2184 goto out;
2185 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2186 if (r)
2187 goto out;
2188 r = -EFAULT;
2189 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2190 goto out;
2191 r = 0;
2192 break;
2193 }
2194 case KVM_SET_FPU: {
2195 fpu = memdup_user(argp, sizeof(*fpu));
2196 if (IS_ERR(fpu)) {
2197 r = PTR_ERR(fpu);
2198 fpu = NULL;
2199 goto out;
2200 }
2201 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2202 break;
2203 }
2204 default:
2205 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2206 }
2207 out:
2208 vcpu_put(vcpu);
2209 kfree(fpu);
2210 kfree(kvm_sregs);
2211 return r;
2212 }
2213
2214 #ifdef CONFIG_COMPAT
2215 static long kvm_vcpu_compat_ioctl(struct file *filp,
2216 unsigned int ioctl, unsigned long arg)
2217 {
2218 struct kvm_vcpu *vcpu = filp->private_data;
2219 void __user *argp = compat_ptr(arg);
2220 int r;
2221
2222 if (vcpu->kvm->mm != current->mm)
2223 return -EIO;
2224
2225 switch (ioctl) {
2226 case KVM_SET_SIGNAL_MASK: {
2227 struct kvm_signal_mask __user *sigmask_arg = argp;
2228 struct kvm_signal_mask kvm_sigmask;
2229 compat_sigset_t csigset;
2230 sigset_t sigset;
2231
2232 if (argp) {
2233 r = -EFAULT;
2234 if (copy_from_user(&kvm_sigmask, argp,
2235 sizeof kvm_sigmask))
2236 goto out;
2237 r = -EINVAL;
2238 if (kvm_sigmask.len != sizeof csigset)
2239 goto out;
2240 r = -EFAULT;
2241 if (copy_from_user(&csigset, sigmask_arg->sigset,
2242 sizeof csigset))
2243 goto out;
2244 sigset_from_compat(&sigset, &csigset);
2245 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2246 } else
2247 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2248 break;
2249 }
2250 default:
2251 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2252 }
2253
2254 out:
2255 return r;
2256 }
2257 #endif
2258
2259 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2260 int (*accessor)(struct kvm_device *dev,
2261 struct kvm_device_attr *attr),
2262 unsigned long arg)
2263 {
2264 struct kvm_device_attr attr;
2265
2266 if (!accessor)
2267 return -EPERM;
2268
2269 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2270 return -EFAULT;
2271
2272 return accessor(dev, &attr);
2273 }
2274
2275 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2276 unsigned long arg)
2277 {
2278 struct kvm_device *dev = filp->private_data;
2279
2280 switch (ioctl) {
2281 case KVM_SET_DEVICE_ATTR:
2282 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2283 case KVM_GET_DEVICE_ATTR:
2284 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2285 case KVM_HAS_DEVICE_ATTR:
2286 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2287 default:
2288 if (dev->ops->ioctl)
2289 return dev->ops->ioctl(dev, ioctl, arg);
2290
2291 return -ENOTTY;
2292 }
2293 }
2294
2295 static int kvm_device_release(struct inode *inode, struct file *filp)
2296 {
2297 struct kvm_device *dev = filp->private_data;
2298 struct kvm *kvm = dev->kvm;
2299
2300 kvm_put_kvm(kvm);
2301 return 0;
2302 }
2303
2304 static const struct file_operations kvm_device_fops = {
2305 .unlocked_ioctl = kvm_device_ioctl,
2306 #ifdef CONFIG_COMPAT
2307 .compat_ioctl = kvm_device_ioctl,
2308 #endif
2309 .release = kvm_device_release,
2310 };
2311
2312 struct kvm_device *kvm_device_from_filp(struct file *filp)
2313 {
2314 if (filp->f_op != &kvm_device_fops)
2315 return NULL;
2316
2317 return filp->private_data;
2318 }
2319
2320 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2321 #ifdef CONFIG_KVM_MPIC
2322 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2323 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2324 #endif
2325
2326 #ifdef CONFIG_KVM_XICS
2327 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2328 #endif
