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 <kvm/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 static unsigned int halt_poll_ns;
70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
71
72 /*
73 * Ordering of locks:
74 *
75 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
76 */
77
78 DEFINE_SPINLOCK(kvm_lock);
79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
80 LIST_HEAD(vm_list);
81
82 static cpumask_var_t cpus_hardware_enabled;
83 static int kvm_usage_count;
84 static atomic_t hardware_enable_failed;
85
86 struct kmem_cache *kvm_vcpu_cache;
87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
88
89 static __read_mostly struct preempt_ops kvm_preempt_ops;
90
91 struct dentry *kvm_debugfs_dir;
92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
93
94 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
95 unsigned long arg);
96 #ifdef CONFIG_KVM_COMPAT
97 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
98 unsigned long arg);
99 #endif
100 static int hardware_enable_all(void);
101 static void hardware_disable_all(void);
102
103 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
104
105 static void kvm_release_pfn_dirty(pfn_t pfn);
106 static void mark_page_dirty_in_slot(struct kvm *kvm,
107 struct kvm_memory_slot *memslot, gfn_t gfn);
108
109 __visible bool kvm_rebooting;
110 EXPORT_SYMBOL_GPL(kvm_rebooting);
111
112 static bool largepages_enabled = true;
113
kvm_is_reserved_pfn(pfn_t pfn)114 bool kvm_is_reserved_pfn(pfn_t pfn)
115 {
116 if (pfn_valid(pfn))
117 return PageReserved(pfn_to_page(pfn));
118
119 return true;
120 }
121
122 /*
123 * Switches to specified vcpu, until a matching vcpu_put()
124 */
vcpu_load(struct kvm_vcpu * vcpu)125 int vcpu_load(struct kvm_vcpu *vcpu)
126 {
127 int cpu;
128
129 if (mutex_lock_killable(&vcpu->mutex))
130 return -EINTR;
131 cpu = get_cpu();
132 preempt_notifier_register(&vcpu->preempt_notifier);
133 kvm_arch_vcpu_load(vcpu, cpu);
134 put_cpu();
135 return 0;
136 }
137
vcpu_put(struct kvm_vcpu * vcpu)138 void vcpu_put(struct kvm_vcpu *vcpu)
139 {
140 preempt_disable();
141 kvm_arch_vcpu_put(vcpu);
142 preempt_notifier_unregister(&vcpu->preempt_notifier);
143 preempt_enable();
144 mutex_unlock(&vcpu->mutex);
145 }
146
ack_flush(void * _completed)147 static void ack_flush(void *_completed)
148 {
149 }
150
kvm_make_all_cpus_request(struct kvm * kvm,unsigned int req)151 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
152 {
153 int i, cpu, me;
154 cpumask_var_t cpus;
155 bool called = true;
156 struct kvm_vcpu *vcpu;
157
158 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
159
160 me = get_cpu();
161 kvm_for_each_vcpu(i, vcpu, kvm) {
162 kvm_make_request(req, vcpu);
163 cpu = vcpu->cpu;
164
165 /* Set ->requests bit before we read ->mode */
166 smp_mb();
167
168 if (cpus != NULL && cpu != -1 && cpu != me &&
169 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
170 cpumask_set_cpu(cpu, cpus);
171 }
172 if (unlikely(cpus == NULL))
173 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
174 else if (!cpumask_empty(cpus))
175 smp_call_function_many(cpus, ack_flush, NULL, 1);
176 else
177 called = false;
178 put_cpu();
179 free_cpumask_var(cpus);
180 return called;
181 }
182
183 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
kvm_flush_remote_tlbs(struct kvm * kvm)184 void kvm_flush_remote_tlbs(struct kvm *kvm)
185 {
186 long dirty_count = kvm->tlbs_dirty;
187
188 smp_mb();
189 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
190 ++kvm->stat.remote_tlb_flush;
191 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
192 }
193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
194 #endif
195
kvm_reload_remote_mmus(struct kvm * kvm)196 void kvm_reload_remote_mmus(struct kvm *kvm)
197 {
198 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
199 }
200
kvm_make_mclock_inprogress_request(struct kvm * kvm)201 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
202 {
203 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
204 }
205
kvm_make_scan_ioapic_request(struct kvm * kvm)206 void kvm_make_scan_ioapic_request(struct kvm *kvm)
207 {
208 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
209 }
210
kvm_vcpu_init(struct kvm_vcpu * vcpu,struct kvm * kvm,unsigned id)211 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
212 {
213 struct page *page;
214 int r;
215
216 mutex_init(&vcpu->mutex);
217 vcpu->cpu = -1;
218 vcpu->kvm = kvm;
219 vcpu->vcpu_id = id;
220 vcpu->pid = NULL;
221 init_waitqueue_head(&vcpu->wq);
222 kvm_async_pf_vcpu_init(vcpu);
223
224 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
225 if (!page) {
226 r = -ENOMEM;
227 goto fail;
228 }
229 vcpu->run = page_address(page);
230
231 kvm_vcpu_set_in_spin_loop(vcpu, false);
232 kvm_vcpu_set_dy_eligible(vcpu, false);
233 vcpu->preempted = false;
234
235 r = kvm_arch_vcpu_init(vcpu);
236 if (r < 0)
237 goto fail_free_run;
238 return 0;
239
240 fail_free_run:
241 free_page((unsigned long)vcpu->run);
242 fail:
243 return r;
244 }
245 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
246
kvm_vcpu_uninit(struct kvm_vcpu * vcpu)247 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
248 {
249 put_pid(vcpu->pid);
250 kvm_arch_vcpu_uninit(vcpu);
251 free_page((unsigned long)vcpu->run);
252 }
253 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
254
255 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
mmu_notifier_to_kvm(struct mmu_notifier * mn)256 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
257 {
258 return container_of(mn, struct kvm, mmu_notifier);
259 }
260
kvm_mmu_notifier_invalidate_page(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long address)261 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
262 struct mm_struct *mm,
263 unsigned long address)
264 {
265 struct kvm *kvm = mmu_notifier_to_kvm(mn);
266 int need_tlb_flush, idx;
267
268 /*
269 * When ->invalidate_page runs, the linux pte has been zapped
270 * already but the page is still allocated until
271 * ->invalidate_page returns. So if we increase the sequence
272 * here the kvm page fault will notice if the spte can't be
273 * established because the page is going to be freed. If
274 * instead the kvm page fault establishes the spte before
275 * ->invalidate_page runs, kvm_unmap_hva will release it
276 * before returning.
277 *
278 * The sequence increase only need to be seen at spin_unlock
279 * time, and not at spin_lock time.
280 *
281 * Increasing the sequence after the spin_unlock would be
282 * unsafe because the kvm page fault could then establish the
283 * pte after kvm_unmap_hva returned, without noticing the page
284 * is going to be freed.
285 */
286 idx = srcu_read_lock(&kvm->srcu);
287 spin_lock(&kvm->mmu_lock);
288
289 kvm->mmu_notifier_seq++;
290 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
291 /* we've to flush the tlb before the pages can be freed */
292 if (need_tlb_flush)
293 kvm_flush_remote_tlbs(kvm);
294
295 spin_unlock(&kvm->mmu_lock);
296
297 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
298
299 srcu_read_unlock(&kvm->srcu, idx);
300 }
301
kvm_mmu_notifier_change_pte(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long address,pte_t pte)302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long address,
305 pte_t pte)
306 {
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
308 int idx;
309
310 idx = srcu_read_lock(&kvm->srcu);
311 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 kvm_set_spte_hva(kvm, address, pte);
314 spin_unlock(&kvm->mmu_lock);
315 srcu_read_unlock(&kvm->srcu, idx);
316 }
317
kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319 struct mm_struct *mm,
320 unsigned long start,
321 unsigned long end)
322 {
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
324 int need_tlb_flush = 0, idx;
325
326 idx = srcu_read_lock(&kvm->srcu);
327 spin_lock(&kvm->mmu_lock);
328 /*
329 * The count increase must become visible at unlock time as no
330 * spte can be established without taking the mmu_lock and
331 * count is also read inside the mmu_lock critical section.
332 */
333 kvm->mmu_notifier_count++;
334 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335 need_tlb_flush |= kvm->tlbs_dirty;
336 /* we've to flush the tlb before the pages can be freed */
337 if (need_tlb_flush)
338 kvm_flush_remote_tlbs(kvm);
339
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
342 }
343
kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
346 unsigned long start,
347 unsigned long end)
348 {
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350
351 spin_lock(&kvm->mmu_lock);
352 /*
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
355 * been freed.
356 */
357 kvm->mmu_notifier_seq++;
358 smp_wmb();
359 /*
360 * The above sequence increase must be visible before the
361 * below count decrease, which is ensured by the smp_wmb above
362 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 */
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
366
367 BUG_ON(kvm->mmu_notifier_count < 0);
368 }
369
kvm_mmu_notifier_clear_flush_young(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long start,unsigned long end)370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long start,
373 unsigned long end)
374 {
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
376 int young, idx;
377
378 idx = srcu_read_lock(&kvm->srcu);
379 spin_lock(&kvm->mmu_lock);
380
381 young = kvm_age_hva(kvm, start, end);
382 if (young)
383 kvm_flush_remote_tlbs(kvm);
384
385 spin_unlock(&kvm->mmu_lock);
386 srcu_read_unlock(&kvm->srcu, idx);
387
388 return young;
389 }
390
kvm_mmu_notifier_test_young(struct mmu_notifier * mn,struct mm_struct * mm,unsigned long address)391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392 struct mm_struct *mm,
393 unsigned long address)
394 {
395 struct kvm *kvm = mmu_notifier_to_kvm(mn);
396 int young, idx;
397
398 idx = srcu_read_lock(&kvm->srcu);
399 spin_lock(&kvm->mmu_lock);
400 young = kvm_test_age_hva(kvm, address);
401 spin_unlock(&kvm->mmu_lock);
402 srcu_read_unlock(&kvm->srcu, idx);
403
404 return young;
405 }
406
kvm_mmu_notifier_release(struct mmu_notifier * mn,struct mm_struct * mm)407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408 struct mm_struct *mm)
409 {
410 struct kvm *kvm = mmu_notifier_to_kvm(mn);
411 int idx;
412
413 idx = srcu_read_lock(&kvm->srcu);
414 kvm_arch_flush_shadow_all(kvm);
415 srcu_read_unlock(&kvm->srcu, idx);
416 }
417
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419 .invalidate_page = kvm_mmu_notifier_invalidate_page,
420 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
422 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
423 .test_young = kvm_mmu_notifier_test_young,
424 .change_pte = kvm_mmu_notifier_change_pte,
425 .release = kvm_mmu_notifier_release,
426 };
427
kvm_init_mmu_notifier(struct kvm * kvm)428 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 {
430 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 }
433
434 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435
kvm_init_mmu_notifier(struct kvm * kvm)436 static int kvm_init_mmu_notifier(struct kvm *kvm)
437 {
438 return 0;
439 }
440
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442
kvm_init_memslots_id(struct kvm * kvm)443 static void kvm_init_memslots_id(struct kvm *kvm)
444 {
445 int i;
446 struct kvm_memslots *slots = kvm->memslots;
447
448 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449 slots->id_to_index[i] = slots->memslots[i].id = i;
450 }
451
kvm_create_vm(unsigned long type)452 static struct kvm *kvm_create_vm(unsigned long type)
453 {
454 int r, i;
455 struct kvm *kvm = kvm_arch_alloc_vm();
456
457 if (!kvm)
458 return ERR_PTR(-ENOMEM);
459
460 spin_lock_init(&kvm->mmu_lock);
461 atomic_inc(¤t->mm->mm_count);
462 kvm->mm = current->mm;
463 kvm_eventfd_init(kvm);
464 mutex_init(&kvm->lock);
465 mutex_init(&kvm->irq_lock);
466 mutex_init(&kvm->slots_lock);
467 atomic_set(&kvm->users_count, 1);
468 INIT_LIST_HEAD(&kvm->devices);
469
470 r = kvm_arch_init_vm(kvm, type);
471 if (r)
472 goto out_err_no_disable;
473
474 r = hardware_enable_all();
475 if (r)
476 goto out_err_no_disable;
477
478 #ifdef CONFIG_HAVE_KVM_IRQFD
479 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
480 #endif
481
482 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
483
484 r = -ENOMEM;
485 kvm->memslots = kvm_kvzalloc(sizeof(struct kvm_memslots));
486 if (!kvm->memslots)
487 goto out_err_no_srcu;
488
489 /*
490 * Init kvm generation close to the maximum to easily test the
491 * code of handling generation number wrap-around.
