~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/virt/kvm/kvm_main.c

Version: ~ [ linux-5.1-rc2 ] ~ [ linux-5.0.4 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.31 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.108 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.165 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.177 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.137 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/signal.h>
 36 #include <linux/sched/mm.h>
 37 #include <linux/sched/stat.h>
 38 #include <linux/cpumask.h>
 39 #include <linux/smp.h>
 40 #include <linux/anon_inodes.h>
 41 #include <linux/profile.h>
 42 #include <linux/kvm_para.h>
 43 #include <linux/pagemap.h>
 44 #include <linux/mman.h>
 45 #include <linux/swap.h>
 46 #include <linux/bitops.h>
 47 #include <linux/spinlock.h>
 48 #include <linux/compat.h>
 49 #include <linux/srcu.h>
 50 #include <linux/hugetlb.h>
 51 #include <linux/slab.h>
 52 #include <linux/sort.h>
 53 #include <linux/bsearch.h>
 54 
 55 #include <asm/processor.h>
 56 #include <asm/io.h>
 57 #include <asm/ioctl.h>
 58 #include <linux/uaccess.h>
 59 #include <asm/pgtable.h>
 60 
 61 #include "coalesced_mmio.h"
 62 #include "async_pf.h"
 63 #include "vfio.h"
 64 
 65 #define CREATE_TRACE_POINTS
 66 #include <trace/events/kvm.h>
 67 
 68 /* Worst case buffer size needed for holding an integer. */
 69 #define ITOA_MAX_LEN 12
 70 
 71 MODULE_AUTHOR("Qumranet");
 72 MODULE_LICENSE("GPL");
 73 
 74 /* Architectures should define their poll value according to the halt latency */
 75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
 76 module_param(halt_poll_ns, uint, 0644);
 77 EXPORT_SYMBOL_GPL(halt_poll_ns);
 78 
 79 /* Default doubles per-vcpu halt_poll_ns. */
 80 unsigned int halt_poll_ns_grow = 2;
 81 module_param(halt_poll_ns_grow, uint, 0644);
 82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
 83 
 84 /* The start value to grow halt_poll_ns from */
 85 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
 86 module_param(halt_poll_ns_grow_start, uint, 0644);
 87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
 88 
 89 /* Default resets per-vcpu halt_poll_ns . */
 90 unsigned int halt_poll_ns_shrink;
 91 module_param(halt_poll_ns_shrink, uint, 0644);
 92 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
 93 
 94 /*
 95  * Ordering of locks:
 96  *
 97  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
 98  */
 99 
100 DEFINE_SPINLOCK(kvm_lock);
101 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 LIST_HEAD(vm_list);
103 
104 static cpumask_var_t cpus_hardware_enabled;
105 static int kvm_usage_count;
106 static atomic_t hardware_enable_failed;
107 
108 struct kmem_cache *kvm_vcpu_cache;
109 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
110 
111 static __read_mostly struct preempt_ops kvm_preempt_ops;
112 
113 struct dentry *kvm_debugfs_dir;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
115 
116 static int kvm_debugfs_num_entries;
117 static const struct file_operations *stat_fops_per_vm[];
118 
119 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120                            unsigned long arg);
121 #ifdef CONFIG_KVM_COMPAT
122 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123                                   unsigned long arg);
124 #define KVM_COMPAT(c)   .compat_ioctl   = (c)
125 #else
126 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
127                                 unsigned long arg) { return -EINVAL; }
128 #define KVM_COMPAT(c)   .compat_ioctl   = kvm_no_compat_ioctl
129 #endif
130 static int hardware_enable_all(void);
131 static void hardware_disable_all(void);
132 
133 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
134 
135 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
136 
137 __visible bool kvm_rebooting;
138 EXPORT_SYMBOL_GPL(kvm_rebooting);
139 
140 static bool largepages_enabled = true;
141 
142 #define KVM_EVENT_CREATE_VM 0
143 #define KVM_EVENT_DESTROY_VM 1
144 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
145 static unsigned long long kvm_createvm_count;
146 static unsigned long long kvm_active_vms;
147 
148 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
149                 unsigned long start, unsigned long end, bool blockable)
150 {
151         return 0;
152 }
153 
154 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
155 {
156         if (pfn_valid(pfn))
157                 return PageReserved(pfn_to_page(pfn));
158 
159         return true;
160 }
161 
162 /*
163  * Switches to specified vcpu, until a matching vcpu_put()
164  */
165 void vcpu_load(struct kvm_vcpu *vcpu)
166 {
167         int cpu = get_cpu();
168         preempt_notifier_register(&vcpu->preempt_notifier);
169         kvm_arch_vcpu_load(vcpu, cpu);
170         put_cpu();
171 }
172 EXPORT_SYMBOL_GPL(vcpu_load);
173 
174 void vcpu_put(struct kvm_vcpu *vcpu)
175 {
176         preempt_disable();
177         kvm_arch_vcpu_put(vcpu);
178         preempt_notifier_unregister(&vcpu->preempt_notifier);
179         preempt_enable();
180 }
181 EXPORT_SYMBOL_GPL(vcpu_put);
182 
183 /* TODO: merge with kvm_arch_vcpu_should_kick */
184 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
185 {
186         int mode = kvm_vcpu_exiting_guest_mode(vcpu);
187 
188         /*
189          * We need to wait for the VCPU to reenable interrupts and get out of
190          * READING_SHADOW_PAGE_TABLES mode.
191          */
192         if (req & KVM_REQUEST_WAIT)
193                 return mode != OUTSIDE_GUEST_MODE;
194 
195         /*
196          * Need to kick a running VCPU, but otherwise there is nothing to do.
197          */
198         return mode == IN_GUEST_MODE;
199 }
200 
201 static void ack_flush(void *_completed)
202 {
203 }
204 
205 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
206 {
207         if (unlikely(!cpus))
208                 cpus = cpu_online_mask;
209 
210         if (cpumask_empty(cpus))
211                 return false;
212 
213         smp_call_function_many(cpus, ack_flush, NULL, wait);
214         return true;
215 }
216 
217 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
218                                  unsigned long *vcpu_bitmap, cpumask_var_t tmp)
219 {
220         int i, cpu, me;
221         struct kvm_vcpu *vcpu;
222         bool called;
223 
224         me = get_cpu();
225 
226         kvm_for_each_vcpu(i, vcpu, kvm) {
227                 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
228                         continue;
229 
230                 kvm_make_request(req, vcpu);
231                 cpu = vcpu->cpu;
232 
233                 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
234                         continue;
235 
236                 if (tmp != NULL && cpu != -1 && cpu != me &&
237                     kvm_request_needs_ipi(vcpu, req))
238                         __cpumask_set_cpu(cpu, tmp);
239         }
240 
241         called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
242         put_cpu();
243 
244         return called;
245 }
246 
247 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
248 {
249         cpumask_var_t cpus;
250         bool called;
251 
252         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
253 
254         called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
255 
256         free_cpumask_var(cpus);
257         return called;
258 }
259 
260 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
261 void kvm_flush_remote_tlbs(struct kvm *kvm)
262 {
263         /*
264          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
265          * kvm_make_all_cpus_request.
266          */
267         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
268 
269         /*
270          * We want to publish modifications to the page tables before reading
271          * mode. Pairs with a memory barrier in arch-specific code.
272          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
273          * and smp_mb in walk_shadow_page_lockless_begin/end.
274          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
275          *
276          * There is already an smp_mb__after_atomic() before
277          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
278          * barrier here.
279          */
280         if (!kvm_arch_flush_remote_tlb(kvm)
281             || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
282                 ++kvm->stat.remote_tlb_flush;
283         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
284 }
285 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
286 #endif
287 
288 void kvm_reload_remote_mmus(struct kvm *kvm)
289 {
290         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
291 }
292 
293 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
294 {
295         struct page *page;
296         int r;
297 
298         mutex_init(&vcpu->mutex);
299         vcpu->cpu = -1;
300         vcpu->kvm = kvm;
301         vcpu->vcpu_id = id;
302         vcpu->pid = NULL;
303         init_swait_queue_head(&vcpu->wq);
304         kvm_async_pf_vcpu_init(vcpu);
305 
306         vcpu->pre_pcpu = -1;
307         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
308 
309         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
310         if (!page) {
311                 r = -ENOMEM;
312                 goto fail;
313         }
314         vcpu->run = page_address(page);
315 
316         kvm_vcpu_set_in_spin_loop(vcpu, false);
317         kvm_vcpu_set_dy_eligible(vcpu, false);
318         vcpu->preempted = false;
319 
320         r = kvm_arch_vcpu_init(vcpu);
321         if (r < 0)
322                 goto fail_free_run;
323         return 0;
324 
325 fail_free_run:
326         free_page((unsigned long)vcpu->run);
327 fail:
328         return r;
329 }
330 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
331 
332 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
333 {
334         /*
335          * no need for rcu_read_lock as VCPU_RUN is the only place that
336          * will change the vcpu->pid pointer and on uninit all file
337          * descriptors are already gone.
338          */
339         put_pid(rcu_dereference_protected(vcpu->pid, 1));
340         kvm_arch_vcpu_uninit(vcpu);
341         free_page((unsigned long)vcpu->run);
342 }
343 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
344 
345 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
346 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
347 {
348         return container_of(mn, struct kvm, mmu_notifier);
349 }
350 
351 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
352                                         struct mm_struct *mm,
353                                         unsigned long address,
354                                         pte_t pte)
355 {
356         struct kvm *kvm = mmu_notifier_to_kvm(mn);
357         int idx;
358 
359         idx = srcu_read_lock(&kvm->srcu);
360         spin_lock(&kvm->mmu_lock);
361         kvm->mmu_notifier_seq++;
362 
363         if (kvm_set_spte_hva(kvm, address, pte))
364                 kvm_flush_remote_tlbs(kvm);
365 
366         spin_unlock(&kvm->mmu_lock);
367         srcu_read_unlock(&kvm->srcu, idx);
368 }
369 
370 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
371                                         const struct mmu_notifier_range *range)
372 {
373         struct kvm *kvm = mmu_notifier_to_kvm(mn);
374         int need_tlb_flush = 0, idx;
375         int ret;
376 
377         idx = srcu_read_lock(&kvm->srcu);
378         spin_lock(&kvm->mmu_lock);
379         /*
380          * The count increase must become visible at unlock time as no
381          * spte can be established without taking the mmu_lock and
382          * count is also read inside the mmu_lock critical section.
383          */
384         kvm->mmu_notifier_count++;
385         need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
386         need_tlb_flush |= kvm->tlbs_dirty;
387         /* we've to flush the tlb before the pages can be freed */
388         if (need_tlb_flush)
389                 kvm_flush_remote_tlbs(kvm);
390 
391         spin_unlock(&kvm->mmu_lock);
392 
393         ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
394                                         range->end, range->blockable);
395 
396         srcu_read_unlock(&kvm->srcu, idx);
397 
398         return ret;
399 }
400 
401 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
402                                         const struct mmu_notifier_range *range)
403 {
404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
405 
406         spin_lock(&kvm->mmu_lock);
407         /*
408          * This sequence increase will notify the kvm page fault that
409          * the page that is going to be mapped in the spte could have
410          * been freed.
411          */
412         kvm->mmu_notifier_seq++;
413         smp_wmb();
414         /*
415          * The above sequence increase must be visible before the
416          * below count decrease, which is ensured by the smp_wmb above
417          * in conjunction with the smp_rmb in mmu_notifier_retry().
418          */
419         kvm->mmu_notifier_count--;
420         spin_unlock(&kvm->mmu_lock);
421 
422         BUG_ON(kvm->mmu_notifier_count < 0);
423 }
424 
425 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
426                                               struct mm_struct *mm,
427                                               unsigned long start,
428                                               unsigned long end)
429 {
430         struct kvm *kvm = mmu_notifier_to_kvm(mn);
431         int young, idx;
432 
433         idx = srcu_read_lock(&kvm->srcu);
434         spin_lock(&kvm->mmu_lock);
435 
436         young = kvm_age_hva(kvm, start, end);
437         if (young)
438                 kvm_flush_remote_tlbs(kvm);
439 
440         spin_unlock(&kvm->mmu_lock);
441         srcu_read_unlock(&kvm->srcu, idx);
442 
443         return young;
444 }
445 
446 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
447                                         struct mm_struct *mm,
448                                         unsigned long start,
449                                         unsigned long end)
450 {
451         struct kvm *kvm = mmu_notifier_to_kvm(mn);
452         int young, idx;
453 
454         idx = srcu_read_lock(&kvm->srcu);
455         spin_lock(&kvm->mmu_lock);
456         /*
457          * Even though we do not flush TLB, this will still adversely
458          * affect performance on pre-Haswell Intel EPT, where there is
459          * no EPT Access Bit to clear so that we have to tear down EPT
460          * tables instead. If we find this unacceptable, we can always
461          * add a parameter to kvm_age_hva so that it effectively doesn't
462          * do anything on clear_young.
463          *
464          * Also note that currently we never issue secondary TLB flushes
465          * from clear_young, leaving this job up to the regular system
466          * cadence. If we find this inaccurate, we might come up with a
467          * more sophisticated heuristic later.
468          */
469         young = kvm_age_hva(kvm, start, end);
470         spin_unlock(&kvm->mmu_lock);
471         srcu_read_unlock(&kvm->srcu, idx);
472 
473         return young;
474 }
475 
476 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
477                                        struct mm_struct *mm,
478                                        unsigned long address)
479 {
480         struct kvm *kvm = mmu_notifier_to_kvm(mn);
481         int young, idx;
482 
483         idx = srcu_read_lock(&kvm->srcu);
484         spin_lock(&kvm->mmu_lock);
485         young = kvm_test_age_hva(kvm, address);
486         spin_unlock(&kvm->mmu_lock);
487         srcu_read_unlock(&kvm->srcu, idx);
488 
489         return young;
490 }
491 
492 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
493                                      struct mm_struct *mm)
494 {
495         struct kvm *kvm = mmu_notifier_to_kvm(mn);
496         int idx;
497 
498         idx = srcu_read_lock(&kvm->srcu);
499         kvm_arch_flush_shadow_all(kvm);
500         srcu_read_unlock(&kvm->srcu, idx);
501 }
502 
503 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
504         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
505         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
506         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
507         .clear_young            = kvm_mmu_notifier_clear_young,
508         .test_young             = kvm_mmu_notifier_test_young,
509         .change_pte             = kvm_mmu_notifier_change_pte,
510         .release                = kvm_mmu_notifier_release,
511 };
512 
513 static int kvm_init_mmu_notifier(struct kvm *kvm)
514 {
515         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
516         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
517 }
518 
519 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
520 
521 static int kvm_init_mmu_notifier(struct kvm *kvm)
522 {
523         return 0;
524 }
525 
526 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
527 
528 static struct kvm_memslots *kvm_alloc_memslots(void)
529 {
530         int i;
531         struct kvm_memslots *slots;
532 
533         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
534         if (!slots)
535                 return NULL;
536 
537         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
538                 slots->id_to_index[i] = slots->memslots[i].id = i;
539 
540         return slots;
541 }
542 
543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
544 {
545         if (!memslot->dirty_bitmap)
546                 return;
547 
548         kvfree(memslot->dirty_bitmap);
549         memslot->dirty_bitmap = NULL;
550 }
551 
552 /*
553  * Free any memory in @free but not in @dont.