2329 };
2330
2331 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2332 {
2333 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2334 return -ENOSPC;
2335
2336 if (kvm_device_ops_table[type] != NULL)
2337 return -EEXIST;
2338
2339 kvm_device_ops_table[type] = ops;
2340 return 0;
2341 }
2342
2343 void kvm_unregister_device_ops(u32 type)
2344 {
2345 if (kvm_device_ops_table[type] != NULL)
2346 kvm_device_ops_table[type] = NULL;
2347 }
2348
2349 static int kvm_ioctl_create_device(struct kvm *kvm,
2350 struct kvm_create_device *cd)
2351 {
2352 struct kvm_device_ops *ops = NULL;
2353 struct kvm_device *dev;
2354 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2355 int ret;
2356
2357 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2358 return -ENODEV;
2359
2360 ops = kvm_device_ops_table[cd->type];
2361 if (ops == NULL)
2362 return -ENODEV;
2363
2364 if (test)
2365 return 0;
2366
2367 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2368 if (!dev)
2369 return -ENOMEM;
2370
2371 dev->ops = ops;
2372 dev->kvm = kvm;
2373
2374 ret = ops->create(dev, cd->type);
2375 if (ret < 0) {
2376 kfree(dev);
2377 return ret;
2378 }
2379
2380 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2381 if (ret < 0) {
2382 ops->destroy(dev);
2383 return ret;
2384 }
2385
2386 list_add(&dev->vm_node, &kvm->devices);
2387 kvm_get_kvm(kvm);
2388 cd->fd = ret;
2389 return 0;
2390 }
2391
2392 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2393 {
2394 switch (arg) {
2395 case KVM_CAP_USER_MEMORY:
2396 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2397 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2398 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2399 case KVM_CAP_SET_BOOT_CPU_ID:
2400 #endif
2401 case KVM_CAP_INTERNAL_ERROR_DATA:
2402 #ifdef CONFIG_HAVE_KVM_MSI
2403 case KVM_CAP_SIGNAL_MSI:
2404 #endif
2405 #ifdef CONFIG_HAVE_KVM_IRQFD
2406 case KVM_CAP_IRQFD_RESAMPLE:
2407 #endif
2408 case KVM_CAP_CHECK_EXTENSION_VM:
2409 return 1;
2410 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2411 case KVM_CAP_IRQ_ROUTING:
2412 return KVM_MAX_IRQ_ROUTES;
2413 #endif
2414 default:
2415 break;
2416 }
2417 return kvm_vm_ioctl_check_extension(kvm, arg);
2418 }
2419
2420 static long kvm_vm_ioctl(struct file *filp,
2421 unsigned int ioctl, unsigned long arg)
2422 {
2423 struct kvm *kvm = filp->private_data;
2424 void __user *argp = (void __user *)arg;
2425 int r;
2426
2427 if (kvm->mm != current->mm)
2428 return -EIO;
2429 switch (ioctl) {
2430 case KVM_CREATE_VCPU:
2431 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2432 break;
2433 case KVM_SET_USER_MEMORY_REGION: {
2434 struct kvm_userspace_memory_region kvm_userspace_mem;
2435
2436 r = -EFAULT;
2437 if (copy_from_user(&kvm_userspace_mem, argp,
2438 sizeof kvm_userspace_mem))
2439 goto out;
2440
2441 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2442 break;
2443 }
2444 case KVM_GET_DIRTY_LOG: {
2445 struct kvm_dirty_log log;
2446
2447 r = -EFAULT;
2448 if (copy_from_user(&log, argp, sizeof log))
2449 goto out;
2450 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2451 break;
2452 }
2453 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2454 case KVM_REGISTER_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_register_coalesced_mmio(kvm, &zone);
2460 break;
2461 }
2462 case KVM_UNREGISTER_COALESCED_MMIO: {
2463 struct kvm_coalesced_mmio_zone zone;
2464 r = -EFAULT;
2465 if (copy_from_user(&zone, argp, sizeof zone))
2466 goto out;
2467 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2468 break;
2469 }
2470 #endif
2471 case KVM_IRQFD: {
2472 struct kvm_irqfd data;
2473
2474 r = -EFAULT;