492 */
493 kvm->memslots->generation = -150;
494
495 kvm_init_memslots_id(kvm);
496 if (init_srcu_struct(&kvm->srcu))
497 goto out_err_no_srcu;
498 if (init_srcu_struct(&kvm->irq_srcu))
499 goto out_err_no_irq_srcu;
500 for (i = 0; i < KVM_NR_BUSES; i++) {
501 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
502 GFP_KERNEL);
503 if (!kvm->buses[i])
504 goto out_err;
505 }
506
507 r = kvm_init_mmu_notifier(kvm);
508 if (r)
509 goto out_err;
510
511 spin_lock(&kvm_lock);
512 list_add(&kvm->vm_list, &vm_list);
513 spin_unlock(&kvm_lock);
514
515 return kvm;
516
517 out_err:
518 cleanup_srcu_struct(&kvm->irq_srcu);
519 out_err_no_irq_srcu:
520 cleanup_srcu_struct(&kvm->srcu);
521 out_err_no_srcu:
522 hardware_disable_all();
523 out_err_no_disable:
524 for (i = 0; i < KVM_NR_BUSES; i++)
525 kfree(kvm->buses[i]);
526 kvfree(kvm->memslots);
527 kvm_arch_free_vm(kvm);
528 mmdrop(current->mm);
529 return ERR_PTR(r);
530 }
531
532 /*
533 * Avoid using vmalloc for a small buffer.
534 * Should not be used when the size is statically known.
535 */
kvm_kvzalloc(unsigned long size)536 void *kvm_kvzalloc(unsigned long size)
537 {
538 if (size > PAGE_SIZE)
539 return vzalloc(size);
540 else
541 return kzalloc(size, GFP_KERNEL);
542 }
543
kvm_destroy_dirty_bitmap(struct kvm_memory_slot * memslot)544 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 {
546 if (!memslot->dirty_bitmap)
547 return;
548
549 kvfree(memslot->dirty_bitmap);
550 memslot->dirty_bitmap = NULL;
551 }
552
553 /*
554 * Free any memory in @free but not in @dont.
555 */
kvm_free_physmem_slot(struct kvm * kvm,struct kvm_memory_slot * free,struct kvm_memory_slot * dont)556 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
557 struct kvm_memory_slot *dont)
558 {
559 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
560 kvm_destroy_dirty_bitmap(free);
561
562 kvm_arch_free_memslot(kvm, free, dont);
563
564 free->npages = 0;
565 }
566
kvm_free_physmem(struct kvm * kvm)567 static void kvm_free_physmem(struct kvm *kvm)
568 {
569 struct kvm_memslots *slots = kvm->memslots;
570 struct kvm_memory_slot *memslot;
571
572 kvm_for_each_memslot(memslot, slots)
573 kvm_free_physmem_slot(kvm, memslot, NULL);
574
575 kvfree(kvm->memslots);
576 }
577
kvm_destroy_devices(struct kvm * kvm)578 static void kvm_destroy_devices(struct kvm *kvm)
579 {
580 struct list_head *node, *tmp;
581
582 list_for_each_safe(node, tmp, &kvm->devices) {
583 struct kvm_device *dev =
584 list_entry(node, struct kvm_device, vm_node);
585
586 list_del(node);
587 dev->ops->destroy(dev);
588 }
589 }
590
kvm_destroy_vm(struct kvm * kvm)591 static void kvm_destroy_vm(struct kvm *kvm)
592 {
593 int i;
594 struct mm_struct *mm = kvm->mm;
595
596 kvm_arch_sync_events(kvm);
597 spin_lock(&kvm_lock);
598 list_del(&kvm->vm_list);
599 spin_unlock(&kvm_lock);
600 kvm_free_irq_routing(kvm);
601 for (i = 0; i < KVM_NR_BUSES; i++)
602 kvm_io_bus_destroy(kvm->buses[i]);
603 kvm_coalesced_mmio_free(kvm);
604 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
605 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
606 #else
607 kvm_arch_flush_shadow_all(kvm);
608 #endif
609 kvm_arch_destroy_vm(kvm);
610 kvm_destroy_devices(kvm);
611 kvm_free_physmem(kvm);
612 cleanup_srcu_struct(&kvm->irq_srcu);
613 cleanup_srcu_struct(&kvm->srcu);
614 kvm_arch_free_vm(kvm);
615 hardware_disable_all();
616 mmdrop(mm);
617 }
618
kvm_get_kvm(struct kvm * kvm)619 void kvm_get_kvm(struct kvm *kvm)
620 {
621 atomic_inc(&kvm->users_count);
622 }
623 EXPORT_SYMBOL_GPL(kvm_get_kvm);
624
kvm_put_kvm(struct kvm * kvm)625 void kvm_put_kvm(struct kvm *kvm)
626 {
627 if (atomic_dec_and_test(&kvm->users_count))
628 kvm_destroy_vm(kvm);
629 }
630 EXPORT_SYMBOL_GPL(kvm_put_kvm);
631
632
kvm_vm_release(struct inode * inode,struct file * filp)633 static int kvm_vm_release(struct inode *inode, struct file *filp)
634 {
635 struct kvm *kvm = filp->private_data;
636
637 kvm_irqfd_release(kvm);
638
639 kvm_put_kvm(kvm);
640 return 0;
641 }
642
643 /*
644 * Allocation size is twice as large as the actual dirty bitmap size.
645 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
646 */
kvm_create_dirty_bitmap(struct kvm_memory_slot * memslot)647 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
648 {
649 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
650
651 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
652 if (!memslot->dirty_bitmap)
653 return -ENOMEM;
654
655 return 0;
656 }
657
658 /*
659 * Insert memslot and re-sort memslots based on their GFN,
660 * so binary search could be used to lookup GFN.
661 * Sorting algorithm takes advantage of having initially
662 * sorted array and known changed memslot position.
663 */
update_memslots(struct kvm_memslots * slots,struct kvm_memory_slot * new)664 static void update_memslots(struct kvm_memslots *slots,
665 struct kvm_memory_slot *new)
666 {
667 int id = new->id;
668 int i = slots->id_to_index[id];
669 struct kvm_memory_slot *mslots = slots->memslots;
670
671 WARN_ON(mslots[i].id != id);
672 if (!new->npages) {
673 WARN_ON(!mslots[i].npages);
674 new->base_gfn = 0;
675 new->flags = 0;
676 if (mslots[i].npages)
677 slots->used_slots--;
678 } else {
679 if (!mslots[i].npages)
680 slots->used_slots++;
681 }
682
683 while (i < KVM_MEM_SLOTS_NUM - 1 &&
684 new->base_gfn <= mslots[i + 1].base_gfn) {
685 if (!mslots[i + 1].npages)
686 break;
687 mslots[i] = mslots[i + 1];
688 slots->id_to_index[mslots[i].id] = i;
689 i++;
690 }
691
692 /*
693 * The ">=" is needed when creating a slot with base_gfn == 0,
694 * so that it moves before all those with base_gfn == npages == 0.
695 *
696 * On the other hand, if new->npages is zero, the above loop has
697 * already left i pointing to the beginning of the empty part of
698 * mslots, and the ">=" would move the hole backwards in this
699 * case---which is wrong. So skip the loop when deleting a slot.
700 */
701 if (new->npages) {
702 while (i > 0 &&
703 new->base_gfn >= mslots[i - 1].base_gfn) {
704 mslots[i] = mslots[i - 1];
705 slots->id_to_index[mslots[i].id] = i;
706 i--;
707 }
708 } else
709 WARN_ON_ONCE(i != slots->used_slots);
710
711 mslots[i] = *new;
712 slots->id_to_index[mslots[i].id] = i;
713 }
714
check_memory_region_flags(struct kvm_userspace_memory_region * mem)715 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
716 {
717 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
718
719 #ifdef __KVM_HAVE_READONLY_MEM
720 valid_flags |= KVM_MEM_READONLY;
721 #endif
722
723 if (mem->flags & ~valid_flags)
724 return -EINVAL;
725
726 return 0;
727 }
728
install_new_memslots(struct kvm * kvm,struct kvm_memslots * slots)729 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
730 struct kvm_memslots *slots)
731 {
732 struct kvm_memslots *old_memslots = kvm->memslots;
733
734 /*
735 * Set the low bit in the generation, which disables SPTE caching
736 * until the end of synchronize_srcu_expedited.
737 */
738 WARN_ON(old_memslots->generation & 1);
739 slots->generation = old_memslots->generation + 1;
740
741 rcu_assign_pointer(kvm->memslots, slots);
742 synchronize_srcu_expedited(&kvm->srcu);
743
744 /*
745 * Increment the new memslot generation a second time. This prevents
746 * vm exits that race with memslot updates from caching a memslot
747 * generation that will (potentially) be valid forever.
748 */
749 slots->generation++;
750
751 kvm_arch_memslots_updated(kvm);
752
753 return old_memslots;
754 }
755
756 /*
757 * Allocate some memory and give it an address in the guest physical address
758 * space.
759 *
760 * Discontiguous memory is allowed, mostly for framebuffers.
761 *
762 * Must be called holding kvm->slots_lock for write.
763 */
__kvm_set_memory_region(struct kvm * kvm,struct kvm_userspace_memory_region * mem)764 int __kvm_set_memory_region(struct kvm *kvm,
765 struct kvm_userspace_memory_region *mem)
766 {
767 int r;
768 gfn_t base_gfn;
769 unsigned long npages;
770 struct kvm_memory_slot *slot;
771 struct kvm_memory_slot old, new;
772 struct kvm_memslots *slots = NULL, *old_memslots;
773 enum kvm_mr_change change;
774
775 r = check_memory_region_flags(mem);
776 if (r)
777 goto out;
778
779 r = -EINVAL;
780 /* General sanity checks */
781 if (mem->memory_size & (PAGE_SIZE - 1))
782 goto out;
783 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
784 goto out;
785 /* We can read the guest memory with __xxx_user() later on. */
786 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
787 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
788 !access_ok(VERIFY_WRITE,
789 (void __user *)(unsigned long)mem->userspace_addr,
790 mem->memory_size)))
791 goto out;
792 if (mem->slot >= KVM_MEM_SLOTS_NUM)
793 goto out;
794 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
795 goto out;
796
797 slot = id_to_memslot(kvm->memslots, mem->slot);
798 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
799 npages = mem->memory_size >> PAGE_SHIFT;
800
801 if (npages > KVM_MEM_MAX_NR_PAGES)
802 goto out;
803
804 if (!npages)
805 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
806
807 new = old = *slot;
808
809 new.id = mem->slot;
810 new.base_gfn = base_gfn;
811 new.npages = npages;
812 new.flags = mem->flags;
813
814 if (npages) {
815 if (!old.npages)
816 change = KVM_MR_CREATE;
817 else { /* Modify an existing slot. */
818 if ((mem->userspace_addr != old.userspace_addr) ||
819 (npages != old.npages) ||
820 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
821 goto out;
822
823 if (base_gfn != old.base_gfn)
824 change = KVM_MR_MOVE;
825 else if (new.flags != old.flags)
826 change = KVM_MR_FLAGS_ONLY;
827 else { /* Nothing to change. */
828 r = 0;
829 goto out;
830 }
831 }
832 } else if (old.npages) {
833 change = KVM_MR_DELETE;
834 } else /* Modify a non-existent slot: disallowed. */
835 goto out;
836
837 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
838 /* Check for overlaps */
839 r = -EEXIST;
840 kvm_for_each_memslot(slot, kvm->memslots) {
841 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
842 (slot->id == mem->slot))
843 continue;
844 if (!((base_gfn + npages <= slot->base_gfn) ||
845 (base_gfn >= slot->base_gfn + slot->npages)))
846 goto out;
847 }
848 }
849
850 /* Free page dirty bitmap if unneeded */
851 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
852 new.dirty_bitmap = NULL;
853
854 r = -ENOMEM;
855 if (change == KVM_MR_CREATE) {
856 new.userspace_addr = mem->userspace_addr;
857
858 if (kvm_arch_create_memslot(kvm, &new, npages))
859 goto out_free;
860 }
861
862 /* Allocate page dirty bitmap if needed */
863 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
864 if (kvm_create_dirty_bitmap(&new) < 0)
865 goto out_free;
866 }
867
868 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
869 if (!slots)
870 goto out_free;
871 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
872
873 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
874 slot = id_to_memslot(slots, mem->slot);
875 slot->flags |= KVM_MEMSLOT_INVALID;
876
877 old_memslots = install_new_memslots(kvm, slots);
878
879 /* slot was deleted or moved, clear iommu mapping */
880 kvm_iommu_unmap_pages(kvm, &old);
881 /* From this point no new shadow pages pointing to a deleted,
882 * or moved, memslot will be created.
883 *
884 * validation of sp->gfn happens in:
885 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
886 * - kvm_is_visible_gfn (mmu_check_roots)
887 */
888 kvm_arch_flush_shadow_memslot(kvm, slot);
889
890 /*
891 * We can re-use the old_memslots from above, the only difference
892 * from the currently installed memslots is the invalid flag. This
893 * will get overwritten by update_memslots anyway.
894 */
895 slots = old_memslots;
896 }
897
898 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
899 if (r)
900 goto out_slots;
901
902 /* actual memory is freed via old in kvm_free_physmem_slot below */
903 if (change == KVM_MR_DELETE) {
904 new.dirty_bitmap = NULL;
905 memset(&new.arch, 0, sizeof(new.arch));
906 }
907
908 update_memslots(slots, &new);
909 old_memslots = install_new_memslots(kvm, slots);
910
911 kvm_arch_commit_memory_region(kvm, mem, &old, change);
912
913 kvm_free_physmem_slot(kvm, &old, &new);
914 kvfree(old_memslots);
915
916 /*
917 * IOMMU mapping: New slots need to be mapped. Old slots need to be
918 * un-mapped and re-mapped if their base changes. Since base change
919 * unmapping is handled above with slot deletion, mapping alone is
920 * needed here. Anything else the iommu might care about for existing
921 * slots (size changes, userspace addr changes and read-only flag
922 * changes) is disallowed above, so any other attribute changes getting
923 * here can be skipped.
924 */
925 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
926 r = kvm_iommu_map_pages(kvm, &new);
927 return r;
928 }
929
930 return 0;
931
932 out_slots:
933 kvfree(slots);
934 out_free:
935 kvm_free_physmem_slot(kvm, &new, &old);
936 out:
937 return r;
938 }
939 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
940
kvm_set_memory_region(struct kvm * kvm,struct kvm_userspace_memory_region * mem)941 int kvm_set_memory_region(struct kvm *kvm,
942 struct kvm_userspace_memory_region *mem)
943 {
944 int r;
945
946 mutex_lock(&kvm->slots_lock);
947 r = __kvm_set_memory_region(kvm, mem);
948 mutex_unlock(&kvm->slots_lock);
949 return r;
950 }
951 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
952
kvm_vm_ioctl_set_memory_region(struct kvm * kvm,struct kvm_userspace_memory_region * mem)953 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
954 struct kvm_userspace_memory_region *mem)
955 {
956 if (mem->slot >= KVM_USER_MEM_SLOTS)
957 return -EINVAL;
958 return kvm_set_memory_region(kvm, mem);
959 }
960
kvm_get_dirty_log(struct kvm * kvm,struct kvm_dirty_log * log,int * is_dirty)961 int kvm_get_dirty_log(struct kvm *kvm,
962 struct kvm_dirty_log *log, int *is_dirty)
963 {
964 struct kvm_memory_slot *memslot;
965 int r, i;
966 unsigned long n;
967 unsigned long any = 0;
968
969 r = -EINVAL;
970 if (log->slot >= KVM_USER_MEM_SLOTS)
971 goto out;
972
973 memslot = id_to_memslot(kvm->memslots, log->slot);
974 r = -ENOENT;
975 if (!memslot->dirty_bitmap)
976 goto out;
977
978 n = kvm_dirty_bitmap_bytes(memslot);
979
980 for (i = 0; !any && i < n/sizeof(long); ++i)
981 any = memslot->dirty_bitmap[i];
982
983 r = -EFAULT;
984 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
985 goto out;
986
987 if (any)
988 *is_dirty = 1;
989
990 r = 0;
991 out:
992 return r;
993 }
994 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
995
996 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
997 /**
998 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
999 * are dirty write protect them for next write.
1000 * @kvm: pointer to kvm instance
1001 * @log: slot id and address to which we copy the log
1002 * @is_dirty: flag set if any page is dirty
1003 *
1004 * We need to keep it in mind that VCPU threads can write to the bitmap
1005 * concurrently. So, to avoid losing track of dirty pages we keep the
1006 * following order:
1007 *
1008 * 1. Take a snapshot of the bit and clear it if needed.
1009 * 2. Write protect the corresponding page.
1010 * 3. Copy the snapshot to the userspace.
1011 * 4. Upon return caller flushes TLB's if needed.
1012 *
1013 * Between 2 and 4, the guest may write to the page using the remaining TLB
1014 * entry. This is not a problem because the page is reported dirty using
1015 * the snapshot taken before and step 4 ensures that writes done after
1016 * exiting to userspace will be logged for the next call.
1017 *
1018 */
kvm_get_dirty_log_protect(struct kvm * kvm,struct kvm_dirty_log * log,bool * is_dirty)1019 int kvm_get_dirty_log_protect(struct kvm *kvm,
1020 struct kvm_dirty_log *log, bool *is_dirty)
1021 {
1022 struct kvm_memory_slot *memslot;
1023 int r, i;
1024 unsigned long n;
1025 unsigned long *dirty_bitmap;
1026 unsigned long *dirty_bitmap_buffer;
1027
1028 r = -EINVAL;
1029 if (log->slot >= KVM_USER_MEM_SLOTS)
1030 goto out;
1031
1032 memslot = id_to_memslot(kvm->memslots, log->slot);
1033
1034 dirty_bitmap = memslot->dirty_bitmap;
1035 r = -ENOENT;
1036 if (!dirty_bitmap)
1037 goto out;
1038
1039 n = kvm_dirty_bitmap_bytes(memslot);
1040
1041 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1042 memset(dirty_bitmap_buffer, 0, n);
1043
1044 spin_lock(&kvm->mmu_lock);
1045 *is_dirty = false;
1046 for (i = 0; i < n / sizeof(long); i++) {
1047 unsigned long mask;
1048 gfn_t offset;
1049
1050 if (!dirty_bitmap[i])
1051 continue;
1052
1053 *is_dirty = true;
1054
1055 mask = xchg(&dirty_bitmap[i], 0);
1056 dirty_bitmap_buffer[i] = mask;
1057
1058 if (mask) {
1059 offset = i * BITS_PER_LONG;
1060 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1061 offset, mask);
1062 }
1063 }
1064
1065 spin_unlock(&kvm->mmu_lock);
1066
1067 r = -EFAULT;
1068 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1069 goto out;
1070
1071 r = 0;
1072 out:
1073 return r;
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1076 #endif
1077
kvm_largepages_enabled(void)1078 bool kvm_largepages_enabled(void)
1079 {
1080 return largepages_enabled;
1081 }
1082
kvm_disable_largepages(void)1083 void kvm_disable_largepages(void)
1084 {
1085 largepages_enabled = false;
1086 }
1087 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1088
gfn_to_memslot(struct kvm * kvm,gfn_t gfn)1089 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1090 {
1091 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1092 }
1093 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1094
kvm_is_visible_gfn(struct kvm * kvm,gfn_t gfn)1095 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1096 {
1097 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1098
1099 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1100 memslot->flags & KVM_MEMSLOT_INVALID)
1101 return 0;
1102
1103 return 1;
1104 }
1105 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1106
kvm_host_page_size(struct kvm * kvm,gfn_t gfn)1107 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1108 {
1109 struct vm_area_struct *vma;
1110 unsigned long addr, size;
1111
1112 size = PAGE_SIZE;
1113
1114 addr = gfn_to_hva(kvm, gfn);
1115 if (kvm_is_error_hva(addr))
1116 return PAGE_SIZE;
1117
1118 down_read(¤t->mm->mmap_sem);
1119 vma = find_vma(current->mm, addr);
1120 if (!vma)
1121 goto out;
1122
1123 size = vma_kernel_pagesize(vma);
1124
1125 out:
1126 up_read(¤t->mm->mmap_sem);
1127
1128 return size;
1129 }
1130
memslot_is_readonly(struct kvm_memory_slot * slot)1131 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1132 {
1133 return slot->flags & KVM_MEM_READONLY;
1134 }
1135
__gfn_to_hva_many(struct kvm_memory_slot * slot,gfn_t gfn,gfn_t * nr_pages,bool write)1136 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1137 gfn_t *nr_pages, bool write)
1138 {
1139 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1140 return KVM_HVA_ERR_BAD;
1141
1142 if (memslot_is_readonly(slot) && write)
1143 return KVM_HVA_ERR_RO_BAD;
1144
1145 if (nr_pages)
1146 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1147
1148 return __gfn_to_hva_memslot(slot, gfn);
1149 }
1150
gfn_to_hva_many(struct kvm_memory_slot * slot,gfn_t gfn,gfn_t * nr_pages)1151 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1152 gfn_t *nr_pages)
1153 {
1154 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1155 }
1156
gfn_to_hva_memslot(struct kvm_memory_slot * slot,gfn_t gfn)1157 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1158 gfn_t gfn)
1159 {
1160 return gfn_to_hva_many(slot, gfn, NULL);
1161 }
1162 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1163
gfn_to_hva(struct kvm * kvm,gfn_t gfn)1164 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1165 {
1166 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1167 }
1168 EXPORT_SYMBOL_GPL(gfn_to_hva);
1169
1170 /*
1171 * If writable is set to false, the hva returned by this function is only
1172 * allowed to be read.
1173 */
gfn_to_hva_memslot_prot(struct kvm_memory_slot * slot,gfn_t gfn,bool * writable)1174 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1175 gfn_t gfn, bool *writable)
1176 {
1177 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1178
1179 if (!kvm_is_error_hva(hva) && writable)
1180 *writable = !memslot_is_readonly(slot);
1181
1182 return hva;
1183 }
1184
gfn_to_hva_prot(struct kvm * kvm,gfn_t gfn,bool * writable)1185 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1186 {
1187 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1188
1189 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1190 }
1191
get_user_page_nowait(struct task_struct * tsk,struct mm_struct * mm,unsigned long start,int write,struct page ** page)1192 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1193 unsigned long start, int write, struct page **page)
1194 {
1195 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1196
1197 if (write)
1198 flags |= FOLL_WRITE;
1199
1200 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1201 }
1202
check_user_page_hwpoison(unsigned long addr)1203 static inline int check_user_page_hwpoison(unsigned long addr)
1204 {
1205 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1206
1207 rc = __get_user_pages(current, current->mm, addr, 1,
1208 flags, NULL, NULL, NULL);
1209 return rc == -EHWPOISON;
1210 }
1211
1212 /*
1213 * The atomic path to get the writable pfn which will be stored in @pfn,
1214 * true indicates success, otherwise false is returned.
1215 */
hva_to_pfn_fast(unsigned long addr,bool atomic,bool * async,bool write_fault,bool * writable,pfn_t * pfn)1216 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1217 bool write_fault, bool *writable, pfn_t *pfn)
1218 {
1219 struct page *page[1];
1220 int npages;
1221
1222 if (!(async || atomic))
1223 return false;
1224
1225 /*
1226 * Fast pin a writable pfn only if it is a write fault request
1227 * or the caller allows to map a writable pfn for a read fault
1228 * request.
1229 */
1230 if (!(write_fault || writable))
1231 return false;
1232
1233 npages = __get_user_pages_fast(addr, 1, 1, page);
1234 if (npages == 1) {
1235 *pfn = page_to_pfn(page[0]);
1236
1237 if (writable)
1238 *writable = true;
1239 return true;
1240 }
1241
1242 return false;
1243 }
1244
1245 /*
1246 * The slow path to get the pfn of the specified host virtual address,
1247 * 1 indicates success, -errno is returned if error is detected.
1248 */
hva_to_pfn_slow(unsigned long addr,bool * async,bool write_fault,bool * writable,pfn_t * pfn)1249 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1250 bool *writable, pfn_t *pfn)
1251 {
1252 struct page *page[1];
1253 int npages = 0;
1254
1255 might_sleep();
1256
1257 if (writable)
1258 *writable = write_fault;
1259
1260 if (async) {
1261 down_read(¤t->mm->mmap_sem);
1262 npages = get_user_page_nowait(current, current->mm,
1263 addr, write_fault, page);
1264 up_read(¤t->mm->mmap_sem);
1265 } else
1266 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1267 write_fault, 0, page,
1268 FOLL_TOUCH|FOLL_HWPOISON);
1269 if (npages != 1)
1270 return npages;
1271
1272 /* map read fault as writable if possible */
1273 if (unlikely(!write_fault) && writable) {
1274 struct page *wpage[1];
1275
1276 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1277 if (npages == 1) {
1278 *writable = true;
1279 put_page(page[0]);
1280 page[0] = wpage[0];
1281 }
1282
1283 npages = 1;
1284 }
1285 *pfn = page_to_pfn(page[0]);
1286 return npages;
1287 }
1288
vma_is_valid(struct vm_area_struct * vma,bool write_fault)1289 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1290 {
1291 if (unlikely(!(vma->vm_flags & VM_READ)))
1292 return false;
1293
1294 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1295 return false;
1296
1297 return true;
1298 }
1299
1300 /*
1301 * Pin guest page in memory and return its pfn.
1302 * @addr: host virtual address which maps memory to the guest
1303 * @atomic: whether this function can sleep
1304 * @async: whether this function need to wait IO complete if the
1305 * host page is not in the memory
1306 * @write_fault: whether we should get a writable host page
1307 * @writable: whether it allows to map a writable host page for !@write_fault
1308 *
1309 * The function will map a writable host page for these two cases:
1310 * 1): @write_fault = true
1311 * 2): @write_fault = false && @writable, @writable will tell the caller
1312 * whether the mapping is writable.