554  */
555 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
556                               struct kvm_memory_slot *dont)
557 {
558         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
559                 kvm_destroy_dirty_bitmap(free);
560 
561         kvm_arch_free_memslot(kvm, free, dont);
562 
563         free->npages = 0;
564 }
565 
566 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
567 {
568         struct kvm_memory_slot *memslot;
569 
570         if (!slots)
571                 return;
572 
573         kvm_for_each_memslot(memslot, slots)
574                 kvm_free_memslot(kvm, memslot, NULL);
575 
576         kvfree(slots);
577 }
578 
579 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
580 {
581         int i;
582 
583         if (!kvm->debugfs_dentry)
584                 return;
585 
586         debugfs_remove_recursive(kvm->debugfs_dentry);
587 
588         if (kvm->debugfs_stat_data) {
589                 for (i = 0; i < kvm_debugfs_num_entries; i++)
590                         kfree(kvm->debugfs_stat_data[i]);
591                 kfree(kvm->debugfs_stat_data);
592         }
593 }
594 
595 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
596 {
597         char dir_name[ITOA_MAX_LEN * 2];
598         struct kvm_stat_data *stat_data;
599         struct kvm_stats_debugfs_item *p;
600 
601         if (!debugfs_initialized())
602                 return 0;
603 
604         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
605         kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
606 
607         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
608                                          sizeof(*kvm->debugfs_stat_data),
609                                          GFP_KERNEL_ACCOUNT);
610         if (!kvm->debugfs_stat_data)
611                 return -ENOMEM;
612 
613         for (p = debugfs_entries; p->name; p++) {
614                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
615                 if (!stat_data)
616                         return -ENOMEM;
617 
618                 stat_data->kvm = kvm;
619                 stat_data->offset = p->offset;
620                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
621                 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
622                                     stat_data, stat_fops_per_vm[p->kind]);
623         }
624         return 0;
625 }
626 
627 static struct kvm *kvm_create_vm(unsigned long type)
628 {
629         int r, i;
630         struct kvm *kvm = kvm_arch_alloc_vm();
631 
632         if (!kvm)
633                 return ERR_PTR(-ENOMEM);
634 
635         spin_lock_init(&kvm->mmu_lock);
636         mmgrab(current->mm);
637         kvm->mm = current->mm;
638         kvm_eventfd_init(kvm);
639         mutex_init(&kvm->lock);
640         mutex_init(&kvm->irq_lock);
641         mutex_init(&kvm->slots_lock);
642         refcount_set(&kvm->users_count, 1);
643         INIT_LIST_HEAD(&kvm->devices);
644 
645         r = kvm_arch_init_vm(kvm, type);
646         if (r)
647                 goto out_err_no_disable;
648 
649         r = hardware_enable_all();
650         if (r)
651                 goto out_err_no_disable;
652 
653 #ifdef CONFIG_HAVE_KVM_IRQFD
654         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
655 #endif
656 
657         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
658 
659         r = -ENOMEM;
660         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
661                 struct kvm_memslots *slots = kvm_alloc_memslots();
662                 if (!slots)
663                         goto out_err_no_srcu;
664                 /* Generations must be different for each address space. */
665                 slots->generation = i;
666                 rcu_assign_pointer(kvm->memslots[i], slots);
667         }
668 
669         if (init_srcu_struct(&kvm->srcu))
670                 goto out_err_no_srcu;
671         if (init_srcu_struct(&kvm->irq_srcu))
672                 goto out_err_no_irq_srcu;
673         for (i = 0; i < KVM_NR_BUSES; i++) {
674                 rcu_assign_pointer(kvm->buses[i],
675                         kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
676                 if (!kvm->buses[i])
677                         goto out_err;
678         }
679 
680         r = kvm_init_mmu_notifier(kvm);
681         if (r)
682                 goto out_err;
683 
684         spin_lock(&kvm_lock);
685         list_add(&kvm->vm_list, &vm_list);
686         spin_unlock(&kvm_lock);
687 
688         preempt_notifier_inc();
689 
690         return kvm;
691 
692 out_err:
693         cleanup_srcu_struct(&kvm->irq_srcu);
694 out_err_no_irq_srcu:
695         cleanup_srcu_struct(&kvm->srcu);
696 out_err_no_srcu:
697         hardware_disable_all();
698 out_err_no_disable:
699         refcount_set(&kvm->users_count, 0);
700         for (i = 0; i < KVM_NR_BUSES; i++)
701                 kfree(kvm_get_bus(kvm, i));
702         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
703                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
704         kvm_arch_free_vm(kvm);
705         mmdrop(current->mm);
706         return ERR_PTR(r);
707 }
708 
709 static void kvm_destroy_devices(struct kvm *kvm)
710 {
711         struct kvm_device *dev, *tmp;
712 
713         /*
714          * We do not need to take the kvm->lock here, because nobody else
715          * has a reference to the struct kvm at this point and therefore
716          * cannot access the devices list anyhow.
717          */
718         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
719                 list_del(&dev->vm_node);
720                 dev->ops->destroy(dev);
721         }
722 }
723 
724 static void kvm_destroy_vm(struct kvm *kvm)
725 {
726         int i;
727         struct mm_struct *mm = kvm->mm;
728 
729         kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
730         kvm_destroy_vm_debugfs(kvm);
731         kvm_arch_sync_events(kvm);
732         spin_lock(&kvm_lock);
733         list_del(&kvm->vm_list);
734         spin_unlock(&kvm_lock);
735         kvm_free_irq_routing(kvm);
736         for (i = 0; i < KVM_NR_BUSES; i++) {
737                 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
738 
739                 if (bus)
740                         kvm_io_bus_destroy(bus);
741                 kvm->buses[i] = NULL;
742         }
743         kvm_coalesced_mmio_free(kvm);
744 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
745         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
746 #else
747         kvm_arch_flush_shadow_all(kvm);
748 #endif
749         kvm_arch_destroy_vm(kvm);
750         kvm_destroy_devices(kvm);
751         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
752                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
753         cleanup_srcu_struct(&kvm->irq_srcu);
754         cleanup_srcu_struct(&kvm->srcu);
755         kvm_arch_free_vm(kvm);
756         preempt_notifier_dec();
757         hardware_disable_all();
758         mmdrop(mm);
759 }
760 
761 void kvm_get_kvm(struct kvm *kvm)
762 {
763         refcount_inc(&kvm->users_count);
764 }
765 EXPORT_SYMBOL_GPL(kvm_get_kvm);
766 
767 void kvm_put_kvm(struct kvm *kvm)
768 {
769         if (refcount_dec_and_test(&kvm->users_count))
770                 kvm_destroy_vm(kvm);
771 }
772 EXPORT_SYMBOL_GPL(kvm_put_kvm);
773 
774 
775 static int kvm_vm_release(struct inode *inode, struct file *filp)
776 {
777         struct kvm *kvm = filp->private_data;
778 
779         kvm_irqfd_release(kvm);
780 
781         kvm_put_kvm(kvm);
782         return 0;
783 }
784 
785 /*
786  * Allocation size is twice as large as the actual dirty bitmap size.
787  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
788  */
789 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
790 {
791         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
792 
793         memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
794         if (!memslot->dirty_bitmap)
795                 return -ENOMEM;
796 
797         return 0;
798 }
799 
800 /*
801  * Insert memslot and re-sort memslots based on their GFN,
802  * so binary search could be used to lookup GFN.
803  * Sorting algorithm takes advantage of having initially
804  * sorted array and known changed memslot position.
805  */
806 static void update_memslots(struct kvm_memslots *slots,
807                             struct kvm_memory_slot *new,
808                             enum kvm_mr_change change)
809 {
810         int id = new->id;
811         int i = slots->id_to_index[id];
812         struct kvm_memory_slot *mslots = slots->memslots;
813 
814         WARN_ON(mslots[i].id != id);
815         switch (change) {
816         case KVM_MR_CREATE:
817                 slots->used_slots++;
818                 WARN_ON(mslots[i].npages || !new->npages);
819                 break;
820         case KVM_MR_DELETE:
821                 slots->used_slots--;
822                 WARN_ON(new->npages || !mslots[i].npages);
823                 break;
824         default:
825                 break;
826         }
827 
828         while (i < KVM_MEM_SLOTS_NUM - 1 &&
829                new->base_gfn <= mslots[i + 1].base_gfn) {
830                 if (!mslots[i + 1].npages)
831                         break;
832                 mslots[i] = mslots[i + 1];
833                 slots->id_to_index[mslots[i].id] = i;
834                 i++;
835         }
836 
837         /*
838          * The ">=" is needed when creating a slot with base_gfn == 0,
839          * so that it moves before all those with base_gfn == npages == 0.
840          *
841          * On the other hand, if new->npages is zero, the above loop has
842          * already left i pointing to the beginning of the empty part of
843          * mslots, and the ">=" would move the hole backwards in this
844          * case---which is wrong.  So skip the loop when deleting a slot.
845          */
846         if (new->npages) {
847                 while (i > 0 &&
848                        new->base_gfn >= mslots[i - 1].base_gfn) {
849                         mslots[i] = mslots[i - 1];
850                         slots->id_to_index[mslots[i].id] = i;
851                         i--;
852                 }
853         } else
854                 WARN_ON_ONCE(i != slots->used_slots);
855 
856         mslots[i] = *new;
857         slots->id_to_index[mslots[i].id] = i;
858 }
859 
860 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
861 {
862         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
863 
864 #ifdef __KVM_HAVE_READONLY_MEM
865         valid_flags |= KVM_MEM_READONLY;
866 #endif
867 
868         if (mem->flags & ~valid_flags)
869                 return -EINVAL;
870 
871         return 0;
872 }
873 
874 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
875                 int as_id, struct kvm_memslots *slots)
876 {
877         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
878         u64 gen = old_memslots->generation;
879 
880         WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
881         slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
882 
883         rcu_assign_pointer(kvm->memslots[as_id], slots);
884         synchronize_srcu_expedited(&kvm->srcu);
885 
886         /*
887          * Increment the new memslot generation a second time, dropping the
888          * update in-progress flag and incrementing then generation based on
889          * the number of address spaces.  This provides a unique and easily
890          * identifiable generation number while the memslots are in flux.
891          */
892         gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
893 
894         /*
895          * Generations must be unique even across address spaces.  We do not need
896          * a global counter for that, instead the generation space is evenly split
897          * across address spaces.  For example, with two address spaces, address
898          * space 0 will use generations 0, 2, 4, ... while address space 1 will
899          * use generations 1, 3, 5, ...
900          */
901         gen += KVM_ADDRESS_SPACE_NUM;
902 
903         kvm_arch_memslots_updated(kvm, gen);
904 
905         slots->generation = gen;
906 
907         return old_memslots;
908 }
909 
910 /*
911  * Allocate some memory and give it an address in the guest physical address
912  * space.
913  *
914  * Discontiguous memory is allowed, mostly for framebuffers.
915  *
916  * Must be called holding kvm->slots_lock for write.
917  */
918 int __kvm_set_memory_region(struct kvm *kvm,
919                             const struct kvm_userspace_memory_region *mem)
920 {
921         int r;
922         gfn_t base_gfn;
923         unsigned long npages;
924         struct kvm_memory_slot *slot;
925         struct kvm_memory_slot old, new;
926         struct kvm_memslots *slots = NULL, *old_memslots;
927         int as_id, id;
928         enum kvm_mr_change change;
929 
930         r = check_memory_region_flags(mem);
931         if (r)
932                 goto out;
933 
934         r = -EINVAL;
935         as_id = mem->slot >> 16;
936         id = (u16)mem->slot;
937 
938         /* General sanity checks */
939         if (mem->memory_size & (PAGE_SIZE - 1))
940                 goto out;
941         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
942                 goto out;
943         /* We can read the guest memory with __xxx_user() later on. */
944         if ((id < KVM_USER_MEM_SLOTS) &&
945             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
946              !access_ok((void __user *)(unsigned long)mem->userspace_addr,
947                         mem->memory_size)))
948                 goto out;
949         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
950                 goto out;
951         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
952                 goto out;
953 
954         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
955         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
956         npages = mem->memory_size >> PAGE_SHIFT;
957 
958         if (npages > KVM_MEM_MAX_NR_PAGES)
959                 goto out;
960 
961         new = old = *slot;
962 
963         new.id = id;
964         new.base_gfn = base_gfn;
965         new.npages = npages;
966         new.flags = mem->flags;
967 
968         if (npages) {
969                 if (!old.npages)
970                         change = KVM_MR_CREATE;
971                 else { /* Modify an existing slot. */
972                         if ((mem->userspace_addr != old.userspace_addr) ||
973                             (npages != old.npages) ||
974                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
975                                 goto out;
976 
977                         if (base_gfn != old.base_gfn)
978                                 change = KVM_MR_MOVE;
979                         else if (new.flags != old.flags)
980                                 change = KVM_MR_FLAGS_ONLY;
981                         else { /* Nothing to change. */
982                                 r = 0;
983                                 goto out;
984                         }
985                 }
986         } else {
987                 if (!old.npages)
988                         goto out;
989 
990                 change = KVM_MR_DELETE;
991                 new.base_gfn = 0;
992                 new.flags = 0;
993         }
994 
995         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
996                 /* Check for overlaps */
997                 r = -EEXIST;
998                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
999                         if (slot->id == id)
1000                                 continue;
1001                         if (!((base_gfn + npages <= slot->base_gfn) ||
1002                               (base_gfn >= slot->base_gfn + slot->npages)))
1003                                 goto out;
1004                 }
1005         }
1006 
1007         /* Free page dirty bitmap if unneeded */
1008         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1009                 new.dirty_bitmap = NULL;
1010 
1011         r = -ENOMEM;
1012         if (change == KVM_MR_CREATE) {
1013                 new.userspace_addr = mem->userspace_addr;
1014 
1015                 if (kvm_arch_create_memslot(kvm, &new, npages))
1016                         goto out_free;
1017         }
1018 
1019         /* Allocate page dirty bitmap if needed */
1020         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1021                 if (kvm_create_dirty_bitmap(&new) < 0)
1022                         goto out_free;
1023         }
1024 
1025         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1026         if (!slots)
1027                 goto out_free;
1028         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1029 
1030         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1031                 slot = id_to_memslot(slots, id);
1032                 slot->flags |= KVM_MEMSLOT_INVALID;
1033 
1034                 old_memslots = install_new_memslots(kvm, as_id, slots);
1035 
1036                 /* From this point no new shadow pages pointing to a deleted,
1037                  * or moved, memslot will be created.
