?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
xen/blkfront: Implement FUA with BLKIF_OP_WRITE_BARRIER
[linux-2.6.git] / virt / kvm / kvm_main.c
blob5225052aebc15d28709359b7ee365f0490035203
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/sysdev.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
51 #include <asm/processor.h>
52 #include <asm/io.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
56
57 #include "coalesced_mmio.h"
58
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
61
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
64
65 /*
66 * Ordering of locks:
67 *
68 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
69 */
70
71 DEFINE_SPINLOCK(kvm_lock);
72 LIST_HEAD(vm_list);
73
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
77
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
80
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
82
83 struct dentry *kvm_debugfs_dir;
84
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
86 unsigned long arg);
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
89
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
91
92 static bool kvm_rebooting;
93
94 static bool largepages_enabled = true;
95
96 static struct page *hwpoison_page;
97 static pfn_t hwpoison_pfn;
98
99 static struct page *fault_page;
100 static pfn_t fault_pfn;
101
102 inline int kvm_is_mmio_pfn(pfn_t pfn)
103 {
104 if (pfn_valid(pfn)) {
105 struct page *page = compound_head(pfn_to_page(pfn));
106 return PageReserved(page);
107 }
108
109 return true;
110 }
111
112 /*
113 * Switches to specified vcpu, until a matching vcpu_put()
114 */
115 void vcpu_load(struct kvm_vcpu *vcpu)
116 {
117 int cpu;
118
119 mutex_lock(&vcpu->mutex);
120 cpu = get_cpu();
121 preempt_notifier_register(&vcpu->preempt_notifier);
122 kvm_arch_vcpu_load(vcpu, cpu);
123 put_cpu();
124 }
125
126 void vcpu_put(struct kvm_vcpu *vcpu)
127 {
128 preempt_disable();
129 kvm_arch_vcpu_put(vcpu);
130 preempt_notifier_unregister(&vcpu->preempt_notifier);
131 preempt_enable();
132 mutex_unlock(&vcpu->mutex);
133 }
134
135 static void ack_flush(void *_completed)
136 {
137 }
138
139 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
140 {
141 int i, cpu, me;
142 cpumask_var_t cpus;
143 bool called = true;
144 struct kvm_vcpu *vcpu;
145
146 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
147
148 raw_spin_lock(&kvm->requests_lock);
149 me = smp_processor_id();
150 kvm_for_each_vcpu(i, vcpu, kvm) {
151 if (kvm_make_check_request(req, vcpu))
152 continue;
153 cpu = vcpu->cpu;
154 if (cpus != NULL && cpu != -1 && cpu != me)
155 cpumask_set_cpu(cpu, cpus);
156 }
157 if (unlikely(cpus == NULL))
158 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
159 else if (!cpumask_empty(cpus))
160 smp_call_function_many(cpus, ack_flush, NULL, 1);
161 else
162 called = false;
163 raw_spin_unlock(&kvm->requests_lock);
164 free_cpumask_var(cpus);
165 return called;
166 }
167
168 void kvm_flush_remote_tlbs(struct kvm *kvm)
169 {
170 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
171 ++kvm->stat.remote_tlb_flush;
172 }
173
174 void kvm_reload_remote_mmus(struct kvm *kvm)
175 {
176 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
177 }
178
179 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
180 {
181 struct page *page;
182 int r;
183
184 mutex_init(&vcpu->mutex);
185 vcpu->cpu = -1;
186 vcpu->kvm = kvm;
187 vcpu->vcpu_id = id;
188 init_waitqueue_head(&vcpu->wq);
189
190 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
191 if (!page) {
192 r = -ENOMEM;
193 goto fail;
194 }
195 vcpu->run = page_address(page);
196
197 r = kvm_arch_vcpu_init(vcpu);
198 if (r < 0)
199 goto fail_free_run;
200 return 0;
201
202 fail_free_run:
203 free_page((unsigned long)vcpu->run);
204 fail:
205 return r;
206 }
207 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
208
209 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
210 {
211 kvm_arch_vcpu_uninit(vcpu);
212 free_page((unsigned long)vcpu->run);
213 }
214 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
215
216 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
217 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
218 {
219 return container_of(mn, struct kvm, mmu_notifier);
220 }
221
222 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
223 struct mm_struct *mm,
224 unsigned long address)
225 {
226 struct kvm *kvm = mmu_notifier_to_kvm(mn);
227 int need_tlb_flush, idx;
228
229 /*
230 * When ->invalidate_page runs, the linux pte has been zapped
231 * already but the page is still allocated until
232 * ->invalidate_page returns. So if we increase the sequence
233 * here the kvm page fault will notice if the spte can't be
234 * established because the page is going to be freed. If
235 * instead the kvm page fault establishes the spte before
236 * ->invalidate_page runs, kvm_unmap_hva will release it
237 * before returning.
238 *
239 * The sequence increase only need to be seen at spin_unlock
240 * time, and not at spin_lock time.
241 *
242 * Increasing the sequence after the spin_unlock would be
243 * unsafe because the kvm page fault could then establish the
244 * pte after kvm_unmap_hva returned, without noticing the page
245 * is going to be freed.
246 */
247 idx = srcu_read_lock(&kvm->srcu);
248 spin_lock(&kvm->mmu_lock);
249 kvm->mmu_notifier_seq++;
250 need_tlb_flush = kvm_unmap_hva(kvm, address);
251 spin_unlock(&kvm->mmu_lock);
252 srcu_read_unlock(&kvm->srcu, idx);
253
254 /* we've to flush the tlb before the pages can be freed */
255 if (need_tlb_flush)
256 kvm_flush_remote_tlbs(kvm);
257
258 }
259
260 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address,
263 pte_t pte)
264 {
265 struct kvm *kvm = mmu_notifier_to_kvm(mn);
266 int idx;
267
268 idx = srcu_read_lock(&kvm->srcu);
269 spin_lock(&kvm->mmu_lock);
270 kvm->mmu_notifier_seq++;
271 kvm_set_spte_hva(kvm, address, pte);
272 spin_unlock(&kvm->mmu_lock);
273 srcu_read_unlock(&kvm->srcu, idx);
274 }
275
276 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
277 struct mm_struct *mm,
278 unsigned long start,
279 unsigned long end)
280 {
281 struct kvm *kvm = mmu_notifier_to_kvm(mn);
282 int need_tlb_flush = 0, idx;
283
284 idx = srcu_read_lock(&kvm->srcu);
285 spin_lock(&kvm->mmu_lock);
286 /*
287 * The count increase must become visible at unlock time as no
288 * spte can be established without taking the mmu_lock and
289 * count is also read inside the mmu_lock critical section.
290 */
291 kvm->mmu_notifier_count++;
292 for (; start < end; start += PAGE_SIZE)
293 need_tlb_flush |= kvm_unmap_hva(kvm, start);
294 spin_unlock(&kvm->mmu_lock);
295 srcu_read_unlock(&kvm->srcu, idx);
296
297 /* we've to flush the tlb before the pages can be freed */
298 if (need_tlb_flush)
299 kvm_flush_remote_tlbs(kvm);
300 }
301
302 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long start,
305 unsigned long end)
306 {
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
308
309 spin_lock(&kvm->mmu_lock);
310 /*
311 * This sequence increase will notify the kvm page fault that
312 * the page that is going to be mapped in the spte could have
313 * been freed.
314 */
315 kvm->mmu_notifier_seq++;
316 /*
317 * The above sequence increase must be visible before the
318 * below count decrease but both values are read by the kvm
319 * page fault under mmu_lock spinlock so we don't need to add
320 * a smb_wmb() here in between the two.