2475 if (copy_from_user(&data, argp, sizeof data))
2476 goto out;
2477 r = kvm_irqfd(kvm, &data);
2478 break;
2479 }
2480 case KVM_IOEVENTFD: {
2481 struct kvm_ioeventfd data;
2482
2483 r = -EFAULT;
2484 if (copy_from_user(&data, argp, sizeof data))
2485 goto out;
2486 r = kvm_ioeventfd(kvm, &data);
2487 break;
2488 }
2489 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2490 case KVM_SET_BOOT_CPU_ID:
2491 r = 0;
2492 mutex_lock(&kvm->lock);
2493 if (atomic_read(&kvm->online_vcpus) != 0)
2494 r = -EBUSY;
2495 else
2496 kvm->bsp_vcpu_id = arg;
2497 mutex_unlock(&kvm->lock);
2498 break;
2499 #endif
2500 #ifdef CONFIG_HAVE_KVM_MSI
2501 case KVM_SIGNAL_MSI: {
2502 struct kvm_msi msi;
2503
2504 r = -EFAULT;
2505 if (copy_from_user(&msi, argp, sizeof msi))
2506 goto out;
2507 r = kvm_send_userspace_msi(kvm, &msi);
2508 break;
2509 }
2510 #endif
2511 #ifdef __KVM_HAVE_IRQ_LINE
2512 case KVM_IRQ_LINE_STATUS:
2513 case KVM_IRQ_LINE: {
2514 struct kvm_irq_level irq_event;
2515
2516 r = -EFAULT;
2517 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2518 goto out;
2519
2520 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2521 ioctl == KVM_IRQ_LINE_STATUS);
2522 if (r)
2523 goto out;
2524
2525 r = -EFAULT;
2526 if (ioctl == KVM_IRQ_LINE_STATUS) {
2527 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2528 goto out;
2529 }
2530
2531 r = 0;
2532 break;
2533 }
2534 #endif
2535 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2536 case KVM_SET_GSI_ROUTING: {
2537 struct kvm_irq_routing routing;
2538 struct kvm_irq_routing __user *urouting;
2539 struct kvm_irq_routing_entry *entries;
2540
2541 r = -EFAULT;
2542 if (copy_from_user(&routing, argp, sizeof(routing)))
2543 goto out;
2544 r = -EINVAL;
2545 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2546 goto out;
2547 if (routing.flags)
2548 goto out;
2549 r = -ENOMEM;
2550 entries = vmalloc(routing.nr * sizeof(*entries));
2551 if (!entries)
2552 goto out;
2553 r = -EFAULT;
2554 urouting = argp;
2555 if (copy_from_user(entries, urouting->entries,
2556 routing.nr * sizeof(*entries)))
2557 goto out_free_irq_routing;
2558 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2559 routing.flags);
2560 out_free_irq_routing:
2561 vfree(entries);
2562 break;
2563 }
2564 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2565 case KVM_CREATE_DEVICE: {
2566 struct kvm_create_device cd;
2567
2568 r = -EFAULT;
2569 if (copy_from_user(&cd, argp, sizeof(cd)))
2570 goto out;
2571
2572 r = kvm_ioctl_create_device(kvm, &cd);
2573 if (r)
2574 goto out;
2575
2576 r = -EFAULT;
2577 if (copy_to_user(argp, &cd, sizeof(cd)))
2578 goto out;
2579
2580 r = 0;
2581 break;
2582 }
2583 case KVM_CHECK_EXTENSION:
2584 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2585 break;
2586 default:
2587 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2588 if (r == -ENOTTY)
2589 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2590 }
2591 out:
2592 return r;
2593 }
2594
2595 #ifdef CONFIG_COMPAT
2596 struct compat_kvm_dirty_log {
2597 __u32 slot;
2598 __u32 padding1;
2599 union {
2600 compat_uptr_t dirty_bitmap; /* one bit per page */
2601 __u64 padding2;
2602 };
2603 };
2604
2605 static long kvm_vm_compat_ioctl(struct file *filp,
2606 unsigned int ioctl, unsigned long arg)
2607 {
2608 struct kvm *kvm = filp->private_data;
2609 int r;
2610
2611 if (kvm->mm != current->mm)
2612 return -EIO;
2613 switch (ioctl) {
2614 case KVM_GET_DIRTY_LOG: {
2615 struct compat_kvm_dirty_log compat_log;
2616 struct kvm_dirty_log log;
2617
2618 r = -EFAULT;
2619 if (copy_from_user(&compat_log, (void __user *)arg,
2620 sizeof(compat_log)))
2621 goto out;
2622 log.slot = compat_log.slot;
2623 log.padding1 = compat_log.padding1;
2624 log.padding2 = compat_log.padding2;
2625 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2626
2627 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2628 break;
2629 }
2630 default:
2631 r = kvm_vm_ioctl(filp, ioctl, arg);
2632 }
2633
2634 out:
2635 return r;
2636 }
2637 #endif
2638
2639 static struct file_operations kvm_vm_fops = {
2640 .release = kvm_vm_release,
2641 .unlocked_ioctl = kvm_vm_ioctl,
2642 #ifdef CONFIG_COMPAT
2643 .compat_ioctl = kvm_vm_compat_ioctl,
2644 #endif
2645 .llseek = noop_llseek,
2646 };
2647
2648 static int kvm_dev_ioctl_create_vm(unsigned long type)
2649 {
2650 int r;
2651 struct kvm *kvm;
2652
2653 kvm = kvm_create_vm(type);
2654 if (IS_ERR(kvm))
2655 return PTR_ERR(kvm);
2656 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2657 r = kvm_coalesced_mmio_init(kvm);
2658 if (r < 0) {
2659 kvm_put_kvm(kvm);
2660 return r;
2661 }
2662 #endif
2663 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2664 if (r < 0)
2665 kvm_put_kvm(kvm);
2666
2667 return r;
2668 }
2669
2670 static long kvm_dev_ioctl(struct file *filp,
2671 unsigned int ioctl, unsigned long arg)
2672 {
2673 long r = -EINVAL;
2674
2675 switch (ioctl) {
2676 case KVM_GET_API_VERSION:
2677 if (arg)
2678 goto out;
2679 r = KVM_API_VERSION;
2680 break;
2681 case KVM_CREATE_VM:
2682 r = kvm_dev_ioctl_create_vm(arg);
2683 break;
2684 case KVM_CHECK_EXTENSION:
2685 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2686 break;
2687 case KVM_GET_VCPU_MMAP_SIZE:
2688 if (arg)
2689 goto out;
2690 r = PAGE_SIZE; /* struct kvm_run */
2691 #ifdef CONFIG_X86
2692 r += PAGE_SIZE; /* pio data page */
2693 #endif
2694 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2695 r += PAGE_SIZE; /* coalesced mmio ring page */
2696 #endif
2697 break;
2698 case KVM_TRACE_ENABLE:
2699 case KVM_TRACE_PAUSE:
2700 case KVM_TRACE_DISABLE:
2701 r = -EOPNOTSUPP;
2702 break;
2703 default:
2704 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2705 }
2706 out:
2707 return r;
2708 }
2709
2710 static struct file_operations kvm_chardev_ops = {
2711 .unlocked_ioctl = kvm_dev_ioctl,
2712 .compat_ioctl = kvm_dev_ioctl,
2713 .llseek = noop_llseek,
2714 };
2715
2716 static struct miscdevice kvm_dev = {
2717 KVM_MINOR,
2718 "kvm",
2719 &kvm_chardev_ops,
2720 };
2721
2722 static void hardware_enable_nolock(void *junk)
2723 {
2724 int cpu = raw_smp_processor_id();
2725 int r;
2726
2727 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2728 return;
2729
2730 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2731
2732 r = kvm_arch_hardware_enable();
2733
2734 if (r) {
2735 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2736 atomic_inc(&hardware_enable_failed);
2737 printk(KERN_INFO "kvm: enabling virtualization on "
2738 "CPU%d failed\n", cpu);
2739 }
2740 }
2741
2742 static void hardware_enable(void)
2743 {
2744 raw_spin_lock(&kvm_count_lock);
2745 if (kvm_usage_count)
2746 hardware_enable_nolock(NULL);
2747 raw_spin_unlock(&kvm_count_lock);
2748 }
2749
2750 static void hardware_disable_nolock(void *junk)
2751 {
2752 int cpu = raw_smp_processor_id();
2753
2754 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2755 return;
2756 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2757 kvm_arch_hardware_disable();
2758 }
2759
2760 static void hardware_disable(void)
2761 {
2762 raw_spin_lock(&kvm_count_lock);
2763 if (kvm_usage_count)
2764 hardware_disable_nolock(NULL);
2765 raw_spin_unlock(&kvm_count_lock);
2766 }
2767
2768 static void hardware_disable_all_nolock(void)
2769 {
2770 BUG_ON(!kvm_usage_count);
2771
2772 kvm_usage_count--;
2773 if (!kvm_usage_count)
2774 on_each_cpu(hardware_disable_nolock, NULL, 1);