1038                  *
1039                  * validation of sp->gfn happens in:
1040                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1041                  *      - kvm_is_visible_gfn (mmu_check_roots)
1042                  */
1043                 kvm_arch_flush_shadow_memslot(kvm, slot);
1044 
1045                 /*
1046                  * We can re-use the old_memslots from above, the only difference
1047                  * from the currently installed memslots is the invalid flag.  This
1048                  * will get overwritten by update_memslots anyway.
1049                  */
1050                 slots = old_memslots;
1051         }
1052 
1053         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1054         if (r)
1055                 goto out_slots;
1056 
1057         /* actual memory is freed via old in kvm_free_memslot below */
1058         if (change == KVM_MR_DELETE) {
1059                 new.dirty_bitmap = NULL;
1060                 memset(&new.arch, 0, sizeof(new.arch));
1061         }
1062 
1063         update_memslots(slots, &new, change);
1064         old_memslots = install_new_memslots(kvm, as_id, slots);
1065 
1066         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1067 
1068         kvm_free_memslot(kvm, &old, &new);
1069         kvfree(old_memslots);
1070         return 0;
1071 
1072 out_slots:
1073         kvfree(slots);
1074 out_free:
1075         kvm_free_memslot(kvm, &new, &old);
1076 out:
1077         return r;
1078 }
1079 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1080 
1081 int kvm_set_memory_region(struct kvm *kvm,
1082                           const struct kvm_userspace_memory_region *mem)
1083 {
1084         int r;
1085 
1086         mutex_lock(&kvm->slots_lock);
1087         r = __kvm_set_memory_region(kvm, mem);
1088         mutex_unlock(&kvm->slots_lock);
1089         return r;
1090 }
1091 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1092 
1093 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1094                                           struct kvm_userspace_memory_region *mem)
1095 {
1096         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1097                 return -EINVAL;
1098 
1099         return kvm_set_memory_region(kvm, mem);
1100 }
1101 
1102 int kvm_get_dirty_log(struct kvm *kvm,
1103                         struct kvm_dirty_log *log, int *is_dirty)
1104 {
1105         struct kvm_memslots *slots;
1106         struct kvm_memory_slot *memslot;
1107         int i, as_id, id;
1108         unsigned long n;
1109         unsigned long any = 0;
1110 
1111         as_id = log->slot >> 16;
1112         id = (u16)log->slot;
1113         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1114                 return -EINVAL;
1115 
1116         slots = __kvm_memslots(kvm, as_id);
1117         memslot = id_to_memslot(slots, id);
1118         if (!memslot->dirty_bitmap)
1119                 return -ENOENT;
1120 
1121         n = kvm_dirty_bitmap_bytes(memslot);
1122 
1123         for (i = 0; !any && i < n/sizeof(long); ++i)
1124                 any = memslot->dirty_bitmap[i];
1125 
1126         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1127                 return -EFAULT;
1128 
1129         if (any)
1130                 *is_dirty = 1;
1131         return 0;
1132 }
1133 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1134 
1135 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1136 /**
1137  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1138  *      and reenable dirty page tracking for the corresponding pages.
1139  * @kvm:        pointer to kvm instance
1140  * @log:        slot id and address to which we copy the log
1141  * @is_dirty:   flag set if any page is dirty
1142  *
1143  * We need to keep it in mind that VCPU threads can write to the bitmap
1144  * concurrently. So, to avoid losing track of dirty pages we keep the
1145  * following order:
1146  *
1147  *    1. Take a snapshot of the bit and clear it if needed.
1148  *    2. Write protect the corresponding page.
1149  *    3. Copy the snapshot to the userspace.
1150  *    4. Upon return caller flushes TLB's if needed.
1151  *
1152  * Between 2 and 4, the guest may write to the page using the remaining TLB
1153  * entry.  This is not a problem because the page is reported dirty using
1154  * the snapshot taken before and step 4 ensures that writes done after
1155  * exiting to userspace will be logged for the next call.
1156  *
1157  */
1158 int kvm_get_dirty_log_protect(struct kvm *kvm,
1159                         struct kvm_dirty_log *log, bool *flush)
1160 {
1161         struct kvm_memslots *slots;
1162         struct kvm_memory_slot *memslot;
1163         int i, as_id, id;
1164         unsigned long n;
1165         unsigned long *dirty_bitmap;
1166         unsigned long *dirty_bitmap_buffer;
1167 
1168         as_id = log->slot >> 16;
1169         id = (u16)log->slot;
1170         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1171                 return -EINVAL;
1172 
1173         slots = __kvm_memslots(kvm, as_id);
1174         memslot = id_to_memslot(slots, id);
1175 
1176         dirty_bitmap = memslot->dirty_bitmap;
1177         if (!dirty_bitmap)
1178                 return -ENOENT;
1179 
1180         n = kvm_dirty_bitmap_bytes(memslot);
1181         *flush = false;
1182         if (kvm->manual_dirty_log_protect) {
1183                 /*
1184                  * Unlike kvm_get_dirty_log, we always return false in *flush,
1185                  * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
1186                  * is some code duplication between this function and
1187                  * kvm_get_dirty_log, but hopefully all architecture
1188                  * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1189                  * can be eliminated.
1190                  */
1191                 dirty_bitmap_buffer = dirty_bitmap;
1192         } else {
1193                 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1194                 memset(dirty_bitmap_buffer, 0, n);
1195 
1196                 spin_lock(&kvm->mmu_lock);
1197                 for (i = 0; i < n / sizeof(long); i++) {
1198                         unsigned long mask;
1199                         gfn_t offset;
1200 
1201                         if (!dirty_bitmap[i])
1202                                 continue;
1203 
1204                         *flush = true;
1205                         mask = xchg(&dirty_bitmap[i], 0);
1206                         dirty_bitmap_buffer[i] = mask;
1207 
1208                         offset = i * BITS_PER_LONG;
1209                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1210                                                                 offset, mask);
1211                 }
1212                 spin_unlock(&kvm->mmu_lock);
1213         }
1214 
1215         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1216                 return -EFAULT;
1217         return 0;
1218 }
1219 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1220 
1221 /**
1222  * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1223  *      and reenable dirty page tracking for the corresponding pages.
1224  * @kvm:        pointer to kvm instance
1225  * @log:        slot id and address from which to fetch the bitmap of dirty pages
1226  */
1227 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1228                                 struct kvm_clear_dirty_log *log, bool *flush)
1229 {
1230         struct kvm_memslots *slots;
1231         struct kvm_memory_slot *memslot;
1232         int as_id, id;
1233         gfn_t offset;
1234         unsigned long i, n;
1235         unsigned long *dirty_bitmap;
1236         unsigned long *dirty_bitmap_buffer;
1237 
1238         as_id = log->slot >> 16;
1239         id = (u16)log->slot;
1240         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1241                 return -EINVAL;
1242 
1243         if ((log->first_page & 63) || (log->num_pages & 63))
1244                 return -EINVAL;
1245 
1246         slots = __kvm_memslots(kvm, as_id);
1247         memslot = id_to_memslot(slots, id);
1248 
1249         dirty_bitmap = memslot->dirty_bitmap;
1250         if (!dirty_bitmap)
1251                 return -ENOENT;
1252 
1253         n = kvm_dirty_bitmap_bytes(memslot);
1254 
1255         if (log->first_page > memslot->npages ||
1256             log->num_pages > memslot->npages - log->first_page)
1257                         return -EINVAL;
1258 
1259         *flush = false;
1260         dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1261         if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1262                 return -EFAULT;
1263 
1264         spin_lock(&kvm->mmu_lock);
1265         for (offset = log->first_page,
1266              i = offset / BITS_PER_LONG, n = log->num_pages / BITS_PER_LONG; n--;
1267              i++, offset += BITS_PER_LONG) {
1268                 unsigned long mask = *dirty_bitmap_buffer++;
1269                 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1270                 if (!mask)
1271                         continue;
1272 
1273                 mask &= atomic_long_fetch_andnot(mask, p);
1274 
1275                 /*
1276                  * mask contains the bits that really have been cleared.  This
1277                  * never includes any bits beyond the length of the memslot (if
1278                  * the length is not aligned to 64 pages), therefore it is not
1279                  * a problem if userspace sets them in log->dirty_bitmap.
1280                 */
1281                 if (mask) {
1282                         *flush = true;
1283                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1284                                                                 offset, mask);
1285                 }
1286         }
1287         spin_unlock(&kvm->mmu_lock);
1288 
1289         return 0;
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1292 #endif
1293 
1294 bool kvm_largepages_enabled(void)
1295 {
1296         return largepages_enabled;
1297 }
1298 
1299 void kvm_disable_largepages(void)
1300 {
1301         largepages_enabled = false;
1302 }
1303 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1304 
1305 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1306 {
1307         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1308 }
1309 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1310 
1311 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1312 {
1313         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1314 }
1315 
1316 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1317 {
1318         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1319 
1320         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1321               memslot->flags & KVM_MEMSLOT_INVALID)
1322                 return false;
1323 
1324         return true;
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1327 
1328 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1329 {
1330         struct vm_area_struct *vma;
1331         unsigned long addr, size;
1332 
1333         size = PAGE_SIZE;
1334 
1335         addr = gfn_to_hva(kvm, gfn);
1336         if (kvm_is_error_hva(addr))
1337                 return PAGE_SIZE;
1338 
1339         down_read(&current->mm->mmap_sem);
1340         vma = find_vma(current->mm, addr);
1341         if (!vma)
1342                 goto out;
1343 
1344         size = vma_kernel_pagesize(vma);
1345 
1346 out:
1347         up_read(&current->mm->mmap_sem);
1348 
1349         return size;
1350 }
1351 
1352 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1353 {
1354         return slot->flags & KVM_MEM_READONLY;
1355 }
1356 
1357 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1358                                        gfn_t *nr_pages, bool write)
1359 {
1360         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1361                 return KVM_HVA_ERR_BAD;
1362 
1363         if (memslot_is_readonly(slot) && write)
1364                 return KVM_HVA_ERR_RO_BAD;
1365 
1366         if (nr_pages)
1367                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1368 
1369         return __gfn_to_hva_memslot(slot, gfn);
1370 }
1371 
1372 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1373                                      gfn_t *nr_pages)
1374 {
1375         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1376 }
1377 
1378 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1379                                         gfn_t gfn)
1380 {
1381         return gfn_to_hva_many(slot, gfn, NULL);
1382 }
1383 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1384 
1385 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1386 {
1387         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1388 }
1389 EXPORT_SYMBOL_GPL(gfn_to_hva);
1390 
1391 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1392 {
1393         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1396 
1397 /*
1398  * Return the hva of a @gfn and the R/W attribute if possible.
1399  *
1400  * @slot: the kvm_memory_slot which contains @gfn
1401  * @gfn: the gfn to be translated
1402  * @writable: used to return the read/write attribute of the @slot if the hva
1403  * is valid and @writable is not NULL
1404  */
1405 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1406                                       gfn_t gfn, bool *writable)
1407 {
1408         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1409 
1410         if (!kvm_is_error_hva(hva) && writable)
1411                 *writable = !memslot_is_readonly(slot);
1412 
1413         return hva;
1414 }
1415 
1416 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1417 {
1418         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1419 
1420         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1421 }
1422 
1423 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1424 {
1425         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1426 
1427         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1428 }
1429 
1430 static inline int check_user_page_hwpoison(unsigned long addr)
1431 {
1432         int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1433 
1434         rc = get_user_pages(addr, 1, flags, NULL, NULL);
1435         return rc == -EHWPOISON;
1436 }
1437 
1438 /*
1439  * The fast path to get the writable pfn which will be stored in @pfn,
1440  * true indicates success, otherwise false is returned.  It's also the
1441  * only part that runs if we can are in atomic context.
1442  */
1443 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1444                             bool *writable, kvm_pfn_t *pfn)
1445 {
1446         struct page *page[1];
1447         int npages;
1448 
1449         /*
1450          * Fast pin a writable pfn only if it is a write fault request
1451          * or the caller allows to map a writable pfn for a read fault
1452          * request.
1453          */
1454         if (!(write_fault || writable))
1455                 return false;
1456 
1457         npages = __get_user_pages_fast(addr, 1, 1, page);
1458         if (npages == 1) {
1459                 *pfn = page_to_pfn(page[0]);
1460 
1461                 if (writable)
1462                         *writable = true;
1463                 return true;
1464         }
1465 
1466         return false;
1467 }
1468 
1469 /*
1470  * The slow path to get the pfn of the specified host virtual address,
1471  * 1 indicates success, -errno is returned if error is detected.
1472  */
1473 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1474                            bool *writable, kvm_pfn_t *pfn)
1475 {
1476         unsigned int flags = FOLL_HWPOISON;
1477         struct page *page;
1478         int npages = 0;
1479 
1480         might_sleep();
1481 
1482         if (writable)
1483                 *writable = write_fault;
1484 
1485         if (write_fault)
1486                 flags |= FOLL_WRITE;
1487         if (async)
1488                 flags |= FOLL_NOWAIT;
1489 
1490         npages = get_user_pages_unlocked(addr, 1, &page, flags);
1491         if (npages != 1)
1492                 return npages;
1493 
1494         /* map read fault as writable if possible */
1495         if (unlikely(!write_fault) && writable) {
1496                 struct page *wpage;
1497 
1498                 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1499                         *writable = true;
1500                         put_page(page);
1501                         page = wpage;
1502                 }
1503         }
1504         *pfn = page_to_pfn(page);
1505         return npages;
1506 }
1507 
1508 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1509 {
1510         if (unlikely(!(vma->vm_flags & VM_READ)))
1511                 return false;
1512 
1513         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1514                 return false;
1515 
1516         return true;
1517 }
1518 
1519 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1520                                unsigned long addr, bool *async,
1521                                bool write_fault, bool *writable,
1522                                kvm_pfn_t *p_pfn)
1523 {
1524         unsigned long pfn;
1525         int r;
1526 
1527         r = follow_pfn(vma, addr, &pfn);
1528         if (r) {
1529                 /*
1530                  * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1531                  * not call the fault handler, so do it here.
1532                  */
1533                 bool unlocked = false;
1534                 r = fixup_user_fault(current, current->mm, addr,
1535                                      (write_fault ? FAULT_FLAG_WRITE : 0),
1536                                      &unlocked);
1537                 if (unlocked)
1538                         return -EAGAIN;
1539                 if (r)
1540                         return r;
1541 
1542                 r = follow_pfn(vma, addr, &pfn);
1543                 if (r)
1544                         return r;
1545 
1546         }
1547 
1548         if (writable)
1549                 *writable = true;
1550 
1551         /*
1552          * Get a reference here because callers of *hva_to_pfn* and
1553          * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1554          * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1555          * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1556          * simply do nothing for reserved pfns.
1557          *
1558          * Whoever called remap_pfn_range is also going to call e.g.
1559          * unmap_mapping_range before the underlying pages are freed,
1560          * causing a call to our MMU notifier.
1561          */ 
1562         kvm_get_pfn(pfn);
1563 
1564         *p_pfn = pfn;
1565         return 0;
1566 }
1567 
1568 /*
1569  * Pin guest page in memory and return its pfn.