321 */
322 kvm->mmu_notifier_count--;
323 spin_unlock(&kvm->mmu_lock);
324
325 BUG_ON(kvm->mmu_notifier_count < 0);
326 }
327
328 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
329 struct mm_struct *mm,
330 unsigned long address)
331 {
332 struct kvm *kvm = mmu_notifier_to_kvm(mn);
333 int young, idx;
334
335 idx = srcu_read_lock(&kvm->srcu);
336 spin_lock(&kvm->mmu_lock);
337 young = kvm_age_hva(kvm, address);
338 spin_unlock(&kvm->mmu_lock);
339 srcu_read_unlock(&kvm->srcu, idx);
340
341 if (young)
342 kvm_flush_remote_tlbs(kvm);
343
344 return young;
345 }
346
347 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
348 struct mm_struct *mm)
349 {
350 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 int idx;
352
353 idx = srcu_read_lock(&kvm->srcu);
354 kvm_arch_flush_shadow(kvm);
355 srcu_read_unlock(&kvm->srcu, idx);
356 }
357
358 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
359 .invalidate_page = kvm_mmu_notifier_invalidate_page,
360 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
361 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
362 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
363 .change_pte = kvm_mmu_notifier_change_pte,
364 .release = kvm_mmu_notifier_release,
365 };
366
367 static int kvm_init_mmu_notifier(struct kvm *kvm)
368 {
369 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
370 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
371 }
372
373 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
374
375 static int kvm_init_mmu_notifier(struct kvm *kvm)
376 {
377 return 0;
378 }
379
380 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
381
382 static struct kvm *kvm_create_vm(void)
383 {
384 int r = 0, i;
385 struct kvm *kvm = kvm_arch_create_vm();
386
387 if (IS_ERR(kvm))
388 goto out;
389
390 r = hardware_enable_all();
391 if (r)
392 goto out_err_nodisable;
393
394 #ifdef CONFIG_HAVE_KVM_IRQCHIP
395 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
396 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
397 #endif
398
399 r = -ENOMEM;
400 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
401 if (!kvm->memslots)
402 goto out_err;
403 if (init_srcu_struct(&kvm->srcu))
404 goto out_err;
405 for (i = 0; i < KVM_NR_BUSES; i++) {
406 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
407 GFP_KERNEL);
408 if (!kvm->buses[i]) {
409 cleanup_srcu_struct(&kvm->srcu);
410 goto out_err;
411 }
412 }
413
414 r = kvm_init_mmu_notifier(kvm);
415 if (r) {
416 cleanup_srcu_struct(&kvm->srcu);
417 goto out_err;
418 }
419
420 kvm->mm = current->mm;
421 atomic_inc(&kvm->mm->mm_count);
422 spin_lock_init(&kvm->mmu_lock);
423 raw_spin_lock_init(&kvm->requests_lock);
424 kvm_eventfd_init(kvm);
425 mutex_init(&kvm->lock);
426 mutex_init(&kvm->irq_lock);
427 mutex_init(&kvm->slots_lock);
428 atomic_set(&kvm->users_count, 1);
429 spin_lock(&kvm_lock);
430 list_add(&kvm->vm_list, &vm_list);
431 spin_unlock(&kvm_lock);
432 out:
433 return kvm;
434
435 out_err:
436 hardware_disable_all();
437 out_err_nodisable:
438 for (i = 0; i < KVM_NR_BUSES; i++)
439 kfree(kvm->buses[i]);
440 kfree(kvm->memslots);
441 kfree(kvm);
442 return ERR_PTR(r);
443 }
444
445 /*
446 * Free any memory in @free but not in @dont.
447 */
448 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
449 struct kvm_memory_slot *dont)
450 {
451 int i;
452
453 if (!dont || free->rmap != dont->rmap)
454 vfree(free->rmap);
455
456 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
457 vfree(free->dirty_bitmap);
458
459
460 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
461 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
462 vfree(free->lpage_info[i]);
463 free->lpage_info[i] = NULL;
464 }
465 }
466
467 free->npages = 0;
468 free->dirty_bitmap = NULL;
469 free->rmap = NULL;
470 }
471
472 void kvm_free_physmem(struct kvm *kvm)
473 {
474 int i;
475 struct kvm_memslots *slots = kvm->memslots;
476
477 for (i = 0; i < slots->nmemslots; ++i)
478 kvm_free_physmem_slot(&slots->memslots[i], NULL);
479
480 kfree(kvm->memslots);
481 }
482
483 static void kvm_destroy_vm(struct kvm *kvm)
484 {
485 int i;
486 struct mm_struct *mm = kvm->mm;
487
488 kvm_arch_sync_events(kvm);
489 spin_lock(&kvm_lock);
490 list_del(&kvm->vm_list);
491 spin_unlock(&kvm_lock);
492 kvm_free_irq_routing(kvm);
493 for (i = 0; i < KVM_NR_BUSES; i++)
494 kvm_io_bus_destroy(kvm->buses[i]);
495 kvm_coalesced_mmio_free(kvm);
496 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
497 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
498 #else
499 kvm_arch_flush_shadow(kvm);
500 #endif
501 kvm_arch_destroy_vm(kvm);
502 hardware_disable_all();
503 mmdrop(mm);
504 }
505
506 void kvm_get_kvm(struct kvm *kvm)
507 {
508 atomic_inc(&kvm->users_count);
509 }
510 EXPORT_SYMBOL_GPL(kvm_get_kvm);
511
512 void kvm_put_kvm(struct kvm *kvm)
513 {
514 if (atomic_dec_and_test(&kvm->users_count))
515 kvm_destroy_vm(kvm);
516 }
517 EXPORT_SYMBOL_GPL(kvm_put_kvm);
518
519
520 static int kvm_vm_release(struct inode *inode, struct file *filp)
521 {
522 struct kvm *kvm = filp->private_data;
523
524 kvm_irqfd_release(kvm);
525
526 kvm_put_kvm(kvm);
527 return 0;
528 }
529
530 /*
531 * Allocate some memory and give it an address in the guest physical address
532 * space.
533 *
534 * Discontiguous memory is allowed, mostly for framebuffers.
535 *
536 * Must be called holding mmap_sem for write.
537 */
538 int __kvm_set_memory_region(struct kvm *kvm,
539 struct kvm_userspace_memory_region *mem,
540 int user_alloc)
541 {
542 int r, flush_shadow = 0;
543 gfn_t base_gfn;
544 unsigned long npages;
545 unsigned long i;
546 struct kvm_memory_slot *memslot;
547 struct kvm_memory_slot old, new;
548 struct kvm_memslots *slots, *old_memslots;
549
550 r = -EINVAL;
551 /* General sanity checks */
552 if (mem->memory_size & (PAGE_SIZE - 1))
553 goto out;
554 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
555 goto out;
556 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
557 goto out;
558 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
559 goto out;
560 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
561 goto out;
562
563 memslot = &kvm->memslots->memslots[mem->slot];
564 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
565 npages = mem->memory_size >> PAGE_SHIFT;
566
567 r = -EINVAL;
568 if (npages > KVM_MEM_MAX_NR_PAGES)
569 goto out;
570
571 if (!npages)
572 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
573
574 new = old = *memslot;
575
576 new.id = mem->slot;
577 new.base_gfn = base_gfn;
578 new.npages = npages;
579 new.flags = mem->flags;
580
581 /* Disallow changing a memory slot's size. */
582 r = -EINVAL;
583 if (npages && old.npages && npages != old.npages)
584 goto out_free;
585
586 /* Check for overlaps */
587 r = -EEXIST;
588 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
589 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
590
591 if (s == memslot || !s->npages)
592 continue;
593 if (!((base_gfn + npages <= s->base_gfn) ||
594 (base_gfn >= s->base_gfn + s->npages)))
595 goto out_free;
596 }
597
598 /* Free page dirty bitmap if unneeded */
599 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
600 new.dirty_bitmap = NULL;
601
602 r = -ENOMEM;
603
604 /* Allocate if a slot is being created */
605 #ifndef CONFIG_S390
606 if (npages && !new.rmap) {
607 new.rmap = vmalloc(npages * sizeof(*new.rmap));
608
609 if (!new.rmap)
610 goto out_free;
611
612 memset(new.rmap, 0, npages * sizeof(*new.rmap));
613
614 new.user_alloc = user_alloc;
615 new.userspace_addr = mem->userspace_addr;
616 }
617 if (!npages)
618 goto skip_lpage;
619
620 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
621 unsigned long ugfn;
622 unsigned long j;
623 int lpages;
624 int level = i + 2;
625
626 /* Avoid unused variable warning if no large pages */
627 (void)level;
628
629 if (new.lpage_info[i])
630 continue;
631
632 lpages = 1 + ((base_gfn + npages - 1)
633 >> KVM_HPAGE_GFN_SHIFT(level));
634 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
635
636 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
637
638 if (!new.lpage_info[i])
639 goto out_free;
640
641 memset(new.lpage_info[i], 0,
642 lpages * sizeof(*new.lpage_info[i]));
643
644 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
645 new.lpage_info[i][0].write_count = 1;
646 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
647 new.lpage_info[i][lpages - 1].write_count = 1;
648 ugfn = new.userspace_addr >> PAGE_SHIFT;
649 /*
650 * If the gfn and userspace address are not aligned wrt each
651 * other, or if explicitly asked to, disable large page
652 * support for this slot
653 */
654 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
655 !largepages_enabled)
656 for (j = 0; j < lpages; ++j)
657 new.lpage_info[i][j].write_count = 1;
658 }
659
660 skip_lpage:
661
662 /* Allocate page dirty bitmap if needed */
663 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
664 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
665
666 new.dirty_bitmap = vmalloc(dirty_bytes);
667 if (!new.dirty_bitmap)
668 goto out_free;
669 memset(new.dirty_bitmap, 0, dirty_bytes);
670 /* destroy any largepage mappings for dirty tracking */
671 if (old.npages)
672 flush_shadow = 1;
673 }
674 #else /* not defined CONFIG_S390 */
675 new.user_alloc = user_alloc;
676 if (user_alloc)
677 new.userspace_addr = mem->userspace_addr;
678 #endif /* not defined CONFIG_S390 */
679
680 if (!npages) {
681 r = -ENOMEM;
682 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
683 if (!slots)
684 goto out_free;
685 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
686 if (mem->slot >= slots->nmemslots)
687 slots->nmemslots = mem->slot + 1;
688 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
689
690 old_memslots = kvm->memslots;
691 rcu_assign_pointer(kvm->memslots, slots);
692 synchronize_srcu_expedited(&kvm->srcu);
693 /* From this point no new shadow pages pointing to a deleted
694 * memslot will be created.
695 *
696 * validation of sp->gfn happens in:
697 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
698 * - kvm_is_visible_gfn (mmu_check_roots)
699 */
700 kvm_arch_flush_shadow(kvm);
701 kfree(old_memslots);
702 }
703
704 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
705 if (r)
706 goto out_free;
707
708 /* map the pages in iommu page table */
709 if (npages) {
710 r = kvm_iommu_map_pages(kvm, &new);
711 if (r)
712 goto out_free;
713 }
714
715 r = -ENOMEM;
716 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
717 if (!slots)
718 goto out_free;
719 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
720 if (mem->slot >= slots->nmemslots)
721 slots->nmemslots = mem->slot + 1;
722
723 /* actual memory is freed via old in kvm_free_physmem_slot below */
724 if (!npages) {
725 new.rmap = NULL;
726 new.dirty_bitmap = NULL;
727 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
728 new.lpage_info[i] = NULL;
729 }
730
731 slots->memslots[mem->slot] = new;
732 old_memslots = kvm->memslots;
733 rcu_assign_pointer(kvm->memslots, slots);
734 synchronize_srcu_expedited(&kvm->srcu);
735
736 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
737
738 kvm_free_physmem_slot(&old, &new);
739 kfree(old_memslots);
740
741 if (flush_shadow)
742 kvm_arch_flush_shadow(kvm);
743
744 return 0;
745
746 out_free:
747 kvm_free_physmem_slot(&new, &old);
748 out:
749 return r;
750
751 }
752 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
753
754 int kvm_set_memory_region(struct kvm *kvm,
755 struct kvm_userspace_memory_region *mem,
756 int user_alloc)
757 {
758 int r;
759
760 mutex_lock(&kvm->slots_lock);
761 r = __kvm_set_memory_region(kvm, mem, user_alloc);
762 mutex_unlock(&kvm->slots_lock);
763 return r;
764 }
765 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
766
767 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
768 struct
769 kvm_userspace_memory_region *mem,
770 int user_alloc)
771 {
772 if (mem->slot >= KVM_MEMORY_SLOTS)
773 return -EINVAL;
774 return kvm_set_memory_region(kvm, mem, user_alloc);
775 }
776
777 int kvm_get_dirty_log(struct kvm *kvm,
778 struct kvm_dirty_log *log, int *is_dirty)
779 {
780 struct kvm_memory_slot *memslot;
781 int r, i;
782 unsigned long n;
783 unsigned long any = 0;
784
785 r = -EINVAL;
786 if (log->slot >= KVM_MEMORY_SLOTS)
787 goto out;
788
789 memslot = &kvm->memslots->memslots[log->slot];
790 r = -ENOENT;
791 if (!memslot->dirty_bitmap)
792 goto out;
793
794 n = kvm_dirty_bitmap_bytes(memslot);
795
796 for (i = 0; !any && i < n/sizeof(long); ++i)
797 any = memslot->dirty_bitmap[i];
798
799 r = -EFAULT;
800 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
801 goto out;
802
803 if (any)
804 *is_dirty = 1;
805
806 r = 0;
807 out:
808 return r;
809 }
810
811 void kvm_disable_largepages(void)
812 {
813 largepages_enabled = false;
814 }
815 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
816
817 int is_error_page(struct page *page)
818 {
819 return page == bad_page || page == hwpoison_page || page == fault_page;
820 }
821 EXPORT_SYMBOL_GPL(is_error_page);
822
823 int is_error_pfn(pfn_t pfn)
824 {
825 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
826 }
827 EXPORT_SYMBOL_GPL(is_error_pfn);
828
829 int is_hwpoison_pfn(pfn_t pfn)
830 {
831 return pfn == hwpoison_pfn;
832 }
833 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
834
835 int is_fault_pfn(pfn_t pfn)
836 {
837 return pfn == fault_pfn;
838 }
839 EXPORT_SYMBOL_GPL(is_fault_pfn);
840
841 static inline unsigned long bad_hva(void)
842 {
843 return PAGE_OFFSET;
844 }
845
846 int kvm_is_error_hva(unsigned long addr)
847 {
848 return addr == bad_hva();
849 }
850 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
851
852 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
853 {
854 int i;
855 struct kvm_memslots *slots = kvm_memslots(kvm);
856
857 for (i = 0; i < slots->nmemslots; ++i) {
858 struct kvm_memory_slot *memslot = &slots->memslots[i];
859
860 if (gfn >= memslot->base_gfn
861 && gfn < memslot->base_gfn + memslot->npages)
862 return memslot;
863 }
864 return NULL;
865 }
866 EXPORT_SYMBOL_GPL(gfn_to_memslot);
867
868 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
869 {
870 int i;
871 struct kvm_memslots *slots = kvm_memslots(kvm);
872
873 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
874 struct kvm_memory_slot *memslot = &slots->memslots[i];
875
876 if (memslot->flags & KVM_MEMSLOT_INVALID)
877 continue;
878
879 if (gfn >= memslot->base_gfn
880 && gfn < memslot->base_gfn + memslot->npages)
881 return 1;
882 }
883 return 0;
884 }
885 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
886
887 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
888 {
889 struct vm_area_struct *vma;
890 unsigned long addr, size;
891
892 size = PAGE_SIZE;
893
894 addr = gfn_to_hva(kvm, gfn);
895 if (kvm_is_error_hva(addr))
896 return PAGE_SIZE;
897
898 down_read(&current->mm->mmap_sem);
899 vma = find_vma(current->mm, addr);
900 if (!vma)
901 goto out;
902
903 size = vma_kernel_pagesize(vma);
904
905 out:
906 up_read(&current->mm->mmap_sem);
907
908 return size;
909 }
910
911 int memslot_id(struct kvm *kvm, gfn_t gfn)
912 {
913 int i;
914 struct kvm_memslots *slots = kvm_memslots(kvm);
915 struct kvm_memory_slot *memslot = NULL;
916
917 for (i = 0; i < slots->nmemslots; ++i) {
918 memslot = &slots->memslots[i];
919
920 if (gfn >= memslot->base_gfn
921 && gfn < memslot->base_gfn + memslot->npages)
922 break;
923 }
924
925 return memslot - slots->memslots;
926 }
927
928 static unsigned long gfn_to_hva_many(struct kvm *kvm, gfn_t gfn,
929 gfn_t *nr_pages)
930 {
931 struct kvm_memory_slot *slot;
932
933 slot = gfn_to_memslot(kvm, gfn);
934 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
935 return bad_hva();
936
937 if (nr_pages)
938 *nr_pages = slot->npages - (gfn - slot->base_gfn);
939
940 return gfn_to_hva_memslot(slot, gfn);
941 }
942
943 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
944 {
945 return gfn_to_hva_many(kvm, gfn, NULL);
946 }
947 EXPORT_SYMBOL_GPL(gfn_to_hva);
948
949 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic)
950 {
951 struct page *page[1];
952 int npages;
953 pfn_t pfn;
954
955 if (atomic)
956 npages = __get_user_pages_fast(addr, 1, 1, page);
957 else {
958 might_sleep();
959 npages = get_user_pages_fast(addr, 1, 1, page);
960 }
961
962 if (unlikely(npages != 1)) {
963 struct vm_area_struct *vma;
964
965 if (atomic)
966 goto return_fault_page;
967
968 down_read(&current->mm->mmap_sem);
969 if (is_hwpoison_address(addr)) {
970 up_read(&current->mm->mmap_sem);
971 get_page(hwpoison_page);
972 return page_to_pfn(hwpoison_page);
973 }
974
975 vma = find_vma(current->mm, addr);
976
977 if (vma == NULL || addr < vma->vm_start ||
978 !(vma->vm_flags & VM_PFNMAP)) {
979 up_read(&current->mm->mmap_sem);
980 return_fault_page:
981 get_page(fault_page);
982 return page_to_pfn(fault_page);
983 }
984
985 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
986 up_read(&current->mm->mmap_sem);
987 BUG_ON(!kvm_is_mmio_pfn(pfn));
988 } else
989 pfn = page_to_pfn(page[0]);
990
991 return pfn;
992 }
993
994 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
995 {
996 return hva_to_pfn(kvm, addr, true);
997 }
998 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
999
1000 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic)
1001 {
1002 unsigned long addr;
1003
1004 addr = gfn_to_hva(kvm, gfn);
1005 if (kvm_is_error_hva(addr)) {
1006 get_page(bad_page);
1007 return page_to_pfn(bad_page);
1008 }
1009
1010 return hva_to_pfn(kvm, addr, atomic);
1011 }
1012
1013 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1014 {
1015 return __gfn_to_pfn(kvm, gfn, true);
1016 }
1017 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1018
1019 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1020 {
1021 return __gfn_to_pfn(kvm, gfn, false);
1022 }
1023 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1024
1025 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1026 struct kvm_memory_slot *slot, gfn_t gfn)
1027 {
1028 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1029 return hva_to_pfn(kvm, addr, false);
1030 }
1031
1032 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1033 int nr_pages)
1034 {
1035 unsigned long addr;
1036 gfn_t entry;
1037
1038 addr = gfn_to_hva_many(kvm, gfn, &entry);
1039 if (kvm_is_error_hva(addr))
1040 return -1;
1041
1042 if (entry < nr_pages)
1043 return 0;
1044
1045 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1046 }
1047 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1048
1049 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1050 {
1051 pfn_t pfn;
1052
1053 pfn = gfn_to_pfn(kvm, gfn);
1054 if (!kvm_is_mmio_pfn(pfn))
1055 return pfn_to_page(pfn);
1056
1057 WARN_ON(kvm_is_mmio_pfn(pfn));
1058
1059 get_page(bad_page);
1060 return bad_page;
1061 }
1062
1063 EXPORT_SYMBOL_GPL(gfn_to_page);
1064
1065 void kvm_release_page_clean(struct page *page)
1066 {
1067 kvm_release_pfn_clean(page_to_pfn(page));
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1070
1071 void kvm_release_pfn_clean(pfn_t pfn)
1072 {
1073 if (!kvm_is_mmio_pfn(pfn))
1074 put_page(pfn_to_page(pfn));
1075 }
1076 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1077
1078 void kvm_release_page_dirty(struct page *page)
1079 {
1080 kvm_release_pfn_dirty(page_to_pfn(page));
1081 }
1082 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1083
1084 void kvm_release_pfn_dirty(pfn_t pfn)
1085 {
1086 kvm_set_pfn_dirty(pfn);
1087 kvm_release_pfn_clean(pfn);
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1090
1091 void kvm_set_page_dirty(struct page *page)
1092 {
1093 kvm_set_pfn_dirty(page_to_pfn(page));
1094 }
1095 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1096
1097 void kvm_set_pfn_dirty(pfn_t pfn)
1098 {
1099 if (!kvm_is_mmio_pfn(pfn)) {
1100 struct page *page = pfn_to_page(pfn);
1101 if (!PageReserved(page))
1102 SetPageDirty(page);
1103 }
1104 }
1105 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1106
1107 void kvm_set_pfn_accessed(pfn_t pfn)
1108 {
1109 if (!kvm_is_mmio_pfn(pfn))
1110 mark_page_accessed(pfn_to_page(pfn));
1111 }
1112 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1113
1114 void kvm_get_pfn(pfn_t pfn)
1115 {
1116 if (!kvm_is_mmio_pfn(pfn))
1117 get_page(pfn_to_page(pfn));
1118 }
1119 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1120
1121 static int next_segment(unsigned long len, int offset)
1122 {
1123 if (len > PAGE_SIZE - offset)
1124 return PAGE_SIZE - offset;
1125 else
1126 return len;
1127 }
1128
1129 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1130 int len)
1131 {
1132 int r;
1133 unsigned long addr;
1134
1135 addr = gfn_to_hva(kvm, gfn);
1136 if (kvm_is_error_hva(addr))
1137 return -EFAULT;
1138 r = copy_from_user(data, (void __user *)addr + offset, len);
1139 if (r)
1140 return -EFAULT;
1141 return 0;
1142 }
1143 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1144
1145 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1146 {
1147 gfn_t gfn = gpa >> PAGE_SHIFT;
1148 int seg;
1149 int offset = offset_in_page(gpa);
1150 int ret;
1151
1152 while ((seg = next_segment(len, offset)) != 0) {
1153 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1154 if (ret < 0)
1155 return ret;
1156 offset = 0;
1157 len -= seg;
1158 data += seg;
1159 ++gfn;
1160 }
1161 return 0;
1162 }
1163 EXPORT_SYMBOL_GPL(kvm_read_guest);
1164
1165 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1166 unsigned long len)
1167 {
1168 int r;
1169 unsigned long addr;
1170 gfn_t gfn = gpa >> PAGE_SHIFT;
1171 int offset = offset_in_page(gpa);
1172
1173 addr = gfn_to_hva(kvm, gfn);
1174 if (kvm_is_error_hva(addr))
1175 return -EFAULT;
1176 pagefault_disable();
1177 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1178 pagefault_enable();
1179 if (r)
1180 return -EFAULT;
1181 return 0;
1182 }
1183 EXPORT_SYMBOL(kvm_read_guest_atomic);
1184
1185 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1186 int offset, int len)
1187 {
1188 int r;
1189 unsigned long addr;
1190
1191 addr = gfn_to_hva(kvm, gfn);
1192 if (kvm_is_error_hva(addr))
1193 return -EFAULT;
1194 r = copy_to_user((void __user *)addr + offset, data, len);
1195 if (r)
1196 return -EFAULT;
1197 mark_page_dirty(kvm, gfn);
1198 return 0;
1199 }
1200 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1201
1202 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1203 unsigned long len)
1204 {
1205 gfn_t gfn = gpa >> PAGE_SHIFT;
1206 int seg;
1207 int offset = offset_in_page(gpa);
1208 int ret;
1209
1210 while ((seg = next_segment(len, offset)) != 0) {
1211 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1212 if (ret < 0)
1213 return ret;
1214 offset = 0;
1215 len -= seg;
1216 data += seg;
1217 ++gfn;
1218 }
1219 return 0;
1220 }
1221
1222 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1223 {
1224 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1225 }
1226 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1227
1228 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1229 {
1230 gfn_t gfn = gpa >> PAGE_SHIFT;
1231 int seg;
1232 int offset = offset_in_page(gpa);
1233 int ret;
1234
1235 while ((seg = next_segment(len, offset)) != 0) {
1236 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1237 if (ret < 0)
1238 return ret;
1239 offset = 0;
1240 len -= seg;
1241 ++gfn;
1242 }
1243 return 0;
1244 }
1245 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1246
1247 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1248 {
1249 struct kvm_memory_slot *memslot;
1250
1251 memslot = gfn_to_memslot(kvm, gfn);
1252 if (memslot && memslot->dirty_bitmap) {
1253 unsigned long rel_gfn = gfn - memslot->base_gfn;
1254
1255 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1256 }
1257 }
1258
1259 /*
1260 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1261 */
1262 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1263 {
1264 DEFINE_WAIT(wait);
1265
1266 for (;;) {
1267 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1268
1269 if (kvm_arch_vcpu_runnable(vcpu)) {
1270 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1271 break;
1272 }
1273 if (kvm_cpu_has_pending_timer(vcpu))
1274 break;
1275 if (signal_pending(current))
1276 break;
1277
1278 schedule();
1279 }
1280
1281 finish_wait(&vcpu->wq, &wait);
1282 }
1283
1284 void kvm_resched(struct kvm_vcpu *vcpu)
1285 {
1286 if (!need_resched())
1287 return;
1288 cond_resched();
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_resched);
1291
1292 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1293 {
1294 ktime_t expires;
1295 DEFINE_WAIT(wait);
1296
1297 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1298
1299 /* Sleep for 100 us, and hope lock-holder got scheduled */
1300 expires = ktime_add_ns(ktime_get(), 100000UL);
1301 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1302
1303 finish_wait(&vcpu->wq, &wait);
1304 }
1305 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1306
1307 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1308 {
1309 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1310 struct page *page;
1311
1312 if (vmf->pgoff == 0)
1313 page = virt_to_page(vcpu->run);
1314 #ifdef CONFIG_X86
1315 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1316 page = virt_to_page(vcpu->arch.pio_data);
1317 #endif
1318 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1319 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1320 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1321 #endif
1322 else
1323 return VM_FAULT_SIGBUS;
1324 get_page(page);
1325 vmf->page = page;
1326 return 0;
1327 }
1328
1329 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1330 .fault = kvm_vcpu_fault,
1331 };
1332
1333 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1334 {
1335 vma->vm_ops = &kvm_vcpu_vm_ops;
1336 return 0;
1337 }
1338
1339 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1340 {
1341 struct kvm_vcpu *vcpu = filp->private_data;
1342
1343 kvm_put_kvm(vcpu->kvm);
1344 return 0;
1345 }
1346
1347 static struct file_operations kvm_vcpu_fops = {
1348 .release = kvm_vcpu_release,
1349 .unlocked_ioctl = kvm_vcpu_ioctl,
1350 .compat_ioctl = kvm_vcpu_ioctl,
1351 .mmap = kvm_vcpu_mmap,
1352 .llseek = noop_llseek,
1353 };
1354
1355 /*
1356 * Allocates an inode for the vcpu.
1357 */
1358 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1359 {
1360 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1361 }
1362
1363 /*
1364 * Creates some virtual cpus. Good luck creating more than one.
1365 */
1366 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1367 {
1368 int r;
1369 struct kvm_vcpu *vcpu, *v;
1370
1371 vcpu = kvm_arch_vcpu_create(kvm, id);
1372 if (IS_ERR(vcpu))
1373 return PTR_ERR(vcpu);
1374
1375 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1376
1377 r = kvm_arch_vcpu_setup(vcpu);
1378 if (r)
1379 return r;
1380
1381 mutex_lock(&kvm->lock);
1382 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1383 r = -EINVAL;
1384 goto vcpu_destroy;
1385 }
1386
1387 kvm_for_each_vcpu(r, v, kvm)
1388 if (v->vcpu_id == id) {
1389 r = -EEXIST;
1390 goto vcpu_destroy;
1391 }
1392
1393 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1394
1395 /* Now it's all set up, let userspace reach it */
1396 kvm_get_kvm(kvm);
1397 r = create_vcpu_fd(vcpu);
1398 if (r < 0) {
1399 kvm_put_kvm(kvm);
1400 goto vcpu_destroy;
1401 }
1402
1403 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1404 smp_wmb();
1405 atomic_inc(&kvm->online_vcpus);
1406
1407 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1408 if (kvm->bsp_vcpu_id == id)
1409 kvm->bsp_vcpu = vcpu;
1410 #endif
1411 mutex_unlock(&kvm->lock);
1412 return r;
1413
1414 vcpu_destroy:
1415 mutex_unlock(&kvm->lock);
1416 kvm_arch_vcpu_destroy(vcpu);
1417 return r;
1418 }
1419
1420 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1421 {
1422 if (sigset) {
1423 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1424 vcpu->sigset_active = 1;
1425 vcpu->sigset = *sigset;
1426 } else
1427 vcpu->sigset_active = 0;
1428 return 0;
1429 }
1430
1431 static long kvm_vcpu_ioctl(struct file *filp,
1432 unsigned int ioctl, unsigned long arg)
1433 {
1434 struct kvm_vcpu *vcpu = filp->private_data;
1435 void __user *argp = (void __user *)arg;
1436 int r;
1437 struct kvm_fpu *fpu = NULL;
1438 struct kvm_sregs *kvm_sregs = NULL;
1439
1440 if (vcpu->kvm->mm != current->mm)
1441 return -EIO;
1442
1443 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1444 /*
1445 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1446 * so vcpu_load() would break it.
1447 */
1448 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1449 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1450 #endif
1451
1452
1453 vcpu_load(vcpu);
1454 switch (ioctl) {
1455 case KVM_RUN:
1456 r = -EINVAL;
1457 if (arg)
1458 goto out;
1459 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1460 break;
1461 case KVM_GET_REGS: {
1462 struct kvm_regs *kvm_regs;
1463
1464 r = -ENOMEM;
1465 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1466 if (!kvm_regs)
1467 goto out;
1468 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1469 if (r)
1470 goto out_free1;
1471 r = -EFAULT;
1472 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1473 goto out_free1;
1474 r = 0;
1475 out_free1:
1476 kfree(kvm_regs);
1477 break;
1478 }
1479 case KVM_SET_REGS: {
1480 struct kvm_regs *kvm_regs;
1481
1482 r = -ENOMEM;
1483 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1484 if (!kvm_regs)
1485 goto out;
1486 r = -EFAULT;
1487 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1488 goto out_free2;
1489 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1490 if (r)
1491 goto out_free2;
1492 r = 0;
1493 out_free2:
1494 kfree(kvm_regs);
1495 break;
1496 }
1497 case KVM_GET_SREGS: {
1498 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1499 r = -ENOMEM;
1500 if (!kvm_sregs)
1501 goto out;
1502 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1503 if (r)
1504 goto out;
1505 r = -EFAULT;
1506 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1507 goto out;
1508 r = 0;
1509 break;
1510 }
1511 case KVM_SET_SREGS: {
1512 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1513 r = -ENOMEM;
1514 if (!kvm_sregs)
1515 goto out;
1516 r = -EFAULT;
1517 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1518 goto out;
1519 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1520 if (r)
1521 goto out;
1522 r = 0;
1523 break;
1524 }
1525 case KVM_GET_MP_STATE: {
1526 struct kvm_mp_state mp_state;
1527
1528 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1529 if (r)
1530 goto out;
1531 r = -EFAULT;
1532 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1533 goto out;
1534 r = 0;
1535 break;
1536 }
1537 case KVM_SET_MP_STATE: {
1538 struct kvm_mp_state mp_state;
1539
1540 r = -EFAULT;
1541 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1542 goto out;
1543 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1544 if (r)
1545 goto out;
1546 r = 0;
1547 break;
1548 }
1549 case KVM_TRANSLATE: {
1550 struct kvm_translation tr;
1551
1552 r = -EFAULT;
1553 if (copy_from_user(&tr, argp, sizeof tr))
1554 goto out;
1555 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1556 if (r)
1557 goto out;
1558 r = -EFAULT;
1559 if (copy_to_user(argp, &tr, sizeof tr))
1560 goto out;
1561 r = 0;
1562 break;
1563 }
1564 case KVM_SET_GUEST_DEBUG: {
1565 struct kvm_guest_debug dbg;
1566
1567 r = -EFAULT;
1568 if (copy_from_user(&dbg, argp, sizeof dbg))
1569 goto out;
1570 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1571 if (r)
1572 goto out;
1573 r = 0;
1574 break;
1575 }
1576 case KVM_SET_SIGNAL_MASK: {
1577 struct kvm_signal_mask __user *sigmask_arg = argp;
1578 struct kvm_signal_mask kvm_sigmask;
1579 sigset_t sigset, *p;
1580
1581 p = NULL;
1582 if (argp) {
1583 r = -EFAULT;
1584 if (copy_from_user(&kvm_sigmask, argp,
1585 sizeof kvm_sigmask))
1586 goto out;
1587 r = -EINVAL;
1588 if (kvm_sigmask.len != sizeof sigset)
1589 goto out;
1590 r = -EFAULT;
1591 if (copy_from_user(&sigset, sigmask_arg->sigset,
1592 sizeof sigset))
1593 goto out;
1594 p = &sigset;
1595 }
1596 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1597 break;
1598 }
1599 case KVM_GET_FPU: {
1600 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1601 r = -ENOMEM;
1602 if (!fpu)
1603 goto out;
1604 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1605 if (r)
1606 goto out;
1607 r = -EFAULT;
1608 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1609 goto out;
1610 r = 0;
1611 break;
1612 }
1613 case KVM_SET_FPU: {
1614 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1615 r = -ENOMEM;
1616 if (!fpu)
1617 goto out;
1618 r = -EFAULT;
1619 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1620 goto out;
1621 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1622 if (r)
1623 goto out;
1624 r = 0;
1625 break;
1626 }
1627 default:
1628 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1629 }
1630 out:
1631 vcpu_put(vcpu);
1632 kfree(fpu);
1633 kfree(kvm_sregs);
1634 return r;
1635 }
1636
1637 static long kvm_vm_ioctl(struct file *filp,
1638 unsigned int ioctl, unsigned long arg)
1639 {
1640 struct kvm *kvm = filp->private_data;
1641 void __user *argp = (void __user *)arg;
1642 int r;
1643
1644 if (kvm->mm != current->mm)
1645 return -EIO;
1646 switch (ioctl) {
1647 case KVM_CREATE_VCPU:
1648 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1649 if (r < 0)
1650 goto out;
1651 break;
1652 case KVM_SET_USER_MEMORY_REGION: {
1653 struct kvm_userspace_memory_region kvm_userspace_mem;
1654
1655 r = -EFAULT;
1656 if (copy_from_user(&kvm_userspace_mem, argp,
1657 sizeof kvm_userspace_mem))
1658 goto out;
1659
1660 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1661 if (r)
1662 goto out;
1663 break;
1664 }
1665 case KVM_GET_DIRTY_LOG: {
1666 struct kvm_dirty_log log;
1667
1668 r = -EFAULT;
1669 if (copy_from_user(&log, argp, sizeof log))
1670 goto out;
1671 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1672 if (r)
1673 goto out;
1674 break;
1675 }
1676 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1677 case KVM_REGISTER_COALESCED_MMIO: {
1678 struct kvm_coalesced_mmio_zone zone;
1679 r = -EFAULT;
1680 if (copy_from_user(&zone, argp, sizeof zone))
1681 goto out;
1682 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1683 if (r)
1684 goto out;
1685 r = 0;
1686 break;
1687 }
1688 case KVM_UNREGISTER_COALESCED_MMIO: {
1689 struct kvm_coalesced_mmio_zone zone;
1690 r = -EFAULT;
1691 if (copy_from_user(&zone, argp, sizeof zone))
1692 goto out;
1693 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1694 if (r)
1695 goto out;
1696 r = 0;
1697 break;
1698 }
1699 #endif
1700 case KVM_IRQFD: {
1701 struct kvm_irqfd data;
1702
1703 r = -EFAULT;
1704 if (copy_from_user(&data, argp, sizeof data))
1705 goto out;
1706 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1707 break;
1708 }
1709 case KVM_IOEVENTFD: {
1710 struct kvm_ioeventfd data;
1711
1712 r = -EFAULT;
1713 if (copy_from_user(&data, argp, sizeof data))
1714 goto out;
1715 r = kvm_ioeventfd(kvm, &data);
1716 break;
1717 }
1718 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1719 case KVM_SET_BOOT_CPU_ID:
1720 r = 0;
1721 mutex_lock(&kvm->lock);
1722 if (atomic_read(&kvm->online_vcpus) != 0)
1723 r = -EBUSY;
1724 else
1725 kvm->bsp_vcpu_id = arg;
1726 mutex_unlock(&kvm->lock);
1727 break;
1728 #endif
1729 default:
1730 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1731 if (r == -ENOTTY)
1732 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1733 }
1734 out:
1735 return r;
1736 }
1737
1738 #ifdef CONFIG_COMPAT
1739 struct compat_kvm_dirty_log {
1740 __u32 slot;
1741 __u32 padding1;
1742 union {
1743 compat_uptr_t dirty_bitmap; /* one bit per page */
1744 __u64 padding2;
1745 };
1746 };
1747
1748 static long kvm_vm_compat_ioctl(struct file *filp,
1749 unsigned int ioctl, unsigned long arg)
1750 {
1751 struct kvm *kvm = filp->private_data;
1752 int r;
1753
1754 if (kvm->mm != current->mm)
1755 return -EIO;
1756 switch (ioctl) {
1757 case KVM_GET_DIRTY_LOG: {
1758 struct compat_kvm_dirty_log compat_log;
1759 struct kvm_dirty_log log;
1760
1761 r = -EFAULT;
1762 if (copy_from_user(&compat_log, (void __user *)arg,
1763 sizeof(compat_log)))
1764 goto out;
1765 log.slot = compat_log.slot;
1766 log.padding1 = compat_log.padding1;
1767 log.padding2 = compat_log.padding2;
1768 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1769
1770 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1771 if (r)
1772 goto out;
1773 break;
1774 }
1775 default:
1776 r = kvm_vm_ioctl(filp, ioctl, arg);
1777 }
1778
1779 out:
1780 return r;
1781 }
1782 #endif
1783
1784 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1785 {
1786 struct page *page[1];
1787 unsigned long addr;
1788 int npages;
1789 gfn_t gfn = vmf->pgoff;
1790 struct kvm *kvm = vma->vm_file->private_data;
1791
1792 addr = gfn_to_hva(kvm, gfn);
1793 if (kvm_is_error_hva(addr))
1794 return VM_FAULT_SIGBUS;
1795
1796 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1797 NULL);
1798 if (unlikely(npages != 1))
1799 return VM_FAULT_SIGBUS;
1800
1801 vmf->page = page[0];
1802 return 0;
1803 }
1804
1805 static const struct vm_operations_struct kvm_vm_vm_ops = {
1806 .fault = kvm_vm_fault,
1807 };
1808
1809 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1810 {
1811 vma->vm_ops = &kvm_vm_vm_ops;
1812 return 0;
1813 }
1814
1815 static struct file_operations kvm_vm_fops = {
1816 .release = kvm_vm_release,
1817 .unlocked_ioctl = kvm_vm_ioctl,
1818 #ifdef CONFIG_COMPAT
1819 .compat_ioctl = kvm_vm_compat_ioctl,
1820 #endif
1821 .mmap = kvm_vm_mmap,
1822 .llseek = noop_llseek,
1823 };
1824
1825 static int kvm_dev_ioctl_create_vm(void)
1826 {
1827 int fd, r;
1828 struct kvm *kvm;
1829
1830 kvm = kvm_create_vm();
1831 if (IS_ERR(kvm))
1832 return PTR_ERR(kvm);
1833 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1834 r = kvm_coalesced_mmio_init(kvm);
1835 if (r < 0) {
1836 kvm_put_kvm(kvm);
1837 return r;
1838 }
1839 #endif
1840 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1841 if (fd < 0)
1842 kvm_put_kvm(kvm);
1843
1844 return fd;
1845 }
1846
1847 static long kvm_dev_ioctl_check_extension_generic(long arg)
1848 {
1849 switch (arg) {
1850 case KVM_CAP_USER_MEMORY:
1851 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1852 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1853 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1854 case KVM_CAP_SET_BOOT_CPU_ID:
1855 #endif
1856 case KVM_CAP_INTERNAL_ERROR_DATA:
1857 return 1;
1858 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1859 case KVM_CAP_IRQ_ROUTING:
1860 return KVM_MAX_IRQ_ROUTES;
1861 #endif
1862 default:
1863 break;
1864 }
1865 return kvm_dev_ioctl_check_extension(arg);
1866 }
1867
1868 static long kvm_dev_ioctl(struct file *filp,
1869 unsigned int ioctl, unsigned long arg)
1870 {
1871 long r = -EINVAL;
1872
1873 switch (ioctl) {
1874 case KVM_GET_API_VERSION:
1875 r = -EINVAL;
1876 if (arg)
1877 goto out;
1878 r = KVM_API_VERSION;
1879 break;
1880 case KVM_CREATE_VM:
1881 r = -EINVAL;
1882 if (arg)
1883 goto out;
1884 r = kvm_dev_ioctl_create_vm();
1885 break;
1886 case KVM_CHECK_EXTENSION:
1887 r = kvm_dev_ioctl_check_extension_generic(arg);
1888 break;
1889 case KVM_GET_VCPU_MMAP_SIZE:
1890 r = -EINVAL;
1891 if (arg)
1892 goto out;
1893 r = PAGE_SIZE; /* struct kvm_run */
1894 #ifdef CONFIG_X86
1895 r += PAGE_SIZE; /* pio data page */
1896 #endif
1897 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1898 r += PAGE_SIZE; /* coalesced mmio ring page */
1899 #endif
1900 break;
1901 case KVM_TRACE_ENABLE:
1902 case KVM_TRACE_PAUSE:
1903 case KVM_TRACE_DISABLE:
1904 r = -EOPNOTSUPP;
1905 break;
1906 default:
1907 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1908 }
1909 out:
1910 return r;
1911 }
1912
1913 static struct file_operations kvm_chardev_ops = {
1914 .unlocked_ioctl = kvm_dev_ioctl,
1915 .compat_ioctl = kvm_dev_ioctl,
1916 .llseek = noop_llseek,
1917 };
1918
1919 static struct miscdevice kvm_dev = {
1920 KVM_MINOR,
1921 "kvm",
1922 &kvm_chardev_ops,
1923 };
1924
1925 static void hardware_enable(void *junk)
1926 {
1927 int cpu = raw_smp_processor_id();
1928 int r;
1929
1930 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1931 return;
1932
1933 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1934
1935 r = kvm_arch_hardware_enable(NULL);
1936
1937 if (r) {
1938 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1939 atomic_inc(&hardware_enable_failed);
1940 printk(KERN_INFO "kvm: enabling virtualization on "
1941 "CPU%d failed\n", cpu);
1942 }
1943 }
1944
1945 static void hardware_disable(void *junk)
1946 {
1947 int cpu = raw_smp_processor_id();
1948
1949 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1950 return;
1951 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1952 kvm_arch_hardware_disable(NULL);
1953 }
1954
1955 static void hardware_disable_all_nolock(void)
1956 {
1957 BUG_ON(!kvm_usage_count);
1958
1959 kvm_usage_count--;
1960 if (!kvm_usage_count)
1961 on_each_cpu(hardware_disable, NULL, 1);
1962 }
1963
1964 static void hardware_disable_all(void)
1965 {
1966 spin_lock(&kvm_lock);
1967 hardware_disable_all_nolock();
1968 spin_unlock(&kvm_lock);
1969 }
1970
1971 static int hardware_enable_all(void)
1972 {
1973 int r = 0;
1974
1975 spin_lock(&kvm_lock);
1976
1977 kvm_usage_count++;
1978 if (kvm_usage_count == 1) {
1979 atomic_set(&hardware_enable_failed, 0);
1980 on_each_cpu(hardware_enable, NULL, 1);
1981
1982 if (atomic_read(&hardware_enable_failed)) {
1983 hardware_disable_all_nolock();
1984 r = -EBUSY;
1985 }
1986 }
1987
1988 spin_unlock(&kvm_lock);
1989
1990 return r;
1991 }
1992
1993 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1994 void *v)
1995 {
1996 int cpu = (long)v;
1997
1998 if (!kvm_usage_count)
1999 return NOTIFY_OK;
2000
2001 val &= ~CPU_TASKS_FROZEN;
2002 switch (val) {
2003 case CPU_DYING:
2004 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2005 cpu);
2006 hardware_disable(NULL);
2007 break;
2008 case CPU_STARTING:
2009 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2010 cpu);
2011 spin_lock(&kvm_lock);
2012 hardware_enable(NULL);
2013 spin_unlock(&kvm_lock);
2014 break;
2015 }
2016 return NOTIFY_OK;
2017 }
2018
2019
2020 asmlinkage void kvm_handle_fault_on_reboot(void)
2021 {
2022 if (kvm_rebooting) {
2023 /* spin while reset goes on */
2024 local_irq_enable();
2025 while (true)
2026 cpu_relax();
2027 }
2028 /* Fault while not rebooting. We want the trace. */
2029 BUG();
2030 }
2031 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2032
2033 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2034 void *v)
2035 {
2036 /*
2037 * Some (well, at least mine) BIOSes hang on reboot if
2038 * in vmx root mode.
2039 *
2040 * And Intel TXT required VMX off for all cpu when system shutdown.
2041 */
2042 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2043 kvm_rebooting = true;
2044 on_each_cpu(hardware_disable, NULL, 1);
2045 return NOTIFY_OK;
2046 }
2047
2048 static struct notifier_block kvm_reboot_notifier = {
2049 .notifier_call = kvm_reboot,
2050 .priority = 0,
2051 };
2052
2053 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2054 {
2055 int i;
2056
2057 for (i = 0; i < bus->dev_count; i++) {
2058 struct kvm_io_device *pos = bus->devs[i];
2059
2060 kvm_iodevice_destructor(pos);
2061 }
2062 kfree(bus);
2063 }
2064
2065 /* kvm_io_bus_write - called under kvm->slots_lock */
2066 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2067 int len, const void *val)
2068 {
2069 int i;
2070 struct kvm_io_bus *bus;
2071
2072 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2073 for (i = 0; i < bus->dev_count; i++)
2074 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2075 return 0;
2076 return -EOPNOTSUPP;
2077 }
2078
2079 /* kvm_io_bus_read - called under kvm->slots_lock */
2080 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2081 int len, void *val)
2082 {
2083 int i;
2084 struct kvm_io_bus *bus;
2085
2086 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2087 for (i = 0; i < bus->dev_count; i++)
2088 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2089 return 0;
2090 return -EOPNOTSUPP;
2091 }
2092
2093 /* Caller must hold slots_lock. */
2094 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2095 struct kvm_io_device *dev)
2096 {
2097 struct kvm_io_bus *new_bus, *bus;
2098
2099 bus = kvm->buses[bus_idx];
2100 if (bus->dev_count > NR_IOBUS_DEVS-1)
2101 return -ENOSPC;
2102
2103 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2104 if (!new_bus)
2105 return -ENOMEM;
2106 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2107 new_bus->devs[new_bus->dev_count++] = dev;
2108 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2109 synchronize_srcu_expedited(&kvm->srcu);
2110 kfree(bus);
2111
2112 return 0;
2113 }
2114
2115 /* Caller must hold slots_lock. */
2116 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2117 struct kvm_io_device *dev)
2118 {
2119 int i, r;
2120 struct kvm_io_bus *new_bus, *bus;
2121
2122 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2123 if (!new_bus)
2124 return -ENOMEM;
2125
2126 bus = kvm->buses[bus_idx];
2127 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2128
2129 r = -ENOENT;
2130 for (i = 0; i < new_bus->dev_count; i++)
2131 if (new_bus->devs[i] == dev) {
2132 r = 0;
2133 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2134 break;
2135 }
2136
2137 if (r) {
2138 kfree(new_bus);
2139 return r;
2140 }
2141
2142 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2143 synchronize_srcu_expedited(&kvm->srcu);
2144 kfree(bus);
2145 return r;
2146 }
2147
2148 static struct notifier_block kvm_cpu_notifier = {
2149 .notifier_call = kvm_cpu_hotplug,
2150 };
2151
2152 static int vm_stat_get(void *_offset, u64 *val)
2153 {
2154 unsigned offset = (long)_offset;
2155 struct kvm *kvm;
2156
2157 *val = 0;
2158 spin_lock(&kvm_lock);
2159 list_for_each_entry(kvm, &vm_list, vm_list)
2160 *val += *(u32 *)((void *)kvm + offset);
2161 spin_unlock(&kvm_lock);
2162 return 0;
2163 }
2164
2165 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2166
2167 static int vcpu_stat_get(void *_offset, u64 *val)
2168 {
2169 unsigned offset = (long)_offset;
2170 struct kvm *kvm;
2171 struct kvm_vcpu *vcpu;
2172 int i;
2173
2174 *val = 0;
2175 spin_lock(&kvm_lock);
2176 list_for_each_entry(kvm, &vm_list, vm_list)
2177 kvm_for_each_vcpu(i, vcpu, kvm)
2178 *val += *(u32 *)((void *)vcpu + offset);
2179
2180 spin_unlock(&kvm_lock);
2181 return 0;
2182 }
2183
2184 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2185
2186 static const struct file_operations *stat_fops[] = {
2187 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2188 [KVM_STAT_VM] = &vm_stat_fops,
2189 };
2190
2191 static void kvm_init_debug(void)
2192 {
2193 struct kvm_stats_debugfs_item *p;
2194
2195 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2196 for (p = debugfs_entries; p->name; ++p)
2197 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2198 (void *)(long)p->offset,
2199 stat_fops[p->kind]);
2200 }
2201
2202 static void kvm_exit_debug(void)
2203 {
2204 struct kvm_stats_debugfs_item *p;
2205
2206 for (p = debugfs_entries; p->name; ++p)
2207 debugfs_remove(p->dentry);
2208 debugfs_remove(kvm_debugfs_dir);
2209 }
2210
2211 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2212 {
2213 if (kvm_usage_count)
2214 hardware_disable(NULL);
2215 return 0;
2216 }
2217
2218 static int kvm_resume(struct sys_device *dev)
2219 {
2220 if (kvm_usage_count) {
2221 WARN_ON(spin_is_locked(&kvm_lock));
2222 hardware_enable(NULL);
2223 }
2224 return 0;
2225 }
2226
2227 static struct sysdev_class kvm_sysdev_class = {
2228 .name = "kvm",
2229 .suspend = kvm_suspend,
2230 .resume = kvm_resume,
2231 };
2232
2233 static struct sys_device kvm_sysdev = {
2234 .id = 0,
2235 .cls = &kvm_sysdev_class,
2236 };
2237
2238 struct page *bad_page;
2239 pfn_t bad_pfn;
2240
2241 static inline
2242 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2243 {
2244 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2245 }
2246
2247 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2248 {
2249 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2250
2251 kvm_arch_vcpu_load(vcpu, cpu);
2252 }
2253
2254 static void kvm_sched_out(struct preempt_notifier *pn,
2255 struct task_struct *next)
2256 {
2257 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2258
2259 kvm_arch_vcpu_put(vcpu);
2260 }
2261
2262 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2263 struct module *module)
2264 {
2265 int r;
2266 int cpu;
2267
2268 r = kvm_arch_init(opaque);
2269 if (r)
2270 goto out_fail;
2271
2272 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2273
2274 if (bad_page == NULL) {
2275 r = -ENOMEM;
2276 goto out;
2277 }
2278
2279 bad_pfn = page_to_pfn(bad_page);
2280
2281 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2282
2283 if (hwpoison_page == NULL) {
2284 r = -ENOMEM;
2285 goto out_free_0;
2286 }
2287
2288 hwpoison_pfn = page_to_pfn(hwpoison_page);
2289
2290 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2291
2292 if (fault_page == NULL) {
2293 r = -ENOMEM;
2294 goto out_free_0;
2295 }
2296
2297 fault_pfn = page_to_pfn(fault_page);
2298
2299 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2300 r = -ENOMEM;
2301 goto out_free_0;
2302 }
2303
2304 r = kvm_arch_hardware_setup();
2305 if (r < 0)
2306 goto out_free_0a;
2307
2308 for_each_online_cpu(cpu) {
2309 smp_call_function_single(cpu,
2310 kvm_arch_check_processor_compat,
2311 &r, 1);
2312 if (r < 0)
2313 goto out_free_1;
2314 }
2315
2316 r = register_cpu_notifier(&kvm_cpu_notifier);
2317 if (r)
2318 goto out_free_2;
2319 register_reboot_notifier(&kvm_reboot_notifier);
2320
2321 r = sysdev_class_register(&kvm_sysdev_class);
2322 if (r)
2323 goto out_free_3;
2324
2325 r = sysdev_register(&kvm_sysdev);
2326 if (r)
2327 goto out_free_4;
2328
2329 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2330 if (!vcpu_align)
2331 vcpu_align = __alignof__(struct kvm_vcpu);
2332 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2333 0, NULL);
2334 if (!kvm_vcpu_cache) {
2335 r = -ENOMEM;
2336 goto out_free_5;
2337 }
2338
2339 kvm_chardev_ops.owner = module;
2340 kvm_vm_fops.owner = module;
2341 kvm_vcpu_fops.owner = module;
2342
2343 r = misc_register(&kvm_dev);
2344 if (r) {
2345 printk(KERN_ERR "kvm: misc device register failed\n");
2346 goto out_free;
2347 }
2348
2349 kvm_preempt_ops.sched_in = kvm_sched_in;
2350 kvm_preempt_ops.sched_out = kvm_sched_out;
2351
2352 kvm_init_debug();
2353
2354 return 0;
2355
2356 out_free:
2357 kmem_cache_destroy(kvm_vcpu_cache);
2358 out_free_5:
2359 sysdev_unregister(&kvm_sysdev);
2360 out_free_4:
2361 sysdev_class_unregister(&kvm_sysdev_class);
2362 out_free_3:
2363 unregister_reboot_notifier(&kvm_reboot_notifier);
2364 unregister_cpu_notifier(&kvm_cpu_notifier);
2365 out_free_2:
2366 out_free_1:
2367 kvm_arch_hardware_unsetup();
2368 out_free_0a:
2369 free_cpumask_var(cpus_hardware_enabled);
2370 out_free_0:
2371 if (fault_page)
2372 __free_page(fault_page);
2373 if (hwpoison_page)
2374 __free_page(hwpoison_page);
2375 __free_page(bad_page);
2376 out:
2377 kvm_arch_exit();
2378 out_fail:
2379 return r;
2380 }
2381 EXPORT_SYMBOL_GPL(kvm_init);
2382
2383 void kvm_exit(void)
2384 {
2385 kvm_exit_debug();
2386 misc_deregister(&kvm_dev);
2387 kmem_cache_destroy(kvm_vcpu_cache);
2388 sysdev_unregister(&kvm_sysdev);
2389 sysdev_class_unregister(&kvm_sysdev_class);
2390 unregister_reboot_notifier(&kvm_reboot_notifier);
2391 unregister_cpu_notifier(&kvm_cpu_notifier);
2392 on_each_cpu(hardware_disable, NULL, 1);
2393 kvm_arch_hardware_unsetup();
2394 kvm_arch_exit();
2395 free_cpumask_var(cpus_hardware_enabled);
2396 __free_page(hwpoison_page);
2397 __free_page(bad_page);
2398 }
2399 EXPORT_SYMBOL_GPL(kvm_exit);
Linus' kernel tree