?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
netfilter: Convert uses of __constant_<foo> to <foo>
[linux/fpc-iii.git] / virt / kvm / kvm_main.c
blob03a0381b1cb79a2b09a565d43883bb89c5855950
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 static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
100
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
104
105 bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
107
108 static bool largepages_enabled = true;
109
110 bool kvm_is_mmio_pfn(pfn_t pfn)
111 {
112 if (pfn_valid(pfn))
113 return PageReserved(pfn_to_page(pfn));
114
115 return true;
116 }
117
118 /*
119 * Switches to specified vcpu, until a matching vcpu_put()
120 */
121 int vcpu_load(struct kvm_vcpu *vcpu)
122 {
123 int cpu;
124
125 if (mutex_lock_killable(&vcpu->mutex))
126 return -EINTR;
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
132 synchronize_rcu();
133 put_pid(oldpid);
134 }
135 cpu = get_cpu();
136 preempt_notifier_register(&vcpu->preempt_notifier);
137 kvm_arch_vcpu_load(vcpu, cpu);
138 put_cpu();
139 return 0;
140 }
141
142 void vcpu_put(struct kvm_vcpu *vcpu)
143 {
144 preempt_disable();
145 kvm_arch_vcpu_put(vcpu);
146 preempt_notifier_unregister(&vcpu->preempt_notifier);
147 preempt_enable();
148 mutex_unlock(&vcpu->mutex);
149 }
150
151 static void ack_flush(void *_completed)
152 {
153 }
154
155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
156 {
157 int i, cpu, me;
158 cpumask_var_t cpus;
159 bool called = true;
160 struct kvm_vcpu *vcpu;
161
162 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
163
164 me = get_cpu();
165 kvm_for_each_vcpu(i, vcpu, kvm) {
166 kvm_make_request(req, vcpu);
167 cpu = vcpu->cpu;
168
169 /* Set ->requests bit before we read ->mode */
170 smp_mb();
171
172 if (cpus != NULL && cpu != -1 && cpu != me &&
173 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
174 cpumask_set_cpu(cpu, cpus);
175 }
176 if (unlikely(cpus == NULL))
177 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
178 else if (!cpumask_empty(cpus))
179 smp_call_function_many(cpus, ack_flush, NULL, 1);
180 else
181 called = false;
182 put_cpu();
183 free_cpumask_var(cpus);
184 return called;
185 }
186
187 void kvm_flush_remote_tlbs(struct kvm *kvm)
188 {
189 long dirty_count = kvm->tlbs_dirty;
190
191 smp_mb();
192 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
193 ++kvm->stat.remote_tlb_flush;
194 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
195 }
196 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
197
198 void kvm_reload_remote_mmus(struct kvm *kvm)
199 {
200 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
201 }
202
203 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
204 {
205 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
206 }
207
208 void kvm_make_scan_ioapic_request(struct kvm *kvm)
209 {
210 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
211 }
212
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214 {
215 struct page *page;
216 int r;
217
218 mutex_init(&vcpu->mutex);
219 vcpu->cpu = -1;
220 vcpu->kvm = kvm;
221 vcpu->vcpu_id = id;
222 vcpu->pid = NULL;
223 init_waitqueue_head(&vcpu->wq);
224 kvm_async_pf_vcpu_init(vcpu);
225
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
227 if (!page) {
228 r = -ENOMEM;
229 goto fail;
230 }
231 vcpu->run = page_address(page);
232
233 kvm_vcpu_set_in_spin_loop(vcpu, false);
234 kvm_vcpu_set_dy_eligible(vcpu, false);
235 vcpu->preempted = false;
236
237 r = kvm_arch_vcpu_init(vcpu);
238 if (r < 0)
239 goto fail_free_run;
240 return 0;
241
242 fail_free_run:
243 free_page((unsigned long)vcpu->run);
244 fail:
245 return r;
246 }
247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248
249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
250 {
251 put_pid(vcpu->pid);
252 kvm_arch_vcpu_uninit(vcpu);
253 free_page((unsigned long)vcpu->run);
254 }
255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256
257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259 {
260 return container_of(mn, struct kvm, mmu_notifier);
261 }
262
263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
264 struct mm_struct *mm,
265 unsigned long address)
266 {
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 int need_tlb_flush, idx;
269
270 /*
271 * When ->invalidate_page runs, the linux pte has been zapped
272 * already but the page is still allocated until
273 * ->invalidate_page returns. So if we increase the sequence
274 * here the kvm page fault will notice if the spte can't be
275 * established because the page is going to be freed. If
276 * instead the kvm page fault establishes the spte before
277 * ->invalidate_page runs, kvm_unmap_hva will release it
278 * before returning.
279 *
280 * The sequence increase only need to be seen at spin_unlock
281 * time, and not at spin_lock time.
282 *
283 * Increasing the sequence after the spin_unlock would be
284 * unsafe because the kvm page fault could then establish the
285 * pte after kvm_unmap_hva returned, without noticing the page
286 * is going to be freed.
287 */
288 idx = srcu_read_lock(&kvm->srcu);
289 spin_lock(&kvm->mmu_lock);
290
291 kvm->mmu_notifier_seq++;
292 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
293 /* we've to flush the tlb before the pages can be freed */
294 if (need_tlb_flush)
295 kvm_flush_remote_tlbs(kvm);
296
297 spin_unlock(&kvm->mmu_lock);
298 srcu_read_unlock(&kvm->srcu, idx);
299 }
300
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
304 pte_t pte)
305 {
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
307 int idx;
308
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
314 srcu_read_unlock(&kvm->srcu, idx);
315 }
316
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
319 unsigned long start,
320 unsigned long end)
321 {
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int need_tlb_flush = 0, idx;
324
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
327 /*
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
331 */
332 kvm->mmu_notifier_count++;
333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334 need_tlb_flush |= kvm->tlbs_dirty;
335 /* we've to flush the tlb before the pages can be freed */
336 if (need_tlb_flush)
337 kvm_flush_remote_tlbs(kvm);
338
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
341 }
342
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
345 unsigned long start,
346 unsigned long end)
347 {
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
349
350 spin_lock(&kvm->mmu_lock);
351 /*
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
354 * been freed.
355 */
356 kvm->mmu_notifier_seq++;
357 smp_wmb();
358 /*
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
362 */
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
365
366 BUG_ON(kvm->mmu_notifier_count < 0);
367 }
368
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
371 unsigned long address)
372 {
373 struct kvm *kvm = mmu_notifier_to_kvm(mn);
374 int young, idx;
375
376 idx = srcu_read_lock(&kvm->srcu);
377 spin_lock(&kvm->mmu_lock);
378
379 young = kvm_age_hva(kvm, address);
380 if (young)
381 kvm_flush_remote_tlbs(kvm);
382
383 spin_unlock(&kvm->mmu_lock);
384 srcu_read_unlock(&kvm->srcu, idx);
385
386 return young;
387 }
388
389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
390 struct mm_struct *mm,
391 unsigned long address)
392 {
393 struct kvm *kvm = mmu_notifier_to_kvm(mn);
394 int young, idx;
395
396 idx = srcu_read_lock(&kvm->srcu);
397 spin_lock(&kvm->mmu_lock);
398 young = kvm_test_age_hva(kvm, address);
399 spin_unlock(&kvm->mmu_lock);
400 srcu_read_unlock(&kvm->srcu, idx);
401
402 return young;
403 }
404
405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
406 struct mm_struct *mm)
407 {
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
409 int idx;
410
411 idx = srcu_read_lock(&kvm->srcu);
412 kvm_arch_flush_shadow_all(kvm);
413 srcu_read_unlock(&kvm->srcu, idx);
414 }
415
416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
417 .invalidate_page = kvm_mmu_notifier_invalidate_page,
418 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
419 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
420 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
421 .test_young = kvm_mmu_notifier_test_young,
422 .change_pte = kvm_mmu_notifier_change_pte,
423 .release = kvm_mmu_notifier_release,
424 };
425
426 static int kvm_init_mmu_notifier(struct kvm *kvm)
427 {
428 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
429 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
430 }
431
432 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
433
434 static int kvm_init_mmu_notifier(struct kvm *kvm)
435 {
436 return 0;
437 }
438
439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
440
441 static void kvm_init_memslots_id(struct kvm *kvm)
442 {
443 int i;
444 struct kvm_memslots *slots = kvm->memslots;
445
446 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
447 slots->id_to_index[i] = slots->memslots[i].id = i;
448 }
449
450 static struct kvm *kvm_create_vm(unsigned long type)
451 {
452 int r, i;
453 struct kvm *kvm = kvm_arch_alloc_vm();
454
455 if (!kvm)
456 return ERR_PTR(-ENOMEM);
457
458 r = kvm_arch_init_vm(kvm, type);
459 if (r)
460 goto out_err_nodisable;
461
462 r = hardware_enable_all();
463 if (r)
464 goto out_err_nodisable;
465
466 #ifdef CONFIG_HAVE_KVM_IRQCHIP
467 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
468 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
469 #endif
470
471 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
472
473 r = -ENOMEM;
474 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
475 if (!kvm->memslots)
476 goto out_err_nosrcu;
477 kvm_init_memslots_id(kvm);
478 if (init_srcu_struct(&kvm->srcu))
479 goto out_err_nosrcu;
480 for (i = 0; i < KVM_NR_BUSES; i++) {
481 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
482 GFP_KERNEL);
483 if (!kvm->buses[i])
484 goto out_err;
485 }
486
487 spin_lock_init(&kvm->mmu_lock);
488 kvm->mm = current->mm;
489 atomic_inc(&kvm->mm->mm_count);
490 kvm_eventfd_init(kvm);
491 mutex_init(&kvm->lock);
492 mutex_init(&kvm->irq_lock);
493 mutex_init(&kvm->slots_lock);
494 atomic_set(&kvm->users_count, 1);
495 INIT_LIST_HEAD(&kvm->devices);
496
497 r = kvm_init_mmu_notifier(kvm);
498 if (r)
499 goto out_err;
500
501 spin_lock(&kvm_lock);
502 list_add(&kvm->vm_list, &vm_list);
503 spin_unlock(&kvm_lock);
504
505 return kvm;
506
507 out_err:
508 cleanup_srcu_struct(&kvm->srcu);
509 out_err_nosrcu:
510 hardware_disable_all();
511 out_err_nodisable:
512 for (i = 0; i < KVM_NR_BUSES; i++)
513 kfree(kvm->buses[i]);
514 kfree(kvm->memslots);
515 kvm_arch_free_vm(kvm);
516 return ERR_PTR(r);
517 }
518
519 /*
520 * Avoid using vmalloc for a small buffer.
521 * Should not be used when the size is statically known.
522 */
523 void *kvm_kvzalloc(unsigned long size)
524 {
525 if (size > PAGE_SIZE)
526 return vzalloc(size);
527 else
528 return kzalloc(size, GFP_KERNEL);
529 }
530
531 void kvm_kvfree(const void *addr)
532 {
533 if (is_vmalloc_addr(addr))
534 vfree(addr);
535 else
536 kfree(addr);
537 }
538
539 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
540 {
541 if (!memslot->dirty_bitmap)
542 return;
543
544 kvm_kvfree(memslot->dirty_bitmap);
545 memslot->dirty_bitmap = NULL;
546 }
547
548 /*
549 * Free any memory in @free but not in @dont.
550 */
551 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
552 struct kvm_memory_slot *dont)
553 {
554 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
555 kvm_destroy_dirty_bitmap(free);
556
557 kvm_arch_free_memslot(kvm, free, dont);
558
559 free->npages = 0;
560 }
561
562 static void kvm_free_physmem(struct kvm *kvm)
563 {
564 struct kvm_memslots *slots = kvm->memslots;
565 struct kvm_memory_slot *memslot;
566
567 kvm_for_each_memslot(memslot, slots)
568 kvm_free_physmem_slot(kvm, memslot, NULL);
569
570 kfree(kvm->memslots);
571 }
572
573 static void kvm_destroy_devices(struct kvm *kvm)
574 {
575 struct list_head *node, *tmp;
576
577 list_for_each_safe(node, tmp, &kvm->devices) {
578 struct kvm_device *dev =
579 list_entry(node, struct kvm_device, vm_node);
580
581 list_del(node);
582 dev->ops->destroy(dev);
583 }
584 }
585
586 static void kvm_destroy_vm(struct kvm *kvm)
587 {
588 int i;
589 struct mm_struct *mm = kvm->mm;
590
591 kvm_arch_sync_events(kvm);
592 spin_lock(&kvm_lock);
593 list_del(&kvm->vm_list);
594 spin_unlock(&kvm_lock);
595 kvm_free_irq_routing(kvm);
596 for (i = 0; i < KVM_NR_BUSES; i++)
597 kvm_io_bus_destroy(kvm->buses[i]);
598 kvm_coalesced_mmio_free(kvm);
599 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
600 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
601 #else
602 kvm_arch_flush_shadow_all(kvm);
603 #endif
604 kvm_arch_destroy_vm(kvm);
605 kvm_destroy_devices(kvm);
606 kvm_free_physmem(kvm);
607 cleanup_srcu_struct(&kvm->srcu);
608 kvm_arch_free_vm(kvm);
609 hardware_disable_all();
610 mmdrop(mm);
611 }
612
613 void kvm_get_kvm(struct kvm *kvm)
614 {
615 atomic_inc(&kvm->users_count);
616 }
617 EXPORT_SYMBOL_GPL(kvm_get_kvm);
618
619 void kvm_put_kvm(struct kvm *kvm)
620 {
621 if (atomic_dec_and_test(&kvm->users_count))
622 kvm_destroy_vm(kvm);
623 }
624 EXPORT_SYMBOL_GPL(kvm_put_kvm);
625
626
627 static int kvm_vm_release(struct inode *inode, struct file *filp)
628 {
629 struct kvm *kvm = filp->private_data;
630
631 kvm_irqfd_release(kvm);
632
633 kvm_put_kvm(kvm);
634 return 0;
635 }
636
637 /*
638 * Allocation size is twice as large as the actual dirty bitmap size.
639 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
640 */
641 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
642 {
643 #ifndef CONFIG_S390
644 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
645
646 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
647 if (!memslot->dirty_bitmap)
648 return -ENOMEM;
649
650 #endif /* !CONFIG_S390 */
651 return 0;
652 }
653
654 static int cmp_memslot(const void *slot1, const void *slot2)
655 {
656 struct kvm_memory_slot *s1, *s2;
657
658 s1 = (struct kvm_memory_slot *)slot1;
659 s2 = (struct kvm_memory_slot *)slot2;
660
661 if (s1->npages < s2->npages)
662 return 1;
663 if (s1->npages > s2->npages)
664 return -1;
665
666 return 0;
667 }
668
669 /*
670 * Sort the memslots base on its size, so the larger slots
671 * will get better fit.
672 */
673 static void sort_memslots(struct kvm_memslots *slots)
674 {
675 int i;
676
677 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
678 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
679
680 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
681 slots->id_to_index[slots->memslots[i].id] = i;
682 }
683
684 static void update_memslots(struct kvm_memslots *slots,
685 struct kvm_memory_slot *new,
686 u64 last_generation)
687 {
688 if (new) {
689 int id = new->id;
690 struct kvm_memory_slot *old = id_to_memslot(slots, id);
691 unsigned long npages = old->npages;
692
693 *old = *new;
694 if (new->npages != npages)
695 sort_memslots(slots);
696 }
697
698 slots->generation = last_generation + 1;
699 }
700
701 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
702 {
703 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
704
705 #ifdef KVM_CAP_READONLY_MEM
706 valid_flags |= KVM_MEM_READONLY;
707 #endif
708
709 if (mem->flags & ~valid_flags)
710 return -EINVAL;
711
712 return 0;
713 }
714
715 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
716 struct kvm_memslots *slots, struct kvm_memory_slot *new)
717 {
718 struct kvm_memslots *old_memslots = kvm->memslots;
719
720 update_memslots(slots, new, kvm->memslots->generation);
721 rcu_assign_pointer(kvm->memslots, slots);
722 synchronize_srcu_expedited(&kvm->srcu);
723
724 kvm_arch_memslots_updated(kvm);
725
726 return old_memslots;
727 }
728
729 /*
730 * Allocate some memory and give it an address in the guest physical address
731 * space.
732 *
733 * Discontiguous memory is allowed, mostly for framebuffers.
734 *
735 * Must be called holding mmap_sem for write.
736 */
737 int __kvm_set_memory_region(struct kvm *kvm,
738 struct kvm_userspace_memory_region *mem)
739 {
740 int r;
741 gfn_t base_gfn;
742 unsigned long npages;
743 struct kvm_memory_slot *slot;
744 struct kvm_memory_slot old, new;
745 struct kvm_memslots *slots = NULL, *old_memslots;
746 enum kvm_mr_change change;
747
748 r = check_memory_region_flags(mem);
749 if (r)
750 goto out;
751
752 r = -EINVAL;
753 /* General sanity checks */
754 if (mem->memory_size & (PAGE_SIZE - 1))
755 goto out;
756 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
757 goto out;
758 /* We can read the guest memory with __xxx_user() later on. */
759 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
760 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
761 !access_ok(VERIFY_WRITE,
762 (void __user *)(unsigned long)mem->userspace_addr,
763 mem->memory_size)))
764 goto out;
765 if (mem->slot >= KVM_MEM_SLOTS_NUM)
766 goto out;
767 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
768 goto out;
769
770 slot = id_to_memslot(kvm->memslots, mem->slot);
771 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
772 npages = mem->memory_size >> PAGE_SHIFT;
773
774 r = -EINVAL;
775 if (npages > KVM_MEM_MAX_NR_PAGES)
776 goto out;
777
778 if (!npages)
779 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
780
781 new = old = *slot;
782
783 new.id = mem->slot;
784 new.base_gfn = base_gfn;
785 new.npages = npages;
786 new.flags = mem->flags;
787
788 r = -EINVAL;
789 if (npages) {
790 if (!old.npages)
791 change = KVM_MR_CREATE;
792 else { /* Modify an existing slot. */
793 if ((mem->userspace_addr != old.userspace_addr) ||
794 (npages != old.npages) ||
795 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
796 goto out;
797
798 if (base_gfn != old.base_gfn)
799 change = KVM_MR_MOVE;
800 else if (new.flags != old.flags)
801 change = KVM_MR_FLAGS_ONLY;
802 else { /* Nothing to change. */
803 r = 0;
804 goto out;
805 }
806 }
807 } else if (old.npages) {
808 change = KVM_MR_DELETE;
809 } else /* Modify a non-existent slot: disallowed. */
810 goto out;
811
812 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
813 /* Check for overlaps */
814 r = -EEXIST;
815 kvm_for_each_memslot(slot, kvm->memslots) {
816 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
817 (slot->id == mem->slot))
818 continue;
819 if (!((base_gfn + npages <= slot->base_gfn) ||
820 (base_gfn >= slot->base_gfn + slot->npages)))
821 goto out;
822 }
823 }
824
825 /* Free page dirty bitmap if unneeded */
826 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
827 new.dirty_bitmap = NULL;
828
829 r = -ENOMEM;
830 if (change == KVM_MR_CREATE) {
831 new.userspace_addr = mem->userspace_addr;
832
833 if (kvm_arch_create_memslot(kvm, &new, npages))
834 goto out_free;
835 }
836
837 /* Allocate page dirty bitmap if needed */
838 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
839 if (kvm_create_dirty_bitmap(&new) < 0)
840 goto out_free;
841 }
842
843 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
844 r = -ENOMEM;
845 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
846 GFP_KERNEL);
847 if (!slots)
848 goto out_free;
849 slot = id_to_memslot(slots, mem->slot);
850 slot->flags |= KVM_MEMSLOT_INVALID;
851
852 old_memslots = install_new_memslots(kvm, slots, NULL);
853
854 /* slot was deleted or moved, clear iommu mapping */
855 kvm_iommu_unmap_pages(kvm, &old);
856 /* From this point no new shadow pages pointing to a deleted,
857 * or moved, memslot will be created.
858 *
859 * validation of sp->gfn happens in:
860 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
861 * - kvm_is_visible_gfn (mmu_check_roots)
862 */
863 kvm_arch_flush_shadow_memslot(kvm, slot);
864 slots = old_memslots;
865 }
866
867 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
868 if (r)
869 goto out_slots;
870
871 r = -ENOMEM;
872 /*
873 * We can re-use the old_memslots from above, the only difference
874 * from the currently installed memslots is the invalid flag. This
875 * will get overwritten by update_memslots anyway.
876 */
877 if (!slots) {
878 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
879 GFP_KERNEL);
880 if (!slots)
881 goto out_free;
882 }
883
884 /* actual memory is freed via old in kvm_free_physmem_slot below */
885 if (change == KVM_MR_DELETE) {
886 new.dirty_bitmap = NULL;
887 memset(&new.arch, 0, sizeof(new.arch));
888 }
889
890 old_memslots = install_new_memslots(kvm, slots, &new);
891
892 kvm_arch_commit_memory_region(kvm, mem, &old, change);
893
894 kvm_free_physmem_slot(kvm, &old, &new);
895 kfree(old_memslots);
896
897 /*
898 * IOMMU mapping: New slots need to be mapped. Old slots need to be
899 * un-mapped and re-mapped if their base changes. Since base change
900 * unmapping is handled above with slot deletion, mapping alone is
901 * needed here. Anything else the iommu might care about for existing
902 * slots (size changes, userspace addr changes and read-only flag
903 * changes) is disallowed above, so any other attribute changes getting
904 * here can be skipped.
905 */
906 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
907 r = kvm_iommu_map_pages(kvm, &new);
908 return r;
909 }
910
911 return 0;
912
913 out_slots:
914 kfree(slots);
915 out_free:
916 kvm_free_physmem_slot(kvm, &new, &old);
917 out:
918 return r;
919 }
920 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
921
922 int kvm_set_memory_region(struct kvm *kvm,
923 struct kvm_userspace_memory_region *mem)
924 {
925 int r;
926
927 mutex_lock(&kvm->slots_lock);
928 r = __kvm_set_memory_region(kvm, mem);
929 mutex_unlock(&kvm->slots_lock);
930 return r;
931 }
932 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
933
934 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
935 struct kvm_userspace_memory_region *mem)
936 {
937 if (mem->slot >= KVM_USER_MEM_SLOTS)
938 return -EINVAL;
939 return kvm_set_memory_region(kvm, mem);
940 }
941
942 int kvm_get_dirty_log(struct kvm *kvm,
943 struct kvm_dirty_log *log, int *is_dirty)
944 {
945 struct kvm_memory_slot *memslot;
946 int r, i;
947 unsigned long n;
948 unsigned long any = 0;
949
950 r = -EINVAL;
951 if (log->slot >= KVM_USER_MEM_SLOTS)
952 goto out;
953
954 memslot = id_to_memslot(kvm->memslots, log->slot);
955 r = -ENOENT;
956 if (!memslot->dirty_bitmap)
957 goto out;
958
959 n = kvm_dirty_bitmap_bytes(memslot);
960
961 for (i = 0; !any && i < n/sizeof(long); ++i)
962 any = memslot->dirty_bitmap[i];
963
964 r = -EFAULT;
965 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
966 goto out;
967
968 if (any)
969 *is_dirty = 1;
970
971 r = 0;
972 out:
973 return r;
974 }
975 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
976
977 bool kvm_largepages_enabled(void)
978 {
979 return largepages_enabled;
980 }
981
982 void kvm_disable_largepages(void)
983 {
984 largepages_enabled = false;
985 }
986 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
987
988 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
989 {
990 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
991 }
992 EXPORT_SYMBOL_GPL(gfn_to_memslot);
993
994 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
995 {
996 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
997
998 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
999 memslot->flags & KVM_MEMSLOT_INVALID)
1000 return 0;
1001
1002 return 1;
1003 }
1004 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1005
1006 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1007 {
1008 struct vm_area_struct *vma;
1009 unsigned long addr, size;
1010
1011 size = PAGE_SIZE;
1012
1013 addr = gfn_to_hva(kvm, gfn);
1014 if (kvm_is_error_hva(addr))
1015 return PAGE_SIZE;
1016
1017 down_read(&current->mm->mmap_sem);
1018 vma = find_vma(current->mm, addr);
1019 if (!vma)
1020 goto out;
1021
1022 size = vma_kernel_pagesize(vma);
1023
1024 out:
1025 up_read(&current->mm->mmap_sem);
1026
1027 return size;
1028 }
1029
1030 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1031 {
1032 return slot->flags & KVM_MEM_READONLY;
1033 }
1034
1035 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1036 gfn_t *nr_pages, bool write)
1037 {
1038 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1039 return KVM_HVA_ERR_BAD;
1040
1041 if (memslot_is_readonly(slot) && write)
1042 return KVM_HVA_ERR_RO_BAD;
1043
1044 if (nr_pages)
1045 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1046
1047 return __gfn_to_hva_memslot(slot, gfn);
1048 }
1049
1050 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1051 gfn_t *nr_pages)
1052 {
1053 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1054 }
1055
1056 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1057 gfn_t gfn)
1058 {
1059 return gfn_to_hva_many(slot, gfn, NULL);
1060 }
1061 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1062
1063 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1064 {
1065 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1066 }
1067 EXPORT_SYMBOL_GPL(gfn_to_hva);
1068
1069 /*
1070 * If writable is set to false, the hva returned by this function is only
1071 * allowed to be read.
1072 */
1073 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1074 {
1075 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1076 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1077
1078 if (!kvm_is_error_hva(hva) && writable)
1079 *writable = !memslot_is_readonly(slot);
1080
1081 return hva;
1082 }
1083
1084 static int kvm_read_hva(void *data, void __user *hva, int len)
1085 {
1086 return __copy_from_user(data, hva, len);
1087 }
1088
1089 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1090 {
1091 return __copy_from_user_inatomic(data, hva, len);
1092 }
1093
1094 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1095 unsigned long start, int write, struct page **page)
1096 {
1097 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1098
1099 if (write)
1100 flags |= FOLL_WRITE;
1101
1102 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1103 }
1104
1105 static inline int check_user_page_hwpoison(unsigned long addr)
1106 {
1107 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1108
1109 rc = __get_user_pages(current, current->mm, addr, 1,
1110 flags, NULL, NULL, NULL);
1111 return rc == -EHWPOISON;
1112 }
1113
1114 /*
1115 * The atomic path to get the writable pfn which will be stored in @pfn,
1116 * true indicates success, otherwise false is returned.
1117 */
1118 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1119 bool write_fault, bool *writable, pfn_t *pfn)
1120 {
1121 struct page *page[1];
1122 int npages;
1123
1124 if (!(async || atomic))
1125 return false;
1126
1127 /*
1128 * Fast pin a writable pfn only if it is a write fault request
1129 * or the caller allows to map a writable pfn for a read fault
1130 * request.
1131 */
1132 if (!(write_fault || writable))
1133 return false;
1134
1135 npages = __get_user_pages_fast(addr, 1, 1, page);
1136 if (npages == 1) {
1137 *pfn = page_to_pfn(page[0]);
1138
1139 if (writable)
1140 *writable = true;
1141 return true;
1142 }
1143
1144 return false;
1145 }
1146
1147 /*
1148 * The slow path to get the pfn of the specified host virtual address,
1149 * 1 indicates success, -errno is returned if error is detected.
1150 */
1151 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1152 bool *writable, pfn_t *pfn)
1153 {
1154 struct page *page[1];
1155 int npages = 0;
1156
1157 might_sleep();
1158
1159 if (writable)
1160 *writable = write_fault;
1161
1162 if (async) {
1163 down_read(&current->mm->mmap_sem);
1164 npages = get_user_page_nowait(current, current->mm,
1165 addr, write_fault, page);
1166 up_read(&current->mm->mmap_sem);
1167 } else
1168 npages = get_user_pages_fast(addr, 1, write_fault,
1169 page);
1170 if (npages != 1)
1171 return npages;
1172
1173 /* map read fault as writable if possible */
1174 if (unlikely(!write_fault) && writable) {
1175 struct page *wpage[1];
1176
1177 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1178 if (npages == 1) {
1179 *writable = true;
1180 put_page(page[0]);
1181 page[0] = wpage[0];
1182 }
1183
1184 npages = 1;
1185 }
1186 *pfn = page_to_pfn(page[0]);
1187 return npages;
1188 }
1189
1190 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1191 {
1192 if (unlikely(!(vma->vm_flags & VM_READ)))
1193 return false;
1194
1195 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1196 return false;
1197
1198 return true;
1199 }
1200
1201 /*
1202 * Pin guest page in memory and return its pfn.
1203 * @addr: host virtual address which maps memory to the guest
1204 * @atomic: whether this function can sleep
1205 * @async: whether this function need to wait IO complete if the
1206 * host page is not in the memory
1207 * @write_fault: whether we should get a writable host page
1208 * @writable: whether it allows to map a writable host page for !@write_fault
1209 *
1210 * The function will map a writable host page for these two cases:
1211 * 1): @write_fault = true
1212 * 2): @write_fault = false && @writable, @writable will tell the caller
1213 * whether the mapping is writable.
1214 */
1215 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1216 bool write_fault, bool *writable)
1217 {
1218 struct vm_area_struct *vma;
1219 pfn_t pfn = 0;
1220 int npages;
1221
1222 /* we can do it either atomically or asynchronously, not both */
1223 BUG_ON(atomic && async);
1224
1225 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1226 return pfn;
1227
1228 if (atomic)
1229 return KVM_PFN_ERR_FAULT;
1230
1231 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1232 if (npages == 1)
1233 return pfn;
1234
1235 down_read(&current->mm->mmap_sem);
1236 if (npages == -EHWPOISON ||
1237 (!async && check_user_page_hwpoison(addr))) {
1238 pfn = KVM_PFN_ERR_HWPOISON;
1239 goto exit;
1240 }
1241
1242 vma = find_vma_intersection(current->mm, addr, addr + 1);
1243
1244 if (vma == NULL)
1245 pfn = KVM_PFN_ERR_FAULT;
1246 else if ((vma->vm_flags & VM_PFNMAP)) {
1247 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1248 vma->vm_pgoff;
1249 BUG_ON(!kvm_is_mmio_pfn(pfn));
1250 } else {
1251 if (async && vma_is_valid(vma, write_fault))
1252 *async = true;
1253 pfn = KVM_PFN_ERR_FAULT;
1254 }
1255 exit:
1256 up_read(&current->mm->mmap_sem);
1257 return pfn;
1258 }
1259
1260 static pfn_t
1261 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1262 bool *async, bool write_fault, bool *writable)
1263 {
1264 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1265
1266 if (addr == KVM_HVA_ERR_RO_BAD)
1267 return KVM_PFN_ERR_RO_FAULT;
1268
1269 if (kvm_is_error_hva(addr))
1270 return KVM_PFN_NOSLOT;
1271
1272 /* Do not map writable pfn in the readonly memslot. */
1273 if (writable && memslot_is_readonly(slot)) {
1274 *writable = false;
1275 writable = NULL;
1276 }
1277
1278 return hva_to_pfn(addr, atomic, async, write_fault,
1279 writable);
1280 }
1281
1282 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1283 bool write_fault, bool *writable)
1284 {
1285 struct kvm_memory_slot *slot;
1286
1287 if (async)
1288 *async = false;
1289
1290 slot = gfn_to_memslot(kvm, gfn);
1291
1292 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1293 writable);
1294 }
1295
1296 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1297 {
1298 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1299 }
1300 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1301
1302 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1303 bool write_fault, bool *writable)
1304 {
1305 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1306 }
1307 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1308
1309 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1310 {
1311 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1312 }
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1314
1315 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1316 bool *writable)
1317 {
1318 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1319 }
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1321
1322 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1323 {
1324 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1325 }
1326
1327 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1328 {
1329 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1330 }
1331 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1332
1333 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1334 int nr_pages)
1335 {
1336 unsigned long addr;
1337 gfn_t entry;
1338
1339 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1340 if (kvm_is_error_hva(addr))
1341 return -1;
1342
1343 if (entry < nr_pages)
1344 return 0;
1345
1346 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1347 }
1348 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1349
1350 static struct page *kvm_pfn_to_page(pfn_t pfn)
1351 {
1352 if (is_error_noslot_pfn(pfn))
1353 return KVM_ERR_PTR_BAD_PAGE;
1354
1355 if (kvm_is_mmio_pfn(pfn)) {
1356 WARN_ON(1);
1357 return KVM_ERR_PTR_BAD_PAGE;
1358 }
1359
1360 return pfn_to_page(pfn);
1361 }
1362
1363 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1364 {
1365 pfn_t pfn;
1366
1367 pfn = gfn_to_pfn(kvm, gfn);
1368
1369 return kvm_pfn_to_page(pfn);
1370 }
1371
1372 EXPORT_SYMBOL_GPL(gfn_to_page);
1373
1374 void kvm_release_page_clean(struct page *page)
1375 {
1376 WARN_ON(is_error_page(page));
1377
1378 kvm_release_pfn_clean(page_to_pfn(page));
1379 }
1380 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1381
1382 void kvm_release_pfn_clean(pfn_t pfn)
1383 {
1384 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1385 put_page(pfn_to_page(pfn));
1386 }
1387 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1388
1389 void kvm_release_page_dirty(struct page *page)
1390 {
1391 WARN_ON(is_error_page(page));
1392
1393 kvm_release_pfn_dirty(page_to_pfn(page));
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1396
1397 static void kvm_release_pfn_dirty(pfn_t pfn)
1398 {
1399 kvm_set_pfn_dirty(pfn);
1400 kvm_release_pfn_clean(pfn);
1401 }
1402
1403 void kvm_set_pfn_dirty(pfn_t pfn)
1404 {
1405 if (!kvm_is_mmio_pfn(pfn)) {
1406 struct page *page = pfn_to_page(pfn);
1407 if (!PageReserved(page))
1408 SetPageDirty(page);
1409 }
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1412
1413 void kvm_set_pfn_accessed(pfn_t pfn)
1414 {
1415 if (!kvm_is_mmio_pfn(pfn))
1416 mark_page_accessed(pfn_to_page(pfn));
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1419
1420 void kvm_get_pfn(pfn_t pfn)
1421 {
1422 if (!kvm_is_mmio_pfn(pfn))
1423 get_page(pfn_to_page(pfn));
1424 }
1425 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1426
1427 static int next_segment(unsigned long len, int offset)
1428 {
1429 if (len > PAGE_SIZE - offset)
1430 return PAGE_SIZE - offset;
1431 else
1432 return len;
1433 }
1434
1435 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1436 int len)
1437 {
1438 int r;
1439 unsigned long addr;
1440
1441 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1442 if (kvm_is_error_hva(addr))
1443 return -EFAULT;
1444 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1445 if (r)
1446 return -EFAULT;
1447 return 0;
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1450
1451 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1452 {
1453 gfn_t gfn = gpa >> PAGE_SHIFT;
1454 int seg;
1455 int offset = offset_in_page(gpa);
1456 int ret;
1457
1458 while ((seg = next_segment(len, offset)) != 0) {
1459 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1460 if (ret < 0)
1461 return ret;
1462 offset = 0;
1463 len -= seg;
1464 data += seg;
1465 ++gfn;
1466 }
1467 return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_read_guest);
1470
1471 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1472 unsigned long len)
1473 {
1474 int r;
1475 unsigned long addr;
1476 gfn_t gfn = gpa >> PAGE_SHIFT;
1477 int offset = offset_in_page(gpa);
1478
1479 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1480 if (kvm_is_error_hva(addr))
1481 return -EFAULT;
1482 pagefault_disable();
1483 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1484 pagefault_enable();
1485 if (r)
1486 return -EFAULT;
1487 return 0;
1488 }
1489 EXPORT_SYMBOL(kvm_read_guest_atomic);
1490
1491 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1492 int offset, int len)
1493 {
1494 int r;
1495 unsigned long addr;
1496
1497 addr = gfn_to_hva(kvm, gfn);
1498 if (kvm_is_error_hva(addr))
1499 return -EFAULT;
1500 r = __copy_to_user((void __user *)addr + offset, data, len);
1501 if (r)
1502 return -EFAULT;
1503 mark_page_dirty(kvm, gfn);
1504 return 0;
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1507
1508 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1509 unsigned long len)
1510 {
1511 gfn_t gfn = gpa >> PAGE_SHIFT;
1512 int seg;
1513 int offset = offset_in_page(gpa);
1514 int ret;
1515
1516 while ((seg = next_segment(len, offset)) != 0) {
1517 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1518 if (ret < 0)
1519 return ret;
1520 offset = 0;
1521 len -= seg;
1522 data += seg;
1523 ++gfn;
1524 }
1525 return 0;
1526 }
1527
1528 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1529 gpa_t gpa, unsigned long len)
1530 {
1531 struct kvm_memslots *slots = kvm_memslots(kvm);
1532 int offset = offset_in_page(gpa);
1533 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1534 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1535 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1536 gfn_t nr_pages_avail;
1537
1538 ghc->gpa = gpa;
1539 ghc->generation = slots->generation;
1540 ghc->len = len;
1541 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1542 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1543 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1544 ghc->hva += offset;
1545 } else {
1546 /*
1547 * If the requested region crosses two memslots, we still
1548 * verify that the entire region is valid here.
1549 */
1550 while (start_gfn <= end_gfn) {
1551 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1552 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1553 &nr_pages_avail);
1554 if (kvm_is_error_hva(ghc->hva))
1555 return -EFAULT;
1556 start_gfn += nr_pages_avail;
1557 }
1558 /* Use the slow path for cross page reads and writes. */
1559 ghc->memslot = NULL;
1560 }
1561 return 0;
1562 }
1563 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1564
1565 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1566 void *data, unsigned long len)
1567 {
1568 struct kvm_memslots *slots = kvm_memslots(kvm);
1569 int r;
1570
1571 BUG_ON(len > ghc->len);
1572
1573 if (slots->generation != ghc->generation)
1574 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1575
1576 if (unlikely(!ghc->memslot))
1577 return kvm_write_guest(kvm, ghc->gpa, data, len);
1578
1579 if (kvm_is_error_hva(ghc->hva))
1580 return -EFAULT;
1581
1582 r = __copy_to_user((void __user *)ghc->hva, data, len);
1583 if (r)
1584 return -EFAULT;
1585 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1586
1587 return 0;
1588 }
1589 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1590
1591 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1592 void *data, unsigned long len)
1593 {
1594 struct kvm_memslots *slots = kvm_memslots(kvm);
1595 int r;
1596
1597 BUG_ON(len > ghc->len);
1598
1599 if (slots->generation != ghc->generation)
1600 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1601
1602 if (unlikely(!ghc->memslot))
1603 return kvm_read_guest(kvm, ghc->gpa, data, len);
1604
1605 if (kvm_is_error_hva(ghc->hva))
1606 return -EFAULT;
1607
1608 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1609 if (r)
1610 return -EFAULT;
1611
1612 return 0;
1613 }
1614 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1615
1616 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1617 {
1618 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1619
1620 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1621 }
1622 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1623
1624 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1625 {
1626 gfn_t gfn = gpa >> PAGE_SHIFT;
1627 int seg;
1628 int offset = offset_in_page(gpa);
1629 int ret;
1630
1631 while ((seg = next_segment(len, offset)) != 0) {
1632 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1633 if (ret < 0)
1634 return ret;
1635 offset = 0;
1636 len -= seg;
1637 ++gfn;
1638 }
1639 return 0;
1640 }
1641 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1642
1643 static void mark_page_dirty_in_slot(struct kvm *kvm,
1644 struct kvm_memory_slot *memslot,
1645 gfn_t gfn)
1646 {
1647 if (memslot && memslot->dirty_bitmap) {
1648 unsigned long rel_gfn = gfn - memslot->base_gfn;
1649
1650 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1651 }
1652 }
1653
1654 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1655 {
1656 struct kvm_memory_slot *memslot;
1657
1658 memslot = gfn_to_memslot(kvm, gfn);
1659 mark_page_dirty_in_slot(kvm, memslot, gfn);
1660 }
1661 EXPORT_SYMBOL_GPL(mark_page_dirty);
1662
1663 /*
1664 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1665 */
1666 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1667 {
1668 DEFINE_WAIT(wait);
1669
1670 for (;;) {
1671 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1672
1673 if (kvm_arch_vcpu_runnable(vcpu)) {
1674 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1675 break;
1676 }
1677 if (kvm_cpu_has_pending_timer(vcpu))
1678 break;
1679 if (signal_pending(current))
1680 break;
1681
1682 schedule();
1683 }
1684
1685 finish_wait(&vcpu->wq, &wait);
1686 }
1687 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1688
1689 #ifndef CONFIG_S390
1690 /*
1691 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1692 */
1693 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1694 {
1695 int me;
1696 int cpu = vcpu->cpu;
1697 wait_queue_head_t *wqp;
1698
1699 wqp = kvm_arch_vcpu_wq(vcpu);
1700 if (waitqueue_active(wqp)) {
1701 wake_up_interruptible(wqp);
1702 ++vcpu->stat.halt_wakeup;
1703 }
1704
1705 me = get_cpu();
1706 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1707 if (kvm_arch_vcpu_should_kick(vcpu))
1708 smp_send_reschedule(cpu);
1709 put_cpu();
1710 }
1711 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1712 #endif /* !CONFIG_S390 */
1713
1714 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1715 {
1716 struct pid *pid;
1717 struct task_struct *task = NULL;
1718 bool ret = false;
1719
1720 rcu_read_lock();
1721 pid = rcu_dereference(target->pid);
1722 if (pid)
1723 task = get_pid_task(target->pid, PIDTYPE_PID);
1724 rcu_read_unlock();
1725 if (!task)
1726 return ret;
1727 if (task->flags & PF_VCPU) {
1728 put_task_struct(task);
1729 return ret;
1730 }
1731 ret = yield_to(task, 1);
1732 put_task_struct(task);
1733
1734 return ret;
1735 }
1736 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1737
1738 /*
1739 * Helper that checks whether a VCPU is eligible for directed yield.
1740 * Most eligible candidate to yield is decided by following heuristics:
1741 *
1742 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1743 * (preempted lock holder), indicated by @in_spin_loop.
1744 * Set at the beiginning and cleared at the end of interception/PLE handler.
1745 *
1746 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1747 * chance last time (mostly it has become eligible now since we have probably
1748 * yielded to lockholder in last iteration. This is done by toggling
1749 * @dy_eligible each time a VCPU checked for eligibility.)
1750 *
1751 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1752 * to preempted lock-holder could result in wrong VCPU selection and CPU
1753 * burning. Giving priority for a potential lock-holder increases lock
1754 * progress.
1755 *
1756 * Since algorithm is based on heuristics, accessing another VCPU data without
1757 * locking does not harm. It may result in trying to yield to same VCPU, fail
1758 * and continue with next VCPU and so on.
1759 */
1760 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1761 {
1762 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1763 bool eligible;
1764
1765 eligible = !vcpu->spin_loop.in_spin_loop ||
1766 (vcpu->spin_loop.in_spin_loop &&
1767 vcpu->spin_loop.dy_eligible);
1768
1769 if (vcpu->spin_loop.in_spin_loop)
1770 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1771
1772 return eligible;
1773 #else
1774 return true;
1775 #endif
1776 }
1777
1778 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1779 {
1780 struct kvm *kvm = me->kvm;
1781 struct kvm_vcpu *vcpu;
1782 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1783 int yielded = 0;
1784 int try = 3;
1785 int pass;
1786 int i;
1787
1788 kvm_vcpu_set_in_spin_loop(me, true);
1789 /*
1790 * We boost the priority of a VCPU that is runnable but not
1791 * currently running, because it got preempted by something
1792 * else and called schedule in __vcpu_run. Hopefully that
1793 * VCPU is holding the lock that we need and will release it.
1794 * We approximate round-robin by starting at the last boosted VCPU.
1795 */
1796 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1797 kvm_for_each_vcpu(i, vcpu, kvm) {
1798 if (!pass && i <= last_boosted_vcpu) {
1799 i = last_boosted_vcpu;
1800 continue;
1801 } else if (pass && i > last_boosted_vcpu)
1802 break;
1803 if (!ACCESS_ONCE(vcpu->preempted))
1804 continue;
1805 if (vcpu == me)
1806 continue;
1807 if (waitqueue_active(&vcpu->wq))
1808 continue;
1809 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1810 continue;
1811
1812 yielded = kvm_vcpu_yield_to(vcpu);
1813 if (yielded > 0) {
1814 kvm->last_boosted_vcpu = i;
1815 break;
1816 } else if (yielded < 0) {
1817 try--;
1818 if (!try)
1819 break;
1820 }
1821 }
1822 }
1823 kvm_vcpu_set_in_spin_loop(me, false);
1824
1825 /* Ensure vcpu is not eligible during next spinloop */
1826 kvm_vcpu_set_dy_eligible(me, false);
1827 }
1828 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1829
1830 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1831 {
1832 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1833 struct page *page;
1834
1835 if (vmf->pgoff == 0)
1836 page = virt_to_page(vcpu->run);
1837 #ifdef CONFIG_X86
1838 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1839 page = virt_to_page(vcpu->arch.pio_data);
1840 #endif
1841 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1842 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1843 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1844 #endif
1845 else
1846 return kvm_arch_vcpu_fault(vcpu, vmf);
1847 get_page(page);
1848 vmf->page = page;
1849 return 0;
1850 }
1851
1852 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1853 .fault = kvm_vcpu_fault,
1854 };
1855
1856 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1857 {
1858 vma->vm_ops = &kvm_vcpu_vm_ops;
1859 return 0;
1860 }
1861
1862 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1863 {
1864 struct kvm_vcpu *vcpu = filp->private_data;
1865
1866 kvm_put_kvm(vcpu->kvm);
1867 return 0;
1868 }
1869
1870 static struct file_operations kvm_vcpu_fops = {
1871 .release = kvm_vcpu_release,
1872 .unlocked_ioctl = kvm_vcpu_ioctl,
1873 #ifdef CONFIG_COMPAT
1874 .compat_ioctl = kvm_vcpu_compat_ioctl,
1875 #endif
1876 .mmap = kvm_vcpu_mmap,
1877 .llseek = noop_llseek,
1878 };
1879
1880 /*
1881 * Allocates an inode for the vcpu.
1882 */
1883 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1884 {
1885 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1886 }
1887
1888 /*
1889 * Creates some virtual cpus. Good luck creating more than one.
1890 */
1891 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1892 {
1893 int r;
1894 struct kvm_vcpu *vcpu, *v;
1895
1896 if (id >= KVM_MAX_VCPUS)
1897 return -EINVAL;
1898
1899 vcpu = kvm_arch_vcpu_create(kvm, id);
1900 if (IS_ERR(vcpu))
1901 return PTR_ERR(vcpu);
1902
1903 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1904
1905 r = kvm_arch_vcpu_setup(vcpu);
1906 if (r)
1907 goto vcpu_destroy;
1908
1909 mutex_lock(&kvm->lock);
1910 if (!kvm_vcpu_compatible(vcpu)) {
1911 r = -EINVAL;
1912 goto unlock_vcpu_destroy;
1913 }
1914 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1915 r = -EINVAL;
1916 goto unlock_vcpu_destroy;
1917 }
1918
1919 kvm_for_each_vcpu(r, v, kvm)
1920 if (v->vcpu_id == id) {
1921 r = -EEXIST;
1922 goto unlock_vcpu_destroy;
1923 }
1924
1925 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1926
1927 /* Now it's all set up, let userspace reach it */
1928 kvm_get_kvm(kvm);
1929 r = create_vcpu_fd(vcpu);
1930 if (r < 0) {
1931 kvm_put_kvm(kvm);
1932 goto unlock_vcpu_destroy;
1933 }
1934
1935 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1936 smp_wmb();
1937 atomic_inc(&kvm->online_vcpus);
1938
1939 mutex_unlock(&kvm->lock);
1940 kvm_arch_vcpu_postcreate(vcpu);
1941 return r;
1942
1943 unlock_vcpu_destroy:
1944 mutex_unlock(&kvm->lock);
1945 vcpu_destroy:
1946 kvm_arch_vcpu_destroy(vcpu);
1947 return r;
1948 }
1949
1950 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1951 {
1952 if (sigset) {
1953 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1954 vcpu->sigset_active = 1;
1955 vcpu->sigset = *sigset;
1956 } else
1957 vcpu->sigset_active = 0;
1958 return 0;
1959 }
1960
1961 static long kvm_vcpu_ioctl(struct file *filp,
1962 unsigned int ioctl, unsigned long arg)
1963 {
1964 struct kvm_vcpu *vcpu = filp->private_data;
1965 void __user *argp = (void __user *)arg;
1966 int r;
1967 struct kvm_fpu *fpu = NULL;
1968 struct kvm_sregs *kvm_sregs = NULL;
1969
1970 if (vcpu->kvm->mm != current->mm)
1971 return -EIO;
1972
1973 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1974 /*
1975 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1976 * so vcpu_load() would break it.
1977 */
1978 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1979 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1980 #endif
1981
1982
1983 r = vcpu_load(vcpu);
1984 if (r)
1985 return r;
1986 switch (ioctl) {
1987 case KVM_RUN:
1988 r = -EINVAL;
1989 if (arg)
1990 goto out;
1991 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1992 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1993 break;
1994 case KVM_GET_REGS: {
1995 struct kvm_regs *kvm_regs;
1996
1997 r = -ENOMEM;
1998 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1999 if (!kvm_regs)
2000 goto out;
2001 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2002 if (r)
2003 goto out_free1;
2004 r = -EFAULT;
2005 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2006 goto out_free1;
2007 r = 0;
2008 out_free1:
2009 kfree(kvm_regs);
2010 break;
2011 }
2012 case KVM_SET_REGS: {
2013 struct kvm_regs *kvm_regs;
2014
2015 r = -ENOMEM;
2016 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2017 if (IS_ERR(kvm_regs)) {
2018 r = PTR_ERR(kvm_regs);
2019 goto out;
2020 }
2021 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2022 kfree(kvm_regs);
2023 break;
2024 }
2025 case KVM_GET_SREGS: {
2026 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2027 r = -ENOMEM;
2028 if (!kvm_sregs)
2029 goto out;
2030 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2031 if (r)
2032 goto out;
2033 r = -EFAULT;
2034 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2035 goto out;
2036 r = 0;
2037 break;
2038 }
2039 case KVM_SET_SREGS: {
2040 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2041 if (IS_ERR(kvm_sregs)) {
2042 r = PTR_ERR(kvm_sregs);
2043 kvm_sregs = NULL;
2044 goto out;
2045 }
2046 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2047 break;
2048 }
2049 case KVM_GET_MP_STATE: {
2050 struct kvm_mp_state mp_state;
2051
2052 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2053 if (r)
2054 goto out;
2055 r = -EFAULT;
2056 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2057 goto out;
2058 r = 0;
2059 break;
2060 }
2061 case KVM_SET_MP_STATE: {
2062 struct kvm_mp_state mp_state;
2063
2064 r = -EFAULT;
2065 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2066 goto out;
2067 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2068 break;
2069 }
2070 case KVM_TRANSLATE: {
2071 struct kvm_translation tr;
2072
2073 r = -EFAULT;
2074 if (copy_from_user(&tr, argp, sizeof tr))
2075 goto out;
2076 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2077 if (r)
2078 goto out;
2079 r = -EFAULT;
2080 if (copy_to_user(argp, &tr, sizeof tr))
2081 goto out;
2082 r = 0;
2083 break;
2084 }
2085 case KVM_SET_GUEST_DEBUG: {
2086 struct kvm_guest_debug dbg;
2087
2088 r = -EFAULT;
2089 if (copy_from_user(&dbg, argp, sizeof dbg))
2090 goto out;
2091 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2092 break;
2093 }
2094 case KVM_SET_SIGNAL_MASK: {
2095 struct kvm_signal_mask __user *sigmask_arg = argp;
2096 struct kvm_signal_mask kvm_sigmask;
2097 sigset_t sigset, *p;
2098
2099 p = NULL;
2100 if (argp) {
2101 r = -EFAULT;
2102 if (copy_from_user(&kvm_sigmask, argp,
2103 sizeof kvm_sigmask))
2104 goto out;
2105 r = -EINVAL;
2106 if (kvm_sigmask.len != sizeof sigset)
2107 goto out;
2108 r = -EFAULT;
2109 if (copy_from_user(&sigset, sigmask_arg->sigset,
2110 sizeof sigset))
2111 goto out;
2112 p = &sigset;
2113 }
2114 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2115 break;
2116 }
2117 case KVM_GET_FPU: {
2118 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2119 r = -ENOMEM;
2120 if (!fpu)
2121 goto out;
2122 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2123 if (r)
2124 goto out;
2125 r = -EFAULT;
2126 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2127 goto out;
2128 r = 0;
2129 break;
2130 }
2131 case KVM_SET_FPU: {
2132 fpu = memdup_user(argp, sizeof(*fpu));
2133 if (IS_ERR(fpu)) {
2134 r = PTR_ERR(fpu);
2135 fpu = NULL;
2136 goto out;
2137 }
2138 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2139 break;
2140 }
2141 default:
2142 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2143 }
2144 out:
2145 vcpu_put(vcpu);
2146 kfree(fpu);
2147 kfree(kvm_sregs);
2148 return r;
2149 }
2150
2151 #ifdef CONFIG_COMPAT
2152 static long kvm_vcpu_compat_ioctl(struct file *filp,
2153 unsigned int ioctl, unsigned long arg)
2154 {
2155 struct kvm_vcpu *vcpu = filp->private_data;
2156 void __user *argp = compat_ptr(arg);
2157 int r;
2158
2159 if (vcpu->kvm->mm != current->mm)
2160 return -EIO;
2161
2162 switch (ioctl) {
2163 case KVM_SET_SIGNAL_MASK: {
2164 struct kvm_signal_mask __user *sigmask_arg = argp;
2165 struct kvm_signal_mask kvm_sigmask;
2166 compat_sigset_t csigset;
2167 sigset_t sigset;
2168
2169 if (argp) {
2170 r = -EFAULT;
2171 if (copy_from_user(&kvm_sigmask, argp,
2172 sizeof kvm_sigmask))
2173 goto out;
2174 r = -EINVAL;
2175 if (kvm_sigmask.len != sizeof csigset)
2176 goto out;
2177 r = -EFAULT;
2178 if (copy_from_user(&csigset, sigmask_arg->sigset,
2179 sizeof csigset))
2180 goto out;
2181 sigset_from_compat(&sigset, &csigset);
2182 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2183 } else
2184 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2185 break;
2186 }
2187 default:
2188 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2189 }
2190
2191 out:
2192 return r;
2193 }
2194 #endif
2195
2196 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2197 int (*accessor)(struct kvm_device *dev,
2198 struct kvm_device_attr *attr),
2199 unsigned long arg)
2200 {
2201 struct kvm_device_attr attr;
2202
2203 if (!accessor)
2204 return -EPERM;
2205
2206 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2207 return -EFAULT;
2208
2209 return accessor(dev, &attr);
2210 }
2211
2212 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2213 unsigned long arg)
2214 {
2215 struct kvm_device *dev = filp->private_data;
2216
2217 switch (ioctl) {
2218 case KVM_SET_DEVICE_ATTR:
2219 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2220 case KVM_GET_DEVICE_ATTR:
2221 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2222 case KVM_HAS_DEVICE_ATTR:
2223 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2224 default:
2225 if (dev->ops->ioctl)
2226 return dev->ops->ioctl(dev, ioctl, arg);
2227
2228 return -ENOTTY;
2229 }
2230 }
2231
2232 static int kvm_device_release(struct inode *inode, struct file *filp)
2233 {
2234 struct kvm_device *dev = filp->private_data;
2235 struct kvm *kvm = dev->kvm;
2236
2237 kvm_put_kvm(kvm);
2238 return 0;
2239 }
2240
2241 static const struct file_operations kvm_device_fops = {
2242 .unlocked_ioctl = kvm_device_ioctl,
2243 #ifdef CONFIG_COMPAT
2244 .compat_ioctl = kvm_device_ioctl,
2245 #endif
2246 .release = kvm_device_release,
2247 };
2248
2249 struct kvm_device *kvm_device_from_filp(struct file *filp)
2250 {
2251 if (filp->f_op != &kvm_device_fops)
2252 return NULL;
2253
2254 return filp->private_data;
2255 }
2256
2257 static int kvm_ioctl_create_device(struct kvm *kvm,
2258 struct kvm_create_device *cd)
2259 {
2260 struct kvm_device_ops *ops = NULL;
2261 struct kvm_device *dev;
2262 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2263 int ret;
2264
2265 switch (cd->type) {
2266 #ifdef CONFIG_KVM_MPIC
2267 case KVM_DEV_TYPE_FSL_MPIC_20:
2268 case KVM_DEV_TYPE_FSL_MPIC_42:
2269 ops = &kvm_mpic_ops;
2270 break;
2271 #endif
2272 #ifdef CONFIG_KVM_XICS
2273 case KVM_DEV_TYPE_XICS:
2274 ops = &kvm_xics_ops;
2275 break;
2276 #endif
2277 #ifdef CONFIG_KVM_VFIO
2278 case KVM_DEV_TYPE_VFIO:
2279 ops = &kvm_vfio_ops;
2280 break;
2281 #endif
2282 #ifdef CONFIG_KVM_ARM_VGIC
2283 case KVM_DEV_TYPE_ARM_VGIC_V2:
2284 ops = &kvm_arm_vgic_v2_ops;
2285 break;
2286 #endif
2287 default:
2288 return -ENODEV;
2289 }
2290
2291 if (test)
2292 return 0;
2293
2294 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2295 if (!dev)
2296 return -ENOMEM;
2297
2298 dev->ops = ops;
2299 dev->kvm = kvm;
2300
2301 ret = ops->create(dev, cd->type);
2302 if (ret < 0) {
2303 kfree(dev);
2304 return ret;
2305 }
2306
2307 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2308 if (ret < 0) {
2309 ops->destroy(dev);
2310 return ret;
2311 }
2312
2313 list_add(&dev->vm_node, &kvm->devices);
2314 kvm_get_kvm(kvm);
2315 cd->fd = ret;
2316 return 0;
2317 }
2318
2319 static long kvm_vm_ioctl(struct file *filp,
2320 unsigned int ioctl, unsigned long arg)
2321 {
2322 struct kvm *kvm = filp->private_data;
2323 void __user *argp = (void __user *)arg;
2324 int r;
2325
2326 if (kvm->mm != current->mm)
2327 return -EIO;
2328 switch (ioctl) {
2329 case KVM_CREATE_VCPU:
2330 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2331 break;
2332 case KVM_SET_USER_MEMORY_REGION: {
2333 struct kvm_userspace_memory_region kvm_userspace_mem;
2334
2335 r = -EFAULT;
2336 if (copy_from_user(&kvm_userspace_mem, argp,
2337 sizeof kvm_userspace_mem))
2338 goto out;
2339
2340 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2341 break;
2342 }
2343 case KVM_GET_DIRTY_LOG: {
2344 struct kvm_dirty_log log;
2345
2346 r = -EFAULT;
2347 if (copy_from_user(&log, argp, sizeof log))
2348 goto out;
2349 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2350 break;
2351 }
2352 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2353 case KVM_REGISTER_COALESCED_MMIO: {
2354 struct kvm_coalesced_mmio_zone zone;
2355 r = -EFAULT;
2356 if (copy_from_user(&zone, argp, sizeof zone))
2357 goto out;
2358 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2359 break;
2360 }
2361 case KVM_UNREGISTER_COALESCED_MMIO: {
2362 struct kvm_coalesced_mmio_zone zone;
2363 r = -EFAULT;
2364 if (copy_from_user(&zone, argp, sizeof zone))
2365 goto out;
2366 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2367 break;
2368 }
2369 #endif
2370 case KVM_IRQFD: {
2371 struct kvm_irqfd data;
2372
2373 r = -EFAULT;
2374 if (copy_from_user(&data, argp, sizeof data))
2375 goto out;
2376 r = kvm_irqfd(kvm, &data);
2377 break;
2378 }
2379 case KVM_IOEVENTFD: {
2380 struct kvm_ioeventfd data;
2381
2382 r = -EFAULT;
2383 if (copy_from_user(&data, argp, sizeof data))
2384 goto out;
2385 r = kvm_ioeventfd(kvm, &data);
2386 break;
2387 }
2388 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2389 case KVM_SET_BOOT_CPU_ID:
2390 r = 0;
2391 mutex_lock(&kvm->lock);
2392 if (atomic_read(&kvm->online_vcpus) != 0)
2393 r = -EBUSY;
2394 else
2395 kvm->bsp_vcpu_id = arg;
2396 mutex_unlock(&kvm->lock);
2397 break;
2398 #endif
2399 #ifdef CONFIG_HAVE_KVM_MSI
2400 case KVM_SIGNAL_MSI: {
2401 struct kvm_msi msi;
2402
2403 r = -EFAULT;
2404 if (copy_from_user(&msi, argp, sizeof msi))
2405 goto out;
2406 r = kvm_send_userspace_msi(kvm, &msi);
2407 break;
2408 }
2409 #endif
2410 #ifdef __KVM_HAVE_IRQ_LINE
2411 case KVM_IRQ_LINE_STATUS:
2412 case KVM_IRQ_LINE: {
2413 struct kvm_irq_level irq_event;
2414
2415 r = -EFAULT;
2416 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2417 goto out;
2418
2419 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2420 ioctl == KVM_IRQ_LINE_STATUS);
2421 if (r)
2422 goto out;
2423
2424 r = -EFAULT;
2425 if (ioctl == KVM_IRQ_LINE_STATUS) {
2426 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2427 goto out;
2428 }
2429
2430 r = 0;
2431 break;
2432 }
2433 #endif
2434 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2435 case KVM_SET_GSI_ROUTING: {
2436 struct kvm_irq_routing routing;
2437 struct kvm_irq_routing __user *urouting;
2438 struct kvm_irq_routing_entry *entries;
2439
2440 r = -EFAULT;
2441 if (copy_from_user(&routing, argp, sizeof(routing)))
2442 goto out;
2443 r = -EINVAL;
2444 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2445 goto out;
2446 if (routing.flags)
2447 goto out;
2448 r = -ENOMEM;
2449 entries = vmalloc(routing.nr * sizeof(*entries));
2450 if (!entries)
2451 goto out;
2452 r = -EFAULT;
2453 urouting = argp;
2454 if (copy_from_user(entries, urouting->entries,
2455 routing.nr * sizeof(*entries)))
2456 goto out_free_irq_routing;
2457 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2458 routing.flags);
2459 out_free_irq_routing:
2460 vfree(entries);
2461 break;
2462 }
2463 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2464 case KVM_CREATE_DEVICE: {
2465 struct kvm_create_device cd;
2466
2467 r = -EFAULT;
2468 if (copy_from_user(&cd, argp, sizeof(cd)))
2469 goto out;
2470
2471 r = kvm_ioctl_create_device(kvm, &cd);
2472 if (r)
2473 goto out;
2474
2475 r = -EFAULT;
2476 if (copy_to_user(argp, &cd, sizeof(cd)))
2477 goto out;
2478
2479 r = 0;
2480 break;
2481 }
2482 default:
2483 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2484 if (r == -ENOTTY)
2485 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2486 }
2487 out:
2488 return r;
2489 }
2490
2491 #ifdef CONFIG_COMPAT
2492 struct compat_kvm_dirty_log {
2493 __u32 slot;
2494 __u32 padding1;
2495 union {
2496 compat_uptr_t dirty_bitmap; /* one bit per page */
2497 __u64 padding2;
2498 };
2499 };
2500
2501 static long kvm_vm_compat_ioctl(struct file *filp,
2502 unsigned int ioctl, unsigned long arg)
2503 {
2504 struct kvm *kvm = filp->private_data;
2505 int r;
2506
2507 if (kvm->mm != current->mm)
2508 return -EIO;
2509 switch (ioctl) {
2510 case KVM_GET_DIRTY_LOG: {
2511 struct compat_kvm_dirty_log compat_log;
2512 struct kvm_dirty_log log;
2513
2514 r = -EFAULT;
2515 if (copy_from_user(&compat_log, (void __user *)arg,
2516 sizeof(compat_log)))
2517 goto out;
2518 log.slot = compat_log.slot;
2519 log.padding1 = compat_log.padding1;
2520 log.padding2 = compat_log.padding2;
2521 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2522
2523 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2524 break;
2525 }
2526 default:
2527 r = kvm_vm_ioctl(filp, ioctl, arg);
2528 }
2529
2530 out:
2531 return r;
2532 }
2533 #endif
2534
2535 static struct file_operations kvm_vm_fops = {
2536 .release = kvm_vm_release,
2537 .unlocked_ioctl = kvm_vm_ioctl,
2538 #ifdef CONFIG_COMPAT
2539 .compat_ioctl = kvm_vm_compat_ioctl,
2540 #endif
2541 .llseek = noop_llseek,
2542 };
2543
2544 static int kvm_dev_ioctl_create_vm(unsigned long type)
2545 {
2546 int r;
2547 struct kvm *kvm;
2548
2549 kvm = kvm_create_vm(type);
2550 if (IS_ERR(kvm))
2551 return PTR_ERR(kvm);
2552 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2553 r = kvm_coalesced_mmio_init(kvm);
2554 if (r < 0) {
2555 kvm_put_kvm(kvm);
2556 return r;
2557 }
2558 #endif
2559 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2560 if (r < 0)
2561 kvm_put_kvm(kvm);
2562
2563 return r;
2564 }
2565
2566 static long kvm_dev_ioctl_check_extension_generic(long arg)
2567 {
2568 switch (arg) {
2569 case KVM_CAP_USER_MEMORY:
2570 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2571 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2572 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2573 case KVM_CAP_SET_BOOT_CPU_ID:
2574 #endif
2575 case KVM_CAP_INTERNAL_ERROR_DATA:
2576 #ifdef CONFIG_HAVE_KVM_MSI
2577 case KVM_CAP_SIGNAL_MSI:
2578 #endif
2579 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2580 case KVM_CAP_IRQFD_RESAMPLE:
2581 #endif
2582 return 1;
2583 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2584 case KVM_CAP_IRQ_ROUTING:
2585 return KVM_MAX_IRQ_ROUTES;
2586 #endif
2587 default:
2588 break;
2589 }
2590 return kvm_dev_ioctl_check_extension(arg);
2591 }
2592
2593 static long kvm_dev_ioctl(struct file *filp,
2594 unsigned int ioctl, unsigned long arg)
2595 {
2596 long r = -EINVAL;
2597
2598 switch (ioctl) {
2599 case KVM_GET_API_VERSION:
2600 r = -EINVAL;
2601 if (arg)
2602 goto out;
2603 r = KVM_API_VERSION;
2604 break;
2605 case KVM_CREATE_VM:
2606 r = kvm_dev_ioctl_create_vm(arg);
2607 break;
2608 case KVM_CHECK_EXTENSION:
2609 r = kvm_dev_ioctl_check_extension_generic(arg);
2610 break;
2611 case KVM_GET_VCPU_MMAP_SIZE:
2612 r = -EINVAL;
2613 if (arg)
2614 goto out;
2615 r = PAGE_SIZE; /* struct kvm_run */
2616 #ifdef CONFIG_X86
2617 r += PAGE_SIZE; /* pio data page */
2618 #endif
2619 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2620 r += PAGE_SIZE; /* coalesced mmio ring page */
2621 #endif
2622 break;
2623 case KVM_TRACE_ENABLE:
2624 case KVM_TRACE_PAUSE:
2625 case KVM_TRACE_DISABLE:
2626 r = -EOPNOTSUPP;
2627 break;
2628 default:
2629 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2630 }
2631 out:
2632 return r;
2633 }
2634
2635 static struct file_operations kvm_chardev_ops = {
2636 .unlocked_ioctl = kvm_dev_ioctl,
2637 .compat_ioctl = kvm_dev_ioctl,
2638 .llseek = noop_llseek,
2639 };
2640
2641 static struct miscdevice kvm_dev = {
2642 KVM_MINOR,
2643 "kvm",
2644 &kvm_chardev_ops,
2645 };
2646
2647 static void hardware_enable_nolock(void *junk)
2648 {
2649 int cpu = raw_smp_processor_id();
2650 int r;
2651
2652 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2653 return;
2654
2655 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2656
2657 r = kvm_arch_hardware_enable(NULL);
2658
2659 if (r) {
2660 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2661 atomic_inc(&hardware_enable_failed);
2662 printk(KERN_INFO "kvm: enabling virtualization on "
2663 "CPU%d failed\n", cpu);
2664 }
2665 }
2666
2667 static void hardware_enable(void)
2668 {
2669 raw_spin_lock(&kvm_count_lock);
2670 if (kvm_usage_count)
2671 hardware_enable_nolock(NULL);
2672 raw_spin_unlock(&kvm_count_lock);
2673 }
2674
2675 static void hardware_disable_nolock(void *junk)
2676 {
2677 int cpu = raw_smp_processor_id();
2678
2679 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2680 return;
2681 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2682 kvm_arch_hardware_disable(NULL);
2683 }
2684
2685 static void hardware_disable(void)
2686 {
2687 raw_spin_lock(&kvm_count_lock);
2688 if (kvm_usage_count)
2689 hardware_disable_nolock(NULL);
2690 raw_spin_unlock(&kvm_count_lock);
2691 }
2692
2693 static void hardware_disable_all_nolock(void)
2694 {
2695 BUG_ON(!kvm_usage_count);
2696
2697 kvm_usage_count--;
2698 if (!kvm_usage_count)
2699 on_each_cpu(hardware_disable_nolock, NULL, 1);
2700 }
2701
2702 static void hardware_disable_all(void)
2703 {
2704 raw_spin_lock(&kvm_count_lock);
2705 hardware_disable_all_nolock();
2706 raw_spin_unlock(&kvm_count_lock);
2707 }
2708
2709 static int hardware_enable_all(void)
2710 {
2711 int r = 0;
2712
2713 raw_spin_lock(&kvm_count_lock);
2714
2715 kvm_usage_count++;
2716 if (kvm_usage_count == 1) {
2717 atomic_set(&hardware_enable_failed, 0);
2718 on_each_cpu(hardware_enable_nolock, NULL, 1);
2719
2720 if (atomic_read(&hardware_enable_failed)) {
2721 hardware_disable_all_nolock();
2722 r = -EBUSY;
2723 }
2724 }
2725
2726 raw_spin_unlock(&kvm_count_lock);
2727
2728 return r;
2729 }
2730
2731 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2732 void *v)
2733 {
2734 int cpu = (long)v;
2735
2736 val &= ~CPU_TASKS_FROZEN;
2737 switch (val) {
2738 case CPU_DYING:
2739 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2740 cpu);
2741 hardware_disable();
2742 break;
2743 case CPU_STARTING:
2744 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2745 cpu);
2746 hardware_enable();
2747 break;
2748 }
2749 return NOTIFY_OK;
2750 }
2751
2752 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2753 void *v)
2754 {
2755 /*
2756 * Some (well, at least mine) BIOSes hang on reboot if
2757 * in vmx root mode.
2758 *
2759 * And Intel TXT required VMX off for all cpu when system shutdown.
2760 */
2761 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2762 kvm_rebooting = true;
2763 on_each_cpu(hardware_disable_nolock, NULL, 1);
2764 return NOTIFY_OK;
2765 }
2766
2767 static struct notifier_block kvm_reboot_notifier = {
2768 .notifier_call = kvm_reboot,
2769 .priority = 0,
2770 };
2771
2772 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2773 {
2774 int i;
2775
2776 for (i = 0; i < bus->dev_count; i++) {
2777 struct kvm_io_device *pos = bus->range[i].dev;
2778
2779 kvm_iodevice_destructor(pos);
2780 }
2781 kfree(bus);
2782 }
2783
2784 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2785 const struct kvm_io_range *r2)
2786 {
2787 if (r1->addr < r2->addr)
2788 return -1;
2789 if (r1->addr + r1->len > r2->addr + r2->len)
2790 return 1;
2791 return 0;
2792 }
2793
2794 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2795 {
2796 return kvm_io_bus_cmp(p1, p2);
2797 }
2798
2799 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2800 gpa_t addr, int len)
2801 {
2802 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2803 .addr = addr,
2804 .len = len,
2805 .dev = dev,
2806 };
2807
2808 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2809 kvm_io_bus_sort_cmp, NULL);
2810
2811 return 0;
2812 }
2813
2814 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2815 gpa_t addr, int len)
2816 {
2817 struct kvm_io_range *range, key;
2818 int off;
2819
2820 key = (struct kvm_io_range) {
2821 .addr = addr,
2822 .len = len,
2823 };
2824
2825 range = bsearch(&key, bus->range, bus->dev_count,
2826 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2827 if (range == NULL)
2828 return -ENOENT;
2829
2830 off = range - bus->range;
2831
2832 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2833 off--;
2834
2835 return off;
2836 }
2837
2838 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2839 struct kvm_io_range *range, const void *val)
2840 {
2841 int idx;
2842
2843 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2844 if (idx < 0)
2845 return -EOPNOTSUPP;
2846
2847 while (idx < bus->dev_count &&
2848 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2849 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2850 range->len, val))
2851 return idx;
2852 idx++;
2853 }
2854
2855 return -EOPNOTSUPP;
2856 }
2857
2858 /* kvm_io_bus_write - called under kvm->slots_lock */
2859 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2860 int len, const void *val)
2861 {
2862 struct kvm_io_bus *bus;
2863 struct kvm_io_range range;
2864 int r;
2865
2866 range = (struct kvm_io_range) {
2867 .addr = addr,
2868 .len = len,
2869 };
2870
2871 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2872 r = __kvm_io_bus_write(bus, &range, val);
2873 return r < 0 ? r : 0;
2874 }
2875
2876 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2877 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2878 int len, const void *val, long cookie)
2879 {
2880 struct kvm_io_bus *bus;
2881 struct kvm_io_range range;
2882
2883 range = (struct kvm_io_range) {
2884 .addr = addr,
2885 .len = len,
2886 };
2887
2888 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2889
2890 /* First try the device referenced by cookie. */
2891 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2892 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2893 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2894 val))
2895 return cookie;
2896
2897 /*
2898 * cookie contained garbage; fall back to search and return the
2899 * correct cookie value.
2900 */
2901 return __kvm_io_bus_write(bus, &range, val);
2902 }
2903
2904 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2905 void *val)
2906 {
2907 int idx;
2908
2909 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2910 if (idx < 0)
2911 return -EOPNOTSUPP;
2912
2913 while (idx < bus->dev_count &&
2914 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2915 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2916 range->len, val))
2917 return idx;
2918 idx++;
2919 }
2920
2921 return -EOPNOTSUPP;
2922 }
2923
2924 /* kvm_io_bus_read - called under kvm->slots_lock */
2925 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2926 int len, void *val)
2927 {
2928 struct kvm_io_bus *bus;
2929 struct kvm_io_range range;
2930 int r;
2931
2932 range = (struct kvm_io_range) {
2933 .addr = addr,
2934 .len = len,
2935 };
2936
2937 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2938 r = __kvm_io_bus_read(bus, &range, val);
2939 return r < 0 ? r : 0;
2940 }
2941
2942
2943 /* Caller must hold slots_lock. */
2944 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2945 int len, struct kvm_io_device *dev)
2946 {
2947 struct kvm_io_bus *new_bus, *bus;
2948
2949 bus = kvm->buses[bus_idx];
2950 /* exclude ioeventfd which is limited by maximum fd */
2951 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2952 return -ENOSPC;
2953
2954 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2955 sizeof(struct kvm_io_range)), GFP_KERNEL);
2956 if (!new_bus)
2957 return -ENOMEM;
2958 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2959 sizeof(struct kvm_io_range)));
2960 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2961 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2962 synchronize_srcu_expedited(&kvm->srcu);
2963 kfree(bus);
2964
2965 return 0;
2966 }
2967
2968 /* Caller must hold slots_lock. */
2969 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2970 struct kvm_io_device *dev)
2971 {
2972 int i, r;
2973 struct kvm_io_bus *new_bus, *bus;
2974
2975 bus = kvm->buses[bus_idx];
2976 r = -ENOENT;
2977 for (i = 0; i < bus->dev_count; i++)
2978 if (bus->range[i].dev == dev) {
2979 r = 0;
2980 break;
2981 }
2982
2983 if (r)
2984 return r;
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
2991 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2992 new_bus->dev_count--;
2993 memcpy(new_bus->range + i, bus->range + i + 1,
2994 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2995
2996 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2997 synchronize_srcu_expedited(&kvm->srcu);
2998 kfree(bus);
2999 return r;
3000 }
3001
3002 static struct notifier_block kvm_cpu_notifier = {
3003 .notifier_call = kvm_cpu_hotplug,
3004 };
3005
3006 static int vm_stat_get(void *_offset, u64 *val)
3007 {
3008 unsigned offset = (long)_offset;
3009 struct kvm *kvm;
3010
3011 *val = 0;
3012 spin_lock(&kvm_lock);
3013 list_for_each_entry(kvm, &vm_list, vm_list)
3014 *val += *(u32 *)((void *)kvm + offset);
3015 spin_unlock(&kvm_lock);
3016 return 0;
3017 }
3018
3019 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3020
3021 static int vcpu_stat_get(void *_offset, u64 *val)
3022 {
3023 unsigned offset = (long)_offset;
3024 struct kvm *kvm;
3025 struct kvm_vcpu *vcpu;
3026 int i;
3027
3028 *val = 0;
3029 spin_lock(&kvm_lock);
3030 list_for_each_entry(kvm, &vm_list, vm_list)
3031 kvm_for_each_vcpu(i, vcpu, kvm)
3032 *val += *(u32 *)((void *)vcpu + offset);
3033
3034 spin_unlock(&kvm_lock);
3035 return 0;
3036 }
3037
3038 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3039
3040 static const struct file_operations *stat_fops[] = {
3041 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3042 [KVM_STAT_VM] = &vm_stat_fops,
3043 };
3044
3045 static int kvm_init_debug(void)
3046 {
3047 int r = -EEXIST;
3048 struct kvm_stats_debugfs_item *p;
3049
3050 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3051 if (kvm_debugfs_dir == NULL)
3052 goto out;
3053
3054 for (p = debugfs_entries; p->name; ++p) {
3055 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3056 (void *)(long)p->offset,
3057 stat_fops[p->kind]);
3058 if (p->dentry == NULL)
3059 goto out_dir;
3060 }
3061
3062 return 0;
3063
3064 out_dir:
3065 debugfs_remove_recursive(kvm_debugfs_dir);
3066 out:
3067 return r;
3068 }
3069
3070 static void kvm_exit_debug(void)
3071 {
3072 struct kvm_stats_debugfs_item *p;
3073
3074 for (p = debugfs_entries; p->name; ++p)
3075 debugfs_remove(p->dentry);
3076 debugfs_remove(kvm_debugfs_dir);
3077 }
3078
3079 static int kvm_suspend(void)
3080 {
3081 if (kvm_usage_count)
3082 hardware_disable_nolock(NULL);
3083 return 0;
3084 }
3085
3086 static void kvm_resume(void)
3087 {
3088 if (kvm_usage_count) {
3089 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3090 hardware_enable_nolock(NULL);
3091 }
3092 }
3093
3094 static struct syscore_ops kvm_syscore_ops = {
3095 .suspend = kvm_suspend,
3096 .resume = kvm_resume,
3097 };
3098
3099 static inline
3100 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3101 {
3102 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3103 }
3104
3105 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3106 {
3107 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3108 if (vcpu->preempted)
3109 vcpu->preempted = false;
3110
3111 kvm_arch_vcpu_load(vcpu, cpu);
3112 }
3113
3114 static void kvm_sched_out(struct preempt_notifier *pn,
3115 struct task_struct *next)
3116 {
3117 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3118
3119 if (current->state == TASK_RUNNING)
3120 vcpu->preempted = true;
3121 kvm_arch_vcpu_put(vcpu);
3122 }
3123
3124 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3125 struct module *module)
3126 {
3127 int r;
3128 int cpu;
3129
3130 r = kvm_arch_init(opaque);
3131 if (r)
3132 goto out_fail;
3133
3134 /*
3135 * kvm_arch_init makes sure there's at most one caller
3136 * for architectures that support multiple implementations,
3137 * like intel and amd on x86.
3138 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3139 * conflicts in case kvm is already setup for another implementation.
3140 */
3141 r = kvm_irqfd_init();
3142 if (r)
3143 goto out_irqfd;
3144
3145 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3146 r = -ENOMEM;
3147 goto out_free_0;
3148 }
3149
3150 r = kvm_arch_hardware_setup();
3151 if (r < 0)
3152 goto out_free_0a;
3153
3154 for_each_online_cpu(cpu) {
3155 smp_call_function_single(cpu,
3156 kvm_arch_check_processor_compat,
3157 &r, 1);
3158 if (r < 0)
3159 goto out_free_1;
3160 }
3161
3162 r = register_cpu_notifier(&kvm_cpu_notifier);
3163 if (r)
3164 goto out_free_2;
3165 register_reboot_notifier(&kvm_reboot_notifier);
3166
3167 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3168 if (!vcpu_align)
3169 vcpu_align = __alignof__(struct kvm_vcpu);
3170 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3171 0, NULL);
3172 if (!kvm_vcpu_cache) {
3173 r = -ENOMEM;
3174 goto out_free_3;
3175 }
3176
3177 r = kvm_async_pf_init();
3178 if (r)
3179 goto out_free;
3180
3181 kvm_chardev_ops.owner = module;
3182 kvm_vm_fops.owner = module;
3183 kvm_vcpu_fops.owner = module;
3184
3185 r = misc_register(&kvm_dev);
3186 if (r) {
3187 printk(KERN_ERR "kvm: misc device register failed\n");
3188 goto out_unreg;
3189 }
3190
3191 register_syscore_ops(&kvm_syscore_ops);
3192
3193 kvm_preempt_ops.sched_in = kvm_sched_in;
3194 kvm_preempt_ops.sched_out = kvm_sched_out;
3195
3196 r = kvm_init_debug();
3197 if (r) {
3198 printk(KERN_ERR "kvm: create debugfs files failed\n");
3199 goto out_undebugfs;
3200 }
3201
3202 return 0;
3203
3204 out_undebugfs:
3205 unregister_syscore_ops(&kvm_syscore_ops);
3206 misc_deregister(&kvm_dev);
3207 out_unreg:
3208 kvm_async_pf_deinit();
3209 out_free:
3210 kmem_cache_destroy(kvm_vcpu_cache);
3211 out_free_3:
3212 unregister_reboot_notifier(&kvm_reboot_notifier);
3213 unregister_cpu_notifier(&kvm_cpu_notifier);
3214 out_free_2:
3215 out_free_1:
3216 kvm_arch_hardware_unsetup();
3217 out_free_0a:
3218 free_cpumask_var(cpus_hardware_enabled);
3219 out_free_0:
3220 kvm_irqfd_exit();
3221 out_irqfd:
3222 kvm_arch_exit();
3223 out_fail:
3224 return r;
3225 }
3226 EXPORT_SYMBOL_GPL(kvm_init);
3227
3228 void kvm_exit(void)
3229 {
3230 kvm_exit_debug();
3231 misc_deregister(&kvm_dev);
3232 kmem_cache_destroy(kvm_vcpu_cache);
3233 kvm_async_pf_deinit();
3234 unregister_syscore_ops(&kvm_syscore_ops);
3235 unregister_reboot_notifier(&kvm_reboot_notifier);
3236 unregister_cpu_notifier(&kvm_cpu_notifier);
3237 on_each_cpu(hardware_disable_nolock, NULL, 1);
3238 kvm_arch_hardware_unsetup();
3239 kvm_arch_exit();
3240 kvm_irqfd_exit();
3241 free_cpumask_var(cpus_hardware_enabled);
3242 }
3243 EXPORT_SYMBOL_GPL(kvm_exit);
Linux with added FPC-III support