1570  * @addr: host virtual address which maps memory to the guest
1571  * @atomic: whether this function can sleep
1572  * @async: whether this function need to wait IO complete if the
1573  *         host page is not in the memory
1574  * @write_fault: whether we should get a writable host page
1575  * @writable: whether it allows to map a writable host page for !@write_fault
1576  *
1577  * The function will map a writable host page for these two cases:
1578  * 1): @write_fault = true
1579  * 2): @write_fault = false && @writable, @writable will tell the caller
1580  *     whether the mapping is writable.
1581  */
1582 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1583                         bool write_fault, bool *writable)
1584 {
1585         struct vm_area_struct *vma;
1586         kvm_pfn_t pfn = 0;
1587         int npages, r;
1588 
1589         /* we can do it either atomically or asynchronously, not both */
1590         BUG_ON(atomic && async);
1591 
1592         if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1593                 return pfn;
1594 
1595         if (atomic)
1596                 return KVM_PFN_ERR_FAULT;
1597 
1598         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1599         if (npages == 1)
1600                 return pfn;
1601 
1602         down_read(&current->mm->mmap_sem);
1603         if (npages == -EHWPOISON ||
1604               (!async && check_user_page_hwpoison(addr))) {
1605                 pfn = KVM_PFN_ERR_HWPOISON;
1606                 goto exit;
1607         }
1608 
1609 retry:
1610         vma = find_vma_intersection(current->mm, addr, addr + 1);
1611 
1612         if (vma == NULL)
1613                 pfn = KVM_PFN_ERR_FAULT;
1614         else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1615                 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1616                 if (r == -EAGAIN)
1617                         goto retry;
1618                 if (r < 0)
1619                         pfn = KVM_PFN_ERR_FAULT;
1620         } else {
1621                 if (async && vma_is_valid(vma, write_fault))
1622                         *async = true;
1623                 pfn = KVM_PFN_ERR_FAULT;
1624         }
1625 exit:
1626         up_read(&current->mm->mmap_sem);
1627         return pfn;
1628 }
1629 
1630 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1631                                bool atomic, bool *async, bool write_fault,
1632                                bool *writable)
1633 {
1634         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1635 
1636         if (addr == KVM_HVA_ERR_RO_BAD) {
1637                 if (writable)
1638                         *writable = false;
1639                 return KVM_PFN_ERR_RO_FAULT;
1640         }
1641 
1642         if (kvm_is_error_hva(addr)) {
1643                 if (writable)
1644                         *writable = false;
1645                 return KVM_PFN_NOSLOT;
1646         }
1647 
1648         /* Do not map writable pfn in the readonly memslot. */
1649         if (writable && memslot_is_readonly(slot)) {
1650                 *writable = false;
1651                 writable = NULL;
1652         }
1653 
1654         return hva_to_pfn(addr, atomic, async, write_fault,
1655                           writable);
1656 }
1657 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1658 
1659 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1660                       bool *writable)
1661 {
1662         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1663                                     write_fault, writable);
1664 }
1665 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1666 
1667 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1668 {
1669         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1670 }
1671 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1672 
1673 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1674 {
1675         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1676 }
1677 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1678 
1679 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1680 {
1681         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1682 }
1683 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1684 
1685 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1686 {
1687         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1688 }
1689 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1690 
1691 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1692 {
1693         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1694 }
1695 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1696 
1697 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1698 {
1699         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1700 }
1701 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1702 
1703 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1704                             struct page **pages, int nr_pages)
1705 {
1706         unsigned long addr;
1707         gfn_t entry = 0;
1708 
1709         addr = gfn_to_hva_many(slot, gfn, &entry);
1710         if (kvm_is_error_hva(addr))
1711                 return -1;
1712 
1713         if (entry < nr_pages)
1714                 return 0;
1715 
1716         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1717 }
1718 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1719 
1720 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1721 {
1722         if (is_error_noslot_pfn(pfn))
1723                 return KVM_ERR_PTR_BAD_PAGE;
1724 
1725         if (kvm_is_reserved_pfn(pfn)) {
1726                 WARN_ON(1);
1727                 return KVM_ERR_PTR_BAD_PAGE;
1728         }
1729 
1730         return pfn_to_page(pfn);
1731 }
1732 
1733 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1734 {
1735         kvm_pfn_t pfn;
1736 
1737         pfn = gfn_to_pfn(kvm, gfn);
1738 
1739         return kvm_pfn_to_page(pfn);
1740 }
1741 EXPORT_SYMBOL_GPL(gfn_to_page);
1742 
1743 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1744 {
1745         kvm_pfn_t pfn;
1746 
1747         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1748 
1749         return kvm_pfn_to_page(pfn);
1750 }
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1752 
1753 void kvm_release_page_clean(struct page *page)
1754 {
1755         WARN_ON(is_error_page(page));
1756 
1757         kvm_release_pfn_clean(page_to_pfn(page));
1758 }
1759 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1760 
1761 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1762 {
1763         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1764                 put_page(pfn_to_page(pfn));
1765 }
1766 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1767 
1768 void kvm_release_page_dirty(struct page *page)
1769 {
1770         WARN_ON(is_error_page(page));
1771 
1772         kvm_release_pfn_dirty(page_to_pfn(page));
1773 }
1774 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1775 
1776 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1777 {
1778         kvm_set_pfn_dirty(pfn);
1779         kvm_release_pfn_clean(pfn);
1780 }
1781 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1782 
1783 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1784 {
1785         if (!kvm_is_reserved_pfn(pfn)) {
1786                 struct page *page = pfn_to_page(pfn);
1787 
1788                 if (!PageReserved(page))
1789                         SetPageDirty(page);
1790         }
1791 }
1792 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1793 
1794 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1795 {
1796         if (!kvm_is_reserved_pfn(pfn))
1797                 mark_page_accessed(pfn_to_page(pfn));
1798 }
1799 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1800 
1801 void kvm_get_pfn(kvm_pfn_t pfn)
1802 {
1803         if (!kvm_is_reserved_pfn(pfn))
1804                 get_page(pfn_to_page(pfn));
1805 }
1806 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1807 
1808 static int next_segment(unsigned long len, int offset)
1809 {
1810         if (len > PAGE_SIZE - offset)
1811                 return PAGE_SIZE - offset;
1812         else
1813                 return len;
1814 }
1815 
1816 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1817                                  void *data, int offset, int len)
1818 {
1819         int r;
1820         unsigned long addr;
1821 
1822         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1823         if (kvm_is_error_hva(addr))
1824                 return -EFAULT;
1825         r = __copy_from_user(data, (void __user *)addr + offset, len);
1826         if (r)
1827                 return -EFAULT;
1828         return 0;
1829 }
1830 
1831 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1832                         int len)
1833 {
1834         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1835 
1836         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1837 }
1838 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1839 
1840 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1841                              int offset, int len)
1842 {
1843         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1844 
1845         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1846 }
1847 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1848 
1849 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1850 {
1851         gfn_t gfn = gpa >> PAGE_SHIFT;
1852         int seg;
1853         int offset = offset_in_page(gpa);
1854         int ret;
1855 
1856         while ((seg = next_segment(len, offset)) != 0) {
1857                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1858                 if (ret < 0)
1859                         return ret;
1860                 offset = 0;
1861                 len -= seg;
1862                 data += seg;
1863                 ++gfn;
1864         }
1865         return 0;
1866 }
1867 EXPORT_SYMBOL_GPL(kvm_read_guest);
1868 
1869 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1870 {
1871         gfn_t gfn = gpa >> PAGE_SHIFT;
1872         int seg;
1873         int offset = offset_in_page(gpa);
1874         int ret;
1875 
1876         while ((seg = next_segment(len, offset)) != 0) {
1877                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1878                 if (ret < 0)
1879                         return ret;
1880                 offset = 0;
1881                 len -= seg;
1882                 data += seg;
1883                 ++gfn;
1884         }
1885         return 0;
1886 }
1887 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1888 
1889 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1890                                    void *data, int offset, unsigned long len)
1891 {
1892         int r;
1893         unsigned long addr;
1894 
1895         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1896         if (kvm_is_error_hva(addr))
1897                 return -EFAULT;
1898         pagefault_disable();
1899         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1900         pagefault_enable();
1901         if (r)
1902                 return -EFAULT;
1903         return 0;
1904 }
1905 
1906 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1907                           unsigned long len)
1908 {
1909         gfn_t gfn = gpa >> PAGE_SHIFT;
1910         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1911         int offset = offset_in_page(gpa);
1912 
1913         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1914 }
1915 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1916 
1917 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1918                                void *data, unsigned long len)
1919 {
1920         gfn_t gfn = gpa >> PAGE_SHIFT;
1921         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1922         int offset = offset_in_page(gpa);
1923 
1924         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1925 }
1926 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1927 
1928 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1929                                   const void *data, int offset, int len)
1930 {
1931         int r;
1932         unsigned long addr;
1933 
1934         addr = gfn_to_hva_memslot(memslot, gfn);
1935         if (kvm_is_error_hva(addr))
1936                 return -EFAULT;
1937         r = __copy_to_user((void __user *)addr + offset, data, len);
1938         if (r)
1939                 return -EFAULT;
1940         mark_page_dirty_in_slot(memslot, gfn);
1941         return 0;
1942 }
1943 
1944 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1945                          const void *data, int offset, int len)
1946 {
1947         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1948 
1949         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1950 }
1951 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1952 
1953 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1954                               const void *data, int offset, int len)
1955 {
1956         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1957 
1958         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1959 }
1960 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1961 
1962 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1963                     unsigned long len)
1964 {
1965         gfn_t gfn = gpa >> PAGE_SHIFT;
1966         int seg;
1967         int offset = offset_in_page(gpa);
1968         int ret;
1969 
1970         while ((seg = next_segment(len, offset)) != 0) {
1971                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1972                 if (ret < 0)
1973                         return ret;
1974                 offset = 0;
1975                 len -= seg;
1976                 data += seg;
1977                 ++gfn;
1978         }
1979         return 0;
1980 }
1981 EXPORT_SYMBOL_GPL(kvm_write_guest);
1982 
1983 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1984                          unsigned long len)
1985 {
1986         gfn_t gfn = gpa >> PAGE_SHIFT;
1987         int seg;
1988         int offset = offset_in_page(gpa);
1989         int ret;
1990 
1991         while ((seg = next_segment(len, offset)) != 0) {
1992                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1993                 if (ret < 0)
1994                         return ret;
1995                 offset = 0;
1996                 len -= seg;
1997                 data += seg;
1998                 ++gfn;
1999         }
2000         return 0;
2001 }
2002 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2003 
2004 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2005                                        struct gfn_to_hva_cache *ghc,
2006                                        gpa_t gpa, unsigned long len)
2007 {
2008         int offset = offset_in_page(gpa);
2009         gfn_t start_gfn = gpa >> PAGE_SHIFT;
2010         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2011         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2012         gfn_t nr_pages_avail;
2013         int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2014 
2015         ghc->gpa = gpa;
2016         ghc->generation = slots->generation;
2017         ghc->len = len;
2018         ghc->hva = KVM_HVA_ERR_BAD;
2019 
2020         /*
2021          * If the requested region crosses two memslots, we still
2022          * verify that the entire region is valid here.
2023          */
2024         while (!r && start_gfn <= end_gfn) {
2025                 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2026                 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2027                                            &nr_pages_avail);
2028                 if (kvm_is_error_hva(ghc->hva))
2029                         r = -EFAULT;
2030                 start_gfn += nr_pages_avail;
2031         }
2032 
2033         /* Use the slow path for cross page reads and writes. */
2034         if (!r && nr_pages_needed == 1)
2035                 ghc->hva += offset;
2036         else
2037                 ghc->memslot = NULL;
2038 
2039         return r;
2040 }
2041 
2042 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2043                               gpa_t gpa, unsigned long len)
2044 {
2045         struct kvm_memslots *slots = kvm_memslots(kvm);
2046         return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2047 }
2048 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2049 
2050 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2051                                   void *data, unsigned int offset,
2052                                   unsigned long len)
2053 {
2054         struct kvm_memslots *slots = kvm_memslots(kvm);
2055         int r;
2056         gpa_t gpa = ghc->gpa + offset;
2057 
2058         BUG_ON(len + offset > ghc->len);
2059 
2060         if (slots->generation != ghc->generation)
2061                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2062 
2063         if (unlikely(!ghc->memslot))
2064                 return kvm_write_guest(kvm, gpa, data, len);
2065 
2066         if (kvm_is_error_hva(ghc->hva))
2067                 return -EFAULT;
2068 
2069         r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2070         if (r)
2071                 return -EFAULT;
2072         mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2073 
2074         return 0;
2075 }
2076 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2077 
2078 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2079                            void *data, unsigned long len)
2080 {
2081         return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2082 }
2083 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2084 
2085 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2086                            void *data, unsigned long len)
2087 {
2088         struct kvm_memslots *slots = kvm_memslots(kvm);
2089         int r;
2090 
2091         BUG_ON(len > ghc->len);
2092 
2093         if (slots->generation != ghc->generation)
2094                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2095 
2096         if (unlikely(!ghc->memslot))
2097                 return kvm_read_guest(kvm, ghc->gpa, data, len);
2098 
2099         if (kvm_is_error_hva(ghc->hva))
2100                 return -EFAULT;
2101 
2102         r = __copy_from_user(data, (void __user *)ghc->hva, len);
2103         if (r)
2104                 return -EFAULT;
2105 
2106         return 0;
2107 }
2108 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2109 
2110 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2111 {
2112         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2113 
2114         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2115 }
2116 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2117 
2118 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2119 {
2120         gfn_t gfn = gpa >> PAGE_SHIFT;
2121         int seg;
2122         int offset = offset_in_page(gpa);
2123         int ret;
2124 
2125         while ((seg = next_segment(len, offset)) != 0) {
2126                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2127                 if (ret < 0)
2128                         return ret;
2129                 offset = 0;
2130                 len -= seg;
2131                 ++gfn;
2132         }
2133         return 0;
2134 }
2135 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2136 
2137 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2138                                     gfn_t gfn)
2139 {
2140         if (memslot && memslot->dirty_bitmap) {
2141                 unsigned long rel_gfn = gfn - memslot->base_gfn;
2142 
2143                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2144         }
2145 }
2146 
2147 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2148 {
2149         struct kvm_memory_slot *memslot;
2150 
2151         memslot = gfn_to_memslot(kvm, gfn);
2152         mark_page_dirty_in_slot(memslot, gfn);
2153 }
2154 EXPORT_SYMBOL_GPL(mark_page_dirty);
2155 
2156 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2157 {
2158         struct kvm_memory_slot *memslot;
2159 
2160         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2161         mark_page_dirty_in_slot(memslot, gfn);
2162 }
2163 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2164 
2165 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2166 {
2167         if (!vcpu->sigset_active)
2168                 return;
2169 
2170         /*
2171          * This does a lockless modification of ->real_blocked, which is fine
2172          * because, only current can change ->real_blocked and all readers of
2173          * ->real_blocked don't care as long ->real_blocked is always a subset
2174          * of ->blocked.