2775 }
2776
2777 static void hardware_disable_all(void)
2778 {
2779 raw_spin_lock(&kvm_count_lock);
2780 hardware_disable_all_nolock();
2781 raw_spin_unlock(&kvm_count_lock);
2782 }
2783
2784 static int hardware_enable_all(void)
2785 {
2786 int r = 0;
2787
2788 raw_spin_lock(&kvm_count_lock);
2789
2790 kvm_usage_count++;
2791 if (kvm_usage_count == 1) {
2792 atomic_set(&hardware_enable_failed, 0);
2793 on_each_cpu(hardware_enable_nolock, NULL, 1);
2794
2795 if (atomic_read(&hardware_enable_failed)) {
2796 hardware_disable_all_nolock();
2797 r = -EBUSY;
2798 }
2799 }
2800
2801 raw_spin_unlock(&kvm_count_lock);
2802
2803 return r;
2804 }
2805
2806 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2807 void *v)
2808 {
2809 int cpu = (long)v;
2810
2811 val &= ~CPU_TASKS_FROZEN;
2812 switch (val) {
2813 case CPU_DYING:
2814 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2815 cpu);
2816 hardware_disable();
2817 break;
2818 case CPU_STARTING:
2819 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2820 cpu);
2821 hardware_enable();
2822 break;
2823 }
2824 return NOTIFY_OK;
2825 }
2826
2827 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2828 void *v)
2829 {
2830 /*
2831 * Some (well, at least mine) BIOSes hang on reboot if
2832 * in vmx root mode.
2833 *
2834 * And Intel TXT required VMX off for all cpu when system shutdown.
2835 */
2836 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2837 kvm_rebooting = true;
2838 on_each_cpu(hardware_disable_nolock, NULL, 1);
2839 return NOTIFY_OK;
2840 }
2841
2842 static struct notifier_block kvm_reboot_notifier = {
2843 .notifier_call = kvm_reboot,
2844 .priority = 0,
2845 };
2846
2847 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2848 {
2849 int i;
2850
2851 for (i = 0; i < bus->dev_count; i++) {
2852 struct kvm_io_device *pos = bus->range[i].dev;
2853
2854 kvm_iodevice_destructor(pos);
2855 }
2856 kfree(bus);
2857 }
2858
2859 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2860 const struct kvm_io_range *r2)
2861 {
2862 if (r1->addr < r2->addr)
2863 return -1;
2864 if (r1->addr + r1->len > r2->addr + r2->len)
2865 return 1;
2866 return 0;
2867 }
2868
2869 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2870 {
2871 return kvm_io_bus_cmp(p1, p2);
2872 }
2873
2874 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2875 gpa_t addr, int len)
2876 {
2877 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2878 .addr = addr,
2879 .len = len,
2880 .dev = dev,
2881 };
2882
2883 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2884 kvm_io_bus_sort_cmp, NULL);
2885
2886 return 0;
2887 }
2888
2889 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2890 gpa_t addr, int len)
2891 {
2892 struct kvm_io_range *range, key;
2893 int off;
2894
2895 key = (struct kvm_io_range) {
2896 .addr = addr,
2897 .len = len,
2898 };
2899
2900 range = bsearch(&key, bus->range, bus->dev_count,
2901 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2902 if (range == NULL)
2903 return -ENOENT;
2904
2905 off = range - bus->range;
2906
2907 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2908 off--;
2909
2910 return off;
2911 }
2912
2913 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2914 struct kvm_io_range *range, const void *val)
2915 {
2916 int idx;
2917
2918 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2919 if (idx < 0)
2920 return -EOPNOTSUPP;
2921
2922 while (idx < bus->dev_count &&
2923 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2924 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2925 range->len, val))
2926 return idx;
2927 idx++;
2928 }
2929
2930 return -EOPNOTSUPP;
2931 }
2932
2933 /* kvm_io_bus_write - called under kvm->slots_lock */
2934 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2935 int len, const void *val)
2936 {
2937 struct kvm_io_bus *bus;
2938 struct kvm_io_range range;
2939 int r;
2940
2941 range = (struct kvm_io_range) {
2942 .addr = addr,
2943 .len = len,
2944 };
2945
2946 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2947 r = __kvm_io_bus_write(bus, &range, val);
2948 return r < 0 ? r : 0;
2949 }
2950
2951 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2952 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2953 int len, const void *val, long cookie)
2954 {
2955 struct kvm_io_bus *bus;
2956 struct kvm_io_range range;
2957
2958 range = (struct kvm_io_range) {
2959 .addr = addr,
2960 .len = len,
2961 };
2962
2963 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2964
2965 /* First try the device referenced by cookie. */
2966 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2967 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2968 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2969 val))
2970 return cookie;
2971
2972 /*
2973 * cookie contained garbage; fall back to search and return the
2974 * correct cookie value.
2975 */
2976 return __kvm_io_bus_write(bus, &range, val);
2977 }
2978
2979 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2980 void *val)
2981 {
2982 int idx;
2983
2984 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2985 if (idx < 0)
2986 return -EOPNOTSUPP;
2987
2988 while (idx < bus->dev_count &&
2989 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2990 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2991 range->len, val))
2992 return idx;
2993 idx++;
2994 }
2995
2996 return -EOPNOTSUPP;
2997 }
2998 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2999
3000 /* kvm_io_bus_read - called under kvm->slots_lock */
3001 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3002 int len, void *val)
3003 {
3004 struct kvm_io_bus *bus;
3005 struct kvm_io_range range;
3006 int r;
3007
3008 range = (struct kvm_io_range) {
3009 .addr = addr,
3010 .len = len,
3011 };
3012
3013 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3014 r = __kvm_io_bus_read(bus, &range, val);
3015 return r < 0 ? r : 0;
3016 }
3017
3018
3019 /* Caller must hold slots_lock. */
3020 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3021 int len, struct kvm_io_device *dev)
3022 {
3023 struct kvm_io_bus *new_bus, *bus;
3024
3025 bus = kvm->buses[bus_idx];
3026 /* exclude ioeventfd which is limited by maximum fd */
3027 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3028 return -ENOSPC;
3029
3030 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3031 sizeof(struct kvm_io_range)), GFP_KERNEL);
3032 if (!new_bus)
3033 return -ENOMEM;
3034 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3035 sizeof(struct kvm_io_range)));
3036 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3037 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3038 synchronize_srcu_expedited(&kvm->srcu);
3039 kfree(bus);
3040
3041 return 0;
3042 }
3043
3044 /* Caller must hold slots_lock. */
3045 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3046 struct kvm_io_device *dev)
3047 {
3048 int i, r;
3049 struct kvm_io_bus *new_bus, *bus;
3050
3051 bus = kvm->buses[bus_idx];
3052 r = -ENOENT;
3053 for (i = 0; i < bus->dev_count; i++)
3054 if (bus->range[i].dev == dev) {
3055 r = 0;
3056 break;
3057 }
3058
3059 if (r)
3060 return r;
3061
3062 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3063 sizeof(struct kvm_io_range)), GFP_KERNEL);
3064 if (!new_bus)
3065 return -ENOMEM;
3066
3067 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3068 new_bus->dev_count--;
3069 memcpy(new_bus->range + i, bus->range + i + 1,
3070 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3071
3072 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3073 synchronize_srcu_expedited(&kvm->srcu);
3074 kfree(bus);
3075 return r;
3076 }
3077
3078 static struct notifier_block kvm_cpu_notifier = {
3079 .notifier_call = kvm_cpu_hotplug,
3080 };
3081
3082 static int vm_stat_get(void *_offset, u64 *val)
3083 {
3084 unsigned offset = (long)_offset;
3085 struct kvm *kvm;
3086
3087 *val = 0;
3088 spin_lock(&kvm_lock);
3089 list_for_each_entry(kvm, &vm_list, vm_list)
3090 *val += *(u32 *)((void *)kvm + offset);
3091 spin_unlock(&kvm_lock);
3092 return 0;
3093 }
3094
3095 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3096
3097 static int vcpu_stat_get(void *_offset, u64 *val)
3098 {
3099 unsigned offset = (long)_offset;
3100 struct kvm *kvm;
3101 struct kvm_vcpu *vcpu;
3102 int i;
3103
3104 *val = 0;
3105 spin_lock(&kvm_lock);
3106 list_for_each_entry(kvm, &vm_list, vm_list)
3107 kvm_for_each_vcpu(i, vcpu, kvm)
3108 *val += *(u32 *)((void *)vcpu + offset);
3109
3110 spin_unlock(&kvm_lock);
3111 return 0;
3112 }
3113
3114 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3115
3116 static const struct file_operations *stat_fops[] = {
3117 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3118 [KVM_STAT_VM] = &vm_stat_fops,
3119 };
3120
3121 static int kvm_init_debug(void)
3122 {
3123 int r = -EEXIST;
3124 struct kvm_stats_debugfs_item *p;
3125
3126 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3127 if (kvm_debugfs_dir == NULL)
3128 goto out;
3129
3130 for (p = debugfs_entries; p->name; ++p) {
3131 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3132 (void *)(long)p->offset,
3133 stat_fops[p->kind]);
3134 if (p->dentry == NULL)
3135 goto out_dir;
3136 }
3137
3138 return 0;
3139
3140 out_dir:
3141 debugfs_remove_recursive(kvm_debugfs_dir);
3142 out:
3143 return r;
3144 }
3145
3146 static void kvm_exit_debug(void)
3147 {
3148 struct kvm_stats_debugfs_item *p;
3149
3150 for (p = debugfs_entries; p->name; ++p)
3151 debugfs_remove(p->dentry);
3152 debugfs_remove(kvm_debugfs_dir);
3153 }
3154
3155 static int kvm_suspend(void)
3156 {
3157 if (kvm_usage_count)
3158 hardware_disable_nolock(NULL);
3159 return 0;
3160 }
3161
3162 static void kvm_resume(void)
3163 {
3164 if (kvm_usage_count) {
3165 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3166 hardware_enable_nolock(NULL);
3167 }
3168 }
3169
3170 static struct syscore_ops kvm_syscore_ops = {
3171 .suspend = kvm_suspend,
3172 .resume = kvm_resume,
3173 };
3174
3175 static inline
3176 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3177 {
3178 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3179 }
3180
3181 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3182 {
3183 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3184 if (vcpu->preempted)
3185 vcpu->preempted = false;
3186
3187 kvm_arch_sched_in(vcpu, cpu);
3188
3189 kvm_arch_vcpu_load(vcpu, cpu);
3190 }
3191
3192 static void kvm_sched_out(struct preempt_notifier *pn,
3193 struct task_struct *next)
3194 {
3195 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3196
3197 if (current->state == TASK_RUNNING)
3198 vcpu->preempted = true;
3199 kvm_arch_vcpu_put(vcpu);
3200 }
3201
3202 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3203 struct module *module)
3204 {
3205 int r;
3206 int cpu;
3207
3208 r = kvm_arch_init(opaque);
3209 if (r)
3210 goto out_fail;
3211
3212 /*
3213 * kvm_arch_init makes sure there's at most one caller
3214 * for architectures that support multiple implementations,
3215 * like intel and amd on x86.
3216 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3217 * conflicts in case kvm is already setup for another implementation.
3218 */
3219 r = kvm_irqfd_init();
3220 if (r)
3221 goto out_irqfd;
3222
3223 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3224 r = -ENOMEM;
3225 goto out_free_0;
3226 }
3227
3228 r = kvm_arch_hardware_setup();
3229 if (r < 0)
3230 goto out_free_0a;
3231
3232 for_each_online_cpu(cpu) {
3233 smp_call_function_single(cpu,
3234 kvm_arch_check_processor_compat,
3235 &r, 1);
3236 if (r < 0)
3237 goto out_free_1;
3238 }
3239
3240 r = register_cpu_notifier(&kvm_cpu_notifier);
3241 if (r)
3242 goto out_free_2;
3243 register_reboot_notifier(&kvm_reboot_notifier);
3244
3245 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3246 if (!vcpu_align)
3247 vcpu_align = __alignof__(struct kvm_vcpu);
3248 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3249 0, NULL);
3250 if (!kvm_vcpu_cache) {
3251 r = -ENOMEM;
3252 goto out_free_3;
3253 }
3254
3255 r = kvm_async_pf_init();
3256 if (r)
3257 goto out_free;
3258
3259 kvm_chardev_ops.owner = module;
3260 kvm_vm_fops.owner = module;
3261 kvm_vcpu_fops.owner = module;
3262
3263 r = misc_register(&kvm_dev);
3264 if (r) {
3265 printk(KERN_ERR "kvm: misc device register failed\n");
3266 goto out_unreg;
3267 }
3268
3269 register_syscore_ops(&kvm_syscore_ops);
3270
3271 kvm_preempt_ops.sched_in = kvm_sched_in;
3272 kvm_preempt_ops.sched_out = kvm_sched_out;
3273
3274 r = kvm_init_debug();
3275 if (r) {
3276 printk(KERN_ERR "kvm: create debugfs files failed\n");
3277 goto out_undebugfs;
3278 }
3279
3280 r = kvm_vfio_ops_init();
3281 WARN_ON(r);
3282
3283 return 0;
3284
3285 out_undebugfs:
3286 unregister_syscore_ops(&kvm_syscore_ops);
3287 misc_deregister(&kvm_dev);
3288 out_unreg:
3289 kvm_async_pf_deinit();
3290 out_free:
3291 kmem_cache_destroy(kvm_vcpu_cache);
3292 out_free_3:
3293 unregister_reboot_notifier(&kvm_reboot_notifier);
3294 unregister_cpu_notifier(&kvm_cpu_notifier);
3295 out_free_2:
3296 out_free_1:
3297 kvm_arch_hardware_unsetup();
3298 out_free_0a:
3299 free_cpumask_var(cpus_hardware_enabled);
3300 out_free_0:
3301 kvm_irqfd_exit();
3302 out_irqfd:
3303 kvm_arch_exit();
3304 out_fail:
3305 return r;
3306 }
3307 EXPORT_SYMBOL_GPL(kvm_init);
3308
3309 void kvm_exit(void)
3310 {
3311 kvm_exit_debug();
3312 misc_deregister(&kvm_dev);
3313 kmem_cache_destroy(kvm_vcpu_cache);
3314 kvm_async_pf_deinit();
3315 unregister_syscore_ops(&kvm_syscore_ops);
3316 unregister_reboot_notifier(&kvm_reboot_notifier);
3317 unregister_cpu_notifier(&kvm_cpu_notifier);
3318 on_each_cpu(hardware_disable_nolock, NULL, 1);
3319 kvm_arch_hardware_unsetup();
3320 kvm_arch_exit();
3321 kvm_irqfd_exit();
3322 free_cpumask_var(cpus_hardware_enabled);
3323 kvm_vfio_ops_exit();
3324 }
3325 EXPORT_SYMBOL_GPL(kvm_exit);
Linux kernel - Deliverables
This page took 0.131876 seconds and 5 git commands to generate.