projects / deliverable / linux.git / blob
? search:


bb8641b5d83ba2b80dc344a165da4db265f08efe
[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/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68 * Ordering of locks:
69 *
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71 */
72
73 DEFINE_SPINLOCK(kvm_lock);
74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
75 LIST_HEAD(vm_list);
76
77 static cpumask_var_t cpus_hardware_enabled;
78 static int kvm_usage_count = 0;
79 static atomic_t hardware_enable_failed;
80
81 struct kmem_cache *kvm_vcpu_cache;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
83
84 static __read_mostly struct preempt_ops kvm_preempt_ops;
85
86 struct dentry *kvm_debugfs_dir;
87
88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
89 unsigned long arg);
90 #ifdef CONFIG_COMPAT
91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
92 unsigned long arg);
93 #endif
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
96
97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
98
99 static void kvm_release_pfn_dirty(pfn_t pfn);
100 static void mark_page_dirty_in_slot(struct kvm *kvm,
101 struct kvm_memory_slot *memslot, gfn_t gfn);
102
103 __visible bool kvm_rebooting;
104 EXPORT_SYMBOL_GPL(kvm_rebooting);
105
106 static bool largepages_enabled = true;
107
108 bool kvm_is_mmio_pfn(pfn_t pfn)
109 {
110 if (pfn_valid(pfn))
111 return PageReserved(pfn_to_page(pfn));
112
113 return true;
114 }
115
116 /*
117 * Switches to specified vcpu, until a matching vcpu_put()
118 */
119 int vcpu_load(struct kvm_vcpu *vcpu)
120 {
121 int cpu;
122
123 if (mutex_lock_killable(&vcpu->mutex))
124 return -EINTR;
125 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
126 /* The thread running this VCPU changed. */
127 struct pid *oldpid = vcpu->pid;
128 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
129 rcu_assign_pointer(vcpu->pid, newpid);
130 if (oldpid)
131 synchronize_rcu();
132 put_pid(oldpid);
133 }
134 cpu = get_cpu();
135 preempt_notifier_register(&vcpu->preempt_notifier);
136 kvm_arch_vcpu_load(vcpu, cpu);
137 put_cpu();
138 return 0;
139 }
140
141 void vcpu_put(struct kvm_vcpu *vcpu)
142 {
143 preempt_disable();
144 kvm_arch_vcpu_put(vcpu);
145 preempt_notifier_unregister(&vcpu->preempt_notifier);
146 preempt_enable();
147 mutex_unlock(&vcpu->mutex);
148 }
149
150 static void ack_flush(void *_completed)
151 {
152 }
153
154 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
155 {
156 int i, cpu, me;
157 cpumask_var_t cpus;
158 bool called = true;
159 struct kvm_vcpu *vcpu;
160
161 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
162
163 me = get_cpu();
164 kvm_for_each_vcpu(i, vcpu, kvm) {
165 kvm_make_request(req, vcpu);
166 cpu = vcpu->cpu;
167
168 /* Set ->requests bit before we read ->mode */
169 smp_mb();
170
171 if (cpus != NULL && cpu != -1 && cpu != me &&
172 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
173 cpumask_set_cpu(cpu, cpus);
174 }
175 if (unlikely(cpus == NULL))
176 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
177 else if (!cpumask_empty(cpus))
178 smp_call_function_many(cpus, ack_flush, NULL, 1);
179 else
180 called = false;
181 put_cpu();
182 free_cpumask_var(cpus);
183 return called;
184 }
185
186 void kvm_flush_remote_tlbs(struct kvm *kvm)
187 {
188 long dirty_count = kvm->tlbs_dirty;
189
190 smp_mb();
191 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
192 ++kvm->stat.remote_tlb_flush;
193 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
194 }
195 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
196
197 void kvm_reload_remote_mmus(struct kvm *kvm)
198 {
199 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
200 }
201
202 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
203 {
204 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
205 }
206
207 void kvm_make_scan_ioapic_request(struct kvm *kvm)
208 {
209 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
210 }
211
212 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
213 {
214 struct page *page;
215 int r;
216
217 mutex_init(&vcpu->mutex);
218 vcpu->cpu = -1;
219 vcpu->kvm = kvm;
220 vcpu->vcpu_id = id;
221 vcpu->pid = NULL;
222 init_waitqueue_head(&vcpu->wq);
223 kvm_async_pf_vcpu_init(vcpu);
224
225 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
226 if (!page) {
227 r = -ENOMEM;
228 goto fail;
229 }
230 vcpu->run = page_address(page);
231
232 kvm_vcpu_set_in_spin_loop(vcpu, false);
233 kvm_vcpu_set_dy_eligible(vcpu, false);
234 vcpu->preempted = false;
235
236 r = kvm_arch_vcpu_init(vcpu);
237 if (r < 0)
238 goto fail_free_run;
239 return 0;
240
241 fail_free_run:
242 free_page((unsigned long)vcpu->run);
243 fail:
244 return r;
245 }
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
247
248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
249 {
250 put_pid(vcpu->pid);
251 kvm_arch_vcpu_uninit(vcpu);
252 free_page((unsigned long)vcpu->run);
253 }
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
255
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
258 {
259 return container_of(mn, struct kvm, mmu_notifier);
260 }
261
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address)
265 {
266 struct kvm *kvm = mmu_notifier_to_kvm(mn);
267 int need_tlb_flush, idx;
268
269 /*
270 * When ->invalidate_page runs, the linux pte has been zapped
271 * already but the page is still allocated until
272 * ->invalidate_page returns. So if we increase the sequence
273 * here the kvm page fault will notice if the spte can't be
274 * established because the page is going to be freed. If
275 * instead the kvm page fault establishes the spte before
276 * ->invalidate_page runs, kvm_unmap_hva will release it
277 * before returning.
278 *
279 * The sequence increase only need to be seen at spin_unlock
280 * time, and not at spin_lock time.
281 *
282 * Increasing the sequence after the spin_unlock would be
283 * unsafe because the kvm page fault could then establish the
284 * pte after kvm_unmap_hva returned, without noticing the page
285 * is going to be freed.
286 */
287 idx = srcu_read_lock(&kvm->srcu);
288 spin_lock(&kvm->mmu_lock);
289
290 kvm->mmu_notifier_seq++;
291 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292 /* we've to flush the tlb before the pages can be freed */
293 if (need_tlb_flush)
294 kvm_flush_remote_tlbs(kvm);
295
296 spin_unlock(&kvm->mmu_lock);
297 srcu_read_unlock(&kvm->srcu, idx);
298 }
299
300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
301 struct mm_struct *mm,
302 unsigned long address,
303 pte_t pte)
304 {
305 struct kvm *kvm = mmu_notifier_to_kvm(mn);
306 int idx;
307
308 idx = srcu_read_lock(&kvm->srcu);
309 spin_lock(&kvm->mmu_lock);
310 kvm->mmu_notifier_seq++;
311 kvm_set_spte_hva(kvm, address, pte);
312 spin_unlock(&kvm->mmu_lock);
313 srcu_read_unlock(&kvm->srcu, idx);
314 }
315
316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
317 struct mm_struct *mm,
318 unsigned long start,
319 unsigned long end)
320 {
321 struct kvm *kvm = mmu_notifier_to_kvm(mn);
322 int need_tlb_flush = 0, idx;
323
324 idx = srcu_read_lock(&kvm->srcu);
325 spin_lock(&kvm->mmu_lock);
326 /*
327 * The count increase must become visible at unlock time as no
328 * spte can be established without taking the mmu_lock and
329 * count is also read inside the mmu_lock critical section.
330 */
331 kvm->mmu_notifier_count++;
332 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
333 need_tlb_flush |= kvm->tlbs_dirty;
334 /* we've to flush the tlb before the pages can be freed */
335 if (need_tlb_flush)
336 kvm_flush_remote_tlbs(kvm);
337
338 spin_unlock(&kvm->mmu_lock);
339 srcu_read_unlock(&kvm->srcu, idx);
340 }
341
342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
343 struct mm_struct *mm,
344 unsigned long start,
345 unsigned long end)
346 {
347 struct kvm *kvm = mmu_notifier_to_kvm(mn);
348
349 spin_lock(&kvm->mmu_lock);
350 /*
351 * This sequence increase will notify the kvm page fault that
352 * the page that is going to be mapped in the spte could have
353 * been freed.
354 */
355 kvm->mmu_notifier_seq++;
356 smp_wmb();
357 /*
358 * The above sequence increase must be visible before the
359 * below count decrease, which is ensured by the smp_wmb above
360 * in conjunction with the smp_rmb in mmu_notifier_retry().
361 */
362 kvm->mmu_notifier_count--;
363 spin_unlock(&kvm->mmu_lock);
364
365 BUG_ON(kvm->mmu_notifier_count < 0);
366 }
367
368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
369 struct mm_struct *mm,
370 unsigned long address)
371 {
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 int young, idx;
374
375 idx = srcu_read_lock(&kvm->srcu);
376 spin_lock(&kvm->mmu_lock);
377
378 young = kvm_age_hva(kvm, address);
379 if (young)
380 kvm_flush_remote_tlbs(kvm);
381
382 spin_unlock(&kvm->mmu_lock);
383 srcu_read_unlock(&kvm->srcu, idx);
384
385 return young;
386 }
387
388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
389 struct mm_struct *mm,
390 unsigned long address)
391 {
392 struct kvm *kvm = mmu_notifier_to_kvm(mn);
393 int young, idx;
394
395 idx = srcu_read_lock(&kvm->srcu);
396 spin_lock(&kvm->mmu_lock);
397 young = kvm_test_age_hva(kvm, address);
398 spin_unlock(&kvm->mmu_lock);
399 srcu_read_unlock(&kvm->srcu, idx);
400
401 return young;
402 }
403
404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
405 struct mm_struct *mm)
406 {
407 struct kvm *kvm = mmu_notifier_to_kvm(mn);
408 int idx;
409
410 idx = srcu_read_lock(&kvm->srcu);
411 kvm_arch_flush_shadow_all(kvm);
412 srcu_read_unlock(&kvm->srcu, idx);
413 }
414
415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
416 .invalidate_page = kvm_mmu_notifier_invalidate_page,
417 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
418 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
419 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
420 .test_young = kvm_mmu_notifier_test_young,
421 .change_pte = kvm_mmu_notifier_change_pte,
422 .release = kvm_mmu_notifier_release,
423 };
424
425 static int kvm_init_mmu_notifier(struct kvm *kvm)
426 {
427 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
428 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
429 }
430
431 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
432
433 static int kvm_init_mmu_notifier(struct kvm *kvm)
434 {
435 return 0;
436 }
437
438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
439
440 static void kvm_init_memslots_id(struct kvm *kvm)
441 {
442 int i;
443 struct kvm_memslots *slots = kvm->memslots;
444
445 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
446 slots->id_to_index[i] = slots->memslots[i].id = i;
447 }
448
449 static struct kvm *kvm_create_vm(unsigned long type)
450 {
451 int r, i;
452 struct kvm *kvm = kvm_arch_alloc_vm();
453
454 if (!kvm)
455 return ERR_PTR(-ENOMEM);
456
457 r = kvm_arch_init_vm(kvm, type);
458 if (r)
459 goto out_err_no_disable;
460
461 r = hardware_enable_all();
462 if (r)
463 goto out_err_no_disable;
464
465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
466 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
467 #endif
468 #ifdef CONFIG_HAVE_KVM_IRQFD
469 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
470 #endif
471
472 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
473
474 r = -ENOMEM;
475 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
476 if (!kvm->memslots)
477 goto out_err_no_srcu;
478
479 /*
480 * Init kvm generation close to the maximum to easily test the
481 * code of handling generation number wrap-around.