2175          */
2176         sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
2177 }
2178 
2179 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2180 {
2181         if (!vcpu->sigset_active)
2182                 return;
2183 
2184         sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
2185         sigemptyset(&current->real_blocked);
2186 }
2187 
2188 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2189 {
2190         unsigned int old, val, grow, grow_start;
2191 
2192         old = val = vcpu->halt_poll_ns;
2193         grow_start = READ_ONCE(halt_poll_ns_grow_start);
2194         grow = READ_ONCE(halt_poll_ns_grow);
2195         if (!grow)
2196                 goto out;
2197 
2198         val *= grow;
2199         if (val < grow_start)
2200                 val = grow_start;
2201 
2202         if (val > halt_poll_ns)
2203                 val = halt_poll_ns;
2204 
2205         vcpu->halt_poll_ns = val;
2206 out:
2207         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2208 }
2209 
2210 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2211 {
2212         unsigned int old, val, shrink;
2213 
2214         old = val = vcpu->halt_poll_ns;
2215         shrink = READ_ONCE(halt_poll_ns_shrink);
2216         if (shrink == 0)
2217                 val = 0;
2218         else
2219                 val /= shrink;
2220 
2221         vcpu->halt_poll_ns = val;
2222         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2223 }
2224 
2225 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2226 {
2227         int ret = -EINTR;
2228         int idx = srcu_read_lock(&vcpu->kvm->srcu);
2229 
2230         if (kvm_arch_vcpu_runnable(vcpu)) {
2231                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2232                 goto out;
2233         }
2234         if (kvm_cpu_has_pending_timer(vcpu))
2235                 goto out;
2236         if (signal_pending(current))
2237                 goto out;
2238 
2239         ret = 0;
2240 out:
2241         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2242         return ret;
2243 }
2244 
2245 /*
2246  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2247  */
2248 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2249 {
2250         ktime_t start, cur;
2251         DECLARE_SWAITQUEUE(wait);
2252         bool waited = false;
2253         u64 block_ns;
2254 
2255         start = cur = ktime_get();
2256         if (vcpu->halt_poll_ns) {
2257                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2258 
2259                 ++vcpu->stat.halt_attempted_poll;
2260                 do {
2261                         /*
2262                          * This sets KVM_REQ_UNHALT if an interrupt
2263                          * arrives.
2264                          */
2265                         if (kvm_vcpu_check_block(vcpu) < 0) {
2266                                 ++vcpu->stat.halt_successful_poll;
2267                                 if (!vcpu_valid_wakeup(vcpu))
2268                                         ++vcpu->stat.halt_poll_invalid;
2269                                 goto out;
2270                         }
2271                         cur = ktime_get();
2272                 } while (single_task_running() && ktime_before(cur, stop));
2273         }
2274 
2275         kvm_arch_vcpu_blocking(vcpu);
2276 
2277         for (;;) {
2278                 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2279 
2280                 if (kvm_vcpu_check_block(vcpu) < 0)
2281                         break;
2282 
2283                 waited = true;
2284                 schedule();
2285         }
2286 
2287         finish_swait(&vcpu->wq, &wait);
2288         cur = ktime_get();
2289 
2290         kvm_arch_vcpu_unblocking(vcpu);
2291 out:
2292         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2293 
2294         if (!vcpu_valid_wakeup(vcpu))
2295                 shrink_halt_poll_ns(vcpu);
2296         else if (halt_poll_ns) {
2297                 if (block_ns <= vcpu->halt_poll_ns)
2298                         ;
2299                 /* we had a long block, shrink polling */
2300                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2301                         shrink_halt_poll_ns(vcpu);
2302                 /* we had a short halt and our poll time is too small */
2303                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2304                         block_ns < halt_poll_ns)
2305                         grow_halt_poll_ns(vcpu);
2306         } else
2307                 vcpu->halt_poll_ns = 0;
2308 
2309         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2310         kvm_arch_vcpu_block_finish(vcpu);
2311 }
2312 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2313 
2314 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2315 {
2316         struct swait_queue_head *wqp;
2317 
2318         wqp = kvm_arch_vcpu_wq(vcpu);
2319         if (swq_has_sleeper(wqp)) {
2320                 swake_up_one(wqp);
2321                 ++vcpu->stat.halt_wakeup;
2322                 return true;
2323         }
2324 
2325         return false;
2326 }
2327 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2328 
2329 #ifndef CONFIG_S390
2330 /*
2331  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2332  */
2333 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2334 {
2335         int me;
2336         int cpu = vcpu->cpu;
2337 
2338         if (kvm_vcpu_wake_up(vcpu))
2339                 return;
2340 
2341         me = get_cpu();
2342         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2343                 if (kvm_arch_vcpu_should_kick(vcpu))
2344                         smp_send_reschedule(cpu);
2345         put_cpu();
2346 }
2347 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2348 #endif /* !CONFIG_S390 */
2349 
2350 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2351 {
2352         struct pid *pid;
2353         struct task_struct *task = NULL;
2354         int ret = 0;
2355 
2356         rcu_read_lock();
2357         pid = rcu_dereference(target->pid);
2358         if (pid)
2359                 task = get_pid_task(pid, PIDTYPE_PID);
2360         rcu_read_unlock();
2361         if (!task)
2362                 return ret;
2363         ret = yield_to(task, 1);
2364         put_task_struct(task);
2365 
2366         return ret;
2367 }
2368 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2369 
2370 /*
2371  * Helper that checks whether a VCPU is eligible for directed yield.
2372  * Most eligible candidate to yield is decided by following heuristics:
2373  *
2374  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2375  *  (preempted lock holder), indicated by @in_spin_loop.
2376  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2377  *
2378  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2379  *  chance last time (mostly it has become eligible now since we have probably
2380  *  yielded to lockholder in last iteration. This is done by toggling
2381  *  @dy_eligible each time a VCPU checked for eligibility.)
2382  *
2383  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2384  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2385  *  burning. Giving priority for a potential lock-holder increases lock
2386  *  progress.
2387  *
2388  *  Since algorithm is based on heuristics, accessing another VCPU data without
2389  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2390  *  and continue with next VCPU and so on.
2391  */
2392 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2393 {
2394 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2395         bool eligible;
2396 
2397         eligible = !vcpu->spin_loop.in_spin_loop ||
2398                     vcpu->spin_loop.dy_eligible;
2399 
2400         if (vcpu->spin_loop.in_spin_loop)
2401                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2402 
2403         return eligible;
2404 #else
2405         return true;
2406 #endif
2407 }
2408 
2409 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2410 {
2411         struct kvm *kvm = me->kvm;
2412         struct kvm_vcpu *vcpu;
2413         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2414         int yielded = 0;
2415         int try = 3;
2416         int pass;
2417         int i;
2418 
2419         kvm_vcpu_set_in_spin_loop(me, true);
2420         /*
2421          * We boost the priority of a VCPU that is runnable but not
2422          * currently running, because it got preempted by something
2423          * else and called schedule in __vcpu_run.  Hopefully that
2424          * VCPU is holding the lock that we need and will release it.
2425          * We approximate round-robin by starting at the last boosted VCPU.
2426          */
2427         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2428                 kvm_for_each_vcpu(i, vcpu, kvm) {
2429                         if (!pass && i <= last_boosted_vcpu) {
2430                                 i = last_boosted_vcpu;
2431                                 continue;
2432                         } else if (pass && i > last_boosted_vcpu)
2433                                 break;
2434                         if (!READ_ONCE(vcpu->preempted))
2435                                 continue;
2436                         if (vcpu == me)
2437                                 continue;
2438                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2439                                 continue;
2440                         if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2441                                 continue;
2442                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2443                                 continue;
2444 
2445                         yielded = kvm_vcpu_yield_to(vcpu);
2446                         if (yielded > 0) {
2447                                 kvm->last_boosted_vcpu = i;
2448                                 break;
2449                         } else if (yielded < 0) {
2450                                 try--;
2451                                 if (!try)
2452                                         break;
2453                         }
2454                 }
2455         }
2456         kvm_vcpu_set_in_spin_loop(me, false);
2457 
2458         /* Ensure vcpu is not eligible during next spinloop */
2459         kvm_vcpu_set_dy_eligible(me, false);
2460 }
2461 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2462 
2463 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2464 {
2465         struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2466         struct page *page;
2467 
2468         if (vmf->pgoff == 0)
2469                 page = virt_to_page(vcpu->run);
2470 #ifdef CONFIG_X86
2471         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2472                 page = virt_to_page(vcpu->arch.pio_data);
2473 #endif
2474 #ifdef CONFIG_KVM_MMIO
2475         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2476                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2477 #endif
2478         else
2479                 return kvm_arch_vcpu_fault(vcpu, vmf);
2480         get_page(page);
2481         vmf->page = page;
2482         return 0;
2483 }
2484 
2485 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2486         .fault = kvm_vcpu_fault,
2487 };
2488 
2489 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2490 {
2491         vma->vm_ops = &kvm_vcpu_vm_ops;
2492         return 0;
2493 }
2494 
2495 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2496 {
2497         struct kvm_vcpu *vcpu = filp->private_data;
2498 
2499         debugfs_remove_recursive(vcpu->debugfs_dentry);
2500         kvm_put_kvm(vcpu->kvm);
2501         return 0;
2502 }
2503 
2504 static struct file_operations kvm_vcpu_fops = {
2505         .release        = kvm_vcpu_release,
2506         .unlocked_ioctl = kvm_vcpu_ioctl,
2507         .mmap           = kvm_vcpu_mmap,
2508         .llseek         = noop_llseek,
2509         KVM_COMPAT(kvm_vcpu_compat_ioctl),
2510 };
2511 
2512 /*
2513  * Allocates an inode for the vcpu.
2514  */
2515 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2516 {
2517         char name[8 + 1 + ITOA_MAX_LEN + 1];
2518 
2519         snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2520         return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2521 }
2522 
2523 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2524 {
2525         char dir_name[ITOA_MAX_LEN * 2];
2526         int ret;
2527 
2528         if (!kvm_arch_has_vcpu_debugfs())
2529                 return 0;
2530 
2531         if (!debugfs_initialized())
2532                 return 0;
2533 
2534         snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2535         vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2536                                                                 vcpu->kvm->debugfs_dentry);
2537         if (!vcpu->debugfs_dentry)
2538                 return -ENOMEM;
2539 
2540         ret = kvm_arch_create_vcpu_debugfs(vcpu);
2541         if (ret < 0) {
2542                 debugfs_remove_recursive(vcpu->debugfs_dentry);
2543                 return ret;
2544         }
2545 
2546         return 0;
2547 }
2548 
2549 /*
2550  * Creates some virtual cpus.  Good luck creating more than one.
2551  */
2552 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2553 {
2554         int r;
2555         struct kvm_vcpu *vcpu;
2556 
2557         if (id >= KVM_MAX_VCPU_ID)
2558                 return -EINVAL;
2559 
2560         mutex_lock(&kvm->lock);
2561         if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2562                 mutex_unlock(&kvm->lock);
2563                 return -EINVAL;
2564         }
2565 
2566         kvm->created_vcpus++;
2567         mutex_unlock(&kvm->lock);
2568 
2569         vcpu = kvm_arch_vcpu_create(kvm, id);
2570         if (IS_ERR(vcpu)) {
2571                 r = PTR_ERR(vcpu);
2572                 goto vcpu_decrement;
2573         }
2574 
2575         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2576 
2577         r = kvm_arch_vcpu_setup(vcpu);
2578         if (r)
2579                 goto vcpu_destroy;
2580 
2581         r = kvm_create_vcpu_debugfs(vcpu);
2582         if (r)
2583                 goto vcpu_destroy;
2584 
2585         mutex_lock(&kvm->lock);
2586         if (kvm_get_vcpu_by_id(kvm, id)) {
2587                 r = -EEXIST;
2588                 goto unlock_vcpu_destroy;
2589         }
2590 
2591         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2592 
2593         /* Now it's all set up, let userspace reach it */
2594         kvm_get_kvm(kvm);
2595         r = create_vcpu_fd(vcpu);
2596         if (r < 0) {
2597                 kvm_put_kvm(kvm);
2598                 goto unlock_vcpu_destroy;
2599         }
2600 
2601         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2602 
2603         /*
2604          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2605          * before kvm->online_vcpu's incremented value.
2606          */
2607         smp_wmb();
2608         atomic_inc(&kvm->online_vcpus);
2609 
2610         mutex_unlock(&kvm->lock);
2611         kvm_arch_vcpu_postcreate(vcpu);
2612         return r;
2613 
2614 unlock_vcpu_destroy:
2615         mutex_unlock(&kvm->lock);
2616         debugfs_remove_recursive(vcpu->debugfs_dentry);
2617 vcpu_destroy:
2618         kvm_arch_vcpu_destroy(vcpu);
2619 vcpu_decrement:
2620         mutex_lock(&kvm->lock);
2621         kvm->created_vcpus--;
2622         mutex_unlock(&kvm->lock);
2623         return r;
2624 }
2625 
2626 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2627 {
2628         if (sigset) {
2629                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2630                 vcpu->sigset_active = 1;
2631                 vcpu->sigset = *sigset;
2632         } else
2633                 vcpu->sigset_active = 0;
2634         return 0;
2635 }
2636 
2637 static long kvm_vcpu_ioctl(struct file *filp,
2638                            unsigned int ioctl, unsigned long arg)
2639 {
2640         struct kvm_vcpu *vcpu = filp->private_data;
2641         void __user *argp = (void __user *)arg;
2642         int r;
2643         struct kvm_fpu *fpu = NULL;
2644         struct kvm_sregs *kvm_sregs = NULL;
2645 
2646         if (vcpu->kvm->mm != current->mm)
2647                 return -EIO;
2648 
2649         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2650                 return -EINVAL;
2651 
2652         /*
2653          * Some architectures have vcpu ioctls that are asynchronous to vcpu
2654          * execution; mutex_lock() would break them.