1313 */
hva_to_pfn(unsigned long addr,bool atomic,bool * async,bool write_fault,bool * writable)1314 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1315 bool write_fault, bool *writable)
1316 {
1317 struct vm_area_struct *vma;
1318 pfn_t pfn = 0;
1319 int npages;
1320
1321 /* we can do it either atomically or asynchronously, not both */
1322 BUG_ON(atomic && async);
1323
1324 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1325 return pfn;
1326
1327 if (atomic)
1328 return KVM_PFN_ERR_FAULT;
1329
1330 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1331 if (npages == 1)
1332 return pfn;
1333
1334 down_read(¤t->mm->mmap_sem);
1335 if (npages == -EHWPOISON ||
1336 (!async && check_user_page_hwpoison(addr))) {
1337 pfn = KVM_PFN_ERR_HWPOISON;
1338 goto exit;
1339 }
1340
1341 vma = find_vma_intersection(current->mm, addr, addr + 1);
1342
1343 if (vma == NULL)
1344 pfn = KVM_PFN_ERR_FAULT;
1345 else if ((vma->vm_flags & VM_PFNMAP)) {
1346 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1347 vma->vm_pgoff;
1348 BUG_ON(!kvm_is_reserved_pfn(pfn));
1349 } else {
1350 if (async && vma_is_valid(vma, write_fault))
1351 *async = true;
1352 pfn = KVM_PFN_ERR_FAULT;
1353 }
1354 exit:
1355 up_read(¤t->mm->mmap_sem);
1356 return pfn;
1357 }
1358
1359 static pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot * slot,gfn_t gfn,bool atomic,bool * async,bool write_fault,bool * writable)1360 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1361 bool *async, bool write_fault, bool *writable)
1362 {
1363 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1364
1365 if (addr == KVM_HVA_ERR_RO_BAD)
1366 return KVM_PFN_ERR_RO_FAULT;
1367
1368 if (kvm_is_error_hva(addr))
1369 return KVM_PFN_NOSLOT;
1370
1371 /* Do not map writable pfn in the readonly memslot. */
1372 if (writable && memslot_is_readonly(slot)) {
1373 *writable = false;
1374 writable = NULL;
1375 }
1376
1377 return hva_to_pfn(addr, atomic, async, write_fault,
1378 writable);
1379 }
1380
__gfn_to_pfn(struct kvm * kvm,gfn_t gfn,bool atomic,bool * async,bool write_fault,bool * writable)1381 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1382 bool write_fault, bool *writable)
1383 {
1384 struct kvm_memory_slot *slot;
1385
1386 if (async)
1387 *async = false;
1388
1389 slot = gfn_to_memslot(kvm, gfn);
1390
1391 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1392 writable);
1393 }
1394
gfn_to_pfn_atomic(struct kvm * kvm,gfn_t gfn)1395 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1396 {
1397 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1398 }
1399 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1400
gfn_to_pfn_async(struct kvm * kvm,gfn_t gfn,bool * async,bool write_fault,bool * writable)1401 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1402 bool write_fault, bool *writable)
1403 {
1404 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1405 }
1406 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1407
gfn_to_pfn(struct kvm * kvm,gfn_t gfn)1408 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1409 {
1410 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1411 }
1412 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1413
gfn_to_pfn_prot(struct kvm * kvm,gfn_t gfn,bool write_fault,bool * writable)1414 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1415 bool *writable)
1416 {
1417 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1418 }
1419 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1420
gfn_to_pfn_memslot(struct kvm_memory_slot * slot,gfn_t gfn)1421 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1422 {
1423 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1424 }
1425
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot * slot,gfn_t gfn)1426 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1427 {
1428 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1429 }
1430 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1431
gfn_to_page_many_atomic(struct kvm * kvm,gfn_t gfn,struct page ** pages,int nr_pages)1432 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1433 int nr_pages)
1434 {
1435 unsigned long addr;
1436 gfn_t entry;
1437
1438 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1439 if (kvm_is_error_hva(addr))
1440 return -1;
1441
1442 if (entry < nr_pages)
1443 return 0;
1444
1445 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1446 }
1447 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1448
kvm_pfn_to_page(pfn_t pfn)1449 static struct page *kvm_pfn_to_page(pfn_t pfn)
1450 {
1451 if (is_error_noslot_pfn(pfn))
1452 return KVM_ERR_PTR_BAD_PAGE;
1453
1454 if (kvm_is_reserved_pfn(pfn)) {
1455 WARN_ON(1);
1456 return KVM_ERR_PTR_BAD_PAGE;
1457 }
1458
1459 return pfn_to_page(pfn);
1460 }
1461
gfn_to_page(struct kvm * kvm,gfn_t gfn)1462 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1463 {
1464 pfn_t pfn;
1465
1466 pfn = gfn_to_pfn(kvm, gfn);
1467
1468 return kvm_pfn_to_page(pfn);
1469 }
1470 EXPORT_SYMBOL_GPL(gfn_to_page);
1471
kvm_release_page_clean(struct page * page)1472 void kvm_release_page_clean(struct page *page)
1473 {
1474 WARN_ON(is_error_page(page));
1475
1476 kvm_release_pfn_clean(page_to_pfn(page));
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1479
kvm_release_pfn_clean(pfn_t pfn)1480 void kvm_release_pfn_clean(pfn_t pfn)
1481 {
1482 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1483 put_page(pfn_to_page(pfn));
1484 }
1485 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1486
kvm_release_page_dirty(struct page * page)1487 void kvm_release_page_dirty(struct page *page)
1488 {
1489 WARN_ON(is_error_page(page));
1490
1491 kvm_release_pfn_dirty(page_to_pfn(page));
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1494
kvm_release_pfn_dirty(pfn_t pfn)1495 static void kvm_release_pfn_dirty(pfn_t pfn)
1496 {
1497 kvm_set_pfn_dirty(pfn);
1498 kvm_release_pfn_clean(pfn);
1499 }
1500
kvm_set_pfn_dirty(pfn_t pfn)1501 void kvm_set_pfn_dirty(pfn_t pfn)
1502 {
1503 if (!kvm_is_reserved_pfn(pfn)) {
1504 struct page *page = pfn_to_page(pfn);
1505
1506 if (!PageReserved(page))
1507 SetPageDirty(page);
1508 }
1509 }
1510 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1511
kvm_set_pfn_accessed(pfn_t pfn)1512 void kvm_set_pfn_accessed(pfn_t pfn)
1513 {
1514 if (!kvm_is_reserved_pfn(pfn))
1515 mark_page_accessed(pfn_to_page(pfn));
1516 }
1517 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1518
kvm_get_pfn(pfn_t pfn)1519 void kvm_get_pfn(pfn_t pfn)
1520 {
1521 if (!kvm_is_reserved_pfn(pfn))
1522 get_page(pfn_to_page(pfn));
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1525
next_segment(unsigned long len,int offset)1526 static int next_segment(unsigned long len, int offset)
1527 {
1528 if (len > PAGE_SIZE - offset)
1529 return PAGE_SIZE - offset;
1530 else
1531 return len;
1532 }
1533
kvm_read_guest_page(struct kvm * kvm,gfn_t gfn,void * data,int offset,int len)1534 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1535 int len)
1536 {
1537 int r;
1538 unsigned long addr;
1539
1540 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1541 if (kvm_is_error_hva(addr))
1542 return -EFAULT;
1543 r = __copy_from_user(data, (void __user *)addr + offset, len);
1544 if (r)
1545 return -EFAULT;
1546 return 0;
1547 }
1548 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1549
kvm_read_guest(struct kvm * kvm,gpa_t gpa,void * data,unsigned long len)1550 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1551 {
1552 gfn_t gfn = gpa >> PAGE_SHIFT;
1553 int seg;
1554 int offset = offset_in_page(gpa);
1555 int ret;
1556
1557 while ((seg = next_segment(len, offset)) != 0) {
1558 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1559 if (ret < 0)
1560 return ret;
1561 offset = 0;
1562 len -= seg;
1563 data += seg;
1564 ++gfn;
1565 }
1566 return 0;
1567 }
1568 EXPORT_SYMBOL_GPL(kvm_read_guest);
1569
kvm_read_guest_atomic(struct kvm * kvm,gpa_t gpa,void * data,unsigned long len)1570 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1571 unsigned long len)
1572 {
1573 int r;
1574 unsigned long addr;
1575 gfn_t gfn = gpa >> PAGE_SHIFT;
1576 int offset = offset_in_page(gpa);
1577
1578 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1579 if (kvm_is_error_hva(addr))
1580 return -EFAULT;
1581 pagefault_disable();
1582 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1583 pagefault_enable();
1584 if (r)
1585 return -EFAULT;
1586 return 0;
1587 }
1588 EXPORT_SYMBOL(kvm_read_guest_atomic);
1589
kvm_write_guest_page(struct kvm * kvm,gfn_t gfn,const void * data,int offset,int len)1590 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1591 int offset, int len)
1592 {
1593 int r;
1594 unsigned long addr;
1595
1596 addr = gfn_to_hva(kvm, gfn);
1597 if (kvm_is_error_hva(addr))
1598 return -EFAULT;
1599 r = __copy_to_user((void __user *)addr + offset, data, len);
1600 if (r)
1601 return -EFAULT;
1602 mark_page_dirty(kvm, gfn);
1603 return 0;
1604 }
1605 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1606
kvm_write_guest(struct kvm * kvm,gpa_t gpa,const void * data,unsigned long len)1607 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1608 unsigned long len)
1609 {
1610 gfn_t gfn = gpa >> PAGE_SHIFT;
1611 int seg;
1612 int offset = offset_in_page(gpa);
1613 int ret;
1614
1615 while ((seg = next_segment(len, offset)) != 0) {
1616 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1617 if (ret < 0)
1618 return ret;
1619 offset = 0;
1620 len -= seg;
1621 data += seg;
1622 ++gfn;
1623 }
1624 return 0;
1625 }
1626 EXPORT_SYMBOL_GPL(kvm_write_guest);
1627
kvm_gfn_to_hva_cache_init(struct kvm * kvm,struct gfn_to_hva_cache * ghc,gpa_t gpa,unsigned long len)1628 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1629 gpa_t gpa, unsigned long len)
1630 {
1631 struct kvm_memslots *slots = kvm_memslots(kvm);
1632 int offset = offset_in_page(gpa);
1633 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1634 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1635 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1636 gfn_t nr_pages_avail;
1637
1638 ghc->gpa = gpa;
1639 ghc->generation = slots->generation;
1640 ghc->len = len;
1641 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1642 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1643 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1644 ghc->hva += offset;
1645 } else {
1646 /*
1647 * If the requested region crosses two memslots, we still
1648 * verify that the entire region is valid here.
1649 */
1650 while (start_gfn <= end_gfn) {
1651 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1652 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1653 &nr_pages_avail);
1654 if (kvm_is_error_hva(ghc->hva))
1655 return -EFAULT;
1656 start_gfn += nr_pages_avail;
1657 }
1658 /* Use the slow path for cross page reads and writes. */
1659 ghc->memslot = NULL;
1660 }
1661 return 0;
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1664
kvm_write_guest_cached(struct kvm * kvm,struct gfn_to_hva_cache * ghc,void * data,unsigned long len)1665 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1666 void *data, unsigned long len)
1667 {
1668 struct kvm_memslots *slots = kvm_memslots(kvm);
1669 int r;
1670
1671 BUG_ON(len > ghc->len);
1672
1673 if (slots->generation != ghc->generation)
1674 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1675
1676 if (unlikely(!ghc->memslot))
1677 return kvm_write_guest(kvm, ghc->gpa, data, len);
1678
1679 if (kvm_is_error_hva(ghc->hva))
1680 return -EFAULT;
1681
1682 r = __copy_to_user((void __user *)ghc->hva, data, len);
1683 if (r)
1684 return -EFAULT;
1685 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1686
1687 return 0;
1688 }
1689 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1690
kvm_read_guest_cached(struct kvm * kvm,struct gfn_to_hva_cache * ghc,void * data,unsigned long len)1691 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1692 void *data, unsigned long len)
1693 {
1694 struct kvm_memslots *slots = kvm_memslots(kvm);
1695 int r;
1696
1697 BUG_ON(len > ghc->len);
1698
1699 if (slots->generation != ghc->generation)
1700 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1701
1702 if (unlikely(!ghc->memslot))
1703 return kvm_read_guest(kvm, ghc->gpa, data, len);
1704
1705 if (kvm_is_error_hva(ghc->hva))
1706 return -EFAULT;
1707
1708 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1709 if (r)
1710 return -EFAULT;
1711
1712 return 0;
1713 }
1714 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1715
kvm_clear_guest_page(struct kvm * kvm,gfn_t gfn,int offset,int len)1716 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1717 {
1718 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1719
1720 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1721 }
1722 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1723
kvm_clear_guest(struct kvm * kvm,gpa_t gpa,unsigned long len)1724 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1725 {
1726 gfn_t gfn = gpa >> PAGE_SHIFT;
1727 int seg;
1728 int offset = offset_in_page(gpa);
1729 int ret;
1730
1731 while ((seg = next_segment(len, offset)) != 0) {
1732 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1733 if (ret < 0)
1734 return ret;
1735 offset = 0;
1736 len -= seg;
1737 ++gfn;
1738 }
1739 return 0;
1740 }
1741 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1742
mark_page_dirty_in_slot(struct kvm * kvm,struct kvm_memory_slot * memslot,gfn_t gfn)1743 static void mark_page_dirty_in_slot(struct kvm *kvm,
1744 struct kvm_memory_slot *memslot,
1745 gfn_t gfn)
1746 {
1747 if (memslot && memslot->dirty_bitmap) {
1748 unsigned long rel_gfn = gfn - memslot->base_gfn;
1749
1750 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1751 }
1752 }
1753
mark_page_dirty(struct kvm * kvm,gfn_t gfn)1754 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1755 {
1756 struct kvm_memory_slot *memslot;
1757
1758 memslot = gfn_to_memslot(kvm, gfn);
1759 mark_page_dirty_in_slot(kvm, memslot, gfn);
1760 }
1761 EXPORT_SYMBOL_GPL(mark_page_dirty);
1762
kvm_vcpu_check_block(struct kvm_vcpu * vcpu)1763 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1764 {
1765 if (kvm_arch_vcpu_runnable(vcpu)) {
1766 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1767 return -EINTR;
1768 }
1769 if (kvm_cpu_has_pending_timer(vcpu))
1770 return -EINTR;
1771 if (signal_pending(current))
1772 return -EINTR;
1773
1774 return 0;
1775 }
1776
1777 /*
1778 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1779 */
kvm_vcpu_block(struct kvm_vcpu * vcpu)1780 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1781 {
1782 ktime_t start, cur;
1783 DEFINE_WAIT(wait);
1784 bool waited = false;
1785
1786 start = cur = ktime_get();
1787 if (halt_poll_ns) {
1788 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1789
1790 do {
1791 /*
1792 * This sets KVM_REQ_UNHALT if an interrupt
1793 * arrives.
1794 */
1795 if (kvm_vcpu_check_block(vcpu) < 0) {
1796 ++vcpu->stat.halt_successful_poll;
1797 goto out;
1798 }
1799 cur = ktime_get();
1800 } while (single_task_running() && ktime_before(cur, stop));
1801 }
1802
1803 for (;;) {
1804 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1805
1806 if (kvm_vcpu_check_block(vcpu) < 0)
1807 break;
1808
1809 waited = true;
1810 schedule();
1811 }
1812
1813 finish_wait(&vcpu->wq, &wait);
1814 cur = ktime_get();
1815
1816 out:
1817 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1818 }
1819 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1820
1821 #ifndef CONFIG_S390
1822 /*
1823 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1824 */
kvm_vcpu_kick(struct kvm_vcpu * vcpu)1825 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1826 {
1827 int me;
1828 int cpu = vcpu->cpu;
1829 wait_queue_head_t *wqp;
1830
1831 wqp = kvm_arch_vcpu_wq(vcpu);
1832 if (waitqueue_active(wqp)) {
1833 wake_up_interruptible(wqp);
1834 ++vcpu->stat.halt_wakeup;
1835 }
1836
1837 me = get_cpu();
1838 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1839 if (kvm_arch_vcpu_should_kick(vcpu))
1840 smp_send_reschedule(cpu);
1841 put_cpu();
1842 }
1843 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1844 #endif /* !CONFIG_S390 */
1845
kvm_vcpu_yield_to(struct kvm_vcpu * target)1846 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1847 {
1848 struct pid *pid;
1849 struct task_struct *task = NULL;
1850 int ret = 0;
1851
1852 rcu_read_lock();
1853 pid = rcu_dereference(target->pid);
1854 if (pid)
1855 task = get_pid_task(pid, PIDTYPE_PID);
1856 rcu_read_unlock();
1857 if (!task)
1858 return ret;
1859 ret = yield_to(task, 1);
1860 put_task_struct(task);
1861
1862 return ret;
1863 }
1864 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1865
1866 /*
1867 * Helper that checks whether a VCPU is eligible for directed yield.
1868 * Most eligible candidate to yield is decided by following heuristics:
1869 *
1870 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1871 * (preempted lock holder), indicated by @in_spin_loop.
1872 * Set at the beiginning and cleared at the end of interception/PLE handler.
1873 *
1874 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1875 * chance last time (mostly it has become eligible now since we have probably
1876 * yielded to lockholder in last iteration. This is done by toggling
1877 * @dy_eligible each time a VCPU checked for eligibility.)
1878 *
1879 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1880 * to preempted lock-holder could result in wrong VCPU selection and CPU
1881 * burning. Giving priority for a potential lock-holder increases lock
1882 * progress.
1883 *
1884 * Since algorithm is based on heuristics, accessing another VCPU data without
1885 * locking does not harm. It may result in trying to yield to same VCPU, fail
1886 * and continue with next VCPU and so on.
1887 */
kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu * vcpu)1888 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1889 {
1890 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1891 bool eligible;
1892
1893 eligible = !vcpu->spin_loop.in_spin_loop ||
1894 vcpu->spin_loop.dy_eligible;
1895
1896 if (vcpu->spin_loop.in_spin_loop)
1897 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1898
1899 return eligible;
1900 #else
1901 return true;
1902 #endif
1903 }
1904
kvm_vcpu_on_spin(struct kvm_vcpu * me)1905 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1906 {
1907 struct kvm *kvm = me->kvm;
1908 struct kvm_vcpu *vcpu;
1909 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1910 int yielded = 0;
1911 int try = 3;
1912 int pass;
1913 int i;
1914
1915 kvm_vcpu_set_in_spin_loop(me, true);
1916 /*
1917 * We boost the priority of a VCPU that is runnable but not
1918 * currently running, because it got preempted by something
1919 * else and called schedule in __vcpu_run. Hopefully that
1920 * VCPU is holding the lock that we need and will release it.
1921 * We approximate round-robin by starting at the last boosted VCPU.
1922 */
1923 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1924 kvm_for_each_vcpu(i, vcpu, kvm) {
1925 if (!pass && i <= last_boosted_vcpu) {
1926 i = last_boosted_vcpu;
1927 continue;
1928 } else if (pass && i > last_boosted_vcpu)
1929 break;
1930 if (!ACCESS_ONCE(vcpu->preempted))
1931 continue;
1932 if (vcpu == me)
1933 continue;
1934 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1935 continue;
1936 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1937 continue;
1938
1939 yielded = kvm_vcpu_yield_to(vcpu);
1940 if (yielded > 0) {
1941 kvm->last_boosted_vcpu = i;
1942 break;
1943 } else if (yielded < 0) {
1944 try--;
1945 if (!try)
1946 break;
1947 }
1948 }
1949 }
1950 kvm_vcpu_set_in_spin_loop(me, false);
1951
1952 /* Ensure vcpu is not eligible during next spinloop */
1953 kvm_vcpu_set_dy_eligible(me, false);
1954 }
1955 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1956
kvm_vcpu_fault(struct vm_area_struct * vma,struct vm_fault * vmf)1957 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1958 {
1959 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1960 struct page *page;
1961
1962 if (vmf->pgoff == 0)
1963 page = virt_to_page(vcpu->run);
1964 #ifdef CONFIG_X86
1965 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1966 page = virt_to_page(vcpu->arch.pio_data);
1967 #endif
1968 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1969 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1970 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1971 #endif
1972 else
1973 return kvm_arch_vcpu_fault(vcpu, vmf);
1974 get_page(page);
1975 vmf->page = page;
1976 return 0;
1977 }
1978
1979 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1980 .fault = kvm_vcpu_fault,
1981 };
1982
kvm_vcpu_mmap(struct file * file,struct vm_area_struct * vma)1983 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1984 {
1985 vma->vm_ops = &kvm_vcpu_vm_ops;
1986 return 0;
1987 }
1988
kvm_vcpu_release(struct inode * inode,struct file * filp)1989 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1990 {
1991 struct kvm_vcpu *vcpu = filp->private_data;
1992
1993 kvm_put_kvm(vcpu->kvm);
1994 return 0;
1995 }
1996
1997 static struct file_operations kvm_vcpu_fops = {
1998 .release = kvm_vcpu_release,
1999 .unlocked_ioctl = kvm_vcpu_ioctl,
2000 #ifdef CONFIG_KVM_COMPAT
2001 .compat_ioctl = kvm_vcpu_compat_ioctl,
2002 #endif
2003 .mmap = kvm_vcpu_mmap,
2004 .llseek = noop_llseek,
2005 };
2006
2007 /*
2008 * Allocates an inode for the vcpu.
2009 */
create_vcpu_fd(struct kvm_vcpu * vcpu)2010 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2011 {
2012 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2013 }
2014
2015 /*
2016 * Creates some virtual cpus. Good luck creating more than one.
2017 */
kvm_vm_ioctl_create_vcpu(struct kvm * kvm,u32 id)2018 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2019 {
2020 int r;
2021 struct kvm_vcpu *vcpu, *v;
2022
2023 if (id >= KVM_MAX_VCPUS)
2024 return -EINVAL;
2025
2026 vcpu = kvm_arch_vcpu_create(kvm, id);
2027 if (IS_ERR(vcpu))
2028 return PTR_ERR(vcpu);
2029
2030 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2031
2032 r = kvm_arch_vcpu_setup(vcpu);
2033 if (r)
2034 goto vcpu_destroy;
2035
2036 mutex_lock(&kvm->lock);
2037 if (!kvm_vcpu_compatible(vcpu)) {
2038 r = -EINVAL;
2039 goto unlock_vcpu_destroy;
2040 }
2041 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2042 r = -EINVAL;
2043 goto unlock_vcpu_destroy;
2044 }
2045
2046 kvm_for_each_vcpu(r, v, kvm)
2047 if (v->vcpu_id == id) {
2048 r = -EEXIST;
2049 goto unlock_vcpu_destroy;
2050 }
2051
2052 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2053
2054 /* Now it's all set up, let userspace reach it */
2055 kvm_get_kvm(kvm);
2056 r = create_vcpu_fd(vcpu);
2057 if (r < 0) {
2058 kvm_put_kvm(kvm);
2059 goto unlock_vcpu_destroy;
2060 }
2061
2062 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2063 smp_wmb();
2064 atomic_inc(&kvm->online_vcpus);
2065
2066 mutex_unlock(&kvm->lock);
2067 kvm_arch_vcpu_postcreate(vcpu);
2068 return r;
2069
2070 unlock_vcpu_destroy:
2071 mutex_unlock(&kvm->lock);
2072 vcpu_destroy:
2073 kvm_arch_vcpu_destroy(vcpu);
2074 return r;
2075 }
2076
kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu * vcpu,sigset_t * sigset)2077 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2078 {
2079 if (sigset) {
2080 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2081 vcpu->sigset_active = 1;
2082 vcpu->sigset = *sigset;
2083 } else
2084 vcpu->sigset_active = 0;
2085 return 0;
2086 }
2087
kvm_vcpu_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2088 static long kvm_vcpu_ioctl(struct file *filp,
2089 unsigned int ioctl, unsigned long arg)
2090 {
2091 struct kvm_vcpu *vcpu = filp->private_data;
2092 void __user *argp = (void __user *)arg;
2093 int r;
2094 struct kvm_fpu *fpu = NULL;
2095 struct kvm_sregs *kvm_sregs = NULL;
2096
2097 if (vcpu->kvm->mm != current->mm)
2098 return -EIO;
2099
2100 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2101 return -EINVAL;
2102
2103 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2104 /*
2105 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2106 * so vcpu_load() would break it.
2107 */
2108 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2109 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2110 #endif
2111
2112
2113 r = vcpu_load(vcpu);
2114 if (r)
2115 return r;
2116 switch (ioctl) {
2117 case KVM_RUN:
2118 r = -EINVAL;
2119 if (arg)
2120 goto out;
2121 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2122 /* The thread running this VCPU changed. */
2123 struct pid *oldpid = vcpu->pid;
2124 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2125
2126 rcu_assign_pointer(vcpu->pid, newpid);
2127 if (oldpid)
2128 synchronize_rcu();
2129 put_pid(oldpid);
2130 }
2131 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2132 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2133 break;
2134 case KVM_GET_REGS: {
2135 struct kvm_regs *kvm_regs;
2136
2137 r = -ENOMEM;
2138 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2139 if (!kvm_regs)
2140 goto out;
2141 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2142 if (r)
2143 goto out_free1;
2144 r = -EFAULT;
2145 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2146 goto out_free1;
2147 r = 0;
2148 out_free1:
2149 kfree(kvm_regs);
2150 break;
2151 }
2152 case KVM_SET_REGS: {
2153 struct kvm_regs *kvm_regs;
2154
2155 r = -ENOMEM;
2156 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2157 if (IS_ERR(kvm_regs)) {
2158 r = PTR_ERR(kvm_regs);
2159 goto out;
2160 }
2161 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2162 kfree(kvm_regs);
2163 break;
2164 }
2165 case KVM_GET_SREGS: {
2166 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2167 r = -ENOMEM;
2168 if (!kvm_sregs)
2169 goto out;
2170 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2171 if (r)
2172 goto out;
2173 r = -EFAULT;
2174 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2175 goto out;
2176 r = 0;
2177 break;
2178 }
2179 case KVM_SET_SREGS: {
2180 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2181 if (IS_ERR(kvm_sregs)) {
2182 r = PTR_ERR(kvm_sregs);
2183 kvm_sregs = NULL;
2184 goto out;
2185 }
2186 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2187 break;
2188 }
2189 case KVM_GET_MP_STATE: {
2190 struct kvm_mp_state mp_state;
2191
2192 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2193 if (r)
2194 goto out;
2195 r = -EFAULT;
2196 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2197 goto out;
2198 r = 0;
2199 break;
2200 }
2201 case KVM_SET_MP_STATE: {
2202 struct kvm_mp_state mp_state;
2203
2204 r = -EFAULT;
2205 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2206 goto out;
2207 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2208 break;
2209 }
2210 case KVM_TRANSLATE: {
2211 struct kvm_translation tr;
2212
2213 r = -EFAULT;
2214 if (copy_from_user(&tr, argp, sizeof(tr)))
2215 goto out;
2216 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2217 if (r)
2218 goto out;
2219 r = -EFAULT;
2220 if (copy_to_user(argp, &tr, sizeof(tr)))
2221 goto out;
2222 r = 0;
2223 break;
2224 }
2225 case KVM_SET_GUEST_DEBUG: {
2226 struct kvm_guest_debug dbg;
2227
2228 r = -EFAULT;
2229 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2230 goto out;
2231 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2232 break;
2233 }
2234 case KVM_SET_SIGNAL_MASK: {
2235 struct kvm_signal_mask __user *sigmask_arg = argp;
2236 struct kvm_signal_mask kvm_sigmask;
2237 sigset_t sigset, *p;
2238
2239 p = NULL;
2240 if (argp) {
2241 r = -EFAULT;
2242 if (copy_from_user(&kvm_sigmask, argp,
2243 sizeof(kvm_sigmask)))
2244 goto out;
2245 r = -EINVAL;
2246 if (kvm_sigmask.len != sizeof(sigset))
2247 goto out;
2248 r = -EFAULT;
2249 if (copy_from_user(&sigset, sigmask_arg->sigset,
2250 sizeof(sigset)))
2251 goto out;
2252 p = &sigset;
2253 }
2254 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2255 break;
2256 }
2257 case KVM_GET_FPU: {
2258 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2259 r = -ENOMEM;
2260 if (!fpu)
2261 goto out;
2262 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2263 if (r)
2264 goto out;
2265 r = -EFAULT;
2266 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2267 goto out;
2268 r = 0;
2269 break;
2270 }
2271 case KVM_SET_FPU: {
2272 fpu = memdup_user(argp, sizeof(*fpu));
2273 if (IS_ERR(fpu)) {
2274 r = PTR_ERR(fpu);
2275 fpu = NULL;
2276 goto out;
2277 }
2278 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2279 break;
2280 }
2281 default:
2282 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2283 }
2284 out:
2285 vcpu_put(vcpu);
2286 kfree(fpu);
2287 kfree(kvm_sregs);
2288 return r;
2289 }
2290
2291 #ifdef CONFIG_KVM_COMPAT
kvm_vcpu_compat_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2292 static long kvm_vcpu_compat_ioctl(struct file *filp,
2293 unsigned int ioctl, unsigned long arg)
2294 {
2295 struct kvm_vcpu *vcpu = filp->private_data;
2296 void __user *argp = compat_ptr(arg);
2297 int r;
2298
2299 if (vcpu->kvm->mm != current->mm)
2300 return -EIO;
2301
2302 switch (ioctl) {
2303 case KVM_SET_SIGNAL_MASK: {
2304 struct kvm_signal_mask __user *sigmask_arg = argp;
2305 struct kvm_signal_mask kvm_sigmask;
2306 compat_sigset_t csigset;
2307 sigset_t sigset;
2308
2309 if (argp) {
2310 r = -EFAULT;
2311 if (copy_from_user(&kvm_sigmask, argp,
2312 sizeof(kvm_sigmask)))
2313 goto out;
2314 r = -EINVAL;
2315 if (kvm_sigmask.len != sizeof(csigset))
2316 goto out;
2317 r = -EFAULT;
2318 if (copy_from_user(&csigset, sigmask_arg->sigset,
2319 sizeof(csigset)))
2320 goto out;
2321 sigset_from_compat(&sigset, &csigset);
2322 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2323 } else
2324 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2325 break;
2326 }
2327 default:
2328 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2329 }
2330
2331 out:
2332 return r;
2333 }
2334 #endif
2335
kvm_device_ioctl_attr(struct kvm_device * dev,int (* accessor)(struct kvm_device * dev,struct kvm_device_attr * attr),unsigned long arg)2336 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2337 int (*accessor)(struct kvm_device *dev,
2338 struct kvm_device_attr *attr),
2339 unsigned long arg)
2340 {
2341 struct kvm_device_attr attr;
2342
2343 if (!accessor)
2344 return -EPERM;
2345
2346 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2347 return -EFAULT;
2348
2349 return accessor(dev, &attr);
2350 }
2351
kvm_device_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2352 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2353 unsigned long arg)
2354 {
2355 struct kvm_device *dev = filp->private_data;
2356
2357 switch (ioctl) {
2358 case KVM_SET_DEVICE_ATTR:
2359 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2360 case KVM_GET_DEVICE_ATTR:
2361 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2362 case KVM_HAS_DEVICE_ATTR:
2363 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2364 default:
2365 if (dev->ops->ioctl)
2366 return dev->ops->ioctl(dev, ioctl, arg);
2367
2368 return -ENOTTY;
2369 }
2370 }
2371
kvm_device_release(struct inode * inode,struct file * filp)2372 static int kvm_device_release(struct inode *inode, struct file *filp)
2373 {
2374 struct kvm_device *dev = filp->private_data;
2375 struct kvm *kvm = dev->kvm;
2376
2377 kvm_put_kvm(kvm);
2378 return 0;
2379 }
2380
2381 static const struct file_operations kvm_device_fops = {
2382 .unlocked_ioctl = kvm_device_ioctl,
2383 #ifdef CONFIG_KVM_COMPAT
2384 .compat_ioctl = kvm_device_ioctl,
2385 #endif
2386 .release = kvm_device_release,
2387 };
2388
kvm_device_from_filp(struct file * filp)2389 struct kvm_device *kvm_device_from_filp(struct file *filp)
2390 {
2391 if (filp->f_op != &kvm_device_fops)
2392 return NULL;
2393
2394 return filp->private_data;
2395 }
2396
2397 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2398 #ifdef CONFIG_KVM_MPIC
2399 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2400 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2401 #endif
2402
2403 #ifdef CONFIG_KVM_XICS
2404 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2405 #endif
2406 };
2407
kvm_register_device_ops(struct kvm_device_ops * ops,u32 type)2408 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2409 {
2410 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2411 return -ENOSPC;
2412
2413 if (kvm_device_ops_table[type] != NULL)
2414 return -EEXIST;
2415
2416 kvm_device_ops_table[type] = ops;
2417 return 0;
2418 }
2419
kvm_unregister_device_ops(u32 type)2420 void kvm_unregister_device_ops(u32 type)
2421 {
2422 if (kvm_device_ops_table[type] != NULL)
2423 kvm_device_ops_table[type] = NULL;
2424 }
2425
kvm_ioctl_create_device(struct kvm * kvm,struct kvm_create_device * cd)2426 static int kvm_ioctl_create_device(struct kvm *kvm,
2427 struct kvm_create_device *cd)
2428 {
2429 struct kvm_device_ops *ops = NULL;
2430 struct kvm_device *dev;
2431 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2432 int ret;
2433
2434 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2435 return -ENODEV;
2436
2437 ops = kvm_device_ops_table[cd->type];
2438 if (ops == NULL)
2439 return -ENODEV;
2440
2441 if (test)
2442 return 0;
2443
2444 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2445 if (!dev)
2446 return -ENOMEM;
2447
2448 dev->ops = ops;
2449 dev->kvm = kvm;
2450
2451 ret = ops->create(dev, cd->type);
2452 if (ret < 0) {
2453 kfree(dev);
2454 return ret;
2455 }
2456
2457 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2458 if (ret < 0) {
2459 ops->destroy(dev);
2460 return ret;
2461 }
2462
2463 list_add(&dev->vm_node, &kvm->devices);
2464 kvm_get_kvm(kvm);
2465 cd->fd = ret;
2466 return 0;
2467 }
2468
kvm_vm_ioctl_check_extension_generic(struct kvm * kvm,long arg)2469 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2470 {
2471 switch (arg) {
2472 case KVM_CAP_USER_MEMORY:
2473 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2474 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2475 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2476 case KVM_CAP_SET_BOOT_CPU_ID:
2477 #endif
2478 case KVM_CAP_INTERNAL_ERROR_DATA:
2479 #ifdef CONFIG_HAVE_KVM_MSI
2480 case KVM_CAP_SIGNAL_MSI:
2481 #endif
2482 #ifdef CONFIG_HAVE_KVM_IRQFD
2483 case KVM_CAP_IRQFD:
2484 case KVM_CAP_IRQFD_RESAMPLE:
2485 #endif
2486 case KVM_CAP_CHECK_EXTENSION_VM:
2487 return 1;
2488 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2489 case KVM_CAP_IRQ_ROUTING:
2490 return KVM_MAX_IRQ_ROUTES;
2491 #endif
2492 default:
2493 break;
2494 }
2495 return kvm_vm_ioctl_check_extension(kvm, arg);
2496 }
2497
kvm_vm_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2498 static long kvm_vm_ioctl(struct file *filp,
2499 unsigned int ioctl, unsigned long arg)
2500 {
2501 struct kvm *kvm = filp->private_data;
2502 void __user *argp = (void __user *)arg;
2503 int r;
2504
2505 if (kvm->mm != current->mm)
2506 return -EIO;
2507 switch (ioctl) {
2508 case KVM_CREATE_VCPU:
2509 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2510 break;
2511 case KVM_SET_USER_MEMORY_REGION: {
2512 struct kvm_userspace_memory_region kvm_userspace_mem;
2513
2514 r = -EFAULT;
2515 if (copy_from_user(&kvm_userspace_mem, argp,
2516 sizeof(kvm_userspace_mem)))
2517 goto out;
2518
2519 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2520 break;
2521 }
2522 case KVM_GET_DIRTY_LOG: {
2523 struct kvm_dirty_log log;
2524
2525 r = -EFAULT;
2526 if (copy_from_user(&log, argp, sizeof(log)))
2527 goto out;
2528 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2529 break;
2530 }
2531 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2532 case KVM_REGISTER_COALESCED_MMIO: {
2533 struct kvm_coalesced_mmio_zone zone;
2534
2535 r = -EFAULT;
2536 if (copy_from_user(&zone, argp, sizeof(zone)))
2537 goto out;
2538 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2539 break;
2540 }
2541 case KVM_UNREGISTER_COALESCED_MMIO: {
2542 struct kvm_coalesced_mmio_zone zone;
2543
2544 r = -EFAULT;
2545 if (copy_from_user(&zone, argp, sizeof(zone)))
2546 goto out;
2547 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2548 break;
2549 }
2550 #endif
2551 case KVM_IRQFD: {
2552 struct kvm_irqfd data;
2553
2554 r = -EFAULT;
2555 if (copy_from_user(&data, argp, sizeof(data)))
2556 goto out;
2557 r = kvm_irqfd(kvm, &data);
2558 break;
2559 }
2560 case KVM_IOEVENTFD: {
2561 struct kvm_ioeventfd data;
2562
2563 r = -EFAULT;
2564 if (copy_from_user(&data, argp, sizeof(data)))
2565 goto out;
2566 r = kvm_ioeventfd(kvm, &data);
2567 break;
2568 }
2569 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2570 case KVM_SET_BOOT_CPU_ID:
2571 r = 0;
2572 mutex_lock(&kvm->lock);
2573 if (atomic_read(&kvm->online_vcpus) != 0)
2574 r = -EBUSY;
2575 else
2576 kvm->bsp_vcpu_id = arg;
2577 mutex_unlock(&kvm->lock);
2578 break;
2579 #endif
2580 #ifdef CONFIG_HAVE_KVM_MSI
2581 case KVM_SIGNAL_MSI: {
2582 struct kvm_msi msi;
2583
2584 r = -EFAULT;
2585 if (copy_from_user(&msi, argp, sizeof(msi)))
2586 goto out;
2587 r = kvm_send_userspace_msi(kvm, &msi);
2588 break;
2589 }
2590 #endif
2591 #ifdef __KVM_HAVE_IRQ_LINE
2592 case KVM_IRQ_LINE_STATUS:
2593 case KVM_IRQ_LINE: {
2594 struct kvm_irq_level irq_event;
2595
2596 r = -EFAULT;
2597 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2598 goto out;
2599
2600 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2601 ioctl == KVM_IRQ_LINE_STATUS);
2602 if (r)
2603 goto out;
2604
2605 r = -EFAULT;
2606 if (ioctl == KVM_IRQ_LINE_STATUS) {
2607 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2608 goto out;
2609 }
2610
2611 r = 0;
2612 break;
2613 }
2614 #endif
2615 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2616 case KVM_SET_GSI_ROUTING: {
2617 struct kvm_irq_routing routing;
2618 struct kvm_irq_routing __user *urouting;
2619 struct kvm_irq_routing_entry *entries;
2620
2621 r = -EFAULT;
2622 if (copy_from_user(&routing, argp, sizeof(routing)))
2623 goto out;
2624 r = -EINVAL;
2625 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2626 goto out;
2627 if (routing.flags)
2628 goto out;
2629 r = -ENOMEM;
2630 entries = vmalloc(routing.nr * sizeof(*entries));
2631 if (!entries)
2632 goto out;
2633 r = -EFAULT;
2634 urouting = argp;
2635 if (copy_from_user(entries, urouting->entries,
2636 routing.nr * sizeof(*entries)))
2637 goto out_free_irq_routing;
2638 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2639 routing.flags);
2640 out_free_irq_routing:
2641 vfree(entries);
2642 break;
2643 }
2644 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2645 case KVM_CREATE_DEVICE: {
2646 struct kvm_create_device cd;
2647
2648 r = -EFAULT;
2649 if (copy_from_user(&cd, argp, sizeof(cd)))
2650 goto out;
2651
2652 r = kvm_ioctl_create_device(kvm, &cd);
2653 if (r)
2654 goto out;
2655
2656 r = -EFAULT;
2657 if (copy_to_user(argp, &cd, sizeof(cd)))
2658 goto out;
2659
2660 r = 0;
2661 break;
2662 }
2663 case KVM_CHECK_EXTENSION:
2664 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2665 break;
2666 default:
2667 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2668 }
2669 out:
2670 return r;
2671 }
2672
2673 #ifdef CONFIG_KVM_COMPAT
2674 struct compat_kvm_dirty_log {
2675 __u32 slot;
2676 __u32 padding1;
2677 union {
2678 compat_uptr_t dirty_bitmap; /* one bit per page */
2679 __u64 padding2;
2680 };
2681 };
2682
kvm_vm_compat_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2683 static long kvm_vm_compat_ioctl(struct file *filp,
2684 unsigned int ioctl, unsigned long arg)
2685 {
2686 struct kvm *kvm = filp->private_data;
2687 int r;
2688
2689 if (kvm->mm != current->mm)
2690 return -EIO;
2691 switch (ioctl) {
2692 case KVM_GET_DIRTY_LOG: {
2693 struct compat_kvm_dirty_log compat_log;
2694 struct kvm_dirty_log log;
2695
2696 r = -EFAULT;
2697 if (copy_from_user(&compat_log, (void __user *)arg,
2698 sizeof(compat_log)))
2699 goto out;
2700 log.slot = compat_log.slot;
2701 log.padding1 = compat_log.padding1;
2702 log.padding2 = compat_log.padding2;
2703 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2704
2705 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2706 break;
2707 }
2708 default:
2709 r = kvm_vm_ioctl(filp, ioctl, arg);
2710 }
2711
2712 out:
2713 return r;
2714 }
2715 #endif
2716
2717 static struct file_operations kvm_vm_fops = {
2718 .release = kvm_vm_release,
2719 .unlocked_ioctl = kvm_vm_ioctl,
2720 #ifdef CONFIG_KVM_COMPAT
2721 .compat_ioctl = kvm_vm_compat_ioctl,
2722 #endif
2723 .llseek = noop_llseek,
2724 };
2725
kvm_dev_ioctl_create_vm(unsigned long type)2726 static int kvm_dev_ioctl_create_vm(unsigned long type)
2727 {
2728 int r;
2729 struct kvm *kvm;
2730
2731 kvm = kvm_create_vm(type);
2732 if (IS_ERR(kvm))
2733 return PTR_ERR(kvm);
2734 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2735 r = kvm_coalesced_mmio_init(kvm);
2736 if (r < 0) {
2737 kvm_put_kvm(kvm);
2738 return r;
2739 }
2740 #endif
2741 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2742 if (r < 0)
2743 kvm_put_kvm(kvm);
2744
2745 return r;
2746 }
2747
kvm_dev_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)2748 static long kvm_dev_ioctl(struct file *filp,
2749 unsigned int ioctl, unsigned long arg)
2750 {
2751 long r = -EINVAL;
2752
2753 switch (ioctl) {
2754 case KVM_GET_API_VERSION:
2755 if (arg)
2756 goto out;
2757 r = KVM_API_VERSION;
2758 break;
2759 case KVM_CREATE_VM:
2760 r = kvm_dev_ioctl_create_vm(arg);
2761 break;
2762 case KVM_CHECK_EXTENSION:
2763 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2764 break;
2765 case KVM_GET_VCPU_MMAP_SIZE:
2766 if (arg)
2767 goto out;
2768 r = PAGE_SIZE; /* struct kvm_run */
2769 #ifdef CONFIG_X86
2770 r += PAGE_SIZE; /* pio data page */
2771 #endif
2772 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2773 r += PAGE_SIZE; /* coalesced mmio ring page */
2774 #endif
2775 break;
2776 case KVM_TRACE_ENABLE:
2777 case KVM_TRACE_PAUSE:
2778 case KVM_TRACE_DISABLE:
2779 r = -EOPNOTSUPP;
2780 break;
2781 default:
2782 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2783 }
2784 out:
2785 return r;
2786 }
2787
2788 static struct file_operations kvm_chardev_ops = {
2789 .unlocked_ioctl = kvm_dev_ioctl,
2790 .compat_ioctl = kvm_dev_ioctl,
2791 .llseek = noop_llseek,
2792 };
2793
2794 static struct miscdevice kvm_dev = {
2795 KVM_MINOR,
2796 "kvm",
2797 &kvm_chardev_ops,
2798 };
2799
hardware_enable_nolock(void * junk)2800 static void hardware_enable_nolock(void *junk)
2801 {
2802 int cpu = raw_smp_processor_id();
2803 int r;
2804
2805 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2806 return;
2807
2808 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2809
2810 r = kvm_arch_hardware_enable();
2811
2812 if (r) {
2813 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2814 atomic_inc(&hardware_enable_failed);
2815 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2816 }
2817 }
2818
hardware_enable(void)2819 static void hardware_enable(void)
2820 {
2821 raw_spin_lock(&kvm_count_lock);
2822 if (kvm_usage_count)
2823 hardware_enable_nolock(NULL);
2824 raw_spin_unlock(&kvm_count_lock);
2825 }
2826
hardware_disable_nolock(void * junk)2827 static void hardware_disable_nolock(void *junk)
2828 {
2829 int cpu = raw_smp_processor_id();
2830
2831 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2832 return;
2833 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2834 kvm_arch_hardware_disable();
2835 }
2836
hardware_disable(void)2837 static void hardware_disable(void)
2838 {
2839 raw_spin_lock(&kvm_count_lock);
2840 if (kvm_usage_count)
2841 hardware_disable_nolock(NULL);
2842 raw_spin_unlock(&kvm_count_lock);
2843 }
2844
hardware_disable_all_nolock(void)2845 static void hardware_disable_all_nolock(void)
2846 {
2847 BUG_ON(!kvm_usage_count);
2848
2849 kvm_usage_count--;
2850 if (!kvm_usage_count)
2851 on_each_cpu(hardware_disable_nolock, NULL, 1);
2852 }
2853
hardware_disable_all(void)2854 static void hardware_disable_all(void)
2855 {
2856 raw_spin_lock(&kvm_count_lock);
2857 hardware_disable_all_nolock();
2858 raw_spin_unlock(&kvm_count_lock);
2859 }
2860
hardware_enable_all(void)2861 static int hardware_enable_all(void)
2862 {
2863 int r = 0;
2864
2865 raw_spin_lock(&kvm_count_lock);
2866
2867 kvm_usage_count++;
2868 if (kvm_usage_count == 1) {
2869 atomic_set(&hardware_enable_failed, 0);
2870 on_each_cpu(hardware_enable_nolock, NULL, 1);
2871
2872 if (atomic_read(&hardware_enable_failed)) {
2873 hardware_disable_all_nolock();
2874 r = -EBUSY;
2875 }
2876 }
2877
2878 raw_spin_unlock(&kvm_count_lock);
2879
2880 return r;
2881 }
2882
kvm_cpu_hotplug(struct notifier_block * notifier,unsigned long val,void * v)2883 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2884 void *v)
2885 {
2886 int cpu = (long)v;
2887
2888 val &= ~CPU_TASKS_FROZEN;
2889 switch (val) {
2890 case CPU_DYING:
2891 pr_info("kvm: disabling virtualization on CPU%d\n",
2892 cpu);
2893 hardware_disable();
2894 break;
2895 case CPU_STARTING:
2896 pr_info("kvm: enabling virtualization on CPU%d\n",
2897 cpu);
2898 hardware_enable();
2899 break;
2900 }
2901 return NOTIFY_OK;
2902 }
2903
kvm_reboot(struct notifier_block * notifier,unsigned long val,void * v)2904 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2905 void *v)
2906 {
2907 /*
2908 * Some (well, at least mine) BIOSes hang on reboot if
2909 * in vmx root mode.
2910 *
2911 * And Intel TXT required VMX off for all cpu when system shutdown.
2912 */
2913 pr_info("kvm: exiting hardware virtualization\n");
2914 kvm_rebooting = true;
2915 on_each_cpu(hardware_disable_nolock, NULL, 1);
2916 return NOTIFY_OK;
2917 }
2918
2919 static struct notifier_block kvm_reboot_notifier = {
2920 .notifier_call = kvm_reboot,
2921 .priority = 0,
2922 };
2923
kvm_io_bus_destroy(struct kvm_io_bus * bus)2924 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2925 {
2926 int i;
2927
2928 for (i = 0; i < bus->dev_count; i++) {
2929 struct kvm_io_device *pos = bus->range[i].dev;
2930
2931 kvm_iodevice_destructor(pos);
2932 }
2933 kfree(bus);
2934 }
2935
kvm_io_bus_cmp(const struct kvm_io_range * r1,const struct kvm_io_range * r2)2936 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2937 const struct kvm_io_range *r2)
2938 {
2939 gpa_t addr1 = r1->addr;
2940 gpa_t addr2 = r2->addr;
2941
2942 if (addr1 < addr2)
2943 return -1;
2944
2945 /* If r2->len == 0, match the exact address. If r2->len != 0,
2946 * accept any overlapping write. Any order is acceptable for
2947 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
2948 * we process all of them.
2949 */
2950 if (r2->len) {
2951 addr1 += r1->len;
2952 addr2 += r2->len;
2953 }
2954
2955 if (addr1 > addr2)
2956 return 1;
2957
2958 return 0;
2959 }
2960
kvm_io_bus_sort_cmp(const void * p1,const void * p2)2961 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2962 {
2963 return kvm_io_bus_cmp(p1, p2);
2964 }
2965
kvm_io_bus_insert_dev(struct kvm_io_bus * bus,struct kvm_io_device * dev,gpa_t addr,int len)2966 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2967 gpa_t addr, int len)
2968 {
2969 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2970 .addr = addr,
2971 .len = len,
2972 .dev = dev,
2973 };
2974
2975 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2976 kvm_io_bus_sort_cmp, NULL);
2977
2978 return 0;
2979 }
2980
kvm_io_bus_get_first_dev(struct kvm_io_bus * bus,gpa_t addr,int len)2981 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2982 gpa_t addr, int len)
2983 {
2984 struct kvm_io_range *range, key;
2985 int off;
2986
2987 key = (struct kvm_io_range) {
2988 .addr = addr,
2989 .len = len,
2990 };
2991
2992 range = bsearch(&key, bus->range, bus->dev_count,
2993 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2994 if (range == NULL)
2995 return -ENOENT;
2996
2997 off = range - bus->range;
2998
2999 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3000 off--;
3001
3002 return off;
3003 }
3004
__kvm_io_bus_write(struct kvm_vcpu * vcpu,struct kvm_io_bus * bus,struct kvm_io_range * range,const void * val)3005 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3006 struct kvm_io_range *range, const void *val)
3007 {
3008 int idx;
3009
3010 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3011 if (idx < 0)
3012 return -EOPNOTSUPP;
3013
3014 while (idx < bus->dev_count &&
3015 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3016 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3017 range->len, val))
3018 return idx;
3019 idx++;
3020 }
3021
3022 return -EOPNOTSUPP;
3023 }
3024
3025 /* kvm_io_bus_write - called under kvm->slots_lock */
kvm_io_bus_write(struct kvm_vcpu * vcpu,enum kvm_bus bus_idx,gpa_t addr,int len,const void * val)3026 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3027 int len, const void *val)
3028 {
3029 struct kvm_io_bus *bus;
3030 struct kvm_io_range range;
3031 int r;
3032
3033 range = (struct kvm_io_range) {
3034 .addr = addr,
3035 .len = len,
3036 };
3037
3038 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3039 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3040 return r < 0 ? r : 0;
3041 }
3042
3043 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
kvm_io_bus_write_cookie(struct kvm_vcpu * vcpu,enum kvm_bus bus_idx,gpa_t addr,int len,const void * val,long cookie)3044 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3045 gpa_t addr, int len, const void *val, long cookie)
3046 {
3047 struct kvm_io_bus *bus;
3048 struct kvm_io_range range;
3049
3050 range = (struct kvm_io_range) {
3051 .addr = addr,
3052 .len = len,
3053 };
3054
3055 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3056
3057 /* First try the device referenced by cookie. */
3058 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3059 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3060 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3061 val))
3062 return cookie;
3063
3064 /*
3065 * cookie contained garbage; fall back to search and return the
3066 * correct cookie value.
3067 */
3068 return __kvm_io_bus_write(vcpu, bus, &range, val);
3069 }
3070
__kvm_io_bus_read(struct kvm_vcpu * vcpu,struct kvm_io_bus * bus,struct kvm_io_range * range,void * val)3071 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3072 struct kvm_io_range *range, void *val)
3073 {
3074 int idx;
3075
3076 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3077 if (idx < 0)
3078 return -EOPNOTSUPP;
3079
3080 while (idx < bus->dev_count &&
3081 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3082 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3083 range->len, val))
3084 return idx;
3085 idx++;
3086 }
3087
3088 return -EOPNOTSUPP;
3089 }
3090 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3091
3092 /* kvm_io_bus_read - called under kvm->slots_lock */
kvm_io_bus_read(struct kvm_vcpu * vcpu,enum kvm_bus bus_idx,gpa_t addr,int len,void * val)3093 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3094 int len, void *val)
3095 {
3096 struct kvm_io_bus *bus;
3097 struct kvm_io_range range;
3098 int r;
3099
3100 range = (struct kvm_io_range) {
3101 .addr = addr,
3102 .len = len,
3103 };
3104
3105 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3106 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3107 return r < 0 ? r : 0;
3108 }
3109
3110
3111 /* Caller must hold slots_lock. */
kvm_io_bus_register_dev(struct kvm * kvm,enum kvm_bus bus_idx,gpa_t addr,int len,struct kvm_io_device * dev)3112 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3113 int len, struct kvm_io_device *dev)
3114 {
3115 struct kvm_io_bus *new_bus, *bus;
3116
3117 bus = kvm->buses[bus_idx];
3118 /* exclude ioeventfd which is limited by maximum fd */
3119 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3120 return -ENOSPC;
3121
3122 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3123 sizeof(struct kvm_io_range)), GFP_KERNEL);
3124 if (!new_bus)
3125 return -ENOMEM;
3126 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3127 sizeof(struct kvm_io_range)));
3128 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3129 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3130 synchronize_srcu_expedited(&kvm->srcu);
3131 kfree(bus);
3132
3133 return 0;
3134 }
3135
3136 /* Caller must hold slots_lock. */
kvm_io_bus_unregister_dev(struct kvm * kvm,enum kvm_bus bus_idx,struct kvm_io_device * dev)3137 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3138 struct kvm_io_device *dev)
3139 {
3140 int i, r;
3141 struct kvm_io_bus *new_bus, *bus;
3142
3143 bus = kvm->buses[bus_idx];
3144 r = -ENOENT;
3145 for (i = 0; i < bus->dev_count; i++)
3146 if (bus->range[i].dev == dev) {
3147 r = 0;
3148 break;
3149 }
3150
3151 if (r)
3152 return r;
3153
3154 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3155 sizeof(struct kvm_io_range)), GFP_KERNEL);
3156 if (!new_bus)
3157 return -ENOMEM;
3158
3159 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3160 new_bus->dev_count--;
3161 memcpy(new_bus->range + i, bus->range + i + 1,
3162 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3163
3164 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3165 synchronize_srcu_expedited(&kvm->srcu);
3166 kfree(bus);
3167 return r;
3168 }
3169
3170 static struct notifier_block kvm_cpu_notifier = {
3171 .notifier_call = kvm_cpu_hotplug,
3172 };
3173
vm_stat_get(void * _offset,u64 * val)3174 static int vm_stat_get(void *_offset, u64 *val)
3175 {
3176 unsigned offset = (long)_offset;
3177 struct kvm *kvm;
3178
3179 *val = 0;
3180 spin_lock(&kvm_lock);
3181 list_for_each_entry(kvm, &vm_list, vm_list)
3182 *val += *(u32 *)((void *)kvm + offset);
3183 spin_unlock(&kvm_lock);
3184 return 0;
3185 }
3186
3187 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3188
vcpu_stat_get(void * _offset,u64 * val)3189 static int vcpu_stat_get(void *_offset, u64 *val)
3190 {
3191 unsigned offset = (long)_offset;
3192 struct kvm *kvm;
3193 struct kvm_vcpu *vcpu;
3194 int i;
3195
3196 *val = 0;
3197 spin_lock(&kvm_lock);
3198 list_for_each_entry(kvm, &vm_list, vm_list)
3199 kvm_for_each_vcpu(i, vcpu, kvm)
3200 *val += *(u32 *)((void *)vcpu + offset);
3201
3202 spin_unlock(&kvm_lock);
3203 return 0;
3204 }
3205
3206 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3207
3208 static const struct file_operations *stat_fops[] = {
3209 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3210 [KVM_STAT_VM] = &vm_stat_fops,
3211 };
3212
kvm_init_debug(void)3213 static int kvm_init_debug(void)
3214 {
3215 int r = -EEXIST;
3216 struct kvm_stats_debugfs_item *p;
3217
3218 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3219 if (kvm_debugfs_dir == NULL)
3220 goto out;
3221
3222 for (p = debugfs_entries; p->name; ++p) {
3223 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3224 (void *)(long)p->offset,
3225 stat_fops[p->kind]);
3226 if (p->dentry == NULL)
3227 goto out_dir;
3228 }
3229
3230 return 0;
3231
3232 out_dir:
3233 debugfs_remove_recursive(kvm_debugfs_dir);
3234 out:
3235 return r;
3236 }
3237
kvm_exit_debug(void)3238 static void kvm_exit_debug(void)
3239 {
3240 struct kvm_stats_debugfs_item *p;
3241
3242 for (p = debugfs_entries; p->name; ++p)
3243 debugfs_remove(p->dentry);
3244 debugfs_remove(kvm_debugfs_dir);
3245 }
3246
kvm_suspend(void)3247 static int kvm_suspend(void)
3248 {
3249 if (kvm_usage_count)
3250 hardware_disable_nolock(NULL);
3251 return 0;
3252 }
3253
kvm_resume(void)3254 static void kvm_resume(void)
3255 {
3256 if (kvm_usage_count) {
3257 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3258 hardware_enable_nolock(NULL);
3259 }
3260 }
3261
3262 static struct syscore_ops kvm_syscore_ops = {
3263 .suspend = kvm_suspend,
3264 .resume = kvm_resume,
3265 };
3266
3267 static inline
preempt_notifier_to_vcpu(struct preempt_notifier * pn)3268 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3269 {
3270 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3271 }
3272
kvm_sched_in(struct preempt_notifier * pn,int cpu)3273 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3274 {
3275 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3276
3277 if (vcpu->preempted)
3278 vcpu->preempted = false;
3279
3280 kvm_arch_sched_in(vcpu, cpu);
3281
3282 kvm_arch_vcpu_load(vcpu, cpu);
3283 }
3284
kvm_sched_out(struct preempt_notifier * pn,struct task_struct * next)3285 static void kvm_sched_out(struct preempt_notifier *pn,
3286 struct task_struct *next)
3287 {
3288 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3289
3290 if (current->state == TASK_RUNNING)
3291 vcpu->preempted = true;
3292 kvm_arch_vcpu_put(vcpu);
3293 }
3294
kvm_init(void * opaque,unsigned vcpu_size,unsigned vcpu_align,struct module * module)3295 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3296 struct module *module)
3297 {
3298 int r;
3299 int cpu;
3300
3301 r = kvm_arch_init(opaque);
3302 if (r)
3303 goto out_fail;
3304
3305 /*
3306 * kvm_arch_init makes sure there's at most one caller
3307 * for architectures that support multiple implementations,
3308 * like intel and amd on x86.
3309 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3310 * conflicts in case kvm is already setup for another implementation.
3311 */
3312 r = kvm_irqfd_init();
3313 if (r)
3314 goto out_irqfd;
3315
3316 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3317 r = -ENOMEM;
3318 goto out_free_0;
3319 }
3320
3321 r = kvm_arch_hardware_setup();
3322 if (r < 0)
3323 goto out_free_0a;
3324
3325 for_each_online_cpu(cpu) {
3326 smp_call_function_single(cpu,
3327 kvm_arch_check_processor_compat,
3328 &r, 1);
3329 if (r < 0)
3330 goto out_free_1;
3331 }
3332
3333 r = register_cpu_notifier(&kvm_cpu_notifier);
3334 if (r)
3335 goto out_free_2;
3336 register_reboot_notifier(&kvm_reboot_notifier);
3337
3338 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3339 if (!vcpu_align)
3340 vcpu_align = __alignof__(struct kvm_vcpu);
3341 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3342 0, NULL);
3343 if (!kvm_vcpu_cache) {
3344 r = -ENOMEM;
3345 goto out_free_3;
3346 }
3347
3348 r = kvm_async_pf_init();
3349 if (r)
3350 goto out_free;
3351
3352 kvm_chardev_ops.owner = module;
3353 kvm_vm_fops.owner = module;
3354 kvm_vcpu_fops.owner = module;
3355
3356 r = misc_register(&kvm_dev);
3357 if (r) {
3358 pr_err("kvm: misc device register failed\n");
3359 goto out_unreg;
3360 }
3361
3362 register_syscore_ops(&kvm_syscore_ops);
3363
3364 kvm_preempt_ops.sched_in = kvm_sched_in;
3365 kvm_preempt_ops.sched_out = kvm_sched_out;
3366
3367 r = kvm_init_debug();
3368 if (r) {
3369 pr_err("kvm: create debugfs files failed\n");
3370 goto out_undebugfs;
3371 }
3372
3373 r = kvm_vfio_ops_init();
3374 WARN_ON(r);
3375
3376 return 0;
3377
3378 out_undebugfs:
3379 unregister_syscore_ops(&kvm_syscore_ops);
3380 misc_deregister(&kvm_dev);
3381 out_unreg:
3382 kvm_async_pf_deinit();
3383 out_free:
3384 kmem_cache_destroy(kvm_vcpu_cache);
3385 out_free_3:
3386 unregister_reboot_notifier(&kvm_reboot_notifier);
3387 unregister_cpu_notifier(&kvm_cpu_notifier);
3388 out_free_2:
3389 out_free_1:
3390 kvm_arch_hardware_unsetup();
3391 out_free_0a:
3392 free_cpumask_var(cpus_hardware_enabled);
3393 out_free_0:
3394 kvm_irqfd_exit();
3395 out_irqfd:
3396 kvm_arch_exit();
3397 out_fail:
3398 return r;
3399 }
3400 EXPORT_SYMBOL_GPL(kvm_init);
3401
kvm_exit(void)3402 void kvm_exit(void)
3403 {
3404 kvm_exit_debug();
3405 misc_deregister(&kvm_dev);
3406 kmem_cache_destroy(kvm_vcpu_cache);
3407 kvm_async_pf_deinit();
3408 unregister_syscore_ops(&kvm_syscore_ops);
3409 unregister_reboot_notifier(&kvm_reboot_notifier);
3410 unregister_cpu_notifier(&kvm_cpu_notifier);
3411 on_each_cpu(hardware_disable_nolock, NULL, 1);
3412 kvm_arch_hardware_unsetup();
3413 kvm_arch_exit();
3414 kvm_irqfd_exit();
3415 free_cpumask_var(cpus_hardware_enabled);
3416 kvm_vfio_ops_exit();
3417 }
3418 EXPORT_SYMBOL_GPL(kvm_exit);
3419