482 */
483 kvm->memslots->generation = -150;
484
485 kvm_init_memslots_id(kvm);
486 if (init_srcu_struct(&kvm->srcu))
487 goto out_err_no_srcu;
488 if (init_srcu_struct(&kvm->irq_srcu))
489 goto out_err_no_irq_srcu;
490 for (i = 0; i < KVM_NR_BUSES; i++) {
491 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
492 GFP_KERNEL);
493 if (!kvm->buses[i])
494 goto out_err;
495 }
496
497 spin_lock_init(&kvm->mmu_lock);
498 kvm->mm = current->mm;
499 atomic_inc(&kvm->mm->mm_count);
500 kvm_eventfd_init(kvm);
501 mutex_init(&kvm->lock);
502 mutex_init(&kvm->irq_lock);
503 mutex_init(&kvm->slots_lock);
504 atomic_set(&kvm->users_count, 1);
505 INIT_LIST_HEAD(&kvm->devices);
506
507 r = kvm_init_mmu_notifier(kvm);
508 if (r)
509 goto out_err;
510
511 spin_lock(&kvm_lock);
512 list_add(&kvm->vm_list, &vm_list);
513 spin_unlock(&kvm_lock);
514
515 return kvm;
516
517 out_err:
518 cleanup_srcu_struct(&kvm->irq_srcu);
519 out_err_no_irq_srcu:
520 cleanup_srcu_struct(&kvm->srcu);
521 out_err_no_srcu:
522 hardware_disable_all();
523 out_err_no_disable:
524 for (i = 0; i < KVM_NR_BUSES; i++)
525 kfree(kvm->buses[i]);
526 kfree(kvm->memslots);
527 kvm_arch_free_vm(kvm);
528 return ERR_PTR(r);
529 }
530
531 /*
532 * Avoid using vmalloc for a small buffer.
533 * Should not be used when the size is statically known.
534 */
535 void *kvm_kvzalloc(unsigned long size)
536 {
537 if (size > PAGE_SIZE)
538 return vzalloc(size);
539 else
540 return kzalloc(size, GFP_KERNEL);
541 }
542
543 void kvm_kvfree(const void *addr)
544 {
545 if (is_vmalloc_addr(addr))
546 vfree(addr);
547 else
548 kfree(addr);
549 }
550
551 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
552 {
553 if (!memslot->dirty_bitmap)
554 return;
555
556 kvm_kvfree(memslot->dirty_bitmap);
557 memslot->dirty_bitmap = NULL;
558 }
559
560 /*
561 * Free any memory in @free but not in @dont.
562 */
563 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
564 struct kvm_memory_slot *dont)
565 {
566 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
567 kvm_destroy_dirty_bitmap(free);
568
569 kvm_arch_free_memslot(kvm, free, dont);
570
571 free->npages = 0;
572 }
573
574 static void kvm_free_physmem(struct kvm *kvm)
575 {
576 struct kvm_memslots *slots = kvm->memslots;
577 struct kvm_memory_slot *memslot;
578
579 kvm_for_each_memslot(memslot, slots)
580 kvm_free_physmem_slot(kvm, memslot, NULL);
581
582 kfree(kvm->memslots);
583 }
584
585 static void kvm_destroy_devices(struct kvm *kvm)
586 {
587 struct list_head *node, *tmp;
588
589 list_for_each_safe(node, tmp, &kvm->devices) {
590 struct kvm_device *dev =
591 list_entry(node, struct kvm_device, vm_node);
592
593 list_del(node);
594 dev->ops->destroy(dev);
595 }
596 }
597
598 static void kvm_destroy_vm(struct kvm *kvm)
599 {
600 int i;
601 struct mm_struct *mm = kvm->mm;
602
603 kvm_arch_sync_events(kvm);
604 spin_lock(&kvm_lock);
605 list_del(&kvm->vm_list);
606 spin_unlock(&kvm_lock);
607 kvm_free_irq_routing(kvm);
608 for (i = 0; i < KVM_NR_BUSES; i++)
609 kvm_io_bus_destroy(kvm->buses[i]);
610 kvm_coalesced_mmio_free(kvm);
611 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
612 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
613 #else
614 kvm_arch_flush_shadow_all(kvm);
615 #endif
616 kvm_arch_destroy_vm(kvm);
617 kvm_destroy_devices(kvm);
618 kvm_free_physmem(kvm);
619 cleanup_srcu_struct(&kvm->irq_srcu);
620 cleanup_srcu_struct(&kvm->srcu);
621 kvm_arch_free_vm(kvm);
622 hardware_disable_all();
623 mmdrop(mm);
624 }
625
626 void kvm_get_kvm(struct kvm *kvm)
627 {
628 atomic_inc(&kvm->users_count);
629 }
630 EXPORT_SYMBOL_GPL(kvm_get_kvm);
631
632 void kvm_put_kvm(struct kvm *kvm)
633 {
634 if (atomic_dec_and_test(&kvm->users_count))
635 kvm_destroy_vm(kvm);
636 }
637 EXPORT_SYMBOL_GPL(kvm_put_kvm);
638
639
640 static int kvm_vm_release(struct inode *inode, struct file *filp)
641 {
642 struct kvm *kvm = filp->private_data;
643
644 kvm_irqfd_release(kvm);
645
646 kvm_put_kvm(kvm);
647 return 0;
648 }
649
650 /*
651 * Allocation size is twice as large as the actual dirty bitmap size.
652 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
653 */
654 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
655 {
656 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
657
658 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
659 if (!memslot->dirty_bitmap)
660 return -ENOMEM;
661
662 return 0;
663 }
664
665 static int cmp_memslot(const void *slot1, const void *slot2)
666 {
667 struct kvm_memory_slot *s1, *s2;
668
669 s1 = (struct kvm_memory_slot *)slot1;
670 s2 = (struct kvm_memory_slot *)slot2;
671
672 if (s1->npages < s2->npages)
673 return 1;
674 if (s1->npages > s2->npages)
675 return -1;
676
677 return 0;
678 }
679
680 /*
681 * Sort the memslots base on its size, so the larger slots
682 * will get better fit.
683 */
684 static void sort_memslots(struct kvm_memslots *slots)
685 {
686 int i;
687
688 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
689 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
690
691 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
692 slots->id_to_index[slots->memslots[i].id] = i;
693 }
694
695 static void update_memslots(struct kvm_memslots *slots,
696 struct kvm_memory_slot *new)
697 {
698 if (new) {
699 int id = new->id;
700 struct kvm_memory_slot *old = id_to_memslot(slots, id);
701 unsigned long npages = old->npages;
702
703 *old = *new;
704 if (new->npages != npages)
705 sort_memslots(slots);
706 }
707 }
708
709 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
710 {
711 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
712
713 #ifdef __KVM_HAVE_READONLY_MEM
714 valid_flags |= KVM_MEM_READONLY;
715 #endif
716
717 if (mem->flags & ~valid_flags)
718 return -EINVAL;
719
720 return 0;
721 }
722
723 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
724 struct kvm_memslots *slots, struct kvm_memory_slot *new)
725 {
726 struct kvm_memslots *old_memslots = kvm->memslots;
727
728 /*
729 * Set the low bit in the generation, which disables SPTE caching
730 * until the end of synchronize_srcu_expedited.
731 */
732 WARN_ON(old_memslots->generation & 1);
733 slots->generation = old_memslots->generation + 1;
734
735 update_memslots(slots, new);
736 rcu_assign_pointer(kvm->memslots, slots);
737 synchronize_srcu_expedited(&kvm->srcu);
738
739 /*
740 * Increment the new memslot generation a second time. This prevents
741 * vm exits that race with memslot updates from caching a memslot
742 * generation that will (potentially) be valid forever.
743 */
744 slots->generation++;
745
746 kvm_arch_memslots_updated(kvm);
747
748 return old_memslots;
749 }
750
751 /*
752 * Allocate some memory and give it an address in the guest physical address
753 * space.
754 *
755 * Discontiguous memory is allowed, mostly for framebuffers.
756 *
757 * Must be called holding mmap_sem for write.
758 */
759 int __kvm_set_memory_region(struct kvm *kvm,
760 struct kvm_userspace_memory_region *mem)
761 {
762 int r;
763 gfn_t base_gfn;
764 unsigned long npages;
765 struct kvm_memory_slot *slot;
766 struct kvm_memory_slot old, new;
767 struct kvm_memslots *slots = NULL, *old_memslots;
768 enum kvm_mr_change change;
769
770 r = check_memory_region_flags(mem);
771 if (r)
772 goto out;
773
774 r = -EINVAL;
775 /* General sanity checks */
776 if (mem->memory_size & (PAGE_SIZE - 1))
777 goto out;
778 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
779 goto out;
780 /* We can read the guest memory with __xxx_user() later on. */
781 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
782 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
783 !access_ok(VERIFY_WRITE,
784 (void __user *)(unsigned long)mem->userspace_addr,
785 mem->memory_size)))
786 goto out;
787 if (mem->slot >= KVM_MEM_SLOTS_NUM)
788 goto out;
789 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
790 goto out;
791
792 slot = id_to_memslot(kvm->memslots, mem->slot);
793 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
794 npages = mem->memory_size >> PAGE_SHIFT;
795
796 r = -EINVAL;
797 if (npages > KVM_MEM_MAX_NR_PAGES)
798 goto out;
799
800 if (!npages)
801 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
802
803 new = old = *slot;
804
805 new.id = mem->slot;
806 new.base_gfn = base_gfn;
807 new.npages = npages;
808 new.flags = mem->flags;
809
810 r = -EINVAL;
811 if (npages) {
812 if (!old.npages)
813 change = KVM_MR_CREATE;
814 else { /* Modify an existing slot. */
815 if ((mem->userspace_addr != old.userspace_addr) ||
816 (npages != old.npages) ||
817 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
818 goto out;
819
820 if (base_gfn != old.base_gfn)
821 change = KVM_MR_MOVE;
822 else if (new.flags != old.flags)
823 change = KVM_MR_FLAGS_ONLY;
824 else { /* Nothing to change. */
825 r = 0;
826 goto out;
827 }
828 }
829 } else if (old.npages) {
830 change = KVM_MR_DELETE;
831 } else /* Modify a non-existent slot: disallowed. */
832 goto out;
833
834 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
835 /* Check for overlaps */
836 r = -EEXIST;
837 kvm_for_each_memslot(slot, kvm->memslots) {
838 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
839 (slot->id == mem->slot))
840 continue;
841 if (!((base_gfn + npages <= slot->base_gfn) ||
842 (base_gfn >= slot->base_gfn + slot->npages)))
843 goto out;
844 }
845 }
846
847 /* Free page dirty bitmap if unneeded */
848 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
849 new.dirty_bitmap = NULL;
850
851 r = -ENOMEM;
852 if (change == KVM_MR_CREATE) {
853 new.userspace_addr = mem->userspace_addr;
854
855 if (kvm_arch_create_memslot(kvm, &new, npages))
856 goto out_free;
857 }
858
859 /* Allocate page dirty bitmap if needed */
860 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
861 if (kvm_create_dirty_bitmap(&new) < 0)
862 goto out_free;
863 }
864
865 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
866 r = -ENOMEM;
867 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
868 GFP_KERNEL);
869 if (!slots)
870 goto out_free;
871 slot = id_to_memslot(slots, mem->slot);
872 slot->flags |= KVM_MEMSLOT_INVALID;
873
874 old_memslots = install_new_memslots(kvm, slots, NULL);
875
876 /* slot was deleted or moved, clear iommu mapping */
877 kvm_iommu_unmap_pages(kvm, &old);
878 /* From this point no new shadow pages pointing to a deleted,
879 * or moved, memslot will be created.
880 *
881 * validation of sp->gfn happens in:
882 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
883 * - kvm_is_visible_gfn (mmu_check_roots)
884 */
885 kvm_arch_flush_shadow_memslot(kvm, slot);
886 slots = old_memslots;
887 }
888
889 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
890 if (r)
891 goto out_slots;
892
893 r = -ENOMEM;
894 /*
895 * We can re-use the old_memslots from above, the only difference
896 * from the currently installed memslots is the invalid flag. This
897 * will get overwritten by update_memslots anyway.
898 */
899 if (!slots) {
900 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
901 GFP_KERNEL);
902 if (!slots)
903 goto out_free;
904 }
905
906 /* actual memory is freed via old in kvm_free_physmem_slot below */
907 if (change == KVM_MR_DELETE) {
908 new.dirty_bitmap = NULL;
909 memset(&new.arch, 0, sizeof(new.arch));
910 }
911
912 old_memslots = install_new_memslots(kvm, slots, &new);
913
914 kvm_arch_commit_memory_region(kvm, mem, &old, change);
915
916 kvm_free_physmem_slot(kvm, &old, &new);
917 kfree(old_memslots);
918
919 /*
920 * IOMMU mapping: New slots need to be mapped. Old slots need to be
921 * un-mapped and re-mapped if their base changes. Since base change
922 * unmapping is handled above with slot deletion, mapping alone is
923 * needed here. Anything else the iommu might care about for existing
924 * slots (size changes, userspace addr changes and read-only flag
925 * changes) is disallowed above, so any other attribute changes getting
926 * here can be skipped.
927 */
928 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
929 r = kvm_iommu_map_pages(kvm, &new);
930 return r;
931 }
932
933 return 0;
934
935 out_slots:
936 kfree(slots);
937 out_free:
938 kvm_free_physmem_slot(kvm, &new, &old);
939 out:
940 return r;
941 }
942 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
943
944 int kvm_set_memory_region(struct kvm *kvm,
945 struct kvm_userspace_memory_region *mem)
946 {
947 int r;
948
949 mutex_lock(&kvm->slots_lock);
950 r = __kvm_set_memory_region(kvm, mem);
951 mutex_unlock(&kvm->slots_lock);
952 return r;
953 }
954 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
955
956 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
957 struct kvm_userspace_memory_region *mem)
958 {
959 if (mem->slot >= KVM_USER_MEM_SLOTS)
960 return -EINVAL;
961 return kvm_set_memory_region(kvm, mem);
962 }
963
964 int kvm_get_dirty_log(struct kvm *kvm,
965 struct kvm_dirty_log *log, int *is_dirty)
966 {
967 struct kvm_memory_slot *memslot;
968 int r, i;
969 unsigned long n;
970 unsigned long any = 0;
971
972 r = -EINVAL;
973 if (log->slot >= KVM_USER_MEM_SLOTS)
974 goto out;
975
976 memslot = id_to_memslot(kvm->memslots, log->slot);
977 r = -ENOENT;
978 if (!memslot->dirty_bitmap)
979 goto out;
980
981 n = kvm_dirty_bitmap_bytes(memslot);
982
983 for (i = 0; !any && i < n/sizeof(long); ++i)
984 any = memslot->dirty_bitmap[i];
985
986 r = -EFAULT;
987 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
988 goto out;
989
990 if (any)
991 *is_dirty = 1;
992
993 r = 0;
994 out:
995 return r;
996 }
997 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
998
999 bool kvm_largepages_enabled(void)
1000 {
1001 return largepages_enabled;
1002 }
1003
1004 void kvm_disable_largepages(void)
1005 {
1006 largepages_enabled = false;
1007 }
1008 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1009
1010 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1011 {
1012 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1015
1016 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1017 {
1018 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1019
1020 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1021 memslot->flags & KVM_MEMSLOT_INVALID)
1022 return 0;
1023
1024 return 1;
1025 }
1026 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1027
1028 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1029 {
1030 struct vm_area_struct *vma;
1031 unsigned long addr, size;
1032
1033 size = PAGE_SIZE;
1034
1035 addr = gfn_to_hva(kvm, gfn);
1036 if (kvm_is_error_hva(addr))
1037 return PAGE_SIZE;
1038
1039 down_read(&current->mm->mmap_sem);
1040 vma = find_vma(current->mm, addr);
1041 if (!vma)
1042 goto out;
1043
1044 size = vma_kernel_pagesize(vma);
1045
1046 out:
1047 up_read(&current->mm->mmap_sem);
1048
1049 return size;
1050 }
1051
1052 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1053 {
1054 return slot->flags & KVM_MEM_READONLY;
1055 }
1056
1057 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1058 gfn_t *nr_pages, bool write)
1059 {
1060 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1061 return KVM_HVA_ERR_BAD;
1062
1063 if (memslot_is_readonly(slot) && write)
1064 return KVM_HVA_ERR_RO_BAD;
1065
1066 if (nr_pages)
1067 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1068
1069 return __gfn_to_hva_memslot(slot, gfn);
1070 }
1071
1072 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1073 gfn_t *nr_pages)
1074 {
1075 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1076 }
1077
1078 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1079 gfn_t gfn)
1080 {
1081 return gfn_to_hva_many(slot, gfn, NULL);
1082 }
1083 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1084
1085 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1086 {
1087 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1088 }
1089 EXPORT_SYMBOL_GPL(gfn_to_hva);
1090
1091 /*
1092 * If writable is set to false, the hva returned by this function is only
1093 * allowed to be read.
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 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1099
1100 if (!kvm_is_error_hva(hva) && writable)
1101 *writable = !memslot_is_readonly(slot);
1102
1103 return hva;
1104 }
1105
1106 static int kvm_read_hva(void *data, void __user *hva, int len)
1107 {
1108 return __copy_from_user(data, hva, len);
1109 }
1110
1111 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1112 {
1113 return __copy_from_user_inatomic(data, hva, len);
1114 }
1115
1116 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1117 unsigned long start, int write, struct page **page)
1118 {
1119 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1120
1121 if (write)
1122 flags |= FOLL_WRITE;
1123
1124 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1125 }
1126
1127 static inline int check_user_page_hwpoison(unsigned long addr)
1128 {
1129 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1130
1131 rc = __get_user_pages(current, current->mm, addr, 1,
1132 flags, NULL, NULL, NULL);
1133 return rc == -EHWPOISON;
1134 }
1135
1136 /*
1137 * The atomic path to get the writable pfn which will be stored in @pfn,
1138 * true indicates success, otherwise false is returned.
1139 */
1140 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1141 bool write_fault, bool *writable, pfn_t *pfn)
1142 {
1143 struct page *page[1];
1144 int npages;
1145
1146 if (!(async || atomic))
1147 return false;
1148
1149 /*
1150 * Fast pin a writable pfn only if it is a write fault request
1151 * or the caller allows to map a writable pfn for a read fault
1152 * request.
1153 */
1154 if (!(write_fault || writable))
1155 return false;
1156
1157 npages = __get_user_pages_fast(addr, 1, 1, page);
1158 if (npages == 1) {
1159 *pfn = page_to_pfn(page[0]);
1160
1161 if (writable)
1162 *writable = true;
1163 return true;
1164 }
1165
1166 return false;
1167 }
1168
1169 /*
1170 * The slow path to get the pfn of the specified host virtual address,
1171 * 1 indicates success, -errno is returned if error is detected.
1172 */
1173 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1174 bool *writable, pfn_t *pfn)
1175 {
1176 struct page *page[1];
1177 int npages = 0;
1178
1179 might_sleep();
1180
1181 if (writable)
1182 *writable = write_fault;
1183
1184 if (async) {
1185 down_read(&current->mm->mmap_sem);
1186 npages = get_user_page_nowait(current, current->mm,
1187 addr, write_fault, page);
1188 up_read(&current->mm->mmap_sem);
1189 } else
1190 npages = get_user_pages_fast(addr, 1, write_fault,
1191 page);
1192 if (npages != 1)
1193 return npages;
1194
1195 /* map read fault as writable if possible */
1196 if (unlikely(!write_fault) && writable) {
1197 struct page *wpage[1];
1198
1199 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1200 if (npages == 1) {
1201 *writable = true;
1202 put_page(page[0]);
1203 page[0] = wpage[0];
1204 }
1205
1206 npages = 1;
1207 }
1208 *pfn = page_to_pfn(page[0]);
1209 return npages;
1210 }
1211
1212 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1213 {
1214 if (unlikely(!(vma->vm_flags & VM_READ)))
1215 return false;
1216
1217 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1218 return false;
1219
1220 return true;
1221 }
1222
1223 /*
1224 * Pin guest page in memory and return its pfn.
1225 * @addr: host virtual address which maps memory to the guest
1226 * @atomic: whether this function can sleep
1227 * @async: whether this function need to wait IO complete if the
1228 * host page is not in the memory
1229 * @write_fault: whether we should get a writable host page
1230 * @writable: whether it allows to map a writable host page for !@write_fault
1231 *
1232 * The function will map a writable host page for these two cases:
1233 * 1): @write_fault = true
1234 * 2): @write_fault = false && @writable, @writable will tell the caller
1235 * whether the mapping is writable.
1236 */
1237 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1238 bool write_fault, bool *writable)
1239 {
1240 struct vm_area_struct *vma;
1241 pfn_t pfn = 0;
1242 int npages;
1243
1244 /* we can do it either atomically or asynchronously, not both */
1245 BUG_ON(atomic && async);
1246
1247 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1248 return pfn;
1249
1250 if (atomic)
1251 return KVM_PFN_ERR_FAULT;
1252
1253 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1254 if (npages == 1)
1255 return pfn;
1256
1257 down_read(&current->mm->mmap_sem);
1258 if (npages == -EHWPOISON ||
1259 (!async && check_user_page_hwpoison(addr))) {
1260 pfn = KVM_PFN_ERR_HWPOISON;
1261 goto exit;
1262 }
1263
1264 vma = find_vma_intersection(current->mm, addr, addr + 1);
1265
1266 if (vma == NULL)
1267 pfn = KVM_PFN_ERR_FAULT;
1268 else if ((vma->vm_flags & VM_PFNMAP)) {
1269 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1270 vma->vm_pgoff;
1271 BUG_ON(!kvm_is_mmio_pfn(pfn));
1272 } else {
1273 if (async && vma_is_valid(vma, write_fault))
1274 *async = true;
1275 pfn = KVM_PFN_ERR_FAULT;
1276 }
1277 exit:
1278 up_read(&current->mm->mmap_sem);
1279 return pfn;
1280 }
1281
1282 static pfn_t
1283 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1284 bool *async, bool write_fault, bool *writable)
1285 {
1286 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1287
1288 if (addr == KVM_HVA_ERR_RO_BAD)
1289 return KVM_PFN_ERR_RO_FAULT;
1290
1291 if (kvm_is_error_hva(addr))
1292 return KVM_PFN_NOSLOT;
1293
1294 /* Do not map writable pfn in the readonly memslot. */
1295 if (writable && memslot_is_readonly(slot)) {
1296 *writable = false;
1297 writable = NULL;
1298 }
1299
1300 return hva_to_pfn(addr, atomic, async, write_fault,
1301 writable);
1302 }
1303
1304 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1305 bool write_fault, bool *writable)
1306 {
1307 struct kvm_memory_slot *slot;
1308
1309 if (async)
1310 *async = false;
1311
1312 slot = gfn_to_memslot(kvm, gfn);
1313
1314 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1315 writable);
1316 }
1317
1318 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1319 {
1320 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1321 }
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1323
1324 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1325 bool write_fault, bool *writable)
1326 {
1327 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1328 }
1329 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1330
1331 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1332 {
1333 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1334 }
1335 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1336
1337 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1338 bool *writable)
1339 {
1340 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1341 }
1342 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1343
1344 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1345 {
1346 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1347 }
1348
1349 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1350 {
1351 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1352 }
1353 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1354
1355 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1356 int nr_pages)
1357 {
1358 unsigned long addr;
1359 gfn_t entry;
1360
1361 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1362 if (kvm_is_error_hva(addr))
1363 return -1;
1364
1365 if (entry < nr_pages)
1366 return 0;
1367
1368 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1369 }
1370 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1371
1372 static struct page *kvm_pfn_to_page(pfn_t pfn)
1373 {
1374 if (is_error_noslot_pfn(pfn))
1375 return KVM_ERR_PTR_BAD_PAGE;
1376
1377 if (kvm_is_mmio_pfn(pfn)) {
1378 WARN_ON(1);
1379 return KVM_ERR_PTR_BAD_PAGE;
1380 }
1381
1382 return pfn_to_page(pfn);
1383 }
1384
1385 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1386 {
1387 pfn_t pfn;
1388
1389 pfn = gfn_to_pfn(kvm, gfn);
1390
1391 return kvm_pfn_to_page(pfn);
1392 }
1393
1394 EXPORT_SYMBOL_GPL(gfn_to_page);
1395
1396 void kvm_release_page_clean(struct page *page)
1397 {
1398 WARN_ON(is_error_page(page));
1399
1400 kvm_release_pfn_clean(page_to_pfn(page));
1401 }
1402 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1403
1404 void kvm_release_pfn_clean(pfn_t pfn)
1405 {
1406 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1407 put_page(pfn_to_page(pfn));
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1410
1411 void kvm_release_page_dirty(struct page *page)
1412 {
1413 WARN_ON(is_error_page(page));
1414
1415 kvm_release_pfn_dirty(page_to_pfn(page));
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1418
1419 static void kvm_release_pfn_dirty(pfn_t pfn)
1420 {
1421 kvm_set_pfn_dirty(pfn);
1422 kvm_release_pfn_clean(pfn);
1423 }
1424
1425 void kvm_set_pfn_dirty(pfn_t pfn)
1426 {
1427 if (!kvm_is_mmio_pfn(pfn)) {
1428 struct page *page = pfn_to_page(pfn);
1429 if (!PageReserved(page))
1430 SetPageDirty(page);
1431 }
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1434
1435 void kvm_set_pfn_accessed(pfn_t pfn)
1436 {
1437 if (!kvm_is_mmio_pfn(pfn))
1438 mark_page_accessed(pfn_to_page(pfn));
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1441
1442 void kvm_get_pfn(pfn_t pfn)
1443 {
1444 if (!kvm_is_mmio_pfn(pfn))
1445 get_page(pfn_to_page(pfn));
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1448
1449 static int next_segment(unsigned long len, int offset)
1450 {
1451 if (len > PAGE_SIZE - offset)
1452 return PAGE_SIZE - offset;
1453 else
1454 return len;
1455 }
1456
1457 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1458 int len)
1459 {
1460 int r;
1461 unsigned long addr;
1462
1463 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1464 if (kvm_is_error_hva(addr))
1465 return -EFAULT;
1466 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1467 if (r)
1468 return -EFAULT;
1469 return 0;
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1472
1473 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1474 {
1475 gfn_t gfn = gpa >> PAGE_SHIFT;
1476 int seg;
1477 int offset = offset_in_page(gpa);
1478 int ret;
1479
1480 while ((seg = next_segment(len, offset)) != 0) {
1481 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1482 if (ret < 0)
1483 return ret;
1484 offset = 0;
1485 len -= seg;
1486 data += seg;
1487 ++gfn;
1488 }
1489 return 0;
1490 }
1491 EXPORT_SYMBOL_GPL(kvm_read_guest);
1492
1493 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1494 unsigned long len)
1495 {
1496 int r;
1497 unsigned long addr;
1498 gfn_t gfn = gpa >> PAGE_SHIFT;
1499 int offset = offset_in_page(gpa);
1500
1501 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1502 if (kvm_is_error_hva(addr))
1503 return -EFAULT;
1504 pagefault_disable();
1505 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1506 pagefault_enable();
1507 if (r)
1508 return -EFAULT;
1509 return 0;
1510 }
1511 EXPORT_SYMBOL(kvm_read_guest_atomic);
1512
1513 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1514 int offset, int len)
1515 {
1516 int r;
1517 unsigned long addr;
1518
1519 addr = gfn_to_hva(kvm, gfn);
1520 if (kvm_is_error_hva(addr))
1521 return -EFAULT;
1522 r = __copy_to_user((void __user *)addr + offset, data, len);
1523 if (r)
1524 return -EFAULT;
1525 mark_page_dirty(kvm, gfn);
1526 return 0;
1527 }
1528 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1529
1530 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1531 unsigned long len)
1532 {
1533 gfn_t gfn = gpa >> PAGE_SHIFT;
1534 int seg;
1535 int offset = offset_in_page(gpa);
1536 int ret;
1537
1538 while ((seg = next_segment(len, offset)) != 0) {
1539 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1540 if (ret < 0)
1541 return ret;
1542 offset = 0;
1543 len -= seg;
1544 data += seg;
1545 ++gfn;
1546 }
1547 return 0;
1548 }
1549
1550 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1551 gpa_t gpa, unsigned long len)
1552 {
1553 struct kvm_memslots *slots = kvm_memslots(kvm);
1554 int offset = offset_in_page(gpa);
1555 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1556 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1557 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1558 gfn_t nr_pages_avail;
1559
1560 ghc->gpa = gpa;
1561 ghc->generation = slots->generation;
1562 ghc->len = len;
1563 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1564 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1565 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1566 ghc->hva += offset;
1567 } else {
1568 /*
1569 * If the requested region crosses two memslots, we still
1570 * verify that the entire region is valid here.
1571 */
1572 while (start_gfn <= end_gfn) {
1573 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1574 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1575 &nr_pages_avail);
1576 if (kvm_is_error_hva(ghc->hva))
1577 return -EFAULT;
1578 start_gfn += nr_pages_avail;
1579 }
1580 /* Use the slow path for cross page reads and writes. */
1581 ghc->memslot = NULL;
1582 }
1583 return 0;
1584 }
1585 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1586
1587 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1588 void *data, unsigned long len)
1589 {
1590 struct kvm_memslots *slots = kvm_memslots(kvm);
1591 int r;
1592
1593 BUG_ON(len > ghc->len);
1594
1595 if (slots->generation != ghc->generation)
1596 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1597
1598 if (unlikely(!ghc->memslot))
1599 return kvm_write_guest(kvm, ghc->gpa, data, len);
1600
1601 if (kvm_is_error_hva(ghc->hva))
1602 return -EFAULT;
1603
1604 r = __copy_to_user((void __user *)ghc->hva, data, len);
1605 if (r)
1606 return -EFAULT;
1607 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1608
1609 return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1612
1613 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1614 void *data, unsigned long len)
1615 {
1616 struct kvm_memslots *slots = kvm_memslots(kvm);
1617 int r;
1618
1619 BUG_ON(len > ghc->len);
1620
1621 if (slots->generation != ghc->generation)
1622 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1623
1624 if (unlikely(!ghc->memslot))
1625 return kvm_read_guest(kvm, ghc->gpa, data, len);
1626
1627 if (kvm_is_error_hva(ghc->hva))
1628 return -EFAULT;
1629
1630 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1631 if (r)
1632 return -EFAULT;
1633
1634 return 0;
1635 }
1636 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1637
1638 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1639 {
1640 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1641
1642 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1643 }
1644 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1645
1646 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1647 {
1648 gfn_t gfn = gpa >> PAGE_SHIFT;
1649 int seg;
1650 int offset = offset_in_page(gpa);
1651 int ret;
1652
1653 while ((seg = next_segment(len, offset)) != 0) {
1654 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1655 if (ret < 0)
1656 return ret;
1657 offset = 0;
1658 len -= seg;
1659 ++gfn;
1660 }
1661 return 0;
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1664
1665 static void mark_page_dirty_in_slot(struct kvm *kvm,
1666 struct kvm_memory_slot *memslot,
1667 gfn_t gfn)
1668 {
1669 if (memslot && memslot->dirty_bitmap) {
1670 unsigned long rel_gfn = gfn - memslot->base_gfn;
1671
1672 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1673 }
1674 }
1675
1676 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1677 {
1678 struct kvm_memory_slot *memslot;
1679
1680 memslot = gfn_to_memslot(kvm, gfn);
1681 mark_page_dirty_in_slot(kvm, memslot, gfn);
1682 }
1683 EXPORT_SYMBOL_GPL(mark_page_dirty);
1684
1685 /*
1686 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1687 */
1688 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1689 {
1690 DEFINE_WAIT(wait);
1691
1692 for (;;) {
1693 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1694
1695 if (kvm_arch_vcpu_runnable(vcpu)) {
1696 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1697 break;
1698 }
1699 if (kvm_cpu_has_pending_timer(vcpu))
1700 break;
1701 if (signal_pending(current))
1702 break;
1703
1704 schedule();
1705 }
1706
1707 finish_wait(&vcpu->wq, &wait);
1708 }
1709 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1710
1711 #ifndef CONFIG_S390
1712 /*
1713 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1714 */
1715 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1716 {
1717 int me;
1718 int cpu = vcpu->cpu;
1719 wait_queue_head_t *wqp;
1720
1721 wqp = kvm_arch_vcpu_wq(vcpu);
1722 if (waitqueue_active(wqp)) {
1723 wake_up_interruptible(wqp);
1724 ++vcpu->stat.halt_wakeup;
1725 }
1726
1727 me = get_cpu();
1728 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1729 if (kvm_arch_vcpu_should_kick(vcpu))
1730 smp_send_reschedule(cpu);
1731 put_cpu();
1732 }
1733 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1734 #endif /* !CONFIG_S390 */
1735
1736 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1737 {
1738 struct pid *pid;
1739 struct task_struct *task = NULL;
1740 int ret = 0;
1741
1742 rcu_read_lock();
1743 pid = rcu_dereference(target->pid);
1744 if (pid)
1745 task = get_pid_task(target->pid, PIDTYPE_PID);
1746 rcu_read_unlock();
1747 if (!task)
1748 return ret;
1749 if (task->flags & PF_VCPU) {
1750 put_task_struct(task);
1751 return ret;
1752 }
1753 ret = yield_to(task, 1);
1754 put_task_struct(task);
1755
1756 return ret;
1757 }
1758 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1759
1760 /*
1761 * Helper that checks whether a VCPU is eligible for directed yield.
1762 * Most eligible candidate to yield is decided by following heuristics:
1763 *
1764 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1765 * (preempted lock holder), indicated by @in_spin_loop.
1766 * Set at the beiginning and cleared at the end of interception/PLE handler.
1767 *
1768 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1769 * chance last time (mostly it has become eligible now since we have probably
1770 * yielded to lockholder in last iteration. This is done by toggling
1771 * @dy_eligible each time a VCPU checked for eligibility.)
1772 *
1773 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1774 * to preempted lock-holder could result in wrong VCPU selection and CPU
1775 * burning. Giving priority for a potential lock-holder increases lock
1776 * progress.
1777 *
1778 * Since algorithm is based on heuristics, accessing another VCPU data without
1779 * locking does not harm. It may result in trying to yield to same VCPU, fail
1780 * and continue with next VCPU and so on.
1781 */
1782 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1783 {
1784 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1785 bool eligible;
1786
1787 eligible = !vcpu->spin_loop.in_spin_loop ||
1788 (vcpu->spin_loop.in_spin_loop &&
1789 vcpu->spin_loop.dy_eligible);
1790
1791 if (vcpu->spin_loop.in_spin_loop)
1792 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1793
1794 return eligible;
1795 #else
1796 return true;
1797 #endif
1798 }
1799
1800 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1801 {
1802 struct kvm *kvm = me->kvm;
1803 struct kvm_vcpu *vcpu;
1804 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1805 int yielded = 0;
1806 int try = 3;
1807 int pass;
1808 int i;
1809
1810 kvm_vcpu_set_in_spin_loop(me, true);
1811 /*
1812 * We boost the priority of a VCPU that is runnable but not
1813 * currently running, because it got preempted by something
1814 * else and called schedule in __vcpu_run. Hopefully that
1815 * VCPU is holding the lock that we need and will release it.
1816 * We approximate round-robin by starting at the last boosted VCPU.
1817 */
1818 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1819 kvm_for_each_vcpu(i, vcpu, kvm) {
1820 if (!pass && i <= last_boosted_vcpu) {
1821 i = last_boosted_vcpu;
1822 continue;
1823 } else if (pass && i > last_boosted_vcpu)
1824 break;
1825 if (!ACCESS_ONCE(vcpu->preempted))
1826 continue;
1827 if (vcpu == me)
1828 continue;
1829 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1830 continue;
1831 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1832 continue;
1833
1834 yielded = kvm_vcpu_yield_to(vcpu);
1835 if (yielded > 0) {
1836 kvm->last_boosted_vcpu = i;
1837 break;
1838 } else if (yielded < 0) {
1839 try--;
1840 if (!try)
1841 break;
1842 }
1843 }
1844 }
1845 kvm_vcpu_set_in_spin_loop(me, false);
1846
1847 /* Ensure vcpu is not eligible during next spinloop */
1848 kvm_vcpu_set_dy_eligible(me, false);
1849 }
1850 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1851
1852 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1853 {
1854 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1855 struct page *page;
1856
1857 if (vmf->pgoff == 0)
1858 page = virt_to_page(vcpu->run);
1859 #ifdef CONFIG_X86
1860 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1861 page = virt_to_page(vcpu->arch.pio_data);
1862 #endif
1863 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1864 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1865 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1866 #endif
1867 else
1868 return kvm_arch_vcpu_fault(vcpu, vmf);
1869 get_page(page);
1870 vmf->page = page;
1871 return 0;
1872 }
1873
1874 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1875 .fault = kvm_vcpu_fault,
1876 };
1877
1878 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1879 {
1880 vma->vm_ops = &kvm_vcpu_vm_ops;
1881 return 0;
1882 }
1883
1884 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1885 {
1886 struct kvm_vcpu *vcpu = filp->private_data;
1887
1888 kvm_put_kvm(vcpu->kvm);
1889 return 0;
1890 }
1891
1892 static struct file_operations kvm_vcpu_fops = {
1893 .release = kvm_vcpu_release,
1894 .unlocked_ioctl = kvm_vcpu_ioctl,
1895 #ifdef CONFIG_COMPAT
1896 .compat_ioctl = kvm_vcpu_compat_ioctl,
1897 #endif
1898 .mmap = kvm_vcpu_mmap,
1899 .llseek = noop_llseek,
1900 };
1901
1902 /*
1903 * Allocates an inode for the vcpu.
1904 */
1905 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1906 {
1907 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1908 }
1909
1910 /*
1911 * Creates some virtual cpus. Good luck creating more than one.
1912 */
1913 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1914 {
1915 int r;
1916 struct kvm_vcpu *vcpu, *v;
1917
1918 if (id >= KVM_MAX_VCPUS)
1919 return -EINVAL;
1920
1921 vcpu = kvm_arch_vcpu_create(kvm, id);
1922 if (IS_ERR(vcpu))
1923 return PTR_ERR(vcpu);
1924
1925 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1926
1927 r = kvm_arch_vcpu_setup(vcpu);
1928 if (r)
1929 goto vcpu_destroy;
1930
1931 mutex_lock(&kvm->lock);
1932 if (!kvm_vcpu_compatible(vcpu)) {
1933 r = -EINVAL;
1934 goto unlock_vcpu_destroy;
1935 }
1936 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1937 r = -EINVAL;
1938 goto unlock_vcpu_destroy;
1939 }
1940
1941 kvm_for_each_vcpu(r, v, kvm)
1942 if (v->vcpu_id == id) {
1943 r = -EEXIST;
1944 goto unlock_vcpu_destroy;
1945 }
1946
1947 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1948
1949 /* Now it's all set up, let userspace reach it */
1950 kvm_get_kvm(kvm);
1951 r = create_vcpu_fd(vcpu);
1952 if (r < 0) {
1953 kvm_put_kvm(kvm);
1954 goto unlock_vcpu_destroy;
1955 }
1956
1957 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1958 smp_wmb();
1959 atomic_inc(&kvm->online_vcpus);
1960
1961 mutex_unlock(&kvm->lock);
1962 kvm_arch_vcpu_postcreate(vcpu);
1963 return r;
1964
1965 unlock_vcpu_destroy:
1966 mutex_unlock(&kvm->lock);
1967 vcpu_destroy:
1968 kvm_arch_vcpu_destroy(vcpu);
1969 return r;
1970 }
1971
1972 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1973 {
1974 if (sigset) {
1975 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1976 vcpu->sigset_active = 1;
1977 vcpu->sigset = *sigset;
1978 } else
1979 vcpu->sigset_active = 0;
1980 return 0;
1981 }
1982
1983 static long kvm_vcpu_ioctl(struct file *filp,
1984 unsigned int ioctl, unsigned long arg)
1985 {
1986 struct kvm_vcpu *vcpu = filp->private_data;
1987 void __user *argp = (void __user *)arg;
1988 int r;
1989 struct kvm_fpu *fpu = NULL;
1990 struct kvm_sregs *kvm_sregs = NULL;
1991
1992 if (vcpu->kvm->mm != current->mm)
1993 return -EIO;
1994
1995 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1996 /*
1997 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1998 * so vcpu_load() would break it.
1999 */
2000 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2001 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2002 #endif
2003
2004
2005 r = vcpu_load(vcpu);
2006 if (r)
2007 return r;
2008 switch (ioctl) {
2009 case KVM_RUN:
2010 r = -EINVAL;
2011 if (arg)
2012 goto out;
2013 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2014 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2015 break;
2016 case KVM_GET_REGS: {
2017 struct kvm_regs *kvm_regs;
2018
2019 r = -ENOMEM;
2020 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2021 if (!kvm_regs)
2022 goto out;
2023 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2024 if (r)
2025 goto out_free1;
2026 r = -EFAULT;
2027 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2028 goto out_free1;
2029 r = 0;
2030 out_free1:
2031 kfree(kvm_regs);
2032 break;
2033 }
2034 case KVM_SET_REGS: {
2035 struct kvm_regs *kvm_regs;
2036
2037 r = -ENOMEM;
2038 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2039 if (IS_ERR(kvm_regs)) {
2040 r = PTR_ERR(kvm_regs);
2041 goto out;
2042 }
2043 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2044 kfree(kvm_regs);
2045 break;
2046 }
2047 case KVM_GET_SREGS: {
2048 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2049 r = -ENOMEM;
2050 if (!kvm_sregs)
2051 goto out;
2052 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2053 if (r)
2054 goto out;
2055 r = -EFAULT;
2056 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2057 goto out;
2058 r = 0;
2059 break;
2060 }
2061 case KVM_SET_SREGS: {
2062 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2063 if (IS_ERR(kvm_sregs)) {
2064 r = PTR_ERR(kvm_sregs);
2065 kvm_sregs = NULL;
2066 goto out;
2067 }
2068 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2069 break;
2070 }
2071 case KVM_GET_MP_STATE: {
2072 struct kvm_mp_state mp_state;
2073
2074 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2075 if (r)
2076 goto out;
2077 r = -EFAULT;
2078 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2079 goto out;
2080 r = 0;
2081 break;
2082 }
2083 case KVM_SET_MP_STATE: {
2084 struct kvm_mp_state mp_state;
2085
2086 r = -EFAULT;
2087 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2088 goto out;
2089 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2090 break;
2091 }
2092 case KVM_TRANSLATE: {
2093 struct kvm_translation tr;
2094
2095 r = -EFAULT;
2096 if (copy_from_user(&tr, argp, sizeof tr))
2097 goto out;
2098 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2099 if (r)
2100 goto out;
2101 r = -EFAULT;
2102 if (copy_to_user(argp, &tr, sizeof tr))
2103 goto out;
2104 r = 0;
2105 break;
2106 }
2107 case KVM_SET_GUEST_DEBUG: {
2108 struct kvm_guest_debug dbg;
2109
2110 r = -EFAULT;
2111 if (copy_from_user(&dbg, argp, sizeof dbg))
2112 goto out;
2113 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2114 break;
2115 }
2116 case KVM_SET_SIGNAL_MASK: {
2117 struct kvm_signal_mask __user *sigmask_arg = argp;
2118 struct kvm_signal_mask kvm_sigmask;
2119 sigset_t sigset, *p;
2120
2121 p = NULL;
2122 if (argp) {
2123 r = -EFAULT;
2124 if (copy_from_user(&kvm_sigmask, argp,
2125 sizeof kvm_sigmask))
2126 goto out;
2127 r = -EINVAL;
2128 if (kvm_sigmask.len != sizeof sigset)
2129 goto out;
2130 r = -EFAULT;
2131 if (copy_from_user(&sigset, sigmask_arg->sigset,
2132 sizeof sigset))
2133 goto out;
2134 p = &sigset;
2135 }
2136 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2137 break;
2138 }
2139 case KVM_GET_FPU: {
2140 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2141 r = -ENOMEM;
2142 if (!fpu)
2143 goto out;
2144 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2145 if (r)
2146 goto out;
2147 r = -EFAULT;
2148 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2149 goto out;
2150 r = 0;
2151 break;
2152 }
2153 case KVM_SET_FPU: {
2154 fpu = memdup_user(argp, sizeof(*fpu));
2155 if (IS_ERR(fpu)) {
2156 r = PTR_ERR(fpu);
2157 fpu = NULL;
2158 goto out;
2159 }
2160 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2161 break;
2162 }
2163 default:
2164 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2165 }
2166 out:
2167 vcpu_put(vcpu);
2168 kfree(fpu);
2169 kfree(kvm_sregs);
2170 return r;
2171 }
2172
2173 #ifdef CONFIG_COMPAT
2174 static long kvm_vcpu_compat_ioctl(struct file *filp,
2175 unsigned int ioctl, unsigned long arg)
2176 {
2177 struct kvm_vcpu *vcpu = filp->private_data;
2178 void __user *argp = compat_ptr(arg);
2179 int r;
2180
2181 if (vcpu->kvm->mm != current->mm)
2182 return -EIO;
2183
2184 switch (ioctl) {
2185 case KVM_SET_SIGNAL_MASK: {
2186 struct kvm_signal_mask __user *sigmask_arg = argp;
2187 struct kvm_signal_mask kvm_sigmask;
2188 compat_sigset_t csigset;
2189 sigset_t sigset;
2190
2191 if (argp) {
2192 r = -EFAULT;
2193 if (copy_from_user(&kvm_sigmask, argp,
2194 sizeof kvm_sigmask))
2195 goto out;
2196 r = -EINVAL;
2197 if (kvm_sigmask.len != sizeof csigset)
2198 goto out;
2199 r = -EFAULT;
2200 if (copy_from_user(&csigset, sigmask_arg->sigset,
2201 sizeof csigset))
2202 goto out;
2203 sigset_from_compat(&sigset, &csigset);
2204 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2205 } else
2206 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2207 break;
2208 }
2209 default:
2210 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2211 }
2212
2213 out:
2214 return r;
2215 }
2216 #endif
2217
2218 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2219 int (*accessor)(struct kvm_device *dev,
2220 struct kvm_device_attr *attr),
2221 unsigned long arg)
2222 {
2223 struct kvm_device_attr attr;
2224
2225 if (!accessor)
2226 return -EPERM;
2227
2228 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2229 return -EFAULT;
2230
2231 return accessor(dev, &attr);
2232 }
2233
2234 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2235 unsigned long arg)
2236 {
2237 struct kvm_device *dev = filp->private_data;
2238
2239 switch (ioctl) {
2240 case KVM_SET_DEVICE_ATTR:
2241 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2242 case KVM_GET_DEVICE_ATTR:
2243 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2244 case KVM_HAS_DEVICE_ATTR:
2245 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2246 default:
2247 if (dev->ops->ioctl)
2248 return dev->ops->ioctl(dev, ioctl, arg);
2249
2250 return -ENOTTY;
2251 }
2252 }
2253
2254 static int kvm_device_release(struct inode *inode, struct file *filp)
2255 {
2256 struct kvm_device *dev = filp->private_data;
2257 struct kvm *kvm = dev->kvm;
2258
2259 kvm_put_kvm(kvm);
2260 return 0;
2261 }
2262
2263 static const struct file_operations kvm_device_fops = {
2264 .unlocked_ioctl = kvm_device_ioctl,
2265 #ifdef CONFIG_COMPAT
2266 .compat_ioctl = kvm_device_ioctl,
2267 #endif
2268 .release = kvm_device_release,
2269 };
2270
2271 struct kvm_device *kvm_device_from_filp(struct file *filp)
2272 {
2273 if (filp->f_op != &kvm_device_fops)
2274 return NULL;
2275
2276 return filp->private_data;
2277 }
2278
2279 static int kvm_ioctl_create_device(struct kvm *kvm,
2280 struct kvm_create_device *cd)
2281 {
2282 struct kvm_device_ops *ops = NULL;
2283 struct kvm_device *dev;
2284 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2285 int ret;
2286
2287 switch (cd->type) {
2288 #ifdef CONFIG_KVM_MPIC
2289 case KVM_DEV_TYPE_FSL_MPIC_20:
2290 case KVM_DEV_TYPE_FSL_MPIC_42:
2291 ops = &kvm_mpic_ops;
2292 break;
2293 #endif
2294 #ifdef CONFIG_KVM_XICS
2295 case KVM_DEV_TYPE_XICS:
2296 ops = &kvm_xics_ops;
2297 break;
2298 #endif
2299 #ifdef CONFIG_KVM_VFIO
2300 case KVM_DEV_TYPE_VFIO:
2301 ops = &kvm_vfio_ops;
2302 break;
2303 #endif
2304 #ifdef CONFIG_KVM_ARM_VGIC
2305 case KVM_DEV_TYPE_ARM_VGIC_V2:
2306 ops = &kvm_arm_vgic_v2_ops;
2307 break;
2308 #endif
2309 #ifdef CONFIG_S390
2310 case KVM_DEV_TYPE_FLIC:
2311 ops = &kvm_flic_ops;
2312 break;
2313 #endif
2314 default:
2315 return -ENODEV;
2316 }
2317
2318 if (test)
2319 return 0;
2320
2321 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2322 if (!dev)
2323 return -ENOMEM;
2324
2325 dev->ops = ops;
2326 dev->kvm = kvm;
2327
2328 ret = ops->create(dev, cd->type);
2329 if (ret < 0) {
2330 kfree(dev);
2331 return ret;
2332 }
2333
2334 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2335 if (ret < 0) {
2336 ops->destroy(dev);
2337 return ret;
2338 }
2339
2340 list_add(&dev->vm_node, &kvm->devices);
2341 kvm_get_kvm(kvm);
2342 cd->fd = ret;
2343 return 0;
2344 }
2345
2346 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2347 {
2348 switch (arg) {
2349 case KVM_CAP_USER_MEMORY:
2350 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2351 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2352 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2353 case KVM_CAP_SET_BOOT_CPU_ID:
2354 #endif
2355 case KVM_CAP_INTERNAL_ERROR_DATA:
2356 #ifdef CONFIG_HAVE_KVM_MSI
2357 case KVM_CAP_SIGNAL_MSI:
2358 #endif
2359 #ifdef CONFIG_HAVE_KVM_IRQFD
2360 case KVM_CAP_IRQFD_RESAMPLE:
2361 #endif
2362 case KVM_CAP_CHECK_EXTENSION_VM:
2363 return 1;
2364 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2365 case KVM_CAP_IRQ_ROUTING:
2366 return KVM_MAX_IRQ_ROUTES;
2367 #endif
2368 default:
2369 break;
2370 }
2371 return kvm_vm_ioctl_check_extension(kvm, arg);
2372 }
2373
2374 static long kvm_vm_ioctl(struct file *filp,
2375 unsigned int ioctl, unsigned long arg)
2376 {
2377 struct kvm *kvm = filp->private_data;
2378 void __user *argp = (void __user *)arg;
2379 int r;
2380
2381 if (kvm->mm != current->mm)
2382 return -EIO;
2383 switch (ioctl) {
2384 case KVM_CREATE_VCPU:
2385 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2386 break;
2387 case KVM_SET_USER_MEMORY_REGION: {
2388 struct kvm_userspace_memory_region kvm_userspace_mem;
2389
2390 r = -EFAULT;
2391 if (copy_from_user(&kvm_userspace_mem, argp,
2392 sizeof kvm_userspace_mem))
2393 goto out;
2394
2395 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2396 break;
2397 }
2398 case KVM_GET_DIRTY_LOG: {
2399 struct kvm_dirty_log log;
2400
2401 r = -EFAULT;
2402 if (copy_from_user(&log, argp, sizeof log))
2403 goto out;
2404 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2405 break;
2406 }
2407 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2408 case KVM_REGISTER_COALESCED_MMIO: {
2409 struct kvm_coalesced_mmio_zone zone;
2410 r = -EFAULT;
2411 if (copy_from_user(&zone, argp, sizeof zone))
2412 goto out;
2413 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2414 break;
2415 }
2416 case KVM_UNREGISTER_COALESCED_MMIO: {
2417 struct kvm_coalesced_mmio_zone zone;
2418 r = -EFAULT;
2419 if (copy_from_user(&zone, argp, sizeof zone))
2420 goto out;
2421 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2422 break;
2423 }
2424 #endif
2425 case KVM_IRQFD: {
2426 struct kvm_irqfd data;
2427
2428 r = -EFAULT;
2429 if (copy_from_user(&data, argp, sizeof data))
2430 goto out;
2431 r = kvm_irqfd(kvm, &data);
2432 break;
2433 }
2434 case KVM_IOEVENTFD: {
2435 struct kvm_ioeventfd data;
2436
2437 r = -EFAULT;
2438 if (copy_from_user(&data, argp, sizeof data))
2439 goto out;
2440 r = kvm_ioeventfd(kvm, &data);
2441 break;
2442 }
2443 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2444 case KVM_SET_BOOT_CPU_ID:
2445 r = 0;
2446 mutex_lock(&kvm->lock);
2447 if (atomic_read(&kvm->online_vcpus) != 0)
2448 r = -EBUSY;
2449 else
2450 kvm->bsp_vcpu_id = arg;
2451 mutex_unlock(&kvm->lock);
2452 break;
2453 #endif
2454 #ifdef CONFIG_HAVE_KVM_MSI
2455 case KVM_SIGNAL_MSI: {
2456 struct kvm_msi msi;
2457
2458 r = -EFAULT;
2459 if (copy_from_user(&msi, argp, sizeof msi))
2460 goto out;
2461 r = kvm_send_userspace_msi(kvm, &msi);
2462 break;
2463 }
2464 #endif
2465 #ifdef __KVM_HAVE_IRQ_LINE
2466 case KVM_IRQ_LINE_STATUS:
2467 case KVM_IRQ_LINE: {
2468 struct kvm_irq_level irq_event;
2469
2470 r = -EFAULT;
2471 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2472 goto out;
2473
2474 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2475 ioctl == KVM_IRQ_LINE_STATUS);
2476 if (r)
2477 goto out;
2478
2479 r = -EFAULT;
2480 if (ioctl == KVM_IRQ_LINE_STATUS) {
2481 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2482 goto out;
2483 }
2484
2485 r = 0;
2486 break;
2487 }
2488 #endif
2489 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2490 case KVM_SET_GSI_ROUTING: {
2491 struct kvm_irq_routing routing;
2492 struct kvm_irq_routing __user *urouting;
2493 struct kvm_irq_routing_entry *entries;
2494
2495 r = -EFAULT;
2496 if (copy_from_user(&routing, argp, sizeof(routing)))
2497 goto out;
2498 r = -EINVAL;
2499 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2500 goto out;
2501 if (routing.flags)
2502 goto out;
2503 r = -ENOMEM;
2504 entries = vmalloc(routing.nr * sizeof(*entries));
2505 if (!entries)
2506 goto out;
2507 r = -EFAULT;
2508 urouting = argp;
2509 if (copy_from_user(entries, urouting->entries,
2510 routing.nr * sizeof(*entries)))
2511 goto out_free_irq_routing;
2512 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2513 routing.flags);
2514 out_free_irq_routing:
2515 vfree(entries);
2516 break;
2517 }
2518 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2519 case KVM_CREATE_DEVICE: {
2520 struct kvm_create_device cd;
2521
2522 r = -EFAULT;
2523 if (copy_from_user(&cd, argp, sizeof(cd)))
2524 goto out;
2525
2526 r = kvm_ioctl_create_device(kvm, &cd);
2527 if (r)
2528 goto out;
2529
2530 r = -EFAULT;
2531 if (copy_to_user(argp, &cd, sizeof(cd)))
2532 goto out;
2533
2534 r = 0;
2535 break;
2536 }
2537 case KVM_CHECK_EXTENSION:
2538 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2539 break;
2540 default:
2541 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2542 if (r == -ENOTTY)
2543 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2544 }
2545 out:
2546 return r;
2547 }
2548
2549 #ifdef CONFIG_COMPAT
2550 struct compat_kvm_dirty_log {
2551 __u32 slot;
2552 __u32 padding1;
2553 union {
2554 compat_uptr_t dirty_bitmap; /* one bit per page */
2555 __u64 padding2;
2556 };
2557 };
2558
2559 static long kvm_vm_compat_ioctl(struct file *filp,
2560 unsigned int ioctl, unsigned long arg)
2561 {
2562 struct kvm *kvm = filp->private_data;
2563 int r;
2564
2565 if (kvm->mm != current->mm)
2566 return -EIO;
2567 switch (ioctl) {
2568 case KVM_GET_DIRTY_LOG: {
2569 struct compat_kvm_dirty_log compat_log;
2570 struct kvm_dirty_log log;
2571
2572 r = -EFAULT;
2573 if (copy_from_user(&compat_log, (void __user *)arg,
2574 sizeof(compat_log)))
2575 goto out;
2576 log.slot = compat_log.slot;
2577 log.padding1 = compat_log.padding1;
2578 log.padding2 = compat_log.padding2;
2579 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2580
2581 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2582 break;
2583 }
2584 default:
2585 r = kvm_vm_ioctl(filp, ioctl, arg);
2586 }
2587
2588 out:
2589 return r;
2590 }
2591 #endif
2592
2593 static struct file_operations kvm_vm_fops = {
2594 .release = kvm_vm_release,
2595 .unlocked_ioctl = kvm_vm_ioctl,
2596 #ifdef CONFIG_COMPAT
2597 .compat_ioctl = kvm_vm_compat_ioctl,
2598 #endif
2599 .llseek = noop_llseek,
2600 };
2601
2602 static int kvm_dev_ioctl_create_vm(unsigned long type)
2603 {
2604 int r;
2605 struct kvm *kvm;
2606
2607 kvm = kvm_create_vm(type);
2608 if (IS_ERR(kvm))
2609 return PTR_ERR(kvm);
2610 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2611 r = kvm_coalesced_mmio_init(kvm);
2612 if (r < 0) {
2613 kvm_put_kvm(kvm);
2614 return r;
2615 }
2616 #endif
2617 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2618 if (r < 0)
2619 kvm_put_kvm(kvm);
2620
2621 return r;
2622 }
2623
2624 static long kvm_dev_ioctl(struct file *filp,
2625 unsigned int ioctl, unsigned long arg)
2626 {
2627 long r = -EINVAL;
2628
2629 switch (ioctl) {
2630 case KVM_GET_API_VERSION:
2631 r = -EINVAL;
2632 if (arg)
2633 goto out;
2634 r = KVM_API_VERSION;
2635 break;
2636 case KVM_CREATE_VM:
2637 r = kvm_dev_ioctl_create_vm(arg);
2638 break;
2639 case KVM_CHECK_EXTENSION:
2640 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2641 break;
2642 case KVM_GET_VCPU_MMAP_SIZE:
2643 r = -EINVAL;
2644 if (arg)
2645 goto out;
2646 r = PAGE_SIZE; /* struct kvm_run */
2647 #ifdef CONFIG_X86
2648 r += PAGE_SIZE; /* pio data page */
2649 #endif
2650 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2651 r += PAGE_SIZE; /* coalesced mmio ring page */
2652 #endif
2653 break;
2654 case KVM_TRACE_ENABLE:
2655 case KVM_TRACE_PAUSE:
2656 case KVM_TRACE_DISABLE:
2657 r = -EOPNOTSUPP;
2658 break;
2659 default:
2660 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2661 }
2662 out:
2663 return r;
2664 }
2665
2666 static struct file_operations kvm_chardev_ops = {
2667 .unlocked_ioctl = kvm_dev_ioctl,
2668 .compat_ioctl = kvm_dev_ioctl,
2669 .llseek = noop_llseek,
2670 };
2671
2672 static struct miscdevice kvm_dev = {
2673 KVM_MINOR,
2674 "kvm",
2675 &kvm_chardev_ops,
2676 };
2677
2678 static void hardware_enable_nolock(void *junk)
2679 {
2680 int cpu = raw_smp_processor_id();
2681 int r;
2682
2683 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2684 return;
2685
2686 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2687
2688 r = kvm_arch_hardware_enable();
2689
2690 if (r) {
2691 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2692 atomic_inc(&hardware_enable_failed);
2693 printk(KERN_INFO "kvm: enabling virtualization on "
2694 "CPU%d failed\n", cpu);
2695 }
2696 }
2697
2698 static void hardware_enable(void)
2699 {
2700 raw_spin_lock(&kvm_count_lock);
2701 if (kvm_usage_count)
2702 hardware_enable_nolock(NULL);
2703 raw_spin_unlock(&kvm_count_lock);
2704 }
2705
2706 static void hardware_disable_nolock(void *junk)
2707 {
2708 int cpu = raw_smp_processor_id();
2709
2710 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2711 return;
2712 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2713 kvm_arch_hardware_disable();
2714 }
2715
2716 static void hardware_disable(void)
2717 {
2718 raw_spin_lock(&kvm_count_lock);
2719 if (kvm_usage_count)
2720 hardware_disable_nolock(NULL);
2721 raw_spin_unlock(&kvm_count_lock);
2722 }
2723
2724 static void hardware_disable_all_nolock(void)
2725 {
2726 BUG_ON(!kvm_usage_count);
2727
2728 kvm_usage_count--;
2729 if (!kvm_usage_count)
2730 on_each_cpu(hardware_disable_nolock, NULL, 1);
2731 }
2732
2733 static void hardware_disable_all(void)
2734 {
2735 raw_spin_lock(&kvm_count_lock);
2736 hardware_disable_all_nolock();
2737 raw_spin_unlock(&kvm_count_lock);
2738 }
2739
2740 static int hardware_enable_all(void)
2741 {
2742 int r = 0;
2743
2744 raw_spin_lock(&kvm_count_lock);
2745
2746 kvm_usage_count++;
2747 if (kvm_usage_count == 1) {
2748 atomic_set(&hardware_enable_failed, 0);
2749 on_each_cpu(hardware_enable_nolock, NULL, 1);
2750
2751 if (atomic_read(&hardware_enable_failed)) {
2752 hardware_disable_all_nolock();
2753 r = -EBUSY;
2754 }
2755 }
2756
2757 raw_spin_unlock(&kvm_count_lock);
2758
2759 return r;
2760 }
2761
2762 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2763 void *v)
2764 {
2765 int cpu = (long)v;
2766
2767 val &= ~CPU_TASKS_FROZEN;
2768 switch (val) {
2769 case CPU_DYING:
2770 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2771 cpu);
2772 hardware_disable();
2773 break;
2774 case CPU_STARTING:
2775 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2776 cpu);
2777 hardware_enable();
2778 break;
2779 }
2780 return NOTIFY_OK;
2781 }
2782
2783 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2784 void *v)
2785 {
2786 /*
2787 * Some (well, at least mine) BIOSes hang on reboot if
2788 * in vmx root mode.
2789 *
2790 * And Intel TXT required VMX off for all cpu when system shutdown.
2791 */
2792 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2793 kvm_rebooting = true;
2794 on_each_cpu(hardware_disable_nolock, NULL, 1);
2795 return NOTIFY_OK;
2796 }
2797
2798 static struct notifier_block kvm_reboot_notifier = {
2799 .notifier_call = kvm_reboot,
2800 .priority = 0,
2801 };
2802
2803 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2804 {
2805 int i;
2806
2807 for (i = 0; i < bus->dev_count; i++) {
2808 struct kvm_io_device *pos = bus->range[i].dev;
2809
2810 kvm_iodevice_destructor(pos);
2811 }
2812 kfree(bus);
2813 }
2814
2815 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2816 const struct kvm_io_range *r2)
2817 {
2818 if (r1->addr < r2->addr)
2819 return -1;
2820 if (r1->addr + r1->len > r2->addr + r2->len)
2821 return 1;
2822 return 0;
2823 }
2824
2825 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2826 {
2827 return kvm_io_bus_cmp(p1, p2);
2828 }
2829
2830 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2831 gpa_t addr, int len)
2832 {
2833 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2834 .addr = addr,
2835 .len = len,
2836 .dev = dev,
2837 };
2838
2839 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2840 kvm_io_bus_sort_cmp, NULL);
2841
2842 return 0;
2843 }
2844
2845 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2846 gpa_t addr, int len)
2847 {
2848 struct kvm_io_range *range, key;
2849 int off;
2850
2851 key = (struct kvm_io_range) {
2852 .addr = addr,
2853 .len = len,
2854 };
2855
2856 range = bsearch(&key, bus->range, bus->dev_count,
2857 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2858 if (range == NULL)
2859 return -ENOENT;
2860
2861 off = range - bus->range;
2862
2863 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2864 off--;
2865
2866 return off;
2867 }
2868
2869 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2870 struct kvm_io_range *range, const void *val)
2871 {
2872 int idx;
2873
2874 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2875 if (idx < 0)
2876 return -EOPNOTSUPP;
2877
2878 while (idx < bus->dev_count &&
2879 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2880 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2881 range->len, val))
2882 return idx;
2883 idx++;
2884 }
2885
2886 return -EOPNOTSUPP;
2887 }
2888
2889 /* kvm_io_bus_write - called under kvm->slots_lock */
2890 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2891 int len, const void *val)
2892 {
2893 struct kvm_io_bus *bus;
2894 struct kvm_io_range range;
2895 int r;
2896
2897 range = (struct kvm_io_range) {
2898 .addr = addr,
2899 .len = len,
2900 };
2901
2902 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2903 r = __kvm_io_bus_write(bus, &range, val);
2904 return r < 0 ? r : 0;
2905 }
2906
2907 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2908 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2909 int len, const void *val, long cookie)
2910 {
2911 struct kvm_io_bus *bus;
2912 struct kvm_io_range range;
2913
2914 range = (struct kvm_io_range) {
2915 .addr = addr,
2916 .len = len,
2917 };
2918
2919 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2920
2921 /* First try the device referenced by cookie. */
2922 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2923 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2924 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2925 val))
2926 return cookie;
2927
2928 /*
2929 * cookie contained garbage; fall back to search and return the
2930 * correct cookie value.
2931 */
2932 return __kvm_io_bus_write(bus, &range, val);
2933 }
2934
2935 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2936 void *val)
2937 {
2938 int idx;
2939
2940 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2941 if (idx < 0)
2942 return -EOPNOTSUPP;
2943
2944 while (idx < bus->dev_count &&
2945 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2946 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2947 range->len, val))
2948 return idx;
2949 idx++;
2950 }
2951
2952 return -EOPNOTSUPP;
2953 }
2954 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2955
2956 /* kvm_io_bus_read - called under kvm->slots_lock */
2957 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2958 int len, void *val)
2959 {
2960 struct kvm_io_bus *bus;
2961 struct kvm_io_range range;
2962 int r;
2963
2964 range = (struct kvm_io_range) {
2965 .addr = addr,
2966 .len = len,
2967 };
2968
2969 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2970 r = __kvm_io_bus_read(bus, &range, val);
2971 return r < 0 ? r : 0;
2972 }
2973
2974
2975 /* Caller must hold slots_lock. */
2976 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2977 int len, struct kvm_io_device *dev)
2978 {
2979 struct kvm_io_bus *new_bus, *bus;
2980
2981 bus = kvm->buses[bus_idx];
2982 /* exclude ioeventfd which is limited by maximum fd */
2983 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2984 return -ENOSPC;
2985
2986 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2987 sizeof(struct kvm_io_range)), GFP_KERNEL);
2988 if (!new_bus)
2989 return -ENOMEM;
2990 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2991 sizeof(struct kvm_io_range)));
2992 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2993 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2994 synchronize_srcu_expedited(&kvm->srcu);
2995 kfree(bus);
2996
2997 return 0;
2998 }
2999
3000 /* Caller must hold slots_lock. */
3001 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3002 struct kvm_io_device *dev)
3003 {
3004 int i, r;
3005 struct kvm_io_bus *new_bus, *bus;
3006
3007 bus = kvm->buses[bus_idx];
3008 r = -ENOENT;
3009 for (i = 0; i < bus->dev_count; i++)
3010 if (bus->range[i].dev == dev) {
3011 r = 0;
3012 break;
3013 }
3014
3015 if (r)
3016 return r;
3017
3018 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3019 sizeof(struct kvm_io_range)), GFP_KERNEL);
3020 if (!new_bus)
3021 return -ENOMEM;
3022
3023 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3024 new_bus->dev_count--;
3025 memcpy(new_bus->range + i, bus->range + i + 1,
3026 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3027
3028 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3029 synchronize_srcu_expedited(&kvm->srcu);
3030 kfree(bus);
3031 return r;
3032 }
3033
3034 static struct notifier_block kvm_cpu_notifier = {
3035 .notifier_call = kvm_cpu_hotplug,
3036 };
3037
3038 static int vm_stat_get(void *_offset, u64 *val)
3039 {
3040 unsigned offset = (long)_offset;
3041 struct kvm *kvm;
3042
3043 *val = 0;
3044 spin_lock(&kvm_lock);
3045 list_for_each_entry(kvm, &vm_list, vm_list)
3046 *val += *(u32 *)((void *)kvm + offset);
3047 spin_unlock(&kvm_lock);
3048 return 0;
3049 }
3050
3051 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3052
3053 static int vcpu_stat_get(void *_offset, u64 *val)
3054 {
3055 unsigned offset = (long)_offset;
3056 struct kvm *kvm;
3057 struct kvm_vcpu *vcpu;
3058 int i;
3059
3060 *val = 0;
3061 spin_lock(&kvm_lock);
3062 list_for_each_entry(kvm, &vm_list, vm_list)
3063 kvm_for_each_vcpu(i, vcpu, kvm)
3064 *val += *(u32 *)((void *)vcpu + offset);
3065
3066 spin_unlock(&kvm_lock);
3067 return 0;
3068 }
3069
3070 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3071
3072 static const struct file_operations *stat_fops[] = {
3073 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3074 [KVM_STAT_VM] = &vm_stat_fops,
3075 };
3076
3077 static int kvm_init_debug(void)
3078 {
3079 int r = -EEXIST;
3080 struct kvm_stats_debugfs_item *p;
3081
3082 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3083 if (kvm_debugfs_dir == NULL)
3084 goto out;
3085
3086 for (p = debugfs_entries; p->name; ++p) {
3087 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3088 (void *)(long)p->offset,
3089 stat_fops[p->kind]);
3090 if (p->dentry == NULL)
3091 goto out_dir;
3092 }
3093
3094 return 0;
3095
3096 out_dir:
3097 debugfs_remove_recursive(kvm_debugfs_dir);
3098 out:
3099 return r;
3100 }
3101
3102 static void kvm_exit_debug(void)
3103 {
3104 struct kvm_stats_debugfs_item *p;
3105
3106 for (p = debugfs_entries; p->name; ++p)
3107 debugfs_remove(p->dentry);
3108 debugfs_remove(kvm_debugfs_dir);
3109 }
3110
3111 static int kvm_suspend(void)
3112 {
3113 if (kvm_usage_count)
3114 hardware_disable_nolock(NULL);
3115 return 0;
3116 }
3117
3118 static void kvm_resume(void)
3119 {
3120 if (kvm_usage_count) {
3121 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3122 hardware_enable_nolock(NULL);
3123 }
3124 }
3125
3126 static struct syscore_ops kvm_syscore_ops = {
3127 .suspend = kvm_suspend,
3128 .resume = kvm_resume,
3129 };
3130
3131 static inline
3132 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3133 {
3134 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3135 }
3136
3137 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3138 {
3139 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3140 if (vcpu->preempted)
3141 vcpu->preempted = false;
3142
3143 kvm_arch_sched_in(vcpu, cpu);
3144
3145 kvm_arch_vcpu_load(vcpu, cpu);
3146 }
3147
3148 static void kvm_sched_out(struct preempt_notifier *pn,
3149 struct task_struct *next)
3150 {
3151 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3152
3153 if (current->state == TASK_RUNNING)
3154 vcpu->preempted = true;
3155 kvm_arch_vcpu_put(vcpu);
3156 }
3157
3158 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3159 struct module *module)
3160 {
3161 int r;
3162 int cpu;
3163
3164 r = kvm_arch_init(opaque);
3165 if (r)
3166 goto out_fail;
3167
3168 /*
3169 * kvm_arch_init makes sure there's at most one caller
3170 * for architectures that support multiple implementations,
3171 * like intel and amd on x86.
3172 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3173 * conflicts in case kvm is already setup for another implementation.
3174 */
3175 r = kvm_irqfd_init();
3176 if (r)
3177 goto out_irqfd;
3178
3179 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3180 r = -ENOMEM;
3181 goto out_free_0;
3182 }
3183
3184 r = kvm_arch_hardware_setup();
3185 if (r < 0)
3186 goto out_free_0a;
3187
3188 for_each_online_cpu(cpu) {
3189 smp_call_function_single(cpu,
3190 kvm_arch_check_processor_compat,
3191 &r, 1);
3192 if (r < 0)
3193 goto out_free_1;
3194 }
3195
3196 r = register_cpu_notifier(&kvm_cpu_notifier);
3197 if (r)
3198 goto out_free_2;
3199 register_reboot_notifier(&kvm_reboot_notifier);
3200
3201 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3202 if (!vcpu_align)
3203 vcpu_align = __alignof__(struct kvm_vcpu);
3204 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3205 0, NULL);
3206 if (!kvm_vcpu_cache) {
3207 r = -ENOMEM;
3208 goto out_free_3;
3209 }
3210
3211 r = kvm_async_pf_init();
3212 if (r)
3213 goto out_free;
3214
3215 kvm_chardev_ops.owner = module;
3216 kvm_vm_fops.owner = module;
3217 kvm_vcpu_fops.owner = module;
3218
3219 r = misc_register(&kvm_dev);
3220 if (r) {
3221 printk(KERN_ERR "kvm: misc device register failed\n");
3222 goto out_unreg;
3223 }
3224
3225 register_syscore_ops(&kvm_syscore_ops);
3226
3227 kvm_preempt_ops.sched_in = kvm_sched_in;
3228 kvm_preempt_ops.sched_out = kvm_sched_out;
3229
3230 r = kvm_init_debug();
3231 if (r) {
3232 printk(KERN_ERR "kvm: create debugfs files failed\n");
3233 goto out_undebugfs;
3234 }
3235
3236 return 0;
3237
3238 out_undebugfs:
3239 unregister_syscore_ops(&kvm_syscore_ops);
3240 misc_deregister(&kvm_dev);
3241 out_unreg:
3242 kvm_async_pf_deinit();
3243 out_free:
3244 kmem_cache_destroy(kvm_vcpu_cache);
3245 out_free_3:
3246 unregister_reboot_notifier(&kvm_reboot_notifier);
3247 unregister_cpu_notifier(&kvm_cpu_notifier);
3248 out_free_2:
3249 out_free_1:
3250 kvm_arch_hardware_unsetup();
3251 out_free_0a:
3252 free_cpumask_var(cpus_hardware_enabled);
3253 out_free_0:
3254 kvm_irqfd_exit();
3255 out_irqfd:
3256 kvm_arch_exit();
3257 out_fail:
3258 return r;
3259 }
3260 EXPORT_SYMBOL_GPL(kvm_init);
3261
3262 void kvm_exit(void)
3263 {
3264 kvm_exit_debug();
3265 misc_deregister(&kvm_dev);
3266 kmem_cache_destroy(kvm_vcpu_cache);
3267 kvm_async_pf_deinit();
3268 unregister_syscore_ops(&kvm_syscore_ops);
3269 unregister_reboot_notifier(&kvm_reboot_notifier);
3270 unregister_cpu_notifier(&kvm_cpu_notifier);
3271 on_each_cpu(hardware_disable_nolock, NULL, 1);
3272 kvm_arch_hardware_unsetup();
3273 kvm_arch_exit();
3274 kvm_irqfd_exit();
3275 free_cpumask_var(cpus_hardware_enabled);
3276 }
3277 EXPORT_SYMBOL_GPL(kvm_exit);
Linux kernel - Deliverables
This page took 0.134964 seconds and 4 git commands to generate.