2655          */
2656         r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2657         if (r != -ENOIOCTLCMD)
2658                 return r;
2659 
2660         if (mutex_lock_killable(&vcpu->mutex))
2661                 return -EINTR;
2662         switch (ioctl) {
2663         case KVM_RUN: {
2664                 struct pid *oldpid;
2665                 r = -EINVAL;
2666                 if (arg)
2667                         goto out;
2668                 oldpid = rcu_access_pointer(vcpu->pid);
2669                 if (unlikely(oldpid != task_pid(current))) {
2670                         /* The thread running this VCPU changed. */
2671                         struct pid *newpid;
2672 
2673                         r = kvm_arch_vcpu_run_pid_change(vcpu);
2674                         if (r)
2675                                 break;
2676 
2677                         newpid = get_task_pid(current, PIDTYPE_PID);
2678                         rcu_assign_pointer(vcpu->pid, newpid);
2679                         if (oldpid)
2680                                 synchronize_rcu();
2681                         put_pid(oldpid);
2682                 }
2683                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2684                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2685                 break;
2686         }
2687         case KVM_GET_REGS: {
2688                 struct kvm_regs *kvm_regs;
2689 
2690                 r = -ENOMEM;
2691                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2692                 if (!kvm_regs)
2693                         goto out;
2694                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2695                 if (r)
2696                         goto out_free1;
2697                 r = -EFAULT;
2698                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2699                         goto out_free1;
2700                 r = 0;
2701 out_free1:
2702                 kfree(kvm_regs);
2703                 break;
2704         }
2705         case KVM_SET_REGS: {
2706                 struct kvm_regs *kvm_regs;
2707 
2708                 r = -ENOMEM;
2709                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2710                 if (IS_ERR(kvm_regs)) {
2711                         r = PTR_ERR(kvm_regs);
2712                         goto out;
2713                 }
2714                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2715                 kfree(kvm_regs);
2716                 break;
2717         }
2718         case KVM_GET_SREGS: {
2719                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2720                                     GFP_KERNEL_ACCOUNT);
2721                 r = -ENOMEM;
2722                 if (!kvm_sregs)
2723                         goto out;
2724                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2725                 if (r)
2726                         goto out;
2727                 r = -EFAULT;
2728                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2729                         goto out;
2730                 r = 0;
2731                 break;
2732         }
2733         case KVM_SET_SREGS: {
2734                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2735                 if (IS_ERR(kvm_sregs)) {
2736                         r = PTR_ERR(kvm_sregs);
2737                         kvm_sregs = NULL;
2738                         goto out;
2739                 }
2740                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2741                 break;
2742         }
2743         case KVM_GET_MP_STATE: {
2744                 struct kvm_mp_state mp_state;
2745 
2746                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2747                 if (r)
2748                         goto out;
2749                 r = -EFAULT;
2750                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2751                         goto out;
2752                 r = 0;
2753                 break;
2754         }
2755         case KVM_SET_MP_STATE: {
2756                 struct kvm_mp_state mp_state;
2757 
2758                 r = -EFAULT;
2759                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2760                         goto out;
2761                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2762                 break;
2763         }
2764         case KVM_TRANSLATE: {
2765                 struct kvm_translation tr;
2766 
2767                 r = -EFAULT;
2768                 if (copy_from_user(&tr, argp, sizeof(tr)))
2769                         goto out;
2770                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2771                 if (r)
2772                         goto out;
2773                 r = -EFAULT;
2774                 if (copy_to_user(argp, &tr, sizeof(tr)))
2775                         goto out;
2776                 r = 0;
2777                 break;
2778         }
2779         case KVM_SET_GUEST_DEBUG: {
2780                 struct kvm_guest_debug dbg;
2781 
2782                 r = -EFAULT;
2783                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2784                         goto out;
2785                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2786                 break;
2787         }
2788         case KVM_SET_SIGNAL_MASK: {
2789                 struct kvm_signal_mask __user *sigmask_arg = argp;
2790                 struct kvm_signal_mask kvm_sigmask;
2791                 sigset_t sigset, *p;
2792 
2793                 p = NULL;
2794                 if (argp) {
2795                         r = -EFAULT;
2796                         if (copy_from_user(&kvm_sigmask, argp,
2797                                            sizeof(kvm_sigmask)))
2798                                 goto out;
2799                         r = -EINVAL;
2800                         if (kvm_sigmask.len != sizeof(sigset))
2801                                 goto out;
2802                         r = -EFAULT;
2803                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2804                                            sizeof(sigset)))
2805                                 goto out;
2806                         p = &sigset;
2807                 }
2808                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2809                 break;
2810         }
2811         case KVM_GET_FPU: {
2812                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2813                 r = -ENOMEM;
2814                 if (!fpu)
2815                         goto out;
2816                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2817                 if (r)
2818                         goto out;
2819                 r = -EFAULT;
2820                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2821                         goto out;
2822                 r = 0;
2823                 break;
2824         }
2825         case KVM_SET_FPU: {
2826                 fpu = memdup_user(argp, sizeof(*fpu));
2827                 if (IS_ERR(fpu)) {
2828                         r = PTR_ERR(fpu);
2829                         fpu = NULL;
2830                         goto out;
2831                 }
2832                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2833                 break;
2834         }
2835         default:
2836                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2837         }
2838 out:
2839         mutex_unlock(&vcpu->mutex);
2840         kfree(fpu);
2841         kfree(kvm_sregs);
2842         return r;
2843 }
2844 
2845 #ifdef CONFIG_KVM_COMPAT
2846 static long kvm_vcpu_compat_ioctl(struct file *filp,
2847                                   unsigned int ioctl, unsigned long arg)
2848 {
2849         struct kvm_vcpu *vcpu = filp->private_data;
2850         void __user *argp = compat_ptr(arg);
2851         int r;
2852 
2853         if (vcpu->kvm->mm != current->mm)
2854                 return -EIO;
2855 
2856         switch (ioctl) {
2857         case KVM_SET_SIGNAL_MASK: {
2858                 struct kvm_signal_mask __user *sigmask_arg = argp;
2859                 struct kvm_signal_mask kvm_sigmask;
2860                 sigset_t sigset;
2861 
2862                 if (argp) {
2863                         r = -EFAULT;
2864                         if (copy_from_user(&kvm_sigmask, argp,
2865                                            sizeof(kvm_sigmask)))
2866                                 goto out;
2867                         r = -EINVAL;
2868                         if (kvm_sigmask.len != sizeof(compat_sigset_t))
2869                                 goto out;
2870                         r = -EFAULT;
2871                         if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2872                                 goto out;
2873                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2874                 } else
2875                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2876                 break;
2877         }
2878         default:
2879                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2880         }
2881 
2882 out:
2883         return r;
2884 }
2885 #endif
2886 
2887 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2888                                  int (*accessor)(struct kvm_device *dev,
2889                                                  struct kvm_device_attr *attr),
2890                                  unsigned long arg)
2891 {
2892         struct kvm_device_attr attr;
2893 
2894         if (!accessor)
2895                 return -EPERM;
2896 
2897         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2898                 return -EFAULT;
2899 
2900         return accessor(dev, &attr);
2901 }
2902 
2903 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2904                              unsigned long arg)
2905 {
2906         struct kvm_device *dev = filp->private_data;
2907 
2908         switch (ioctl) {
2909         case KVM_SET_DEVICE_ATTR:
2910                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2911         case KVM_GET_DEVICE_ATTR:
2912                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2913         case KVM_HAS_DEVICE_ATTR:
2914                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2915         default:
2916                 if (dev->ops->ioctl)
2917                         return dev->ops->ioctl(dev, ioctl, arg);
2918 
2919                 return -ENOTTY;
2920         }
2921 }
2922 
2923 static int kvm_device_release(struct inode *inode, struct file *filp)
2924 {
2925         struct kvm_device *dev = filp->private_data;
2926         struct kvm *kvm = dev->kvm;
2927 
2928         kvm_put_kvm(kvm);
2929         return 0;
2930 }
2931 
2932 static const struct file_operations kvm_device_fops = {
2933         .unlocked_ioctl = kvm_device_ioctl,
2934         .release = kvm_device_release,
2935         KVM_COMPAT(kvm_device_ioctl),
2936 };
2937 
2938 struct kvm_device *kvm_device_from_filp(struct file *filp)
2939 {
2940         if (filp->f_op != &kvm_device_fops)
2941                 return NULL;
2942 
2943         return filp->private_data;
2944 }
2945 
2946 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2947 #ifdef CONFIG_KVM_MPIC
2948         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2949         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2950 #endif
2951 };
2952 
2953 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2954 {
2955         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2956                 return -ENOSPC;
2957 
2958         if (kvm_device_ops_table[type] != NULL)
2959                 return -EEXIST;
2960 
2961         kvm_device_ops_table[type] = ops;
2962         return 0;
2963 }
2964 
2965 void kvm_unregister_device_ops(u32 type)
2966 {
2967         if (kvm_device_ops_table[type] != NULL)
2968                 kvm_device_ops_table[type] = NULL;
2969 }
2970 
2971 static int kvm_ioctl_create_device(struct kvm *kvm,
2972                                    struct kvm_create_device *cd)
2973 {
2974         struct kvm_device_ops *ops = NULL;
2975         struct kvm_device *dev;
2976         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2977         int ret;
2978 
2979         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2980                 return -ENODEV;
2981 
2982         ops = kvm_device_ops_table[cd->type];
2983         if (ops == NULL)
2984                 return -ENODEV;
2985 
2986         if (test)
2987                 return 0;
2988 
2989         dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
2990         if (!dev)
2991                 return -ENOMEM;
2992 
2993         dev->ops = ops;
2994         dev->kvm = kvm;
2995 
2996         mutex_lock(&kvm->lock);
2997         ret = ops->create(dev, cd->type);
2998         if (ret < 0) {
2999                 mutex_unlock(&kvm->lock);
3000                 kfree(dev);
3001                 return ret;
3002         }
3003         list_add(&dev->vm_node, &kvm->devices);
3004         mutex_unlock(&kvm->lock);
3005 
3006         if (ops->init)
3007                 ops->init(dev);
3008 
3009         kvm_get_kvm(kvm);
3010         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3011         if (ret < 0) {
3012                 kvm_put_kvm(kvm);
3013                 mutex_lock(&kvm->lock);
3014                 list_del(&dev->vm_node);
3015                 mutex_unlock(&kvm->lock);
3016                 ops->destroy(dev);
3017                 return ret;
3018         }
3019 
3020         cd->fd = ret;
3021         return 0;
3022 }
3023 
3024 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3025 {
3026         switch (arg) {
3027         case KVM_CAP_USER_MEMORY:
3028         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3029         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3030         case KVM_CAP_INTERNAL_ERROR_DATA:
3031 #ifdef CONFIG_HAVE_KVM_MSI
3032         case KVM_CAP_SIGNAL_MSI:
3033 #endif
3034 #ifdef CONFIG_HAVE_KVM_IRQFD
3035         case KVM_CAP_IRQFD:
3036         case KVM_CAP_IRQFD_RESAMPLE:
3037 #endif
3038         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3039         case KVM_CAP_CHECK_EXTENSION_VM:
3040         case KVM_CAP_ENABLE_CAP_VM:
3041 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3042         case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
3043 #endif
3044                 return 1;
3045 #ifdef CONFIG_KVM_MMIO
3046         case KVM_CAP_COALESCED_MMIO:
3047                 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3048         case KVM_CAP_COALESCED_PIO:
3049                 return 1;
3050 #endif
3051 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3052         case KVM_CAP_IRQ_ROUTING:
3053                 return KVM_MAX_IRQ_ROUTES;
3054 #endif
3055 #if KVM_ADDRESS_SPACE_NUM > 1
3056         case KVM_CAP_MULTI_ADDRESS_SPACE:
3057                 return KVM_ADDRESS_SPACE_NUM;
3058 #endif
3059         case KVM_CAP_MAX_VCPU_ID:
3060                 return KVM_MAX_VCPU_ID;
3061         default:
3062                 break;
3063         }
3064         return kvm_vm_ioctl_check_extension(kvm, arg);
3065 }
3066 
3067 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3068                                                   struct kvm_enable_cap *cap)
3069 {
3070         return -EINVAL;
3071 }
3072 
3073 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3074                                            struct kvm_enable_cap *cap)
3075 {
3076         switch (cap->cap) {
3077 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3078         case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
3079                 if (cap->flags || (cap->args[0] & ~1))
3080                         return -EINVAL;
3081                 kvm->manual_dirty_log_protect = cap->args[0];
3082                 return 0;
3083 #endif
3084         default:
3085                 return kvm_vm_ioctl_enable_cap(kvm, cap);
3086         }
3087 }
3088 
3089 static long kvm_vm_ioctl(struct file *filp,
3090                            unsigned int ioctl, unsigned long arg)
3091 {
3092         struct kvm *kvm = filp->private_data;
3093         void __user *argp = (void __user *)arg;
3094         int r;
3095 
3096         if (kvm->mm != current->mm)
3097                 return -EIO;
3098         switch (ioctl) {
3099         case KVM_CREATE_VCPU:
3100                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3101                 break;
3102         case KVM_ENABLE_CAP: {
3103                 struct kvm_enable_cap cap;
3104 
3105                 r = -EFAULT;
3106                 if (copy_from_user(&cap, argp, sizeof(cap)))
3107                         goto out;
3108                 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3109                 break;
3110         }
3111         case KVM_SET_USER_MEMORY_REGION: {
3112                 struct kvm_userspace_memory_region kvm_userspace_mem;
3113 
3114                 r = -EFAULT;
3115                 if (copy_from_user(&kvm_userspace_mem, argp,
3116                                                 sizeof(kvm_userspace_mem)))
3117                         goto out;
3118 
3119                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3120                 break;
3121         }
3122         case KVM_GET_DIRTY_LOG: {
3123                 struct kvm_dirty_log log;
3124 
3125                 r = -EFAULT;
3126                 if (copy_from_user(&log, argp, sizeof(log)))
3127                         goto out;
3128                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3129                 break;
3130         }
3131 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3132         case KVM_CLEAR_DIRTY_LOG: {
3133                 struct kvm_clear_dirty_log log;
3134 
3135                 r = -EFAULT;
