projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
kvm: optimize GFN to memslot lookup with large slots amount
[deliverable/linux.git] / virt / kvm / kvm_main.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
58
59 #include "coalesced_mmio.h"
60 #include "async_pf.h"
61 #include "vfio.h"
62
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
65
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
68
69 /*
70 * Ordering of locks:
71 *
72 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 */
74
75 DEFINE_SPINLOCK(kvm_lock);
76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
77 LIST_HEAD(vm_list);
78
79 static cpumask_var_t cpus_hardware_enabled;
80 static int kvm_usage_count = 0;
81 static atomic_t hardware_enable_failed;
82
83 struct kmem_cache *kvm_vcpu_cache;
84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
85
86 static __read_mostly struct preempt_ops kvm_preempt_ops;
87
88 struct dentry *kvm_debugfs_dir;
89
90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92 #ifdef CONFIG_COMPAT
93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 unsigned long arg);
95 #endif
96 static int hardware_enable_all(void);
97 static void hardware_disable_all(void);
98
99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
100
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
104
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
107
108 static bool largepages_enabled = true;
109
110 bool kvm_is_mmio_pfn(pfn_t pfn)
111 {
112 if (pfn_valid(pfn))
113 return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn));
114
115 return true;
116 }
117
118 /*
119 * Switches to specified vcpu, until a matching vcpu_put()
120 */
121 int vcpu_load(struct kvm_vcpu *vcpu)
122 {
123 int cpu;
124
125 if (mutex_lock_killable(&vcpu->mutex))
126 return -EINTR;
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
132 if (oldpid)
133 synchronize_rcu();
134 put_pid(oldpid);
135 }
136 cpu = get_cpu();
137 preempt_notifier_register(&vcpu->preempt_notifier);
138 kvm_arch_vcpu_load(vcpu, cpu);
139 put_cpu();
140 return 0;
141 }
142
143 void vcpu_put(struct kvm_vcpu *vcpu)
144 {
145 preempt_disable();
146 kvm_arch_vcpu_put(vcpu);
147 preempt_notifier_unregister(&vcpu->preempt_notifier);
148 preempt_enable();
149 mutex_unlock(&vcpu->mutex);
150 }
151
152 static void ack_flush(void *_completed)
153 {
154 }
155
156 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
157 {
158 int i, cpu, me;
159 cpumask_var_t cpus;
160 bool called = true;
161 struct kvm_vcpu *vcpu;
162
163 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
164
165 me = get_cpu();
166 kvm_for_each_vcpu(i, vcpu, kvm) {
167 kvm_make_request(req, vcpu);
168 cpu = vcpu->cpu;
169
170 /* Set ->requests bit before we read ->mode */
171 smp_mb();
172
173 if (cpus != NULL && cpu != -1 && cpu != me &&
174 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
175 cpumask_set_cpu(cpu, cpus);
176 }
177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
181 else
182 called = false;
183 put_cpu();
184 free_cpumask_var(cpus);
185 return called;
186 }
187
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
189 {
190 long dirty_count = kvm->tlbs_dirty;
191
192 smp_mb();
193 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194 ++kvm->stat.remote_tlb_flush;
195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
196 }
197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
198
199 void kvm_reload_remote_mmus(struct kvm *kvm)
200 {
201 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
202 }
203
204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
205 {
206 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
207 }
208
209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
210 {
211 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
212 }
213
214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215 {
216 struct page *page;
217 int r;
218
219 mutex_init(&vcpu->mutex);
220 vcpu->cpu = -1;
221 vcpu->kvm = kvm;
222 vcpu->vcpu_id = id;
223 vcpu->pid = NULL;
224 init_waitqueue_head(&vcpu->wq);
225 kvm_async_pf_vcpu_init(vcpu);
226
227 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 if (!page) {
229 r = -ENOMEM;
230 goto fail;
231 }
232 vcpu->run = page_address(page);
233
234 kvm_vcpu_set_in_spin_loop(vcpu, false);
235 kvm_vcpu_set_dy_eligible(vcpu, false);
236 vcpu->preempted = false;
237
238 r = kvm_arch_vcpu_init(vcpu);
239 if (r < 0)
240 goto fail_free_run;
241 return 0;
242
243 fail_free_run:
244 free_page((unsigned long)vcpu->run);
245 fail:
246 return r;
247 }
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 {
252 put_pid(vcpu->pid);
253 kvm_arch_vcpu_uninit(vcpu);
254 free_page((unsigned long)vcpu->run);
255 }
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 {
261 return container_of(mn, struct kvm, mmu_notifier);
262 }
263
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
267 {
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
269 int need_tlb_flush, idx;
270
271 /*
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * before returning.
280 *
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
283 *
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
288 */
289 idx = srcu_read_lock(&kvm->srcu);
290 spin_lock(&kvm->mmu_lock);
291
292 kvm->mmu_notifier_seq++;
293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294 /* we've to flush the tlb before the pages can be freed */
295 if (need_tlb_flush)
296 kvm_flush_remote_tlbs(kvm);
297
298 spin_unlock(&kvm->mmu_lock);
299
300 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
301
302 srcu_read_unlock(&kvm->srcu, idx);
303 }
304
305 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
306 struct mm_struct *mm,
307 unsigned long address,
308 pte_t pte)
309 {
310 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 int idx;
312
313 idx = srcu_read_lock(&kvm->srcu);
314 spin_lock(&kvm->mmu_lock);
315 kvm->mmu_notifier_seq++;
316 kvm_set_spte_hva(kvm, address, pte);
317 spin_unlock(&kvm->mmu_lock);
318 srcu_read_unlock(&kvm->srcu, idx);
319 }
320
321 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
322 struct mm_struct *mm,
323 unsigned long start,
324 unsigned long end)
325 {
326 struct kvm *kvm = mmu_notifier_to_kvm(mn);
327 int need_tlb_flush = 0, idx;
328
329 idx = srcu_read_lock(&kvm->srcu);
330 spin_lock(&kvm->mmu_lock);
331 /*
332 * The count increase must become visible at unlock time as no
333 * spte can be established without taking the mmu_lock and
334 * count is also read inside the mmu_lock critical section.
335 */
336 kvm->mmu_notifier_count++;
337 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
338 need_tlb_flush |= kvm->tlbs_dirty;
339 /* we've to flush the tlb before the pages can be freed */
340 if (need_tlb_flush)
341 kvm_flush_remote_tlbs(kvm);
342
343 spin_unlock(&kvm->mmu_lock);
344 srcu_read_unlock(&kvm->srcu, idx);
345 }
346
347 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
348 struct mm_struct *mm,
349 unsigned long start,
350 unsigned long end)
351 {
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
353
354 spin_lock(&kvm->mmu_lock);
355 /*
356 * This sequence increase will notify the kvm page fault that
357 * the page that is going to be mapped in the spte could have
358 * been freed.
359 */
360 kvm->mmu_notifier_seq++;
361 smp_wmb();
362 /*
363 * The above sequence increase must be visible before the
364 * below count decrease, which is ensured by the smp_wmb above
365 * in conjunction with the smp_rmb in mmu_notifier_retry().
366 */
367 kvm->mmu_notifier_count--;
368 spin_unlock(&kvm->mmu_lock);
369
370 BUG_ON(kvm->mmu_notifier_count < 0);
371 }
372
373 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
374 struct mm_struct *mm,
375 unsigned long start,
376 unsigned long end)
377 {
378 struct kvm *kvm = mmu_notifier_to_kvm(mn);
379 int young, idx;
380
381 idx = srcu_read_lock(&kvm->srcu);
382 spin_lock(&kvm->mmu_lock);
383
384 young = kvm_age_hva(kvm, start, end);
385 if (young)
386 kvm_flush_remote_tlbs(kvm);
387
388 spin_unlock(&kvm->mmu_lock);
389 srcu_read_unlock(&kvm->srcu, idx);
390
391 return young;
392 }
393
394 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
395 struct mm_struct *mm,
396 unsigned long address)
397 {
398 struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 int young, idx;
400
401 idx = srcu_read_lock(&kvm->srcu);
402 spin_lock(&kvm->mmu_lock);
403 young = kvm_test_age_hva(kvm, address);
404 spin_unlock(&kvm->mmu_lock);
405 srcu_read_unlock(&kvm->srcu, idx);
406
407 return young;
408 }
409
410 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
411 struct mm_struct *mm)
412 {
413 struct kvm *kvm = mmu_notifier_to_kvm(mn);
414 int idx;
415
416 idx = srcu_read_lock(&kvm->srcu);
417 kvm_arch_flush_shadow_all(kvm);
418 srcu_read_unlock(&kvm->srcu, idx);
419 }
420
421 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
422 .invalidate_page = kvm_mmu_notifier_invalidate_page,
423 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
424 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
425 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
426 .test_young = kvm_mmu_notifier_test_young,
427 .change_pte = kvm_mmu_notifier_change_pte,
428 .release = kvm_mmu_notifier_release,
429 };
430
431 static int kvm_init_mmu_notifier(struct kvm *kvm)
432 {
433 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
434 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
435 }
436
437 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
438
439 static int kvm_init_mmu_notifier(struct kvm *kvm)
440 {
441 return 0;
442 }
443
444 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
445
446 static void kvm_init_memslots_id(struct kvm *kvm)
447 {
448 int i;
449 struct kvm_memslots *slots = kvm->memslots;
450
451 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
452 slots->id_to_index[i] = slots->memslots[i].id = i;
453 }
454
455 static struct kvm *kvm_create_vm(unsigned long type)
456 {
457 int r, i;
458 struct kvm *kvm = kvm_arch_alloc_vm();
459
460 if (!kvm)
461 return ERR_PTR(-ENOMEM);
462
463 r = kvm_arch_init_vm(kvm, type);
464 if (r)
465 goto out_err_no_disable;
466
467 r = hardware_enable_all();
468 if (r)
469 goto out_err_no_disable;
470
471 #ifdef CONFIG_HAVE_KVM_IRQFD
472 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
473 #endif
474
475 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
476
477 r = -ENOMEM;
478 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
479 if (!kvm->memslots)
480 goto out_err_no_srcu;
481
482 /*
483 * Init kvm generation close to the maximum to easily test the
484 * code of handling generation number wrap-around.
485 */
486 kvm->memslots->generation = -150;
487
488 kvm_init_memslots_id(kvm);
489 if (init_srcu_struct(&kvm->srcu))
490 goto out_err_no_srcu;
491 if (init_srcu_struct(&kvm->irq_srcu))
492 goto out_err_no_irq_srcu;
493 for (i = 0; i < KVM_NR_BUSES; i++) {
494 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
495 GFP_KERNEL);
496 if (!kvm->buses[i])
497 goto out_err;
498 }
499
500 spin_lock_init(&kvm->mmu_lock);
501 kvm->mm = current->mm;
502 atomic_inc(&kvm->mm->mm_count);
503 kvm_eventfd_init(kvm);
504 mutex_init(&kvm->lock);
505 mutex_init(&kvm->irq_lock);
506 mutex_init(&kvm->slots_lock);
507 atomic_set(&kvm->users_count, 1);
508 INIT_LIST_HEAD(&kvm->devices);
509
510 r = kvm_init_mmu_notifier(kvm);
511 if (r)
512 goto out_err;
513
514 spin_lock(&kvm_lock);
515 list_add(&kvm->vm_list, &vm_list);
516 spin_unlock(&kvm_lock);
517
518 return kvm;
519
520 out_err:
521 cleanup_srcu_struct(&kvm->irq_srcu);
522 out_err_no_irq_srcu:
523 cleanup_srcu_struct(&kvm->srcu);
524 out_err_no_srcu:
525 hardware_disable_all();
526 out_err_no_disable:
527 for (i = 0; i < KVM_NR_BUSES; i++)
528 kfree(kvm->buses[i]);
529 kfree(kvm->memslots);
530 kvm_arch_free_vm(kvm);
531 return ERR_PTR(r);
532 }
533
534 /*
535 * Avoid using vmalloc for a small buffer.
536 * Should not be used when the size is statically known.
537 */
538 void *kvm_kvzalloc(unsigned long size)
539 {
540 if (size > PAGE_SIZE)
541 return vzalloc(size);
542 else
543 return kzalloc(size, GFP_KERNEL);
544 }
545
546 void kvm_kvfree(const void *addr)
547 {
548 if (is_vmalloc_addr(addr))
549 vfree(addr);
550 else
551 kfree(addr);
552 }
553
554 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
555 {
556 if (!memslot->dirty_bitmap)
557 return;
558
559 kvm_kvfree(memslot->dirty_bitmap);
560 memslot->dirty_bitmap = NULL;
561 }
562
563 /*
564 * Free any memory in @free but not in @dont.
565 */
566 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
567 struct kvm_memory_slot *dont)
568 {
569 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
570 kvm_destroy_dirty_bitmap(free);
571
572 kvm_arch_free_memslot(kvm, free, dont);
573
574 free->npages = 0;
575 }
576
577 static void kvm_free_physmem(struct kvm *kvm)
578 {
579 struct kvm_memslots *slots = kvm->memslots;
580 struct kvm_memory_slot *memslot;
581
582 kvm_for_each_memslot(memslot, slots)
583 kvm_free_physmem_slot(kvm, memslot, NULL);
584
585 kfree(kvm->memslots);
586 }
587
588 static void kvm_destroy_devices(struct kvm *kvm)
589 {
590 struct list_head *node, *tmp;
591
592 list_for_each_safe(node, tmp, &kvm->devices) {
593 struct kvm_device *dev =
594 list_entry(node, struct kvm_device, vm_node);
595
596 list_del(node);
597 dev->ops->destroy(dev);
598 }
599 }
600
601 static void kvm_destroy_vm(struct kvm *kvm)
602 {
603 int i;
604 struct mm_struct *mm = kvm->mm;
605
606 kvm_arch_sync_events(kvm);
607 spin_lock(&kvm_lock);
608 list_del(&kvm->vm_list);
609 spin_unlock(&kvm_lock);
610 kvm_free_irq_routing(kvm);
611 for (i = 0; i < KVM_NR_BUSES; i++)
612 kvm_io_bus_destroy(kvm->buses[i]);
613 kvm_coalesced_mmio_free(kvm);
614 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
615 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
616 #else
617 kvm_arch_flush_shadow_all(kvm);
618 #endif
619 kvm_arch_destroy_vm(kvm);
620 kvm_destroy_devices(kvm);
621 kvm_free_physmem(kvm);
622 cleanup_srcu_struct(&kvm->irq_srcu);
623 cleanup_srcu_struct(&kvm->srcu);
624 kvm_arch_free_vm(kvm);
625 hardware_disable_all();
626 mmdrop(mm);
627 }
628
629 void kvm_get_kvm(struct kvm *kvm)
630 {
631 atomic_inc(&kvm->users_count);
632 }
633 EXPORT_SYMBOL_GPL(kvm_get_kvm);
634
635 void kvm_put_kvm(struct kvm *kvm)
636 {
637 if (atomic_dec_and_test(&kvm->users_count))
638 kvm_destroy_vm(kvm);
639 }
640 EXPORT_SYMBOL_GPL(kvm_put_kvm);
641
642
643 static int kvm_vm_release(struct inode *inode, struct file *filp)
644 {
645 struct kvm *kvm = filp->private_data;
646
647 kvm_irqfd_release(kvm);
648
649 kvm_put_kvm(kvm);
650 return 0;
651 }
652
653 /*
654 * Allocation size is twice as large as the actual dirty bitmap size.
655 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
656 */
657 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
658 {
659 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
660
661 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
662 if (!memslot->dirty_bitmap)
663 return -ENOMEM;
664
665 return 0;
666 }
667
668 /*
669 * Insert memslot and re-sort memslots based on their GFN,
670 * so binary search could be used to lookup GFN.
671 * Sorting algorithm takes advantage of having initially
672 * sorted array and known changed memslot position.
673 */
674 static void update_memslots(struct kvm_memslots *slots,
675 struct kvm_memory_slot *new)
676 {
677 int id = new->id;
678 int i = slots->id_to_index[id];
679 struct kvm_memory_slot *mslots = slots->memslots;
680
681 WARN_ON(mslots[i].id != id);
682 if (!new->npages) {
683 new->base_gfn = 0;
684 if (mslots[i].npages)
685 slots->used_slots--;
686 } else {
687 if (!mslots[i].npages)
688 slots->used_slots++;
689 }
690
691 while (i < KVM_MEM_SLOTS_NUM - 1 &&
692 new->base_gfn <= mslots[i + 1].base_gfn) {
693 if (!mslots[i + 1].npages)
694 break;
695 mslots[i] = mslots[i + 1];
696 slots->id_to_index[mslots[i].id] = i;
697 i++;
698 }
699 while (i > 0 &&
700 new->base_gfn > mslots[i - 1].base_gfn) {
701 mslots[i] = mslots[i - 1];
702 slots->id_to_index[mslots[i].id] = i;
703 i--;
704 }
705
706 mslots[i] = *new;
707 slots->id_to_index[mslots[i].id] = i;
708 }
709
710 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
711 {
712 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
713
714 #ifdef __KVM_HAVE_READONLY_MEM
715 valid_flags |= KVM_MEM_READONLY;
716 #endif
717
718 if (mem->flags & ~valid_flags)
719 return -EINVAL;
720
721 return 0;
722 }
723
724 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
725 struct kvm_memslots *slots)
726 {
727 struct kvm_memslots *old_memslots = kvm->memslots;
728
729 /*
730 * Set the low bit in the generation, which disables SPTE caching
731 * until the end of synchronize_srcu_expedited.
732 */
733 WARN_ON(old_memslots->generation & 1);
734 slots->generation = old_memslots->generation + 1;
735
736 rcu_assign_pointer(kvm->memslots, slots);
737 synchronize_srcu_expedited(&kvm->srcu);
738
739 /*
740 * Increment the new memslot generation a second time. This prevents
741 * vm exits that race with memslot updates from caching a memslot
742 * generation that will (potentially) be valid forever.
743 */
744 slots->generation++;
745
746 kvm_arch_memslots_updated(kvm);
747
748 return old_memslots;
749 }
750
751 /*
752 * Allocate some memory and give it an address in the guest physical address
753 * space.
754 *
755 * Discontiguous memory is allowed, mostly for framebuffers.
756 *
757 * Must be called holding kvm->slots_lock for write.
758 */
759 int __kvm_set_memory_region(struct kvm *kvm,
760 struct kvm_userspace_memory_region *mem)
761 {
762 int r;
763 gfn_t base_gfn;
764 unsigned long npages;
765 struct kvm_memory_slot *slot;
766 struct kvm_memory_slot old, new;
767 struct kvm_memslots *slots = NULL, *old_memslots;
768 enum kvm_mr_change change;
769
770 r = check_memory_region_flags(mem);
771 if (r)
772 goto out;
773
774 r = -EINVAL;
775 /* General sanity checks */
776 if (mem->memory_size & (PAGE_SIZE - 1))
777 goto out;
778 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
779 goto out;
780 /* We can read the guest memory with __xxx_user() later on. */
781 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
782 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
783 !access_ok(VERIFY_WRITE,
784 (void __user *)(unsigned long)mem->userspace_addr,
785 mem->memory_size)))
786 goto out;
787 if (mem->slot >= KVM_MEM_SLOTS_NUM)
788 goto out;
789 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
790 goto out;
791
792 slot = id_to_memslot(kvm->memslots, mem->slot);
793 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
794 npages = mem->memory_size >> PAGE_SHIFT;
795
796 if (npages > KVM_MEM_MAX_NR_PAGES)
797 goto out;
798
799 if (!npages)
800 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
801
802 new = old = *slot;
803
804 new.id = mem->slot;
805 new.base_gfn = base_gfn;
806 new.npages = npages;
807 new.flags = mem->flags;
808
809 if (npages) {
810 if (!old.npages)
811 change = KVM_MR_CREATE;
812 else { /* Modify an existing slot. */
813 if ((mem->userspace_addr != old.userspace_addr) ||
814 (npages != old.npages) ||
815 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
816 goto out;
817
818 if (base_gfn != old.base_gfn)
819 change = KVM_MR_MOVE;
820 else if (new.flags != old.flags)
821 change = KVM_MR_FLAGS_ONLY;
822 else { /* Nothing to change. */
823 r = 0;
824 goto out;
825 }
826 }
827 } else if (old.npages) {
828 change = KVM_MR_DELETE;
829 } else /* Modify a non-existent slot: disallowed. */
830 goto out;
831
832 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
833 /* Check for overlaps */
834 r = -EEXIST;
835 kvm_for_each_memslot(slot, kvm->memslots) {
836 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
837 (slot->id == mem->slot))
838 continue;
839 if (!((base_gfn + npages <= slot->base_gfn) ||
840 (base_gfn >= slot->base_gfn + slot->npages)))
841 goto out;
842 }
843 }
844
845 /* Free page dirty bitmap if unneeded */
846 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
847 new.dirty_bitmap = NULL;
848
849 r = -ENOMEM;
850 if (change == KVM_MR_CREATE) {
851 new.userspace_addr = mem->userspace_addr;
852
853 if (kvm_arch_create_memslot(kvm, &new, npages))
854 goto out_free;
855 }
856
857 /* Allocate page dirty bitmap if needed */
858 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
859 if (kvm_create_dirty_bitmap(&new) < 0)
860 goto out_free;
861 }
862
863 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
864 GFP_KERNEL);
865 if (!slots)
866 goto out_free;
867
868 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
869 slot = id_to_memslot(slots, mem->slot);
870 slot->flags |= KVM_MEMSLOT_INVALID;
871
872 old_memslots = install_new_memslots(kvm, slots);
873
874 /* slot was deleted or moved, clear iommu mapping */
875 kvm_iommu_unmap_pages(kvm, &old);
876 /* From this point no new shadow pages pointing to a deleted,
877 * or moved, memslot will be created.
878 *
879 * validation of sp->gfn happens in:
880 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
881 * - kvm_is_visible_gfn (mmu_check_roots)
882 */
883 kvm_arch_flush_shadow_memslot(kvm, slot);
884
885 /*
886 * We can re-use the old_memslots from above, the only difference
887 * from the currently installed memslots is the invalid flag. This
888 * will get overwritten by update_memslots anyway.
889 */
890 slots = old_memslots;
891 }
892
893 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
894 if (r)
895 goto out_slots;
896
897 /* actual memory is freed via old in kvm_free_physmem_slot below */
898 if (change == KVM_MR_DELETE) {
899 new.dirty_bitmap = NULL;
900 memset(&new.arch, 0, sizeof(new.arch));
901 }
902
903 update_memslots(slots, &new);
904 old_memslots = install_new_memslots(kvm, slots);
905
906 kvm_arch_commit_memory_region(kvm, mem, &old, change);
907
908 kvm_free_physmem_slot(kvm, &old, &new);
909 kfree(old_memslots);
910
911 /*
912 * IOMMU mapping: New slots need to be mapped. Old slots need to be
913 * un-mapped and re-mapped if their base changes. Since base change
914 * unmapping is handled above with slot deletion, mapping alone is
915 * needed here. Anything else the iommu might care about for existing
916 * slots (size changes, userspace addr changes and read-only flag
917 * changes) is disallowed above, so any other attribute changes getting
918 * here can be skipped.
919 */
920 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
921 r = kvm_iommu_map_pages(kvm, &new);
922 return r;
923 }
924
925 return 0;
926
927 out_slots:
928 kfree(slots);
929 out_free:
930 kvm_free_physmem_slot(kvm, &new, &old);
931 out:
932 return r;
933 }
934 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
935
936 int kvm_set_memory_region(struct kvm *kvm,
937 struct kvm_userspace_memory_region *mem)
938 {
939 int r;
940
941 mutex_lock(&kvm->slots_lock);
942 r = __kvm_set_memory_region(kvm, mem);
943 mutex_unlock(&kvm->slots_lock);
944 return r;
945 }
946 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
947
948 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
949 struct kvm_userspace_memory_region *mem)
950 {
951 if (mem->slot >= KVM_USER_MEM_SLOTS)
952 return -EINVAL;
953 return kvm_set_memory_region(kvm, mem);
954 }
955
956 int kvm_get_dirty_log(struct kvm *kvm,
957 struct kvm_dirty_log *log, int *is_dirty)
958 {
959 struct kvm_memory_slot *memslot;
960 int r, i;
961 unsigned long n;
962 unsigned long any = 0;
963
964 r = -EINVAL;
965 if (log->slot >= KVM_USER_MEM_SLOTS)
966 goto out;
967
968 memslot = id_to_memslot(kvm->memslots, log->slot);
969 r = -ENOENT;
970 if (!memslot->dirty_bitmap)
971 goto out;
972
973 n = kvm_dirty_bitmap_bytes(memslot);
974
975 for (i = 0; !any && i < n/sizeof(long); ++i)
976 any = memslot->dirty_bitmap[i];
977
978 r = -EFAULT;
979 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
980 goto out;
981
982 if (any)
983 *is_dirty = 1;
984
985 r = 0;
986 out:
987 return r;
988 }
989 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
990
991 bool kvm_largepages_enabled(void)
992 {
993 return largepages_enabled;
994 }
995
996 void kvm_disable_largepages(void)
997 {
998 largepages_enabled = false;
999 }
1000 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1001
1002 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1003 {
1004 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1005 }
1006 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1007
1008 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1009 {
1010 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1011
1012 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1013 memslot->flags & KVM_MEMSLOT_INVALID)
1014 return 0;
1015
1016 return 1;
1017 }
1018 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1019
1020 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1021 {
1022 struct vm_area_struct *vma;
1023 unsigned long addr, size;
1024
1025 size = PAGE_SIZE;
1026
1027 addr = gfn_to_hva(kvm, gfn);
1028 if (kvm_is_error_hva(addr))
1029 return PAGE_SIZE;
1030
1031 down_read(&current->mm->mmap_sem);
1032 vma = find_vma(current->mm, addr);
1033 if (!vma)
1034 goto out;
1035
1036 size = vma_kernel_pagesize(vma);
1037
1038 out:
1039 up_read(&current->mm->mmap_sem);
1040
1041 return size;
1042 }
1043
1044 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1045 {
1046 return slot->flags & KVM_MEM_READONLY;
1047 }
1048
1049 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1050 gfn_t *nr_pages, bool write)
1051 {
1052 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1053 return KVM_HVA_ERR_BAD;
1054
1055 if (memslot_is_readonly(slot) && write)
1056 return KVM_HVA_ERR_RO_BAD;
1057
1058 if (nr_pages)
1059 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1060
1061 return __gfn_to_hva_memslot(slot, gfn);
1062 }
1063
1064 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1065 gfn_t *nr_pages)
1066 {
1067 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1068 }
1069
1070 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1071 gfn_t gfn)
1072 {
1073 return gfn_to_hva_many(slot, gfn, NULL);
1074 }
1075 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1076
1077 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1078 {
1079 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1080 }
1081 EXPORT_SYMBOL_GPL(gfn_to_hva);
1082
1083 /*
1084 * If writable is set to false, the hva returned by this function is only
1085 * allowed to be read.
1086 */
1087 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1088 gfn_t gfn, bool *writable)
1089 {
1090 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1091
1092 if (!kvm_is_error_hva(hva) && writable)
1093 *writable = !memslot_is_readonly(slot);
1094
1095 return hva;
1096 }
1097
1098 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1099 {
1100 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1101
1102 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1103 }
1104
1105 static int kvm_read_hva(void *data, void __user *hva, int len)
1106 {
1107 return __copy_from_user(data, hva, len);
1108 }
1109
1110 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1111 {
1112 return __copy_from_user_inatomic(data, hva, len);
1113 }
1114
1115 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1116 unsigned long start, int write, struct page **page)
1117 {
1118 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1119
1120 if (write)
1121 flags |= FOLL_WRITE;
1122
1123 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1124 }
1125
1126 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1127 unsigned long addr, bool write_fault,
1128 struct page **pagep)
1129 {
1130 int npages;
1131 int locked = 1;
1132 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1133 (pagep ? FOLL_GET : 0) |
1134 (write_fault ? FOLL_WRITE : 0);
1135
1136 /*
1137 * If retrying the fault, we get here *not* having allowed the filemap
1138 * to wait on the page lock. We should now allow waiting on the IO with
1139 * the mmap semaphore released.
1140 */
1141 down_read(&mm->mmap_sem);
1142 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1143 &locked);
1144 if (!locked) {
1145 VM_BUG_ON(npages);
1146
1147 if (!pagep)
1148 return 0;
1149
1150 /*
1151 * The previous call has now waited on the IO. Now we can
1152 * retry and complete. Pass TRIED to ensure we do not re
1153 * schedule async IO (see e.g. filemap_fault).
1154 */
1155 down_read(&mm->mmap_sem);
1156 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1157 pagep, NULL, NULL);
1158 }
1159 up_read(&mm->mmap_sem);
1160 return npages;
1161 }
1162
1163 static inline int check_user_page_hwpoison(unsigned long addr)
1164 {
1165 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1166
1167 rc = __get_user_pages(current, current->mm, addr, 1,
1168 flags, NULL, NULL, NULL);
1169 return rc == -EHWPOISON;
1170 }
1171
1172 /*
1173 * The atomic path to get the writable pfn which will be stored in @pfn,
1174 * true indicates success, otherwise false is returned.
1175 */
1176 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1177 bool write_fault, bool *writable, pfn_t *pfn)
1178 {
1179 struct page *page[1];
1180 int npages;
1181
1182 if (!(async || atomic))
1183 return false;
1184
1185 /*
1186 * Fast pin a writable pfn only if it is a write fault request
1187 * or the caller allows to map a writable pfn for a read fault
1188 * request.
1189 */
1190 if (!(write_fault || writable))
1191 return false;
1192
1193 npages = __get_user_pages_fast(addr, 1, 1, page);
1194 if (npages == 1) {
1195 *pfn = page_to_pfn(page[0]);
1196
1197 if (writable)
1198 *writable = true;
1199 return true;
1200 }
1201
1202 return false;
1203 }
1204
1205 /*
1206 * The slow path to get the pfn of the specified host virtual address,
1207 * 1 indicates success, -errno is returned if error is detected.
1208 */
1209 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1210 bool *writable, pfn_t *pfn)
1211 {
1212 struct page *page[1];
1213 int npages = 0;
1214
1215 might_sleep();
1216
1217 if (writable)
1218 *writable = write_fault;
1219
1220 if (async) {
1221 down_read(&current->mm->mmap_sem);
1222 npages = get_user_page_nowait(current, current->mm,
1223 addr, write_fault, page);
1224 up_read(&current->mm->mmap_sem);
1225 } else {
1226 /*
1227 * By now we have tried gup_fast, and possibly async_pf, and we
1228 * are certainly not atomic. Time to retry the gup, allowing
1229 * mmap semaphore to be relinquished in the case of IO.
1230 */
1231 npages = kvm_get_user_page_io(current, current->mm, addr,
1232 write_fault, page);
1233 }
1234 if (npages != 1)
1235 return npages;
1236
1237 /* map read fault as writable if possible */
1238 if (unlikely(!write_fault) && writable) {
1239 struct page *wpage[1];
1240
1241 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1242 if (npages == 1) {
1243 *writable = true;
1244 put_page(page[0]);
1245 page[0] = wpage[0];
1246 }
1247
1248 npages = 1;
1249 }
1250 *pfn = page_to_pfn(page[0]);
1251 return npages;
1252 }
1253
1254 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1255 {
1256 if (unlikely(!(vma->vm_flags & VM_READ)))
1257 return false;
1258
1259 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1260 return false;
1261
1262 return true;
1263 }
1264
1265 /*
1266 * Pin guest page in memory and return its pfn.
1267 * @addr: host virtual address which maps memory to the guest
1268 * @atomic: whether this function can sleep
1269 * @async: whether this function need to wait IO complete if the
1270 * host page is not in the memory
1271 * @write_fault: whether we should get a writable host page
1272 * @writable: whether it allows to map a writable host page for !@write_fault
1273 *
1274 * The function will map a writable host page for these two cases:
1275 * 1): @write_fault = true
1276 * 2): @write_fault = false && @writable, @writable will tell the caller
1277 * whether the mapping is writable.
1278 */
1279 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1280 bool write_fault, bool *writable)
1281 {
1282 struct vm_area_struct *vma;
1283 pfn_t pfn = 0;
1284 int npages;
1285
1286 /* we can do it either atomically or asynchronously, not both */
1287 BUG_ON(atomic && async);
1288
1289 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1290 return pfn;
1291
1292 if (atomic)
1293 return KVM_PFN_ERR_FAULT;
1294
1295 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1296 if (npages == 1)
1297 return pfn;
1298
1299 down_read(&current->mm->mmap_sem);
1300 if (npages == -EHWPOISON ||
1301 (!async && check_user_page_hwpoison(addr))) {
1302 pfn = KVM_PFN_ERR_HWPOISON;
1303 goto exit;
1304 }
1305
1306 vma = find_vma_intersection(current->mm, addr, addr + 1);
1307
1308 if (vma == NULL)
1309 pfn = KVM_PFN_ERR_FAULT;
1310 else if ((vma->vm_flags & VM_PFNMAP)) {
1311 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1312 vma->vm_pgoff;
1313 BUG_ON(!kvm_is_mmio_pfn(pfn));
1314 } else {
1315 if (async && vma_is_valid(vma, write_fault))
1316 *async = true;
1317 pfn = KVM_PFN_ERR_FAULT;
1318 }
1319 exit:
1320 up_read(&current->mm->mmap_sem);
1321 return pfn;
1322 }
1323
1324 static pfn_t
1325 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1326 bool *async, bool write_fault, bool *writable)
1327 {
1328 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1329
1330 if (addr == KVM_HVA_ERR_RO_BAD)
1331 return KVM_PFN_ERR_RO_FAULT;
1332
1333 if (kvm_is_error_hva(addr))
1334 return KVM_PFN_NOSLOT;
1335
1336 /* Do not map writable pfn in the readonly memslot. */
1337 if (writable && memslot_is_readonly(slot)) {
1338 *writable = false;
1339 writable = NULL;
1340 }
1341
1342 return hva_to_pfn(addr, atomic, async, write_fault,
1343 writable);
1344 }
1345
1346 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1347 bool write_fault, bool *writable)
1348 {
1349 struct kvm_memory_slot *slot;
1350
1351 if (async)
1352 *async = false;
1353
1354 slot = gfn_to_memslot(kvm, gfn);
1355
1356 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1357 writable);
1358 }
1359
1360 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1361 {
1362 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1363 }
1364 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1365
1366 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1367 bool write_fault, bool *writable)
1368 {
1369 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1370 }
1371 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1372
1373 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1374 {
1375 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1376 }
1377 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1378
1379 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1380 bool *writable)
1381 {
1382 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1383 }
1384 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1385
1386 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1387 {
1388 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1389 }
1390
1391 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1392 {
1393 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1394 }
1395 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1396
1397 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1398 int nr_pages)
1399 {
1400 unsigned long addr;
1401 gfn_t entry;
1402
1403 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1404 if (kvm_is_error_hva(addr))
1405 return -1;
1406
1407 if (entry < nr_pages)
1408 return 0;
1409
1410 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1411 }
1412 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1413
1414 static struct page *kvm_pfn_to_page(pfn_t pfn)
1415 {
1416 if (is_error_noslot_pfn(pfn))
1417 return KVM_ERR_PTR_BAD_PAGE;
1418
1419 if (kvm_is_mmio_pfn(pfn)) {
1420 WARN_ON(1);
1421 return KVM_ERR_PTR_BAD_PAGE;
1422 }
1423
1424 return pfn_to_page(pfn);
1425 }
1426
1427 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1428 {
1429 pfn_t pfn;
1430
1431 pfn = gfn_to_pfn(kvm, gfn);
1432
1433 return kvm_pfn_to_page(pfn);
1434 }
1435
1436 EXPORT_SYMBOL_GPL(gfn_to_page);
1437
1438 void kvm_release_page_clean(struct page *page)
1439 {
1440 WARN_ON(is_error_page(page));
1441
1442 kvm_release_pfn_clean(page_to_pfn(page));
1443 }
1444 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1445
1446 void kvm_release_pfn_clean(pfn_t pfn)
1447 {
1448 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1449 put_page(pfn_to_page(pfn));
1450 }
1451 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1452
1453 void kvm_release_page_dirty(struct page *page)
1454 {
1455 WARN_ON(is_error_page(page));
1456
1457 kvm_release_pfn_dirty(page_to_pfn(page));
1458 }
1459 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1460
1461 static void kvm_release_pfn_dirty(pfn_t pfn)
1462 {
1463 kvm_set_pfn_dirty(pfn);
1464 kvm_release_pfn_clean(pfn);
1465 }
1466
1467 void kvm_set_pfn_dirty(pfn_t pfn)
1468 {
1469 if (!kvm_is_mmio_pfn(pfn)) {
1470 struct page *page = pfn_to_page(pfn);
1471 if (!PageReserved(page))
1472 SetPageDirty(page);
1473 }
1474 }
1475 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1476
1477 void kvm_set_pfn_accessed(pfn_t pfn)
1478 {
1479 if (!kvm_is_mmio_pfn(pfn))
1480 mark_page_accessed(pfn_to_page(pfn));
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1483
1484 void kvm_get_pfn(pfn_t pfn)
1485 {
1486 if (!kvm_is_mmio_pfn(pfn))
1487 get_page(pfn_to_page(pfn));
1488 }
1489 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1490
1491 static int next_segment(unsigned long len, int offset)
1492 {
1493 if (len > PAGE_SIZE - offset)
1494 return PAGE_SIZE - offset;
1495 else
1496 return len;
1497 }
1498
1499 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1500 int len)
1501 {
1502 int r;
1503 unsigned long addr;
1504
1505 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1506 if (kvm_is_error_hva(addr))
1507 return -EFAULT;
1508 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1509 if (r)
1510 return -EFAULT;
1511 return 0;
1512 }
1513 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1514
1515 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1516 {
1517 gfn_t gfn = gpa >> PAGE_SHIFT;
1518 int seg;
1519 int offset = offset_in_page(gpa);
1520 int ret;
1521
1522 while ((seg = next_segment(len, offset)) != 0) {
1523 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1524 if (ret < 0)
1525 return ret;
1526 offset = 0;
1527 len -= seg;
1528 data += seg;
1529 ++gfn;
1530 }
1531 return 0;
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_read_guest);
1534
1535 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1536 unsigned long len)
1537 {
1538 int r;
1539 unsigned long addr;
1540 gfn_t gfn = gpa >> PAGE_SHIFT;
1541 int offset = offset_in_page(gpa);
1542
1543 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1544 if (kvm_is_error_hva(addr))
1545 return -EFAULT;
1546 pagefault_disable();
1547 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1548 pagefault_enable();
1549 if (r)
1550 return -EFAULT;
1551 return 0;
1552 }
1553 EXPORT_SYMBOL(kvm_read_guest_atomic);
1554
1555 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1556 int offset, int len)
1557 {
1558 int r;
1559 unsigned long addr;
1560
1561 addr = gfn_to_hva(kvm, gfn);
1562 if (kvm_is_error_hva(addr))
1563 return -EFAULT;
1564 r = __copy_to_user((void __user *)addr + offset, data, len);
1565 if (r)
1566 return -EFAULT;
1567 mark_page_dirty(kvm, gfn);
1568 return 0;
1569 }
1570 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1571
1572 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1573 unsigned long len)
1574 {
1575 gfn_t gfn = gpa >> PAGE_SHIFT;
1576 int seg;
1577 int offset = offset_in_page(gpa);
1578 int ret;
1579
1580 while ((seg = next_segment(len, offset)) != 0) {
1581 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1582 if (ret < 0)
1583 return ret;
1584 offset = 0;
1585 len -= seg;
1586 data += seg;
1587 ++gfn;
1588 }
1589 return 0;
1590 }
1591
1592 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1593 gpa_t gpa, unsigned long len)
1594 {
1595 struct kvm_memslots *slots = kvm_memslots(kvm);
1596 int offset = offset_in_page(gpa);
1597 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1598 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1599 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1600 gfn_t nr_pages_avail;
1601
1602 ghc->gpa = gpa;
1603 ghc->generation = slots->generation;
1604 ghc->len = len;
1605 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1606 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1607 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1608 ghc->hva += offset;
1609 } else {
1610 /*
1611 * If the requested region crosses two memslots, we still
1612 * verify that the entire region is valid here.
1613 */
1614 while (start_gfn <= end_gfn) {
1615 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1616 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1617 &nr_pages_avail);
1618 if (kvm_is_error_hva(ghc->hva))
1619 return -EFAULT;
1620 start_gfn += nr_pages_avail;
1621 }
1622 /* Use the slow path for cross page reads and writes. */
1623 ghc->memslot = NULL;
1624 }
1625 return 0;
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1628
1629 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1630 void *data, unsigned long len)
1631 {
1632 struct kvm_memslots *slots = kvm_memslots(kvm);
1633 int r;
1634
1635 BUG_ON(len > ghc->len);
1636
1637 if (slots->generation != ghc->generation)
1638 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1639
1640 if (unlikely(!ghc->memslot))
1641 return kvm_write_guest(kvm, ghc->gpa, data, len);
1642
1643 if (kvm_is_error_hva(ghc->hva))
1644 return -EFAULT;
1645
1646 r = __copy_to_user((void __user *)ghc->hva, data, len);
1647 if (r)
1648 return -EFAULT;
1649 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1650
1651 return 0;
1652 }
1653 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1654
1655 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1656 void *data, unsigned long len)
1657 {
1658 struct kvm_memslots *slots = kvm_memslots(kvm);
1659 int r;
1660
1661 BUG_ON(len > ghc->len);
1662
1663 if (slots->generation != ghc->generation)
1664 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1665
1666 if (unlikely(!ghc->memslot))
1667 return kvm_read_guest(kvm, ghc->gpa, data, len);
1668
1669 if (kvm_is_error_hva(ghc->hva))
1670 return -EFAULT;
1671
1672 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1673 if (r)
1674 return -EFAULT;
1675
1676 return 0;
1677 }
1678 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1679
1680 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1681 {
1682 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1683
1684 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1685 }
1686 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1687
1688 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1689 {
1690 gfn_t gfn = gpa >> PAGE_SHIFT;
1691 int seg;
1692 int offset = offset_in_page(gpa);
1693 int ret;
1694
1695 while ((seg = next_segment(len, offset)) != 0) {
1696 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1697 if (ret < 0)
1698 return ret;
1699 offset = 0;
1700 len -= seg;
1701 ++gfn;
1702 }
1703 return 0;
1704 }
1705 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1706
1707 static void mark_page_dirty_in_slot(struct kvm *kvm,
1708 struct kvm_memory_slot *memslot,
1709 gfn_t gfn)
1710 {
1711 if (memslot && memslot->dirty_bitmap) {
1712 unsigned long rel_gfn = gfn - memslot->base_gfn;
1713
1714 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1715 }
1716 }
1717
1718 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1719 {
1720 struct kvm_memory_slot *memslot;
1721
1722 memslot = gfn_to_memslot(kvm, gfn);
1723 mark_page_dirty_in_slot(kvm, memslot, gfn);
1724 }
1725 EXPORT_SYMBOL_GPL(mark_page_dirty);
1726
1727 /*
1728 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1729 */
1730 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1731 {
1732 DEFINE_WAIT(wait);
1733
1734 for (;;) {
1735 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1736
1737 if (kvm_arch_vcpu_runnable(vcpu)) {
1738 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1739 break;
1740 }
1741 if (kvm_cpu_has_pending_timer(vcpu))
1742 break;
1743 if (signal_pending(current))
1744 break;
1745
1746 schedule();
1747 }
1748
1749 finish_wait(&vcpu->wq, &wait);
1750 }
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1752
1753 #ifndef CONFIG_S390
1754 /*
1755 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1756 */
1757 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1758 {
1759 int me;
1760 int cpu = vcpu->cpu;
1761 wait_queue_head_t *wqp;
1762
1763 wqp = kvm_arch_vcpu_wq(vcpu);
1764 if (waitqueue_active(wqp)) {
1765 wake_up_interruptible(wqp);
1766 ++vcpu->stat.halt_wakeup;
1767 }
1768
1769 me = get_cpu();
1770 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1771 if (kvm_arch_vcpu_should_kick(vcpu))
1772 smp_send_reschedule(cpu);
1773 put_cpu();
1774 }
1775 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1776 #endif /* !CONFIG_S390 */
1777
1778 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1779 {
1780 struct pid *pid;
1781 struct task_struct *task = NULL;
1782 int ret = 0;
1783
1784 rcu_read_lock();
1785 pid = rcu_dereference(target->pid);
1786 if (pid)
1787 task = get_pid_task(pid, PIDTYPE_PID);
1788 rcu_read_unlock();
1789 if (!task)
1790 return ret;
1791 if (task->flags & PF_VCPU) {
1792 put_task_struct(task);
1793 return ret;
1794 }
1795 ret = yield_to(task, 1);
1796 put_task_struct(task);
1797
1798 return ret;
1799 }
1800 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1801
1802 /*
1803 * Helper that checks whether a VCPU is eligible for directed yield.
1804 * Most eligible candidate to yield is decided by following heuristics:
1805 *
1806 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1807 * (preempted lock holder), indicated by @in_spin_loop.
1808 * Set at the beiginning and cleared at the end of interception/PLE handler.
1809 *
1810 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1811 * chance last time (mostly it has become eligible now since we have probably
1812 * yielded to lockholder in last iteration. This is done by toggling
1813 * @dy_eligible each time a VCPU checked for eligibility.)
1814 *
1815 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1816 * to preempted lock-holder could result in wrong VCPU selection and CPU
1817 * burning. Giving priority for a potential lock-holder increases lock
1818 * progress.
1819 *
1820 * Since algorithm is based on heuristics, accessing another VCPU data without
1821 * locking does not harm. It may result in trying to yield to same VCPU, fail
1822 * and continue with next VCPU and so on.
1823 */
1824 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1825 {
1826 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1827 bool eligible;
1828
1829 eligible = !vcpu->spin_loop.in_spin_loop ||
1830 vcpu->spin_loop.dy_eligible;
1831
1832 if (vcpu->spin_loop.in_spin_loop)
1833 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1834
1835 return eligible;
1836 #else
1837 return true;
1838 #endif
1839 }
1840
1841 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1842 {
1843 struct kvm *kvm = me->kvm;
1844 struct kvm_vcpu *vcpu;
1845 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1846 int yielded = 0;
1847 int try = 3;
1848 int pass;
1849 int i;
1850
1851 kvm_vcpu_set_in_spin_loop(me, true);
1852 /*
1853 * We boost the priority of a VCPU that is runnable but not
1854 * currently running, because it got preempted by something
1855 * else and called schedule in __vcpu_run. Hopefully that
1856 * VCPU is holding the lock that we need and will release it.
1857 * We approximate round-robin by starting at the last boosted VCPU.
1858 */
1859 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1860 kvm_for_each_vcpu(i, vcpu, kvm) {
1861 if (!pass && i <= last_boosted_vcpu) {
1862 i = last_boosted_vcpu;
1863 continue;
1864 } else if (pass && i > last_boosted_vcpu)
1865 break;
1866 if (!ACCESS_ONCE(vcpu->preempted))
1867 continue;
1868 if (vcpu == me)
1869 continue;
1870 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1871 continue;
1872 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1873 continue;
1874
1875 yielded = kvm_vcpu_yield_to(vcpu);
1876 if (yielded > 0) {
1877 kvm->last_boosted_vcpu = i;
1878 break;
1879 } else if (yielded < 0) {
1880 try--;
1881 if (!try)
1882 break;
1883 }
1884 }
1885 }
1886 kvm_vcpu_set_in_spin_loop(me, false);
1887
1888 /* Ensure vcpu is not eligible during next spinloop */
1889 kvm_vcpu_set_dy_eligible(me, false);
1890 }
1891 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1892
1893 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1894 {
1895 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1896 struct page *page;
1897
1898 if (vmf->pgoff == 0)
1899 page = virt_to_page(vcpu->run);
1900 #ifdef CONFIG_X86
1901 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1902 page = virt_to_page(vcpu->arch.pio_data);
1903 #endif
1904 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1905 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1906 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1907 #endif
1908 else
1909 return kvm_arch_vcpu_fault(vcpu, vmf);
1910 get_page(page);
1911 vmf->page = page;
1912 return 0;
1913 }
1914
1915 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1916 .fault = kvm_vcpu_fault,
1917 };
1918
1919 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1920 {
1921 vma->vm_ops = &kvm_vcpu_vm_ops;
1922 return 0;
1923 }
1924
1925 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1926 {
1927 struct kvm_vcpu *vcpu = filp->private_data;
1928
1929 kvm_put_kvm(vcpu->kvm);
1930 return 0;
1931 }
1932
1933 static struct file_operations kvm_vcpu_fops = {
1934 .release = kvm_vcpu_release,
1935 .unlocked_ioctl = kvm_vcpu_ioctl,
1936 #ifdef CONFIG_COMPAT
1937 .compat_ioctl = kvm_vcpu_compat_ioctl,
1938 #endif
1939 .mmap = kvm_vcpu_mmap,
1940 .llseek = noop_llseek,
1941 };
1942
1943 /*
1944 * Allocates an inode for the vcpu.
1945 */
1946 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1947 {
1948 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1949 }
1950
1951 /*
1952 * Creates some virtual cpus. Good luck creating more than one.
1953 */
1954 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1955 {
1956 int r;
1957 struct kvm_vcpu *vcpu, *v;
1958
1959 if (id >= KVM_MAX_VCPUS)
1960 return -EINVAL;
1961
1962 vcpu = kvm_arch_vcpu_create(kvm, id);
1963 if (IS_ERR(vcpu))
1964 return PTR_ERR(vcpu);
1965
1966 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1967
1968 r = kvm_arch_vcpu_setup(vcpu);
1969 if (r)
1970 goto vcpu_destroy;
1971
1972 mutex_lock(&kvm->lock);
1973 if (!kvm_vcpu_compatible(vcpu)) {
1974 r = -EINVAL;
1975 goto unlock_vcpu_destroy;
1976 }
1977 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1978 r = -EINVAL;
1979 goto unlock_vcpu_destroy;
1980 }
1981
1982 kvm_for_each_vcpu(r, v, kvm)
1983 if (v->vcpu_id == id) {
1984 r = -EEXIST;
1985 goto unlock_vcpu_destroy;
1986 }
1987
1988 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1989
1990 /* Now it's all set up, let userspace reach it */
1991 kvm_get_kvm(kvm);
1992 r = create_vcpu_fd(vcpu);
1993 if (r < 0) {
1994 kvm_put_kvm(kvm);
1995 goto unlock_vcpu_destroy;
1996 }
1997
1998 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1999 smp_wmb();
2000 atomic_inc(&kvm->online_vcpus);
2001
2002 mutex_unlock(&kvm->lock);
2003 kvm_arch_vcpu_postcreate(vcpu);
2004 return r;
2005
2006 unlock_vcpu_destroy:
2007 mutex_unlock(&kvm->lock);
2008 vcpu_destroy:
2009 kvm_arch_vcpu_destroy(vcpu);
2010 return r;
2011 }
2012
2013 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2014 {
2015 if (sigset) {
2016 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2017 vcpu->sigset_active = 1;
2018 vcpu->sigset = *sigset;
2019 } else
2020 vcpu->sigset_active = 0;
2021 return 0;
2022 }
2023
2024 static long kvm_vcpu_ioctl(struct file *filp,
2025 unsigned int ioctl, unsigned long arg)
2026 {
2027 struct kvm_vcpu *vcpu = filp->private_data;
2028 void __user *argp = (void __user *)arg;
2029 int r;
2030 struct kvm_fpu *fpu = NULL;
2031 struct kvm_sregs *kvm_sregs = NULL;
2032
2033 if (vcpu->kvm->mm != current->mm)
2034 return -EIO;
2035
2036 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2037 return -EINVAL;
2038
2039 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2040 /*
2041 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2042 * so vcpu_load() would break it.
2043 */
2044 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2045 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2046 #endif
2047
2048
2049 r = vcpu_load(vcpu);
2050 if (r)
2051 return r;
2052 switch (ioctl) {
2053 case KVM_RUN:
2054 r = -EINVAL;
2055 if (arg)
2056 goto out;
2057 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2058 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2059 break;
2060 case KVM_GET_REGS: {
2061 struct kvm_regs *kvm_regs;
2062
2063 r = -ENOMEM;
2064 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2065 if (!kvm_regs)
2066 goto out;
2067 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2068 if (r)
2069 goto out_free1;
2070 r = -EFAULT;
2071 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2072 goto out_free1;
2073 r = 0;
2074 out_free1:
2075 kfree(kvm_regs);
2076 break;
2077 }
2078 case KVM_SET_REGS: {
2079 struct kvm_regs *kvm_regs;
2080
2081 r = -ENOMEM;
2082 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2083 if (IS_ERR(kvm_regs)) {
2084 r = PTR_ERR(kvm_regs);
2085 goto out;
2086 }
2087 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2088 kfree(kvm_regs);
2089 break;
2090 }
2091 case KVM_GET_SREGS: {
2092 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2093 r = -ENOMEM;
2094 if (!kvm_sregs)
2095 goto out;
2096 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2097 if (r)
2098 goto out;
2099 r = -EFAULT;
2100 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2101 goto out;
2102 r = 0;
2103 break;
2104 }
2105 case KVM_SET_SREGS: {
2106 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2107 if (IS_ERR(kvm_sregs)) {
2108 r = PTR_ERR(kvm_sregs);
2109 kvm_sregs = NULL;
2110 goto out;
2111 }
2112 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2113 break;
2114 }
2115 case KVM_GET_MP_STATE: {
2116 struct kvm_mp_state mp_state;
2117
2118 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2119 if (r)
2120 goto out;
2121 r = -EFAULT;
2122 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2123 goto out;
2124 r = 0;
2125 break;
2126 }
2127 case KVM_SET_MP_STATE: {
2128 struct kvm_mp_state mp_state;
2129
2130 r = -EFAULT;
2131 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2132 goto out;
2133 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2134 break;
2135 }
2136 case KVM_TRANSLATE: {
2137 struct kvm_translation tr;
2138
2139 r = -EFAULT;
2140 if (copy_from_user(&tr, argp, sizeof tr))
2141 goto out;
2142 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2143 if (r)
2144 goto out;
2145 r = -EFAULT;
2146 if (copy_to_user(argp, &tr, sizeof tr))
2147 goto out;
2148 r = 0;
2149 break;
2150 }
2151 case KVM_SET_GUEST_DEBUG: {
2152 struct kvm_guest_debug dbg;
2153
2154 r = -EFAULT;
2155 if (copy_from_user(&dbg, argp, sizeof dbg))
2156 goto out;
2157 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2158 break;
2159 }
2160 case KVM_SET_SIGNAL_MASK: {
2161 struct kvm_signal_mask __user *sigmask_arg = argp;
2162 struct kvm_signal_mask kvm_sigmask;
2163 sigset_t sigset, *p;
2164
2165 p = NULL;
2166 if (argp) {
2167 r = -EFAULT;
2168 if (copy_from_user(&kvm_sigmask, argp,
2169 sizeof kvm_sigmask))
2170 goto out;
2171 r = -EINVAL;
2172 if (kvm_sigmask.len != sizeof sigset)
2173 goto out;
2174 r = -EFAULT;
2175 if (copy_from_user(&sigset, sigmask_arg->sigset,
2176 sizeof sigset))
2177 goto out;
2178 p = &sigset;
2179 }
2180 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2181 break;
2182 }
2183 case KVM_GET_FPU: {
2184 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2185 r = -ENOMEM;
2186 if (!fpu)
2187 goto out;
2188 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2189 if (r)
2190 goto out;
2191 r = -EFAULT;
2192 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2193 goto out;
2194 r = 0;
2195 break;
2196 }
2197 case KVM_SET_FPU: {
2198 fpu = memdup_user(argp, sizeof(*fpu));
2199 if (IS_ERR(fpu)) {
2200 r = PTR_ERR(fpu);
2201 fpu = NULL;
2202 goto out;
2203 }
2204 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2205 break;
2206 }
2207 default:
2208 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2209 }
2210 out:
2211 vcpu_put(vcpu);
2212 kfree(fpu);
2213 kfree(kvm_sregs);
2214 return r;
2215 }
2216
2217 #ifdef CONFIG_COMPAT
2218 static long kvm_vcpu_compat_ioctl(struct file *filp,
2219 unsigned int ioctl, unsigned long arg)
2220 {
2221 struct kvm_vcpu *vcpu = filp->private_data;
2222 void __user *argp = compat_ptr(arg);
2223 int r;
2224
2225 if (vcpu->kvm->mm != current->mm)
2226 return -EIO;
2227
2228 switch (ioctl) {
2229 case KVM_SET_SIGNAL_MASK: {
2230 struct kvm_signal_mask __user *sigmask_arg = argp;
2231 struct kvm_signal_mask kvm_sigmask;
2232 compat_sigset_t csigset;
2233 sigset_t sigset;
2234
2235 if (argp) {
2236 r = -EFAULT;
2237 if (copy_from_user(&kvm_sigmask, argp,
2238 sizeof kvm_sigmask))
2239 goto out;
2240 r = -EINVAL;
2241 if (kvm_sigmask.len != sizeof csigset)
2242 goto out;
2243 r = -EFAULT;
2244 if (copy_from_user(&csigset, sigmask_arg->sigset,
2245 sizeof csigset))
2246 goto out;
2247 sigset_from_compat(&sigset, &csigset);
2248 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2249 } else
2250 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2251 break;
2252 }
2253 default:
2254 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2255 }
2256
2257 out:
2258 return r;
2259 }
2260 #endif
2261
2262 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2263 int (*accessor)(struct kvm_device *dev,
2264 struct kvm_device_attr *attr),
2265 unsigned long arg)
2266 {
2267 struct kvm_device_attr attr;
2268
2269 if (!accessor)
2270 return -EPERM;
2271
2272 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2273 return -EFAULT;
2274
2275 return accessor(dev, &attr);
2276 }
2277
2278 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2279 unsigned long arg)
2280 {
2281 struct kvm_device *dev = filp->private_data;
2282
2283 switch (ioctl) {
2284 case KVM_SET_DEVICE_ATTR:
2285 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2286 case KVM_GET_DEVICE_ATTR:
2287 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2288 case KVM_HAS_DEVICE_ATTR:
2289 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2290 default:
2291 if (dev->ops->ioctl)
2292 return dev->ops->ioctl(dev, ioctl, arg);
2293
2294 return -ENOTTY;
2295 }
2296 }
2297
2298 static int kvm_device_release(struct inode *inode, struct file *filp)
2299 {
2300 struct kvm_device *dev = filp->private_data;
2301 struct kvm *kvm = dev->kvm;
2302
2303 kvm_put_kvm(kvm);
2304 return 0;
2305 }
2306
2307 static const struct file_operations kvm_device_fops = {
2308 .unlocked_ioctl = kvm_device_ioctl,
2309 #ifdef CONFIG_COMPAT
2310 .compat_ioctl = kvm_device_ioctl,
2311 #endif
2312 .release = kvm_device_release,
2313 };
2314
2315 struct kvm_device *kvm_device_from_filp(struct file *filp)
2316 {
2317 if (filp->f_op != &kvm_device_fops)
2318 return NULL;
2319
2320 return filp->private_data;
2321 }
2322
2323 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2324 #ifdef CONFIG_KVM_MPIC
2325 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2326 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2327 #endif
2328
2329 #ifdef CONFIG_KVM_XICS
2330 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2331 #endif
2332 };
2333
2334 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2335 {
2336 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2337 return -ENOSPC;
2338
2339 if (kvm_device_ops_table[type] != NULL)
2340 return -EEXIST;
2341
2342 kvm_device_ops_table[type] = ops;
2343 return 0;
2344 }
2345
2346 void kvm_unregister_device_ops(u32 type)
2347 {
2348 if (kvm_device_ops_table[type] != NULL)
2349 kvm_device_ops_table[type] = NULL;
2350 }
2351
2352 static int kvm_ioctl_create_device(struct kvm *kvm,
2353 struct kvm_create_device *cd)
2354 {
2355 struct kvm_device_ops *ops = NULL;
2356 struct kvm_device *dev;
2357 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2358 int ret;
2359
2360 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2361 return -ENODEV;
2362
2363 ops = kvm_device_ops_table[cd->type];
2364 if (ops == NULL)
2365 return -ENODEV;
2366
2367 if (test)
2368 return 0;
2369
2370 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2371 if (!dev)
2372 return -ENOMEM;
2373
2374 dev->ops = ops;
2375 dev->kvm = kvm;
2376
2377 ret = ops->create(dev, cd->type);
2378 if (ret < 0) {
2379 kfree(dev);
2380 return ret;
2381 }
2382
2383 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2384 if (ret < 0) {
2385 ops->destroy(dev);
2386 return ret;
2387 }
2388
2389 list_add(&dev->vm_node, &kvm->devices);
2390 kvm_get_kvm(kvm);
2391 cd->fd = ret;
2392 return 0;
2393 }
2394
2395 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2396 {
2397 switch (arg) {
2398 case KVM_CAP_USER_MEMORY:
2399 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2400 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2401 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2402 case KVM_CAP_SET_BOOT_CPU_ID:
2403 #endif
2404 case KVM_CAP_INTERNAL_ERROR_DATA:
2405 #ifdef CONFIG_HAVE_KVM_MSI
2406 case KVM_CAP_SIGNAL_MSI:
2407 #endif
2408 #ifdef CONFIG_HAVE_KVM_IRQFD
2409 case KVM_CAP_IRQFD_RESAMPLE:
2410 #endif
2411 case KVM_CAP_CHECK_EXTENSION_VM:
2412 return 1;
2413 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2414 case KVM_CAP_IRQ_ROUTING:
2415 return KVM_MAX_IRQ_ROUTES;
2416 #endif
2417 default:
2418 break;
2419 }
2420 return kvm_vm_ioctl_check_extension(kvm, arg);
2421 }
2422
2423 static long kvm_vm_ioctl(struct file *filp,
2424 unsigned int ioctl, unsigned long arg)
2425 {
2426 struct kvm *kvm = filp->private_data;
2427 void __user *argp = (void __user *)arg;
2428 int r;
2429
2430 if (kvm->mm != current->mm)
2431 return -EIO;
2432 switch (ioctl) {
2433 case KVM_CREATE_VCPU:
2434 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2435 break;
2436 case KVM_SET_USER_MEMORY_REGION: {
2437 struct kvm_userspace_memory_region kvm_userspace_mem;
2438
2439 r = -EFAULT;
2440 if (copy_from_user(&kvm_userspace_mem, argp,
2441 sizeof kvm_userspace_mem))
2442 goto out;
2443
2444 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2445 break;
2446 }
2447 case KVM_GET_DIRTY_LOG: {
2448 struct kvm_dirty_log log;
2449
2450 r = -EFAULT;
2451 if (copy_from_user(&log, argp, sizeof log))
2452 goto out;
2453 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2454 break;
2455 }
2456 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2457 case KVM_REGISTER_COALESCED_MMIO: {
2458 struct kvm_coalesced_mmio_zone zone;
2459 r = -EFAULT;
2460 if (copy_from_user(&zone, argp, sizeof zone))
2461 goto out;
2462 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2463 break;
2464 }
2465 case KVM_UNREGISTER_COALESCED_MMIO: {
2466 struct kvm_coalesced_mmio_zone zone;
2467 r = -EFAULT;
2468 if (copy_from_user(&zone, argp, sizeof zone))
2469 goto out;
2470 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2471 break;
2472 }
2473 #endif
2474 case KVM_IRQFD: {
2475 struct kvm_irqfd data;
2476
2477 r = -EFAULT;
2478 if (copy_from_user(&data, argp, sizeof data))
2479 goto out;
2480 r = kvm_irqfd(kvm, &data);
2481 break;
2482 }
2483 case KVM_IOEVENTFD: {
2484 struct kvm_ioeventfd data;
2485
2486 r = -EFAULT;
2487 if (copy_from_user(&data, argp, sizeof data))
2488 goto out;
2489 r = kvm_ioeventfd(kvm, &data);
2490 break;
2491 }
2492 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2493 case KVM_SET_BOOT_CPU_ID:
2494 r = 0;
2495 mutex_lock(&kvm->lock);
2496 if (atomic_read(&kvm->online_vcpus) != 0)
2497 r = -EBUSY;
2498 else
2499 kvm->bsp_vcpu_id = arg;
2500 mutex_unlock(&kvm->lock);
2501 break;
2502 #endif
2503 #ifdef CONFIG_HAVE_KVM_MSI
2504 case KVM_SIGNAL_MSI: {
2505 struct kvm_msi msi;
2506
2507 r = -EFAULT;
2508 if (copy_from_user(&msi, argp, sizeof msi))
2509 goto out;
2510 r = kvm_send_userspace_msi(kvm, &msi);
2511 break;
2512 }
2513 #endif
2514 #ifdef __KVM_HAVE_IRQ_LINE
2515 case KVM_IRQ_LINE_STATUS:
2516 case KVM_IRQ_LINE: {
2517 struct kvm_irq_level irq_event;
2518
2519 r = -EFAULT;
2520 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2521 goto out;
2522
2523 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2524 ioctl == KVM_IRQ_LINE_STATUS);
2525 if (r)
2526 goto out;
2527
2528 r = -EFAULT;
2529 if (ioctl == KVM_IRQ_LINE_STATUS) {
2530 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2531 goto out;
2532 }
2533
2534 r = 0;
2535 break;
2536 }
2537 #endif
2538 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2539 case KVM_SET_GSI_ROUTING: {
2540 struct kvm_irq_routing routing;
2541 struct kvm_irq_routing __user *urouting;
2542 struct kvm_irq_routing_entry *entries;
2543
2544 r = -EFAULT;
2545 if (copy_from_user(&routing, argp, sizeof(routing)))
2546 goto out;
2547 r = -EINVAL;
2548 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2549 goto out;
2550 if (routing.flags)
2551 goto out;
2552 r = -ENOMEM;
2553 entries = vmalloc(routing.nr * sizeof(*entries));
2554 if (!entries)
2555 goto out;
2556 r = -EFAULT;
2557 urouting = argp;
2558 if (copy_from_user(entries, urouting->entries,
2559 routing.nr * sizeof(*entries)))
2560 goto out_free_irq_routing;
2561 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2562 routing.flags);
2563 out_free_irq_routing:
2564 vfree(entries);
2565 break;
2566 }
2567 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2568 case KVM_CREATE_DEVICE: {
2569 struct kvm_create_device cd;
2570
2571 r = -EFAULT;
2572 if (copy_from_user(&cd, argp, sizeof(cd)))
2573 goto out;
2574
2575 r = kvm_ioctl_create_device(kvm, &cd);
2576 if (r)
2577 goto out;
2578
2579 r = -EFAULT;
2580 if (copy_to_user(argp, &cd, sizeof(cd)))
2581 goto out;
2582
2583 r = 0;
2584 break;
2585 }
2586 case KVM_CHECK_EXTENSION:
2587 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2588 break;
2589 default:
2590 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2591 }
2592 out:
2593 return r;
2594 }
2595
2596 #ifdef CONFIG_COMPAT
2597 struct compat_kvm_dirty_log {
2598 __u32 slot;
2599 __u32 padding1;
2600 union {
2601 compat_uptr_t dirty_bitmap; /* one bit per page */
2602 __u64 padding2;
2603 };
2604 };
2605
2606 static long kvm_vm_compat_ioctl(struct file *filp,
2607 unsigned int ioctl, unsigned long arg)
2608 {
2609 struct kvm *kvm = filp->private_data;
2610 int r;
2611
2612 if (kvm->mm != current->mm)
2613 return -EIO;
2614 switch (ioctl) {
2615 case KVM_GET_DIRTY_LOG: {
2616 struct compat_kvm_dirty_log compat_log;
2617 struct kvm_dirty_log log;
2618
2619 r = -EFAULT;
2620 if (copy_from_user(&compat_log, (void __user *)arg,
2621 sizeof(compat_log)))
2622 goto out;
2623 log.slot = compat_log.slot;
2624 log.padding1 = compat_log.padding1;
2625 log.padding2 = compat_log.padding2;
2626 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2627
2628 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2629 break;
2630 }
2631 default:
2632 r = kvm_vm_ioctl(filp, ioctl, arg);
2633 }
2634
2635 out:
2636 return r;
2637 }
2638 #endif
2639
2640 static struct file_operations kvm_vm_fops = {
2641 .release = kvm_vm_release,
2642 .unlocked_ioctl = kvm_vm_ioctl,
2643 #ifdef CONFIG_COMPAT
2644 .compat_ioctl = kvm_vm_compat_ioctl,
2645 #endif
2646 .llseek = noop_llseek,
2647 };
2648
2649 static int kvm_dev_ioctl_create_vm(unsigned long type)
2650 {
2651 int r;
2652 struct kvm *kvm;
2653
2654 kvm = kvm_create_vm(type);
2655 if (IS_ERR(kvm))
2656 return PTR_ERR(kvm);
2657 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2658 r = kvm_coalesced_mmio_init(kvm);
2659 if (r < 0) {
2660 kvm_put_kvm(kvm);
2661 return r;
2662 }
2663 #endif
2664 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2665 if (r < 0)
2666 kvm_put_kvm(kvm);
2667
2668 return r;
2669 }
2670
2671 static long kvm_dev_ioctl(struct file *filp,
2672 unsigned int ioctl, unsigned long arg)
2673 {
2674 long r = -EINVAL;
2675
2676 switch (ioctl) {
2677 case KVM_GET_API_VERSION:
2678 if (arg)
2679 goto out;
2680 r = KVM_API_VERSION;
2681 break;
2682 case KVM_CREATE_VM:
2683 r = kvm_dev_ioctl_create_vm(arg);
2684 break;
2685 case KVM_CHECK_EXTENSION:
2686 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2687 break;
2688 case KVM_GET_VCPU_MMAP_SIZE:
2689 if (arg)
2690 goto out;
2691 r = PAGE_SIZE; /* struct kvm_run */
2692 #ifdef CONFIG_X86
2693 r += PAGE_SIZE; /* pio data page */
2694 #endif
2695 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2696 r += PAGE_SIZE; /* coalesced mmio ring page */
2697 #endif
2698 break;
2699 case KVM_TRACE_ENABLE:
2700 case KVM_TRACE_PAUSE:
2701 case KVM_TRACE_DISABLE:
2702 r = -EOPNOTSUPP;
2703 break;
2704 default:
2705 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2706 }
2707 out:
2708 return r;
2709 }
2710
2711 static struct file_operations kvm_chardev_ops = {
2712 .unlocked_ioctl = kvm_dev_ioctl,
2713 .compat_ioctl = kvm_dev_ioctl,
2714 .llseek = noop_llseek,
2715 };
2716
2717 static struct miscdevice kvm_dev = {
2718 KVM_MINOR,
2719 "kvm",
2720 &kvm_chardev_ops,
2721 };
2722
2723 static void hardware_enable_nolock(void *junk)
2724 {
2725 int cpu = raw_smp_processor_id();
2726 int r;
2727
2728 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2729 return;
2730
2731 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2732
2733 r = kvm_arch_hardware_enable();
2734
2735 if (r) {
2736 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2737 atomic_inc(&hardware_enable_failed);
2738 printk(KERN_INFO "kvm: enabling virtualization on "
2739 "CPU%d failed\n", cpu);
2740 }
2741 }
2742
2743 static void hardware_enable(void)
2744 {
2745 raw_spin_lock(&kvm_count_lock);
2746 if (kvm_usage_count)
2747 hardware_enable_nolock(NULL);
2748 raw_spin_unlock(&kvm_count_lock);
2749 }
2750
2751 static void hardware_disable_nolock(void *junk)
2752 {
2753 int cpu = raw_smp_processor_id();
2754
2755 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2756 return;
2757 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2758 kvm_arch_hardware_disable();
2759 }
2760
2761 static void hardware_disable(void)
2762 {
2763 raw_spin_lock(&kvm_count_lock);
2764 if (kvm_usage_count)
2765 hardware_disable_nolock(NULL);
2766 raw_spin_unlock(&kvm_count_lock);
2767 }
2768
2769 static void hardware_disable_all_nolock(void)
2770 {
2771 BUG_ON(!kvm_usage_count);
2772
2773 kvm_usage_count--;
2774 if (!kvm_usage_count)
2775 on_each_cpu(hardware_disable_nolock, NULL, 1);
2776 }
2777
2778 static void hardware_disable_all(void)
2779 {
2780 raw_spin_lock(&kvm_count_lock);
2781 hardware_disable_all_nolock();
2782 raw_spin_unlock(&kvm_count_lock);
2783 }
2784
2785 static int hardware_enable_all(void)
2786 {
2787 int r = 0;
2788
2789 raw_spin_lock(&kvm_count_lock);
2790
2791 kvm_usage_count++;
2792 if (kvm_usage_count == 1) {
2793 atomic_set(&hardware_enable_failed, 0);
2794 on_each_cpu(hardware_enable_nolock, NULL, 1);
2795
2796 if (atomic_read(&hardware_enable_failed)) {
2797 hardware_disable_all_nolock();
2798 r = -EBUSY;
2799 }
2800 }
2801
2802 raw_spin_unlock(&kvm_count_lock);
2803
2804 return r;
2805 }
2806
2807 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2808 void *v)
2809 {
2810 int cpu = (long)v;
2811
2812 val &= ~CPU_TASKS_FROZEN;
2813 switch (val) {
2814 case CPU_DYING:
2815 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2816 cpu);
2817 hardware_disable();
2818 break;
2819 case CPU_STARTING:
2820 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2821 cpu);
2822 hardware_enable();
2823 break;
2824 }
2825 return NOTIFY_OK;
2826 }
2827
2828 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2829 void *v)
2830 {
2831 /*
2832 * Some (well, at least mine) BIOSes hang on reboot if
2833 * in vmx root mode.
2834 *
2835 * And Intel TXT required VMX off for all cpu when system shutdown.
2836 */
2837 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2838 kvm_rebooting = true;
2839 on_each_cpu(hardware_disable_nolock, NULL, 1);
2840 return NOTIFY_OK;
2841 }
2842
2843 static struct notifier_block kvm_reboot_notifier = {
2844 .notifier_call = kvm_reboot,
2845 .priority = 0,
2846 };
2847
2848 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2849 {
2850 int i;
2851
2852 for (i = 0; i < bus->dev_count; i++) {
2853 struct kvm_io_device *pos = bus->range[i].dev;
2854
2855 kvm_iodevice_destructor(pos);
2856 }
2857 kfree(bus);
2858 }
2859
2860 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2861 const struct kvm_io_range *r2)
2862 {
2863 if (r1->addr < r2->addr)
2864 return -1;
2865 if (r1->addr + r1->len > r2->addr + r2->len)
2866 return 1;
2867 return 0;
2868 }
2869
2870 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2871 {
2872 return kvm_io_bus_cmp(p1, p2);
2873 }
2874
2875 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2876 gpa_t addr, int len)
2877 {
2878 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2879 .addr = addr,
2880 .len = len,
2881 .dev = dev,
2882 };
2883
2884 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2885 kvm_io_bus_sort_cmp, NULL);
2886
2887 return 0;
2888 }
2889
2890 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2891 gpa_t addr, int len)
2892 {
2893 struct kvm_io_range *range, key;
2894 int off;
2895
2896 key = (struct kvm_io_range) {
2897 .addr = addr,
2898 .len = len,
2899 };
2900
2901 range = bsearch(&key, bus->range, bus->dev_count,
2902 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2903 if (range == NULL)
2904 return -ENOENT;
2905
2906 off = range - bus->range;
2907
2908 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2909 off--;
2910
2911 return off;
2912 }
2913
2914 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2915 struct kvm_io_range *range, const void *val)
2916 {
2917 int idx;
2918
2919 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2920 if (idx < 0)
2921 return -EOPNOTSUPP;
2922
2923 while (idx < bus->dev_count &&
2924 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2925 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2926 range->len, val))
2927 return idx;
2928 idx++;
2929 }
2930
2931 return -EOPNOTSUPP;
2932 }
2933
2934 /* kvm_io_bus_write - called under kvm->slots_lock */
2935 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2936 int len, const void *val)
2937 {
2938 struct kvm_io_bus *bus;
2939 struct kvm_io_range range;
2940 int r;
2941
2942 range = (struct kvm_io_range) {
2943 .addr = addr,
2944 .len = len,
2945 };
2946
2947 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2948 r = __kvm_io_bus_write(bus, &range, val);
2949 return r < 0 ? r : 0;
2950 }
2951
2952 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2953 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2954 int len, const void *val, long cookie)
2955 {
2956 struct kvm_io_bus *bus;
2957 struct kvm_io_range range;
2958
2959 range = (struct kvm_io_range) {
2960 .addr = addr,
2961 .len = len,
2962 };
2963
2964 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2965
2966 /* First try the device referenced by cookie. */
2967 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2968 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2969 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2970 val))
2971 return cookie;
2972
2973 /*
2974 * cookie contained garbage; fall back to search and return the
2975 * correct cookie value.
2976 */
2977 return __kvm_io_bus_write(bus, &range, val);
2978 }
2979
2980 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2981 void *val)
2982 {
2983 int idx;
2984
2985 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2986 if (idx < 0)
2987 return -EOPNOTSUPP;
2988
2989 while (idx < bus->dev_count &&
2990 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2991 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2992 range->len, val))
2993 return idx;
2994 idx++;
2995 }
2996
2997 return -EOPNOTSUPP;
2998 }
2999 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3000
3001 /* kvm_io_bus_read - called under kvm->slots_lock */
3002 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3003 int len, void *val)
3004 {
3005 struct kvm_io_bus *bus;
3006 struct kvm_io_range range;
3007 int r;
3008
3009 range = (struct kvm_io_range) {
3010 .addr = addr,
3011 .len = len,
3012 };
3013
3014 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3015 r = __kvm_io_bus_read(bus, &range, val);
3016 return r < 0 ? r : 0;
3017 }
3018
3019
3020 /* Caller must hold slots_lock. */
3021 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3022 int len, struct kvm_io_device *dev)
3023 {
3024 struct kvm_io_bus *new_bus, *bus;
3025
3026 bus = kvm->buses[bus_idx];
3027 /* exclude ioeventfd which is limited by maximum fd */
3028 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3029 return -ENOSPC;
3030
3031 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3032 sizeof(struct kvm_io_range)), GFP_KERNEL);
3033 if (!new_bus)
3034 return -ENOMEM;
3035 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3036 sizeof(struct kvm_io_range)));
3037 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3038 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3039 synchronize_srcu_expedited(&kvm->srcu);
3040 kfree(bus);
3041
3042 return 0;
3043 }
3044
3045 /* Caller must hold slots_lock. */
3046 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3047 struct kvm_io_device *dev)
3048 {
3049 int i, r;
3050 struct kvm_io_bus *new_bus, *bus;
3051
3052 bus = kvm->buses[bus_idx];
3053 r = -ENOENT;
3054 for (i = 0; i < bus->dev_count; i++)
3055 if (bus->range[i].dev == dev) {
3056 r = 0;
3057 break;
3058 }
3059
3060 if (r)
3061 return r;
3062
3063 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3064 sizeof(struct kvm_io_range)), GFP_KERNEL);
3065 if (!new_bus)
3066 return -ENOMEM;
3067
3068 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3069 new_bus->dev_count--;
3070 memcpy(new_bus->range + i, bus->range + i + 1,
3071 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3072
3073 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3074 synchronize_srcu_expedited(&kvm->srcu);
3075 kfree(bus);
3076 return r;
3077 }
3078
3079 static struct notifier_block kvm_cpu_notifier = {
3080 .notifier_call = kvm_cpu_hotplug,
3081 };
3082
3083 static int vm_stat_get(void *_offset, u64 *val)
3084 {
3085 unsigned offset = (long)_offset;
3086 struct kvm *kvm;
3087
3088 *val = 0;
3089 spin_lock(&kvm_lock);
3090 list_for_each_entry(kvm, &vm_list, vm_list)
3091 *val += *(u32 *)((void *)kvm + offset);
3092 spin_unlock(&kvm_lock);
3093 return 0;
3094 }
3095
3096 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3097
3098 static int vcpu_stat_get(void *_offset, u64 *val)
3099 {
3100 unsigned offset = (long)_offset;
3101 struct kvm *kvm;
3102 struct kvm_vcpu *vcpu;
3103 int i;
3104
3105 *val = 0;
3106 spin_lock(&kvm_lock);
3107 list_for_each_entry(kvm, &vm_list, vm_list)
3108 kvm_for_each_vcpu(i, vcpu, kvm)
3109 *val += *(u32 *)((void *)vcpu + offset);
3110
3111 spin_unlock(&kvm_lock);
3112 return 0;
3113 }
3114
3115 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3116
3117 static const struct file_operations *stat_fops[] = {
3118 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3119 [KVM_STAT_VM] = &vm_stat_fops,
3120 };
3121
3122 static int kvm_init_debug(void)
3123 {
3124 int r = -EEXIST;
3125 struct kvm_stats_debugfs_item *p;
3126
3127 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3128 if (kvm_debugfs_dir == NULL)
3129 goto out;
3130
3131 for (p = debugfs_entries; p->name; ++p) {
3132 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3133 (void *)(long)p->offset,
3134 stat_fops[p->kind]);
3135 if (p->dentry == NULL)
3136 goto out_dir;
3137 }
3138
3139 return 0;
3140
3141 out_dir:
3142 debugfs_remove_recursive(kvm_debugfs_dir);
3143 out:
3144 return r;
3145 }
3146
3147 static void kvm_exit_debug(void)
3148 {
3149 struct kvm_stats_debugfs_item *p;
3150
3151 for (p = debugfs_entries; p->name; ++p)
3152 debugfs_remove(p->dentry);
3153 debugfs_remove(kvm_debugfs_dir);
3154 }
3155
3156 static int kvm_suspend(void)
3157 {
3158 if (kvm_usage_count)
3159 hardware_disable_nolock(NULL);
3160 return 0;
3161 }
3162
3163 static void kvm_resume(void)
3164 {
3165 if (kvm_usage_count) {
3166 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3167 hardware_enable_nolock(NULL);
3168 }
3169 }
3170
3171 static struct syscore_ops kvm_syscore_ops = {
3172 .suspend = kvm_suspend,
3173 .resume = kvm_resume,
3174 };
3175
3176 static inline
3177 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3178 {
3179 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3180 }
3181
3182 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3183 {
3184 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3185 if (vcpu->preempted)
3186 vcpu->preempted = false;
3187
3188 kvm_arch_sched_in(vcpu, cpu);
3189
3190 kvm_arch_vcpu_load(vcpu, cpu);
3191 }
3192
3193 static void kvm_sched_out(struct preempt_notifier *pn,
3194 struct task_struct *next)
3195 {
3196 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3197
3198 if (current->state == TASK_RUNNING)
3199 vcpu->preempted = true;
3200 kvm_arch_vcpu_put(vcpu);
3201 }
3202
3203 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3204 struct module *module)
3205 {
3206 int r;
3207 int cpu;
3208
3209 r = kvm_arch_init(opaque);
3210 if (r)
3211 goto out_fail;
3212
3213 /*
3214 * kvm_arch_init makes sure there's at most one caller
3215 * for architectures that support multiple implementations,
3216 * like intel and amd on x86.
3217 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3218 * conflicts in case kvm is already setup for another implementation.
3219 */
3220 r = kvm_irqfd_init();
3221 if (r)
3222 goto out_irqfd;
3223
3224 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3225 r = -ENOMEM;
3226 goto out_free_0;
3227 }
3228
3229 r = kvm_arch_hardware_setup();
3230 if (r < 0)
3231 goto out_free_0a;
3232
3233 for_each_online_cpu(cpu) {
3234 smp_call_function_single(cpu,
3235 kvm_arch_check_processor_compat,
3236 &r, 1);
3237 if (r < 0)
3238 goto out_free_1;
3239 }
3240
3241 r = register_cpu_notifier(&kvm_cpu_notifier);
3242 if (r)
3243 goto out_free_2;
3244 register_reboot_notifier(&kvm_reboot_notifier);
3245
3246 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3247 if (!vcpu_align)
3248 vcpu_align = __alignof__(struct kvm_vcpu);
3249 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3250 0, NULL);
3251 if (!kvm_vcpu_cache) {
3252 r = -ENOMEM;
3253 goto out_free_3;
3254 }
3255
3256 r = kvm_async_pf_init();
3257 if (r)
3258 goto out_free;
3259
3260 kvm_chardev_ops.owner = module;
3261 kvm_vm_fops.owner = module;
3262 kvm_vcpu_fops.owner = module;
3263
3264 r = misc_register(&kvm_dev);
3265 if (r) {
3266 printk(KERN_ERR "kvm: misc device register failed\n");
3267 goto out_unreg;
3268 }
3269
3270 register_syscore_ops(&kvm_syscore_ops);
3271
3272 kvm_preempt_ops.sched_in = kvm_sched_in;
3273 kvm_preempt_ops.sched_out = kvm_sched_out;
3274
3275 r = kvm_init_debug();
3276 if (r) {
3277 printk(KERN_ERR "kvm: create debugfs files failed\n");
3278 goto out_undebugfs;
3279 }
3280
3281 r = kvm_vfio_ops_init();
3282 WARN_ON(r);
3283
3284 return 0;
3285
3286 out_undebugfs:
3287 unregister_syscore_ops(&kvm_syscore_ops);
3288 misc_deregister(&kvm_dev);
3289 out_unreg:
3290 kvm_async_pf_deinit();
3291 out_free:
3292 kmem_cache_destroy(kvm_vcpu_cache);
3293 out_free_3:
3294 unregister_reboot_notifier(&kvm_reboot_notifier);
3295 unregister_cpu_notifier(&kvm_cpu_notifier);
3296 out_free_2:
3297 out_free_1:
3298 kvm_arch_hardware_unsetup();
3299 out_free_0a:
3300 free_cpumask_var(cpus_hardware_enabled);
3301 out_free_0:
3302 kvm_irqfd_exit();
3303 out_irqfd:
3304 kvm_arch_exit();
3305 out_fail:
3306 return r;
3307 }
3308 EXPORT_SYMBOL_GPL(kvm_init);
3309
3310 void kvm_exit(void)
3311 {
3312 kvm_exit_debug();
3313 misc_deregister(&kvm_dev);
3314 kmem_cache_destroy(kvm_vcpu_cache);
3315 kvm_async_pf_deinit();
3316 unregister_syscore_ops(&kvm_syscore_ops);
3317 unregister_reboot_notifier(&kvm_reboot_notifier);
3318 unregister_cpu_notifier(&kvm_cpu_notifier);
3319 on_each_cpu(hardware_disable_nolock, NULL, 1);
3320 kvm_arch_hardware_unsetup();
3321 kvm_arch_exit();
3322 kvm_irqfd_exit();
3323 free_cpumask_var(cpus_hardware_enabled);
3324 kvm_vfio_ops_exit();
3325 }
3326 EXPORT_SYMBOL_GPL(kvm_exit);
Linux kernel - Deliverables
This page took 0.102096 seconds and 5 git commands to generate.