3136                 if (copy_from_user(&log, argp, sizeof(log)))
3137                         goto out;
3138                 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3139                 break;
3140         }
3141 #endif
3142 #ifdef CONFIG_KVM_MMIO
3143         case KVM_REGISTER_COALESCED_MMIO: {
3144                 struct kvm_coalesced_mmio_zone zone;
3145 
3146                 r = -EFAULT;
3147                 if (copy_from_user(&zone, argp, sizeof(zone)))
3148                         goto out;
3149                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3150                 break;
3151         }
3152         case KVM_UNREGISTER_COALESCED_MMIO: {
3153                 struct kvm_coalesced_mmio_zone zone;
3154 
3155                 r = -EFAULT;
3156                 if (copy_from_user(&zone, argp, sizeof(zone)))
3157                         goto out;
3158                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3159                 break;
3160         }
3161 #endif
3162         case KVM_IRQFD: {
3163                 struct kvm_irqfd data;
3164 
3165                 r = -EFAULT;
3166                 if (copy_from_user(&data, argp, sizeof(data)))
3167                         goto out;
3168                 r = kvm_irqfd(kvm, &data);
3169                 break;
3170         }
3171         case KVM_IOEVENTFD: {
3172                 struct kvm_ioeventfd data;
3173 
3174                 r = -EFAULT;
3175                 if (copy_from_user(&data, argp, sizeof(data)))
3176                         goto out;
3177                 r = kvm_ioeventfd(kvm, &data);
3178                 break;
3179         }
3180 #ifdef CONFIG_HAVE_KVM_MSI
3181         case KVM_SIGNAL_MSI: {
3182                 struct kvm_msi msi;
3183 
3184                 r = -EFAULT;
3185                 if (copy_from_user(&msi, argp, sizeof(msi)))
3186                         goto out;
3187                 r = kvm_send_userspace_msi(kvm, &msi);
3188                 break;
3189         }
3190 #endif
3191 #ifdef __KVM_HAVE_IRQ_LINE
3192         case KVM_IRQ_LINE_STATUS:
3193         case KVM_IRQ_LINE: {
3194                 struct kvm_irq_level irq_event;
3195 
3196                 r = -EFAULT;
3197                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3198                         goto out;
3199 
3200                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3201                                         ioctl == KVM_IRQ_LINE_STATUS);
3202                 if (r)
3203                         goto out;
3204 
3205                 r = -EFAULT;
3206                 if (ioctl == KVM_IRQ_LINE_STATUS) {
3207                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3208                                 goto out;
3209                 }
3210 
3211                 r = 0;
3212                 break;
3213         }
3214 #endif
3215 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3216         case KVM_SET_GSI_ROUTING: {
3217                 struct kvm_irq_routing routing;
3218                 struct kvm_irq_routing __user *urouting;
3219                 struct kvm_irq_routing_entry *entries = NULL;
3220 
3221                 r = -EFAULT;
3222                 if (copy_from_user(&routing, argp, sizeof(routing)))
3223                         goto out;
3224                 r = -EINVAL;
3225                 if (!kvm_arch_can_set_irq_routing(kvm))
3226                         goto out;
3227                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3228                         goto out;
3229                 if (routing.flags)
3230                         goto out;
3231                 if (routing.nr) {
3232                         r = -ENOMEM;
3233                         entries = vmalloc(array_size(sizeof(*entries),
3234                                                      routing.nr));
3235                         if (!entries)
3236                                 goto out;
3237                         r = -EFAULT;
3238                         urouting = argp;
3239                         if (copy_from_user(entries, urouting->entries,
3240                                            routing.nr * sizeof(*entries)))
3241                                 goto out_free_irq_routing;
3242                 }
3243                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3244                                         routing.flags);
3245 out_free_irq_routing:
3246                 vfree(entries);
3247                 break;
3248         }
3249 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3250         case KVM_CREATE_DEVICE: {
3251                 struct kvm_create_device cd;
3252 
3253                 r = -EFAULT;
3254                 if (copy_from_user(&cd, argp, sizeof(cd)))
3255                         goto out;
3256 
3257                 r = kvm_ioctl_create_device(kvm, &cd);
3258                 if (r)
3259                         goto out;
3260 
3261                 r = -EFAULT;
3262                 if (copy_to_user(argp, &cd, sizeof(cd)))
3263                         goto out;
3264 
3265                 r = 0;
3266                 break;
3267         }
3268         case KVM_CHECK_EXTENSION:
3269                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3270                 break;
3271         default:
3272                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3273         }
3274 out:
3275         return r;
3276 }
3277 
3278 #ifdef CONFIG_KVM_COMPAT
3279 struct compat_kvm_dirty_log {
3280         __u32 slot;
3281         __u32 padding1;
3282         union {
3283                 compat_uptr_t dirty_bitmap; /* one bit per page */
3284                 __u64 padding2;
3285         };
3286 };
3287 
3288 static long kvm_vm_compat_ioctl(struct file *filp,
3289                            unsigned int ioctl, unsigned long arg)
3290 {
3291         struct kvm *kvm = filp->private_data;
3292         int r;
3293 
3294         if (kvm->mm != current->mm)
3295                 return -EIO;
3296         switch (ioctl) {
3297         case KVM_GET_DIRTY_LOG: {
3298                 struct compat_kvm_dirty_log compat_log;
3299                 struct kvm_dirty_log log;
3300 
3301                 if (copy_from_user(&compat_log, (void __user *)arg,
3302                                    sizeof(compat_log)))
3303                         return -EFAULT;
3304                 log.slot         = compat_log.slot;
3305                 log.padding1     = compat_log.padding1;
3306                 log.padding2     = compat_log.padding2;
3307                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3308 
3309                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3310                 break;
3311         }
3312         default:
3313                 r = kvm_vm_ioctl(filp, ioctl, arg);
3314         }
3315         return r;
3316 }
3317 #endif
3318 
3319 static struct file_operations kvm_vm_fops = {
3320         .release        = kvm_vm_release,
3321         .unlocked_ioctl = kvm_vm_ioctl,
3322         .llseek         = noop_llseek,
3323         KVM_COMPAT(kvm_vm_compat_ioctl),
3324 };
3325 
3326 static int kvm_dev_ioctl_create_vm(unsigned long type)
3327 {
3328         int r;
3329         struct kvm *kvm;
3330         struct file *file;
3331 
3332         kvm = kvm_create_vm(type);
3333         if (IS_ERR(kvm))
3334                 return PTR_ERR(kvm);
3335 #ifdef CONFIG_KVM_MMIO
3336         r = kvm_coalesced_mmio_init(kvm);
3337         if (r < 0)
3338                 goto put_kvm;
3339 #endif
3340         r = get_unused_fd_flags(O_CLOEXEC);
3341         if (r < 0)
3342                 goto put_kvm;
3343 
3344         file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3345         if (IS_ERR(file)) {
3346                 put_unused_fd(r);
3347                 r = PTR_ERR(file);
3348                 goto put_kvm;
3349         }
3350 
3351         /*
3352          * Don't call kvm_put_kvm anymore at this point; file->f_op is
3353          * already set, with ->release() being kvm_vm_release().  In error
3354          * cases it will be called by the final fput(file) and will take
3355          * care of doing kvm_put_kvm(kvm).
3356          */
3357         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3358                 put_unused_fd(r);
3359                 fput(file);
3360                 return -ENOMEM;
3361         }
3362         kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3363 
3364         fd_install(r, file);
3365         return r;
3366 
3367 put_kvm:
3368         kvm_put_kvm(kvm);
3369         return r;
3370 }
3371 
3372 static long kvm_dev_ioctl(struct file *filp,
3373                           unsigned int ioctl, unsigned long arg)
3374 {
3375         long r = -EINVAL;
3376 
3377         switch (ioctl) {
3378         case KVM_GET_API_VERSION:
3379                 if (arg)
3380                         goto out;
3381                 r = KVM_API_VERSION;
3382                 break;
3383         case KVM_CREATE_VM:
3384                 r = kvm_dev_ioctl_create_vm(arg);
3385                 break;
3386         case KVM_CHECK_EXTENSION:
3387                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3388                 break;
3389         case KVM_GET_VCPU_MMAP_SIZE:
3390                 if (arg)
3391                         goto out;
3392                 r = PAGE_SIZE;     /* struct kvm_run */
3393 #ifdef CONFIG_X86
3394                 r += PAGE_SIZE;    /* pio data page */
3395 #endif
3396 #ifdef CONFIG_KVM_MMIO
3397                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3398 #endif
3399                 break;
3400         case KVM_TRACE_ENABLE:
3401         case KVM_TRACE_PAUSE:
3402         case KVM_TRACE_DISABLE:
3403                 r = -EOPNOTSUPP;
3404                 break;
3405         default:
3406                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3407         }
3408 out:
3409         return r;
3410 }
3411 
3412 static struct file_operations kvm_chardev_ops = {
3413         .unlocked_ioctl = kvm_dev_ioctl,
3414         .llseek         = noop_llseek,
3415         KVM_COMPAT(kvm_dev_ioctl),
3416 };
3417 
3418 static struct miscdevice kvm_dev = {
3419         KVM_MINOR,
3420         "kvm",
3421         &kvm_chardev_ops,
3422 };
3423 
3424 static void hardware_enable_nolock(void *junk)
3425 {
3426         int cpu = raw_smp_processor_id();
3427         int r;
3428 
3429         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3430                 return;
3431 
3432         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3433 
3434         r = kvm_arch_hardware_enable();
3435 
3436         if (r) {
3437                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3438                 atomic_inc(&hardware_enable_failed);
3439                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3440         }
3441 }
3442 
3443 static int kvm_starting_cpu(unsigned int cpu)
3444 {
3445         raw_spin_lock(&kvm_count_lock);
3446         if (kvm_usage_count)
3447                 hardware_enable_nolock(NULL);
3448         raw_spin_unlock(&kvm_count_lock);
3449         return 0;
3450 }
3451 
3452 static void hardware_disable_nolock(void *junk)
3453 {
3454         int cpu = raw_smp_processor_id();
3455 
3456         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3457                 return;
3458         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3459         kvm_arch_hardware_disable();
3460 }
3461 
3462 static int kvm_dying_cpu(unsigned int cpu)
3463 {
3464         raw_spin_lock(&kvm_count_lock);
3465         if (kvm_usage_count)
3466                 hardware_disable_nolock(NULL);
3467         raw_spin_unlock(&kvm_count_lock);
3468         return 0;
3469 }
3470 
3471 static void hardware_disable_all_nolock(void)
3472 {
3473         BUG_ON(!kvm_usage_count);
3474 
3475         kvm_usage_count--;
3476         if (!kvm_usage_count)
3477                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3478 }
3479 
3480 static void hardware_disable_all(void)
3481 {
3482         raw_spin_lock(&kvm_count_lock);
3483         hardware_disable_all_nolock();
3484         raw_spin_unlock(&kvm_count_lock);
3485 }
3486 
3487 static int hardware_enable_all(void)
3488 {
3489         int r = 0;
3490 
3491         raw_spin_lock(&kvm_count_lock);
3492 
3493         kvm_usage_count++;
3494         if (kvm_usage_count == 1) {
3495                 atomic_set(&hardware_enable_failed, 0);
3496                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3497 
3498                 if (atomic_read(&hardware_enable_failed)) {
3499                         hardware_disable_all_nolock();
3500                         r = -EBUSY;
3501                 }
3502         }
3503 
3504         raw_spin_unlock(&kvm_count_lock);
3505 
3506         return r;
3507 }
3508 
3509 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3510                       void *v)
3511 {
3512         /*
3513          * Some (well, at least mine) BIOSes hang on reboot if
3514          * in vmx root mode.
3515          *
3516          * And Intel TXT required VMX off for all cpu when system shutdown.
3517          */
3518         pr_info("kvm: exiting hardware virtualization\n");
3519         kvm_rebooting = true;
3520         on_each_cpu(hardware_disable_nolock, NULL, 1);
3521         return NOTIFY_OK;
3522 }
3523 
3524 static struct notifier_block kvm_reboot_notifier = {
3525         .notifier_call = kvm_reboot,
3526         .priority = 0,
3527 };
3528 
3529 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3530 {
3531         int i;
3532 
3533         for (i = 0; i < bus->dev_count; i++) {
3534                 struct kvm_io_device *pos = bus->range[i].dev;
3535 
3536                 kvm_iodevice_destructor(pos);
3537         }
3538         kfree(bus);
3539 }
3540 
3541 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3542                                  const struct kvm_io_range *r2)
3543 {
3544         gpa_t addr1 = r1->addr;
3545         gpa_t addr2 = r2->addr;
3546 
3547         if (addr1 < addr2)
3548                 return -1;
3549 
3550         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3551          * accept any overlapping write.  Any order is acceptable for
3552          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3553          * we process all of them.
3554          */
3555         if (r2->len) {
3556                 addr1 += r1->len;
3557                 addr2 += r2->len;
3558         }
3559 
3560         if (addr1 > addr2)
3561                 return 1;
3562 
3563         return 0;
3564 }
3565 
3566 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3567 {
3568         return kvm_io_bus_cmp(p1, p2);
3569 }
3570 
3571 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3572                              gpa_t addr, int len)
3573 {
3574         struct kvm_io_range *range, key;
3575         int off;
3576 
3577         key = (struct kvm_io_range) {
3578                 .addr = addr,
3579                 .len = len,
3580         };
3581 
3582         range = bsearch(&key, bus->range, bus->dev_count,
3583                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3584         if (range == NULL)
3585                 return -ENOENT;
3586 
3587         off = range - bus->range;
3588 
3589         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3590                 off--;
3591 
3592         return off;
3593 }
3594 
3595 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3596                               struct kvm_io_range *range, const void *val)
3597 {
3598         int idx;
3599 
3600         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3601         if (idx < 0)
3602                 return -EOPNOTSUPP;
3603 
3604         while (idx < bus->dev_count &&
3605                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3606                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3607                                         range->len, val))
3608                         return idx;
3609                 idx++;
3610         }
3611 
3612         return -EOPNOTSUPP;
3613 }
3614 
3615 /* kvm_io_bus_write - called under kvm->slots_lock */
3616 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3617                      int len, const void *val)
3618 {
3619         struct kvm_io_bus *bus;
3620         struct kvm_io_range range;
3621         int r;
3622 
3623         range = (struct kvm_io_range) {
3624                 .addr = addr,
3625                 .len = len,
3626         };
3627 
3628         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3629         if (!bus)
3630                 return -ENOMEM;
3631         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3632         return r < 0 ? r : 0;
3633 }
3634 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3635 
3636 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3637 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3638                             gpa_t addr, int len, const void *val, long cookie)
3639 {
3640         struct kvm_io_bus *bus;
3641         struct kvm_io_range range;
3642 
3643         range = (struct kvm_io_range) {
3644                 .addr = addr,
3645                 .len = len,
3646         };
3647 
3648         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3649         if (!bus)
3650                 return -ENOMEM;
3651 
3652         /* First try the device referenced by cookie. */
3653         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3654             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3655                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3656                                         val))
3657                         return cookie;
3658 
3659         /*
3660          * cookie contained garbage; fall back to search and return the
3661          * correct cookie value.
3662          */
3663         return __kvm_io_bus_write(vcpu, bus, &range, val);
3664 }
3665 
3666 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3667                              struct kvm_io_range *range, void *val)
3668 {
3669         int idx;
3670 
3671         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3672         if (idx < 0)
3673                 return -EOPNOTSUPP;
3674 
3675         while (idx < bus->dev_count &&
3676                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3677                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3678                                        range->len, val))
3679                         return idx;
3680                 idx++;
3681         }
3682 
3683         return -EOPNOTSUPP;
3684 }
3685 
3686 /* kvm_io_bus_read - called under kvm->slots_lock */
3687 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3688                     int len, void *val)
3689 {
3690         struct kvm_io_bus *bus;
3691         struct kvm_io_range range;
3692         int r;
3693 
3694         range = (struct kvm_io_range) {
3695                 .addr = addr,
3696                 .len = len,
3697         };
3698 
3699         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3700         if (!bus)
3701                 return -ENOMEM;
3702         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3703         return r < 0 ? r : 0;
3704 }
3705 
3706 /* Caller must hold slots_lock. */
3707 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3708                             int len, struct kvm_io_device *dev)
3709 {
3710         int i;
3711         struct kvm_io_bus *new_bus, *bus;
3712         struct kvm_io_range range;
3713 
3714         bus = kvm_get_bus(kvm, bus_idx);
3715         if (!bus)
3716                 return -ENOMEM;
3717 
3718         /* exclude ioeventfd which is limited by maximum fd */
3719         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3720                 return -ENOSPC;
3721 
3722         new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3723                           GFP_KERNEL_ACCOUNT);
3724         if (!new_bus)
3725                 return -ENOMEM;
3726 
3727         range = (struct kvm_io_range) {
3728                 .addr = addr,
3729                 .len = len,
3730                 .dev = dev,
3731         };
3732 
3733         for (i = 0; i < bus->dev_count; i++)
3734                 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3735                         break;
3736 
3737         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3738         new_bus->dev_count++;
3739         new_bus->range[i] = range;
3740         memcpy(new_bus->range + i + 1, bus->range + i,
3741                 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3742         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3743         synchronize_srcu_expedited(&kvm->srcu);
3744         kfree(bus);
3745 
3746         return 0;
3747 }
3748 
3749 /* Caller must hold slots_lock. */
3750 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3751                                struct kvm_io_device *dev)
3752 {
3753         int i;
3754         struct kvm_io_bus *new_bus, *bus;
3755 
3756         bus = kvm_get_bus(kvm, bus_idx);
3757         if (!bus)
3758                 return;
3759 
3760         for (i = 0; i < bus->dev_count; i++)
3761                 if (bus->range[i].dev == dev) {
3762                         break;
3763                 }
3764 
3765         if (i == bus->dev_count)
3766                 return;
3767 
3768         new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3769                           GFP_KERNEL_ACCOUNT);
3770         if (!new_bus)  {
3771                 pr_err("kvm: failed to shrink bus, removing it completely\n");
3772                 goto broken;
3773         }
3774 
3775         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3776         new_bus->dev_count--;
3777         memcpy(new_bus->range + i, bus->range + i + 1,
3778                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3779 
3780 broken:
3781         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3782         synchronize_srcu_expedited(&kvm->srcu);
3783         kfree(bus);
3784         return;
3785 }
3786 
3787 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3788                                          gpa_t addr)
3789 {
3790         struct kvm_io_bus *bus;
3791         int dev_idx, srcu_idx;
3792         struct kvm_io_device *iodev = NULL;
3793 
3794         srcu_idx = srcu_read_lock(&kvm->srcu);
3795 
3796         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3797         if (!bus)
3798                 goto out_unlock;
3799 
3800         dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3801         if (dev_idx < 0)
3802                 goto out_unlock;
3803 
3804         iodev = bus->range[dev_idx].dev;
3805 
3806 out_unlock:
3807         srcu_read_unlock(&kvm->srcu, srcu_idx);
3808 
3809         return iodev;
3810 }
3811 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3812 
3813 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3814                            int (*get)(void *, u64 *), int (*set)(void *, u64),
3815                            const char *fmt)
3816 {
3817         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3818                                           inode->i_private;
3819 
3820         /* The debugfs files are a reference to the kvm struct which
3821          * is still valid when kvm_destroy_vm is called.
3822          * To avoid the race between open and the removal of the debugfs
3823          * directory we test against the users count.
3824          */
3825         if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3826                 return -ENOENT;
3827 
3828         if (simple_attr_open(inode, file, get, set, fmt)) {
3829                 kvm_put_kvm(stat_data->kvm);
3830                 return -ENOMEM;
3831         }
3832 
3833         return 0;
3834 }
3835 
3836 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3837 {
3838         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3839                                           inode->i_private;
3840 
3841         simple_attr_release(inode, file);
3842         kvm_put_kvm(stat_data->kvm);
3843 
3844         return 0;
3845 }
3846 
3847 static int vm_stat_get_per_vm(void *data, u64 *val)
3848 {
3849         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3850 
3851         *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3852 
3853         return 0;
3854 }
3855 
3856 static int vm_stat_clear_per_vm(void *data, u64 val)
3857 {
3858         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3859 
3860         if (val)
3861                 return -EINVAL;
3862 
3863         *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3864 
3865         return 0;
3866 }
3867 
3868 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3869 {
3870         __simple_attr_check_format("%llu\n", 0ull);
3871         return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3872                                 vm_stat_clear_per_vm, "%llu\n");
3873 }
3874 
3875 static const struct file_operations vm_stat_get_per_vm_fops = {
3876         .owner   = THIS_MODULE,
3877         .open    = vm_stat_get_per_vm_open,
3878         .release = kvm_debugfs_release,
3879         .read    = simple_attr_read,
3880         .write   = simple_attr_write,
3881         .llseek  = no_llseek,
3882 };
3883 
3884 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3885 {
3886         int i;
3887         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3888         struct kvm_vcpu *vcpu;
3889 
3890         *val = 0;
3891 
3892         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3893                 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3894 
3895         return 0;
3896 }
3897 
3898 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3899 {
3900         int i;
3901         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3902         struct kvm_vcpu *vcpu;
3903 
3904         if (val)
3905                 return -EINVAL;
3906 
3907         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3908                 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3909 
3910         return 0;
3911 }
3912 
3913 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3914 {
3915         __simple_attr_check_format("%llu\n", 0ull);
3916         return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3917                                  vcpu_stat_clear_per_vm, "%llu\n");
3918 }
3919 
3920 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3921         .owner   = THIS_MODULE,
3922         .open    = vcpu_stat_get_per_vm_open,
3923         .release = kvm_debugfs_release,
3924         .read    = simple_attr_read,
3925         .write   = simple_attr_write,
3926         .llseek  = no_llseek,
3927 };
3928 
3929 static const struct file_operations *stat_fops_per_vm[] = {
3930         [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3931         [KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
3932 };
3933 
3934 static int vm_stat_get(void *_offset, u64 *val)
3935 {
3936         unsigned offset = (long)_offset;
3937         struct kvm *kvm;
3938         struct kvm_stat_data stat_tmp = {.offset = offset};
3939         u64 tmp_val;
3940 
3941         *val = 0;
3942         spin_lock(&kvm_lock);
3943         list_for_each_entry(kvm, &vm_list, vm_list) {
3944                 stat_tmp.kvm = kvm;
3945                 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3946                 *val += tmp_val;
3947         }
3948         spin_unlock(&kvm_lock);
3949         return 0;
3950 }
3951 
3952 static int vm_stat_clear(void *_offset, u64 val)
3953 {
3954         unsigned offset = (long)_offset;
3955         struct kvm *kvm;
3956         struct kvm_stat_data stat_tmp = {.offset = offset};
3957 
3958         if (val)
3959                 return -EINVAL;
3960 
3961         spin_lock(&kvm_lock);
3962         list_for_each_entry(kvm, &vm_list, vm_list) {
3963                 stat_tmp.kvm = kvm;
3964                 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3965         }
3966         spin_unlock(&kvm_lock);
3967 
3968         return 0;
3969 }
3970 
3971 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3972 
3973 static int vcpu_stat_get(void *_offset, u64 *val)
3974 {
3975         unsigned offset = (long)_offset;
3976         struct kvm *kvm;
3977         struct kvm_stat_data stat_tmp = {.offset = offset};
3978         u64 tmp_val;
3979 
3980         *val = 0;
3981         spin_lock(&kvm_lock);
3982         list_for_each_entry(kvm, &vm_list, vm_list) {
3983                 stat_tmp.kvm = kvm;
3984                 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3985                 *val += tmp_val;
3986         }
3987         spin_unlock(&kvm_lock);
3988         return 0;
3989 }
3990 
3991 static int vcpu_stat_clear(void *_offset, u64 val)
3992 {
3993         unsigned offset = (long)_offset;
3994         struct kvm *kvm;
3995         struct kvm_stat_data stat_tmp = {.offset = offset};
3996 
3997         if (val)
3998                 return -EINVAL;
3999 
4000         spin_lock(&kvm_lock);
4001         list_for_each_entry(kvm, &vm_list, vm_list) {
4002                 stat_tmp.kvm = kvm;
4003                 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4004         }
4005         spin_unlock(&kvm_lock);
4006 
4007         return 0;
4008 }
4009 
4010 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4011                         "%llu\n");
4012 
4013 static const struct file_operations *stat_fops[] = {
4014         [KVM_STAT_VCPU] = &vcpu_stat_fops,
4015         [KVM_STAT_VM]   = &vm_stat_fops,
4016 };
4017 
4018 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4019 {
4020         struct kobj_uevent_env *env;
4021         unsigned long long created, active;
4022 
4023         if (!kvm_dev.this_device || !kvm)
4024                 return;
4025 
4026         spin_lock(&kvm_lock);
4027         if (type == KVM_EVENT_CREATE_VM) {
4028                 kvm_createvm_count++;
4029                 kvm_active_vms++;
4030         } else if (type == KVM_EVENT_DESTROY_VM) {
4031                 kvm_active_vms--;
4032         }
4033         created = kvm_createvm_count;
4034         active = kvm_active_vms;
4035         spin_unlock(&kvm_lock);
4036 
4037         env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4038         if (!env)
4039                 return;
4040 
4041         add_uevent_var(env, "CREATED=%llu", created);
4042         add_uevent_var(env, "COUNT=%llu", active);
4043 
4044         if (type == KVM_EVENT_CREATE_VM) {
4045                 add_uevent_var(env, "EVENT=create");
4046                 kvm->userspace_pid = task_pid_nr(current);
4047         } else if (type == KVM_EVENT_DESTROY_VM) {
4048                 add_uevent_var(env, "EVENT=destroy");
4049         }
4050         add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4051 
4052         if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4053                 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4054 
4055                 if (p) {
4056                         tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4057                         if (!IS_ERR(tmp))
4058                                 add_uevent_var(env, "STATS_PATH=%s", tmp);
4059                         kfree(p);
4060                 }
4061         }
4062         /* no need for checks, since we are adding at most only 5 keys */
4063         env->envp[env->envp_idx++] = NULL;
4064         kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4065         kfree(env);
4066 }
4067 
4068 static void kvm_init_debug(void)
4069 {
4070         struct kvm_stats_debugfs_item *p;
4071 
4072         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4073 
4074         kvm_debugfs_num_entries = 0;
4075         for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4076                 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
4077                                     (void *)(long)p->offset,
4078                                     stat_fops[p->kind]);
4079         }
4080 }
4081 
4082 static int kvm_suspend(void)
4083 {
4084         if (kvm_usage_count)
4085                 hardware_disable_nolock(NULL);
4086         return 0;
4087 }
4088 
4089 static void kvm_resume(void)
4090 {
4091         if (kvm_usage_count) {
4092                 lockdep_assert_held(&kvm_count_lock);
4093                 hardware_enable_nolock(NULL);
4094         }
4095 }
4096 
4097 static struct syscore_ops kvm_syscore_ops = {
4098         .suspend = kvm_suspend,
4099         .resume = kvm_resume,
4100 };
4101 
4102 static inline
4103 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4104 {
4105         return container_of(pn, struct kvm_vcpu, preempt_notifier);
4106 }
4107 
4108 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4109 {
4110         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4111 
4112         if (vcpu->preempted)
4113                 vcpu->preempted = false;
4114 
4115         kvm_arch_sched_in(vcpu, cpu);
4116 
4117         kvm_arch_vcpu_load(vcpu, cpu);
4118 }
4119 
4120 static void kvm_sched_out(struct preempt_notifier *pn,
4121                           struct task_struct *next)
4122 {
4123         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4124 
4125         if (current->state == TASK_RUNNING)
4126                 vcpu->preempted = true;
4127         kvm_arch_vcpu_put(vcpu);
4128 }
4129 
4130 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4131                   struct module *module)
4132 {
4133         int r;
4134         int cpu;
4135 
4136         r = kvm_arch_init(opaque);
4137         if (r)
4138                 goto out_fail;
4139 
4140         /*
4141          * kvm_arch_init makes sure there's at most one caller
4142          * for architectures that support multiple implementations,
4143          * like intel and amd on x86.
4144          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4145          * conflicts in case kvm is already setup for another implementation.
4146          */
4147         r = kvm_irqfd_init();
4148         if (r)
4149                 goto out_irqfd;
4150 
4151         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4152                 r = -ENOMEM;
4153                 goto out_free_0;
4154         }
4155 
4156         r = kvm_arch_hardware_setup();
4157         if (r < 0)
4158                 goto out_free_0a;
4159 
4160         for_each_online_cpu(cpu) {
4161                 smp_call_function_single(cpu,
4162                                 kvm_arch_check_processor_compat,
4163                                 &r, 1);
4164                 if (r < 0)
4165                         goto out_free_1;
4166         }
4167 
4168         r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4169                                       kvm_starting_cpu, kvm_dying_cpu);
4170         if (r)
4171                 goto out_free_2;
4172         register_reboot_notifier(&kvm_reboot_notifier);
4173 
4174         /* A kmem cache lets us meet the alignment requirements of fx_save. */
4175         if (!vcpu_align)
4176                 vcpu_align = __alignof__(struct kvm_vcpu);
4177         kvm_vcpu_cache =
4178                 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4179                                            SLAB_ACCOUNT,
4180                                            offsetof(struct kvm_vcpu, arch),
4181                                            sizeof_field(struct kvm_vcpu, arch),
4182                                            NULL);
4183         if (!kvm_vcpu_cache) {
4184                 r = -ENOMEM;
4185                 goto out_free_3;
4186         }
4187 
4188         r = kvm_async_pf_init();
4189         if (r)
4190                 goto out_free;
4191 
4192         kvm_chardev_ops.owner = module;
4193         kvm_vm_fops.owner = module;
4194         kvm_vcpu_fops.owner = module;
4195 
4196         r = misc_register(&kvm_dev);
4197         if (r) {
4198                 pr_err("kvm: misc device register failed\n");
4199                 goto out_unreg;
4200         }
4201 
4202         register_syscore_ops(&kvm_syscore_ops);
4203 
4204         kvm_preempt_ops.sched_in = kvm_sched_in;
4205         kvm_preempt_ops.sched_out = kvm_sched_out;
4206 
4207         kvm_init_debug();
4208 
4209         r = kvm_vfio_ops_init();
4210         WARN_ON(r);
4211 
4212         return 0;
4213 
4214 out_unreg:
4215         kvm_async_pf_deinit();
4216 out_free:
4217         kmem_cache_destroy(kvm_vcpu_cache);
4218 out_free_3:
4219         unregister_reboot_notifier(&kvm_reboot_notifier);
4220         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4221 out_free_2:
4222 out_free_1:
4223         kvm_arch_hardware_unsetup();
4224 out_free_0a:
4225         free_cpumask_var(cpus_hardware_enabled);
4226 out_free_0:
4227         kvm_irqfd_exit();
4228 out_irqfd:
4229         kvm_arch_exit();
4230 out_fail:
4231         return r;
4232 }
4233 EXPORT_SYMBOL_GPL(kvm_init);
4234 
4235 void kvm_exit(void)
4236 {
4237         debugfs_remove_recursive(kvm_debugfs_dir);
4238         misc_deregister(&kvm_dev);
4239         kmem_cache_destroy(kvm_vcpu_cache);
4240         kvm_async_pf_deinit();
4241         unregister_syscore_ops(&kvm_syscore_ops);
4242         unregister_reboot_notifier(&kvm_reboot_notifier);
4243         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4244         on_each_cpu(hardware_disable_nolock, NULL, 1);
4245         kvm_arch_hardware_unsetup();
4246         kvm_arch_exit();
4247         kvm_irqfd_exit();
4248         free_cpumask_var(cpus_hardware_enabled);
4249         kvm_vfio_ops_exit();
4250 }
4251 EXPORT_SYMBOL_GPL(kvm_exit);
4252 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp