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

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

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ 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-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 /* Default resets per-vcpu halt_poll_ns . */
 85 unsigned int halt_poll_ns_shrink;
 86 module_param(halt_poll_ns_shrink, uint, 0644);
 87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
 88 
 89 /*
 90  * Ordering of locks:
 91  *
 92  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
 93  */
 94 
 95 DEFINE_SPINLOCK(kvm_lock);
 96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 97 LIST_HEAD(vm_list);
 98 
 99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
102 
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
105 
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
107 
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
110 
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
113 
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115                            unsigned long arg);
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
118                                   unsigned long arg);
119 #endif
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
122 
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
124 
125 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
126 
127 __visible bool kvm_rebooting;
128 EXPORT_SYMBOL_GPL(kvm_rebooting);
129 
130 static bool largepages_enabled = true;
131 
132 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
135 static unsigned long long kvm_createvm_count;
136 static unsigned long long kvm_active_vms;
137 
138 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
139                 unsigned long start, unsigned long end)
140 {
141 }
142 
143 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
144 {
145         if (pfn_valid(pfn))
146                 return PageReserved(pfn_to_page(pfn));
147 
148         return true;
149 }
150 
151 /*
152  * Switches to specified vcpu, until a matching vcpu_put()
153  */
154 void vcpu_load(struct kvm_vcpu *vcpu)
155 {
156         int cpu = get_cpu();
157         preempt_notifier_register(&vcpu->preempt_notifier);
158         kvm_arch_vcpu_load(vcpu, cpu);
159         put_cpu();
160 }
161 EXPORT_SYMBOL_GPL(vcpu_load);
162 
163 void vcpu_put(struct kvm_vcpu *vcpu)
164 {
165         preempt_disable();
166         kvm_arch_vcpu_put(vcpu);
167         preempt_notifier_unregister(&vcpu->preempt_notifier);
168         preempt_enable();
169 }
170 EXPORT_SYMBOL_GPL(vcpu_put);
171 
172 /* TODO: merge with kvm_arch_vcpu_should_kick */
173 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
174 {
175         int mode = kvm_vcpu_exiting_guest_mode(vcpu);
176 
177         /*
178          * We need to wait for the VCPU to reenable interrupts and get out of
179          * READING_SHADOW_PAGE_TABLES mode.
180          */
181         if (req & KVM_REQUEST_WAIT)
182                 return mode != OUTSIDE_GUEST_MODE;
183 
184         /*
185          * Need to kick a running VCPU, but otherwise there is nothing to do.
186          */
187         return mode == IN_GUEST_MODE;
188 }
189 
190 static void ack_flush(void *_completed)
191 {
192 }
193 
194 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
195 {
196         if (unlikely(!cpus))
197                 cpus = cpu_online_mask;
198 
199         if (cpumask_empty(cpus))
200                 return false;
201 
202         smp_call_function_many(cpus, ack_flush, NULL, wait);
203         return true;
204 }
205 
206 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
207 {
208         int i, cpu, me;
209         cpumask_var_t cpus;
210         bool called;
211         struct kvm_vcpu *vcpu;
212 
213         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
214 
215         me = get_cpu();
216         kvm_for_each_vcpu(i, vcpu, kvm) {
217                 kvm_make_request(req, vcpu);
218                 cpu = vcpu->cpu;
219 
220                 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
221                         continue;
222 
223                 if (cpus != NULL && cpu != -1 && cpu != me &&
224                     kvm_request_needs_ipi(vcpu, req))
225                         __cpumask_set_cpu(cpu, cpus);
226         }
227         called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
228         put_cpu();
229         free_cpumask_var(cpus);
230         return called;
231 }
232 
233 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
234 void kvm_flush_remote_tlbs(struct kvm *kvm)
235 {
236         /*
237          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
238          * kvm_make_all_cpus_request.
239          */
240         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
241 
242         /*
243          * We want to publish modifications to the page tables before reading
244          * mode. Pairs with a memory barrier in arch-specific code.
245          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
246          * and smp_mb in walk_shadow_page_lockless_begin/end.
247          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
248          *
249          * There is already an smp_mb__after_atomic() before
250          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
251          * barrier here.
252          */
253         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
254                 ++kvm->stat.remote_tlb_flush;
255         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
256 }
257 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
258 #endif
259 
260 void kvm_reload_remote_mmus(struct kvm *kvm)
261 {
262         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
263 }
264 
265 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
266 {
267         struct page *page;
268         int r;
269 
270         mutex_init(&vcpu->mutex);
271         vcpu->cpu = -1;
272         vcpu->kvm = kvm;
273         vcpu->vcpu_id = id;
274         vcpu->pid = NULL;
275         init_swait_queue_head(&vcpu->wq);
276         kvm_async_pf_vcpu_init(vcpu);
277 
278         vcpu->pre_pcpu = -1;
279         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
280 
281         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
282         if (!page) {
283                 r = -ENOMEM;
284                 goto fail;
285         }
286         vcpu->run = page_address(page);
287 
288         kvm_vcpu_set_in_spin_loop(vcpu, false);
289         kvm_vcpu_set_dy_eligible(vcpu, false);
290         vcpu->preempted = false;
291 
292         r = kvm_arch_vcpu_init(vcpu);
293         if (r < 0)
294                 goto fail_free_run;
295         return 0;
296 
297 fail_free_run:
298         free_page((unsigned long)vcpu->run);
299 fail:
300         return r;
301 }
302 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
303 
304 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
305 {
306         /*
307          * no need for rcu_read_lock as VCPU_RUN is the only place that
308          * will change the vcpu->pid pointer and on uninit all file
309          * descriptors are already gone.
310          */
311         put_pid(rcu_dereference_protected(vcpu->pid, 1));
312         kvm_arch_vcpu_uninit(vcpu);
313         free_page((unsigned long)vcpu->run);
314 }
315 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
316 
317 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
318 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
319 {
320         return container_of(mn, struct kvm, mmu_notifier);
321 }
322 
323 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
324                                         struct mm_struct *mm,
325                                         unsigned long address,
326                                         pte_t pte)
327 {
328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
329         int idx;
330 
331         idx = srcu_read_lock(&kvm->srcu);
332         spin_lock(&kvm->mmu_lock);
333         kvm->mmu_notifier_seq++;
334         kvm_set_spte_hva(kvm, address, pte);
335         spin_unlock(&kvm->mmu_lock);
336         srcu_read_unlock(&kvm->srcu, idx);
337 }
338 
339 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
340                                                     struct mm_struct *mm,
341                                                     unsigned long start,
342                                                     unsigned long end)
343 {
344         struct kvm *kvm = mmu_notifier_to_kvm(mn);
345         int need_tlb_flush = 0, idx;
346 
347         idx = srcu_read_lock(&kvm->srcu);
348         spin_lock(&kvm->mmu_lock);
349         /*
350          * The count increase must become visible at unlock time as no
351          * spte can be established without taking the mmu_lock and
352          * count is also read inside the mmu_lock critical section.
353          */
354         kvm->mmu_notifier_count++;
355         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
356         need_tlb_flush |= kvm->tlbs_dirty;
357         /* we've to flush the tlb before the pages can be freed */
358         if (need_tlb_flush)
359                 kvm_flush_remote_tlbs(kvm);
360 
361         spin_unlock(&kvm->mmu_lock);
362 
363         kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
364 
365         srcu_read_unlock(&kvm->srcu, idx);
366 }
367 
368 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
369                                                   struct mm_struct *mm,
370                                                   unsigned long start,
371                                                   unsigned long end)
372 {
373         struct kvm *kvm = mmu_notifier_to_kvm(mn);
374 
375         spin_lock(&kvm->mmu_lock);
376         /*
377          * This sequence increase will notify the kvm page fault that
378          * the page that is going to be mapped in the spte could have
379          * been freed.
380          */
381         kvm->mmu_notifier_seq++;
382         smp_wmb();
383         /*
384          * The above sequence increase must be visible before the
385          * below count decrease, which is ensured by the smp_wmb above
386          * in conjunction with the smp_rmb in mmu_notifier_retry().
387          */
388         kvm->mmu_notifier_count--;
389         spin_unlock(&kvm->mmu_lock);
390 
391         BUG_ON(kvm->mmu_notifier_count < 0);
392 }
393 
394 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
395                                               struct mm_struct *mm,
396                                               unsigned long start,
397                                               unsigned long end)
398 {
399         struct kvm *kvm = mmu_notifier_to_kvm(mn);
400         int young, idx;
401 
402         idx = srcu_read_lock(&kvm->srcu);
403         spin_lock(&kvm->mmu_lock);
404 
405         young = kvm_age_hva(kvm, start, end);
406         if (young)
407                 kvm_flush_remote_tlbs(kvm);
408 
409         spin_unlock(&kvm->mmu_lock);
410         srcu_read_unlock(&kvm->srcu, idx);
411 
412         return young;
413 }
414 
415 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
416                                         struct mm_struct *mm,
417                                         unsigned long start,
418                                         unsigned long end)
419 {
420         struct kvm *kvm = mmu_notifier_to_kvm(mn);
421         int young, idx;
422 
423         idx = srcu_read_lock(&kvm->srcu);
424         spin_lock(&kvm->mmu_lock);
425         /*
426          * Even though we do not flush TLB, this will still adversely
427          * affect performance on pre-Haswell Intel EPT, where there is
428          * no EPT Access Bit to clear so that we have to tear down EPT
429          * tables instead. If we find this unacceptable, we can always
430          * add a parameter to kvm_age_hva so that it effectively doesn't
431          * do anything on clear_young.
432          *
433          * Also note that currently we never issue secondary TLB flushes
434          * from clear_young, leaving this job up to the regular system
435          * cadence. If we find this inaccurate, we might come up with a
436          * more sophisticated heuristic later.
437          */
438         young = kvm_age_hva(kvm, start, end);
439         spin_unlock(&kvm->mmu_lock);
440         srcu_read_unlock(&kvm->srcu, idx);
441 
442         return young;
443 }
444 
445 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
446                                        struct mm_struct *mm,
447                                        unsigned long address)
448 {
449         struct kvm *kvm = mmu_notifier_to_kvm(mn);
450         int young, idx;
451 
452         idx = srcu_read_lock(&kvm->srcu);
453         spin_lock(&kvm->mmu_lock);
454         young = kvm_test_age_hva(kvm, address);
455         spin_unlock(&kvm->mmu_lock);
456         srcu_read_unlock(&kvm->srcu, idx);
457 
458         return young;
459 }
460 
461 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
462                                      struct mm_struct *mm)
463 {
464         struct kvm *kvm = mmu_notifier_to_kvm(mn);
465         int idx;
466 
467         idx = srcu_read_lock(&kvm->srcu);
468         kvm_arch_flush_shadow_all(kvm);
469         srcu_read_unlock(&kvm->srcu, idx);
470 }
471 
472 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
473         .flags                  = MMU_INVALIDATE_DOES_NOT_BLOCK,
474         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
475         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
476         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
477         .clear_young            = kvm_mmu_notifier_clear_young,
478         .test_young             = kvm_mmu_notifier_test_young,
479         .change_pte             = kvm_mmu_notifier_change_pte,
480         .release                = kvm_mmu_notifier_release,
481 };
482 
483 static int kvm_init_mmu_notifier(struct kvm *kvm)
484 {
485         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
486         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
487 }
488 
489 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
490 
491 static int kvm_init_mmu_notifier(struct kvm *kvm)
492 {
493         return 0;
494 }
495 
496 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
497 
498 static struct kvm_memslots *kvm_alloc_memslots(void)
499 {
500         int i;
501         struct kvm_memslots *slots;
502 
503         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
504         if (!slots)
505                 return NULL;
506 
507         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
508                 slots->id_to_index[i] = slots->memslots[i].id = i;
509 
510         return slots;
511 }
512 
513 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
514 {
515         if (!memslot->dirty_bitmap)
516                 return;
517 
518         kvfree(memslot->dirty_bitmap);
519         memslot->dirty_bitmap = NULL;
520 }
521 
522 /*
523  * Free any memory in @free but not in @dont.
524  */
525 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
526                               struct kvm_memory_slot *dont)
527 {
528         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
529                 kvm_destroy_dirty_bitmap(free);
530 
531         kvm_arch_free_memslot(kvm, free, dont);
532 
533         free->npages = 0;
534 }
535 
536 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
537 {
538         struct kvm_memory_slot *memslot;
539 
540         if (!slots)
541                 return;
542 
543         kvm_for_each_memslot(memslot, slots)
544                 kvm_free_memslot(kvm, memslot, NULL);
545 
546         kvfree(slots);
547 }
548 
549 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
550 {
551         int i;
552 
553         if (!kvm->debugfs_dentry)
554                 return;
555 
556         debugfs_remove_recursive(kvm->debugfs_dentry);
557 
558         if (kvm->debugfs_stat_data) {
559                 for (i = 0; i < kvm_debugfs_num_entries; i++)
560                         kfree(kvm->debugfs_stat_data[i]);
561                 kfree(kvm->debugfs_stat_data);
562         }
563 }
564 
565 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
566 {
567         char dir_name[ITOA_MAX_LEN * 2];
568         struct kvm_stat_data *stat_data;
569         struct kvm_stats_debugfs_item *p;
570 
571         if (!debugfs_initialized())
572                 return 0;
573 
574         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
575         kvm->debugfs_dentry = debugfs_create_dir(dir_name,
576                                                  kvm_debugfs_dir);
577         if (!kvm->debugfs_dentry)
578                 return -ENOMEM;
579 
580         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
581                                          sizeof(*kvm->debugfs_stat_data),
582                                          GFP_KERNEL);
583         if (!kvm->debugfs_stat_data)
584                 return -ENOMEM;
585 
586         for (p = debugfs_entries; p->name; p++) {
587                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
588                 if (!stat_data)
589                         return -ENOMEM;
590 
591                 stat_data->kvm = kvm;
592                 stat_data->offset = p->offset;
593                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
594                 if (!debugfs_create_file(p->name, 0644,
595                                          kvm->debugfs_dentry,
596                                          stat_data,
597                                          stat_fops_per_vm[p->kind]))
598                         return -ENOMEM;
599         }
600         return 0;
601 }
602 
603 static struct kvm *kvm_create_vm(unsigned long type)
604 {
605         int r, i;
606         struct kvm *kvm = kvm_arch_alloc_vm();
607 
608         if (!kvm)
609                 return ERR_PTR(-ENOMEM);
610 
611         spin_lock_init(&kvm->mmu_lock);
612         mmgrab(current->mm);
613         kvm->mm = current->mm;
614         kvm_eventfd_init(kvm);
615         mutex_init(&kvm->lock);
616         mutex_init(&kvm->irq_lock);
617         mutex_init(&kvm->slots_lock);
618         refcount_set(&kvm->users_count, 1);
619         INIT_LIST_HEAD(&kvm->devices);
620 
621         r = kvm_arch_init_vm(kvm, type);
622         if (r)
623                 goto out_err_no_disable;
624 
625         r = hardware_enable_all();
626         if (r)
627                 goto out_err_no_disable;
628 
629 #ifdef CONFIG_HAVE_KVM_IRQFD
630         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
631 #endif
632 
633         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
634 
635         r = -ENOMEM;
636         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
637                 struct kvm_memslots *slots = kvm_alloc_memslots();
638                 if (!slots)
639                         goto out_err_no_srcu;
640                 /*
641                  * Generations must be different for each address space.
642                  * Init kvm generation close to the maximum to easily test the
643                  * code of handling generation number wrap-around.
644                  */
645                 slots->generation = i * 2 - 150;
646                 rcu_assign_pointer(kvm->memslots[i], slots);
647         }
648 
649         if (init_srcu_struct(&kvm->srcu))
650                 goto out_err_no_srcu;
651         if (init_srcu_struct(&kvm->irq_srcu))
652                 goto out_err_no_irq_srcu;
653         for (i = 0; i < KVM_NR_BUSES; i++) {
654                 rcu_assign_pointer(kvm->buses[i],
655                         kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
656                 if (!kvm->buses[i])
657                         goto out_err;
658         }
659 
660         r = kvm_init_mmu_notifier(kvm);
661         if (r)
662                 goto out_err;
663 
664         spin_lock(&kvm_lock);
665         list_add(&kvm->vm_list, &vm_list);
666         spin_unlock(&kvm_lock);
667 
668         preempt_notifier_inc();
669 
670         return kvm;
671 
672 out_err:
673         cleanup_srcu_struct(&kvm->irq_srcu);
674 out_err_no_irq_srcu:
675         cleanup_srcu_struct(&kvm->srcu);
676 out_err_no_srcu:
677         hardware_disable_all();
678 out_err_no_disable:
679         refcount_set(&kvm->users_count, 0);
680         for (i = 0; i < KVM_NR_BUSES; i++)
681                 kfree(kvm_get_bus(kvm, i));
682         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
683                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
684         kvm_arch_free_vm(kvm);
685         mmdrop(current->mm);
686         return ERR_PTR(r);
687 }
688 
689 static void kvm_destroy_devices(struct kvm *kvm)
690 {
691         struct kvm_device *dev, *tmp;
692 
693         /*
694          * We do not need to take the kvm->lock here, because nobody else
695          * has a reference to the struct kvm at this point and therefore
696          * cannot access the devices list anyhow.
697          */
698         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
699                 list_del(&dev->vm_node);
700                 dev->ops->destroy(dev);
701         }
702 }
703 
704 static void kvm_destroy_vm(struct kvm *kvm)
705 {
706         int i;
707         struct mm_struct *mm = kvm->mm;
708 
709         kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
710         kvm_destroy_vm_debugfs(kvm);
711         kvm_arch_sync_events(kvm);
712         spin_lock(&kvm_lock);
713         list_del(&kvm->vm_list);
714         spin_unlock(&kvm_lock);
715         kvm_free_irq_routing(kvm);
716         for (i = 0; i < KVM_NR_BUSES; i++) {
717                 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
718 
719                 if (bus)
720                         kvm_io_bus_destroy(bus);
721                 kvm->buses[i] = NULL;
722         }
723         kvm_coalesced_mmio_free(kvm);
724 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
725         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
726 #else
727         kvm_arch_flush_shadow_all(kvm);
728 #endif
729         kvm_arch_destroy_vm(kvm);
730         kvm_destroy_devices(kvm);
731         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
732                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
733         cleanup_srcu_struct(&kvm->irq_srcu);
734         cleanup_srcu_struct(&kvm->srcu);
735         kvm_arch_free_vm(kvm);
736         preempt_notifier_dec();
737         hardware_disable_all();
738         mmdrop(mm);
739 }
740 
741 void kvm_get_kvm(struct kvm *kvm)
742 {
743         refcount_inc(&kvm->users_count);
744 }
745 EXPORT_SYMBOL_GPL(kvm_get_kvm);
746 
747 void kvm_put_kvm(struct kvm *kvm)
748 {
749         if (refcount_dec_and_test(&kvm->users_count))
750                 kvm_destroy_vm(kvm);
751 }
752 EXPORT_SYMBOL_GPL(kvm_put_kvm);
753 
754 
755 static int kvm_vm_release(struct inode *inode, struct file *filp)
756 {
757         struct kvm *kvm = filp->private_data;
758 
759         kvm_irqfd_release(kvm);
760 
761         kvm_put_kvm(kvm);
762         return 0;
763 }
764 
765 /*
766  * Allocation size is twice as large as the actual dirty bitmap size.
767  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
768  */
769 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
770 {
771         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
772 
773         memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
774         if (!memslot->dirty_bitmap)
775                 return -ENOMEM;
776 
777         return 0;
778 }
779 
780 /*
781  * Insert memslot and re-sort memslots based on their GFN,
782  * so binary search could be used to lookup GFN.
783  * Sorting algorithm takes advantage of having initially
784  * sorted array and known changed memslot position.
785  */
786 static void update_memslots(struct kvm_memslots *slots,
787                             struct kvm_memory_slot *new)
788 {
789         int id = new->id;
790         int i = slots->id_to_index[id];
791         struct kvm_memory_slot *mslots = slots->memslots;
792 
793         WARN_ON(mslots[i].id != id);
794         if (!new->npages) {
795                 WARN_ON(!mslots[i].npages);
796                 if (mslots[i].npages)
797                         slots->used_slots--;
798         } else {
799                 if (!mslots[i].npages)
800                         slots->used_slots++;
801         }
802 
803         while (i < KVM_MEM_SLOTS_NUM - 1 &&
804                new->base_gfn <= mslots[i + 1].base_gfn) {
805                 if (!mslots[i + 1].npages)
806                         break;
807                 mslots[i] = mslots[i + 1];
808                 slots->id_to_index[mslots[i].id] = i;
809                 i++;
810         }
811 
812         /*
813          * The ">=" is needed when creating a slot with base_gfn == 0,
814          * so that it moves before all those with base_gfn == npages == 0.
815          *
816          * On the other hand, if new->npages is zero, the above loop has
817          * already left i pointing to the beginning of the empty part of
818          * mslots, and the ">=" would move the hole backwards in this
819          * case---which is wrong.  So skip the loop when deleting a slot.
820          */
821         if (new->npages) {
822                 while (i > 0 &&
823                        new->base_gfn >= mslots[i - 1].base_gfn) {
824                         mslots[i] = mslots[i - 1];
825                         slots->id_to_index[mslots[i].id] = i;
826                         i--;
827                 }
828         } else
829                 WARN_ON_ONCE(i != slots->used_slots);
830 
831         mslots[i] = *new;
832         slots->id_to_index[mslots[i].id] = i;
833 }
834 
835 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
836 {
837         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
838 
839 #ifdef __KVM_HAVE_READONLY_MEM
840         valid_flags |= KVM_MEM_READONLY;
841 #endif
842 
843         if (mem->flags & ~valid_flags)
844                 return -EINVAL;
845 
846         return 0;
847 }
848 
849 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
850                 int as_id, struct kvm_memslots *slots)
851 {
852         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
853 
854         /*
855          * Set the low bit in the generation, which disables SPTE caching
856          * until the end of synchronize_srcu_expedited.
857          */
858         WARN_ON(old_memslots->generation & 1);
859         slots->generation = old_memslots->generation + 1;
860 
861         rcu_assign_pointer(kvm->memslots[as_id], slots);
862         synchronize_srcu_expedited(&kvm->srcu);
863 
864         /*
865          * Increment the new memslot generation a second time. This prevents
866          * vm exits that race with memslot updates from caching a memslot
867          * generation that will (potentially) be valid forever.
868          *
869          * Generations must be unique even across address spaces.  We do not need
870          * a global counter for that, instead the generation space is evenly split
871          * across address spaces.  For example, with two address spaces, address
872          * space 0 will use generations 0, 4, 8, ... while * address space 1 will
873          * use generations 2, 6, 10, 14, ...
874          */
875         slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
876 
877         kvm_arch_memslots_updated(kvm, slots);
878 
879         return old_memslots;
880 }
881 
882 /*
883  * Allocate some memory and give it an address in the guest physical address
884  * space.
885  *
886  * Discontiguous memory is allowed, mostly for framebuffers.
887  *
888  * Must be called holding kvm->slots_lock for write.
889  */
890 int __kvm_set_memory_region(struct kvm *kvm,
891                             const struct kvm_userspace_memory_region *mem)
892 {
893         int r;
894         gfn_t base_gfn;
895         unsigned long npages;
896         struct kvm_memory_slot *slot;
897         struct kvm_memory_slot old, new;
898         struct kvm_memslots *slots = NULL, *old_memslots;
899         int as_id, id;
900         enum kvm_mr_change change;
901 
902         r = check_memory_region_flags(mem);
903         if (r)
904                 goto out;
905 
906         r = -EINVAL;
907         as_id = mem->slot >> 16;
908         id = (u16)mem->slot;
909 
910         /* General sanity checks */
911         if (mem->memory_size & (PAGE_SIZE - 1))
912                 goto out;
913         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
914                 goto out;
915         /* We can read the guest memory with __xxx_user() later on. */
916         if ((id < KVM_USER_MEM_SLOTS) &&
917             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
918              !access_ok(VERIFY_WRITE,
919                         (void __user *)(unsigned long)mem->userspace_addr,
920                         mem->memory_size)))
921                 goto out;
922         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
923                 goto out;
924         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
925                 goto out;
926 
927         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
928         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
929         npages = mem->memory_size >> PAGE_SHIFT;
930 
931         if (npages > KVM_MEM_MAX_NR_PAGES)
932                 goto out;
933 
934         new = old = *slot;
935 
936         new.id = id;
937         new.base_gfn = base_gfn;
938         new.npages = npages;
939         new.flags = mem->flags;
940 
941         if (npages) {
942                 if (!old.npages)
943                         change = KVM_MR_CREATE;
944                 else { /* Modify an existing slot. */
945                         if ((mem->userspace_addr != old.userspace_addr) ||
946                             (npages != old.npages) ||
947                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
948                                 goto out;
949 
950                         if (base_gfn != old.base_gfn)
951                                 change = KVM_MR_MOVE;
952                         else if (new.flags != old.flags)
953                                 change = KVM_MR_FLAGS_ONLY;
954                         else { /* Nothing to change. */
955                                 r = 0;
956                                 goto out;
957                         }
958                 }
959         } else {
960                 if (!old.npages)
961                         goto out;
962 
963                 change = KVM_MR_DELETE;
964                 new.base_gfn = 0;
965                 new.flags = 0;
966         }
967 
968         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
969                 /* Check for overlaps */
970                 r = -EEXIST;
971                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
972                         if (slot->id == id)
973                                 continue;
974                         if (!((base_gfn + npages <= slot->base_gfn) ||
975                               (base_gfn >= slot->base_gfn + slot->npages)))
976                                 goto out;
977                 }
978         }
979 
980         /* Free page dirty bitmap if unneeded */
981         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
982                 new.dirty_bitmap = NULL;
983 
984         r = -ENOMEM;
985         if (change == KVM_MR_CREATE) {
986                 new.userspace_addr = mem->userspace_addr;
987 
988                 if (kvm_arch_create_memslot(kvm, &new, npages))
989                         goto out_free;
990         }
991 
992         /* Allocate page dirty bitmap if needed */
993         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
994                 if (kvm_create_dirty_bitmap(&new) < 0)
995                         goto out_free;
996         }
997 
998         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
999         if (!slots)
1000                 goto out_free;
1001         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1002 
1003         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1004                 slot = id_to_memslot(slots, id);
1005                 slot->flags |= KVM_MEMSLOT_INVALID;
1006 
1007                 old_memslots = install_new_memslots(kvm, as_id, slots);
1008 
1009                 /* From this point no new shadow pages pointing to a deleted,
1010                  * or moved, memslot will be created.
1011                  *
1012                  * validation of sp->gfn happens in:
1013                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1014                  *      - kvm_is_visible_gfn (mmu_check_roots)
1015                  */
1016                 kvm_arch_flush_shadow_memslot(kvm, slot);
1017 
1018                 /*
1019                  * We can re-use the old_memslots from above, the only difference
1020                  * from the currently installed memslots is the invalid flag.  This
1021                  * will get overwritten by update_memslots anyway.
1022                  */
1023                 slots = old_memslots;
1024         }
1025 
1026         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1027         if (r)
1028                 goto out_slots;
1029 
1030         /* actual memory is freed via old in kvm_free_memslot below */
1031         if (change == KVM_MR_DELETE) {
1032                 new.dirty_bitmap = NULL;
1033                 memset(&new.arch, 0, sizeof(new.arch));
1034         }
1035 
1036         update_memslots(slots, &new);
1037         old_memslots = install_new_memslots(kvm, as_id, slots);
1038 
1039         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1040 
1041         kvm_free_memslot(kvm, &old, &new);
1042         kvfree(old_memslots);
1043         return 0;
1044 
1045 out_slots:
1046         kvfree(slots);
1047 out_free:
1048         kvm_free_memslot(kvm, &new, &old);
1049 out:
1050         return r;
1051 }
1052 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1053 
1054 int kvm_set_memory_region(struct kvm *kvm,
1055                           const struct kvm_userspace_memory_region *mem)
1056 {
1057         int r;
1058 
1059         mutex_lock(&kvm->slots_lock);
1060         r = __kvm_set_memory_region(kvm, mem);
1061         mutex_unlock(&kvm->slots_lock);
1062         return r;
1063 }
1064 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1065 
1066 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1067                                           struct kvm_userspace_memory_region *mem)
1068 {
1069         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1070                 return -EINVAL;
1071 
1072         return kvm_set_memory_region(kvm, mem);
1073 }
1074 
1075 int kvm_get_dirty_log(struct kvm *kvm,
1076                         struct kvm_dirty_log *log, int *is_dirty)
1077 {
1078         struct kvm_memslots *slots;
1079         struct kvm_memory_slot *memslot;
1080         int i, as_id, id;
1081         unsigned long n;
1082         unsigned long any = 0;
1083 
1084         as_id = log->slot >> 16;
1085         id = (u16)log->slot;
1086         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1087                 return -EINVAL;
1088 
1089         slots = __kvm_memslots(kvm, as_id);
1090         memslot = id_to_memslot(slots, id);
1091         if (!memslot->dirty_bitmap)
1092                 return -ENOENT;
1093 
1094         n = kvm_dirty_bitmap_bytes(memslot);
1095 
1096         for (i = 0; !any && i < n/sizeof(long); ++i)
1097                 any = memslot->dirty_bitmap[i];
1098 
1099         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1100                 return -EFAULT;
1101 
1102         if (any)
1103                 *is_dirty = 1;
1104         return 0;
1105 }
1106 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1107 
1108 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1109 /**
1110  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1111  *      are dirty write protect them for next write.
1112  * @kvm:        pointer to kvm instance
1113  * @log:        slot id and address to which we copy the log
1114  * @is_dirty:   flag set if any page is dirty
1115  *
1116  * We need to keep it in mind that VCPU threads can write to the bitmap
1117  * concurrently. So, to avoid losing track of dirty pages we keep the
1118  * following order:
1119  *
1120  *    1. Take a snapshot of the bit and clear it if needed.
1121  *    2. Write protect the corresponding page.
1122  *    3. Copy the snapshot to the userspace.
1123  *    4. Upon return caller flushes TLB's if needed.
1124  *
1125  * Between 2 and 4, the guest may write to the page using the remaining TLB
1126  * entry.  This is not a problem because the page is reported dirty using
1127  * the snapshot taken before and step 4 ensures that writes done after
1128  * exiting to userspace will be logged for the next call.
1129  *
1130  */
1131 int kvm_get_dirty_log_protect(struct kvm *kvm,
1132                         struct kvm_dirty_log *log, bool *is_dirty)
1133 {
1134         struct kvm_memslots *slots;
1135         struct kvm_memory_slot *memslot;
1136         int i, as_id, id;
1137         unsigned long n;
1138         unsigned long *dirty_bitmap;
1139         unsigned long *dirty_bitmap_buffer;
1140 
1141         as_id = log->slot >> 16;
1142         id = (u16)log->slot;
1143         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1144                 return -EINVAL;
1145 
1146         slots = __kvm_memslots(kvm, as_id);
1147         memslot = id_to_memslot(slots, id);
1148 
1149         dirty_bitmap = memslot->dirty_bitmap;
1150         if (!dirty_bitmap)
1151                 return -ENOENT;
1152 
1153         n = kvm_dirty_bitmap_bytes(memslot);
1154 
1155         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1156         memset(dirty_bitmap_buffer, 0, n);
1157 
1158         spin_lock(&kvm->mmu_lock);
1159         *is_dirty = false;
1160         for (i = 0; i < n / sizeof(long); i++) {
1161                 unsigned long mask;
1162                 gfn_t offset;
1163 
1164                 if (!dirty_bitmap[i])
1165                         continue;
1166 
1167                 *is_dirty = true;
1168 
1169                 mask = xchg(&dirty_bitmap[i], 0);
1170                 dirty_bitmap_buffer[i] = mask;
1171 
1172                 if (mask) {
1173                         offset = i * BITS_PER_LONG;
1174                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1175                                                                 offset, mask);
1176                 }
1177         }
1178 
1179         spin_unlock(&kvm->mmu_lock);
1180         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1181                 return -EFAULT;
1182         return 0;
1183 }
1184 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1185 #endif
1186 
1187 bool kvm_largepages_enabled(void)
1188 {
1189         return largepages_enabled;
1190 }
1191 
1192 void kvm_disable_largepages(void)
1193 {
1194         largepages_enabled = false;
1195 }
1196 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1197 
1198 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1199 {
1200         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1201 }
1202 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1203 
1204 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1205 {
1206         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1207 }
1208 
1209 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1210 {
1211         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1212 
1213         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1214               memslot->flags & KVM_MEMSLOT_INVALID)
1215                 return false;
1216 
1217         return true;
1218 }
1219 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1220 
1221 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1222 {
1223         struct vm_area_struct *vma;
1224         unsigned long addr, size;
1225 
1226         size = PAGE_SIZE;
1227 
1228         addr = gfn_to_hva(kvm, gfn);
1229         if (kvm_is_error_hva(addr))
1230                 return PAGE_SIZE;
1231 
1232         down_read(&current->mm->mmap_sem);
1233         vma = find_vma(current->mm, addr);
1234         if (!vma)
1235                 goto out;
1236 
1237         size = vma_kernel_pagesize(vma);
1238 
1239 out:
1240         up_read(&current->mm->mmap_sem);
1241 
1242         return size;
1243 }
1244 
1245 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1246 {
1247         return slot->flags & KVM_MEM_READONLY;
1248 }
1249 
1250 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1251                                        gfn_t *nr_pages, bool write)
1252 {
1253         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1254                 return KVM_HVA_ERR_BAD;
1255 
1256         if (memslot_is_readonly(slot) && write)
1257                 return KVM_HVA_ERR_RO_BAD;
1258 
1259         if (nr_pages)
1260                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1261 
1262         return __gfn_to_hva_memslot(slot, gfn);
1263 }
1264 
1265 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1266                                      gfn_t *nr_pages)
1267 {
1268         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1269 }
1270 
1271 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1272                                         gfn_t gfn)
1273 {
1274         return gfn_to_hva_many(slot, gfn, NULL);
1275 }
1276 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1277 
1278 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1279 {
1280         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1281 }
1282 EXPORT_SYMBOL_GPL(gfn_to_hva);
1283 
1284 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1285 {
1286         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1287 }
1288 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1289 
1290 /*
1291  * If writable is set to false, the hva returned by this function is only
1292  * allowed to be read.
1293  */
1294 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1295                                       gfn_t gfn, bool *writable)
1296 {
1297         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1298 
1299         if (!kvm_is_error_hva(hva) && writable)
1300                 *writable = !memslot_is_readonly(slot);
1301 
1302         return hva;
1303 }
1304 
1305 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1306 {
1307         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1308 
1309         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1310 }
1311 
1312 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1313 {
1314         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1315 
1316         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1317 }
1318 
1319 static inline int check_user_page_hwpoison(unsigned long addr)
1320 {
1321         int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1322 
1323         rc = get_user_pages(addr, 1, flags, NULL, NULL);
1324         return rc == -EHWPOISON;
1325 }
1326 
1327 /*
1328  * The atomic path to get the writable pfn which will be stored in @pfn,
1329  * true indicates success, otherwise false is returned.
1330  */
1331 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1332                             bool write_fault, bool *writable, kvm_pfn_t *pfn)
1333 {
1334         struct page *page[1];
1335         int npages;
1336 
1337         if (!(async || atomic))
1338                 return false;
1339 
1340         /*
1341          * Fast pin a writable pfn only if it is a write fault request
1342          * or the caller allows to map a writable pfn for a read fault
1343          * request.
1344          */
1345         if (!(write_fault || writable))
1346                 return false;
1347 
1348         npages = __get_user_pages_fast(addr, 1, 1, page);
1349         if (npages == 1) {
1350                 *pfn = page_to_pfn(page[0]);
1351 
1352                 if (writable)
1353                         *writable = true;
1354                 return true;
1355         }
1356 
1357         return false;
1358 }
1359 
1360 /*
1361  * The slow path to get the pfn of the specified host virtual address,
1362  * 1 indicates success, -errno is returned if error is detected.
1363  */
1364 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1365                            bool *writable, kvm_pfn_t *pfn)
1366 {
1367         unsigned int flags = FOLL_HWPOISON;
1368         struct page *page;
1369         int npages = 0;
1370 
1371         might_sleep();
1372 
1373         if (writable)
1374                 *writable = write_fault;
1375 
1376         if (write_fault)
1377                 flags |= FOLL_WRITE;
1378         if (async)
1379                 flags |= FOLL_NOWAIT;
1380 
1381         npages = get_user_pages_unlocked(addr, 1, &page, flags);
1382         if (npages != 1)
1383                 return npages;
1384 
1385         /* map read fault as writable if possible */
1386         if (unlikely(!write_fault) && writable) {
1387                 struct page *wpage;
1388 
1389                 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1390                         *writable = true;
1391                         put_page(page);
1392                         page = wpage;
1393                 }
1394         }
1395         *pfn = page_to_pfn(page);
1396         return npages;
1397 }
1398 
1399 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1400 {
1401         if (unlikely(!(vma->vm_flags & VM_READ)))
1402                 return false;
1403 
1404         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1405                 return false;
1406 
1407         return true;
1408 }
1409 
1410 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1411                                unsigned long addr, bool *async,
1412                                bool write_fault, bool *writable,
1413                                kvm_pfn_t *p_pfn)
1414 {
1415         unsigned long pfn;
1416         int r;
1417 
1418         r = follow_pfn(vma, addr, &pfn);
1419         if (r) {
1420                 /*
1421                  * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1422                  * not call the fault handler, so do it here.
1423                  */
1424                 bool unlocked = false;
1425                 r = fixup_user_fault(current, current->mm, addr,
1426                                      (write_fault ? FAULT_FLAG_WRITE : 0),
1427                                      &unlocked);
1428                 if (unlocked)
1429                         return -EAGAIN;
1430                 if (r)
1431                         return r;
1432 
1433                 r = follow_pfn(vma, addr, &pfn);
1434                 if (r)
1435                         return r;
1436 
1437         }
1438 
1439         if (writable)
1440                 *writable = true;
1441 
1442         /*
1443          * Get a reference here because callers of *hva_to_pfn* and
1444          * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1445          * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1446          * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1447          * simply do nothing for reserved pfns.
1448          *
1449          * Whoever called remap_pfn_range is also going to call e.g.
1450          * unmap_mapping_range before the underlying pages are freed,
1451          * causing a call to our MMU notifier.
1452          */ 
1453         kvm_get_pfn(pfn);
1454 
1455         *p_pfn = pfn;
1456         return 0;
1457 }
1458 
1459 /*
1460  * Pin guest page in memory and return its pfn.
1461  * @addr: host virtual address which maps memory to the guest
1462  * @atomic: whether this function can sleep
1463  * @async: whether this function need to wait IO complete if the
1464  *         host page is not in the memory
1465  * @write_fault: whether we should get a writable host page
1466  * @writable: whether it allows to map a writable host page for !@write_fault
1467  *
1468  * The function will map a writable host page for these two cases:
1469  * 1): @write_fault = true
1470  * 2): @write_fault = false && @writable, @writable will tell the caller
1471  *     whether the mapping is writable.
1472  */
1473 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1474                         bool write_fault, bool *writable)
1475 {
1476         struct vm_area_struct *vma;
1477         kvm_pfn_t pfn = 0;
1478         int npages, r;
1479 
1480         /* we can do it either atomically or asynchronously, not both */
1481         BUG_ON(atomic && async);
1482 
1483         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1484                 return pfn;
1485 
1486         if (atomic)
1487                 return KVM_PFN_ERR_FAULT;
1488 
1489         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1490         if (npages == 1)
1491                 return pfn;
1492 
1493         down_read(&current->mm->mmap_sem);
1494         if (npages == -EHWPOISON ||
1495               (!async && check_user_page_hwpoison(addr))) {
1496                 pfn = KVM_PFN_ERR_HWPOISON;
1497                 goto exit;
1498         }
1499 
1500 retry:
1501         vma = find_vma_intersection(current->mm, addr, addr + 1);
1502 
1503         if (vma == NULL)
1504                 pfn = KVM_PFN_ERR_FAULT;
1505         else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1506                 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1507                 if (r == -EAGAIN)
1508                         goto retry;
1509                 if (r < 0)
1510                         pfn = KVM_PFN_ERR_FAULT;
1511         } else {
1512                 if (async && vma_is_valid(vma, write_fault))
1513                         *async = true;
1514                 pfn = KVM_PFN_ERR_FAULT;
1515         }
1516 exit:
1517         up_read(&current->mm->mmap_sem);
1518         return pfn;
1519 }
1520 
1521 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1522                                bool atomic, bool *async, bool write_fault,
1523                                bool *writable)
1524 {
1525         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1526 
1527         if (addr == KVM_HVA_ERR_RO_BAD) {
1528                 if (writable)
1529                         *writable = false;
1530                 return KVM_PFN_ERR_RO_FAULT;
1531         }
1532 
1533         if (kvm_is_error_hva(addr)) {
1534                 if (writable)
1535                         *writable = false;
1536                 return KVM_PFN_NOSLOT;
1537         }
1538 
1539         /* Do not map writable pfn in the readonly memslot. */
1540         if (writable && memslot_is_readonly(slot)) {
1541                 *writable = false;
1542                 writable = NULL;
1543         }
1544 
1545         return hva_to_pfn(addr, atomic, async, write_fault,
1546                           writable);
1547 }
1548 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1549 
1550 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1551                       bool *writable)
1552 {
1553         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1554                                     write_fault, writable);
1555 }
1556 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1557 
1558 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1559 {
1560         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1561 }
1562 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1563 
1564 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1565 {
1566         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1567 }
1568 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1569 
1570 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1571 {
1572         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1573 }
1574 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1575 
1576 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1577 {
1578         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1579 }
1580 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1581 
1582 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1583 {
1584         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1585 }
1586 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1587 
1588 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1589 {
1590         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1591 }
1592 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1593 
1594 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1595                             struct page **pages, int nr_pages)
1596 {
1597         unsigned long addr;
1598         gfn_t entry = 0;
1599 
1600         addr = gfn_to_hva_many(slot, gfn, &entry);
1601         if (kvm_is_error_hva(addr))
1602                 return -1;
1603 
1604         if (entry < nr_pages)
1605                 return 0;
1606 
1607         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1608 }
1609 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1610 
1611 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1612 {
1613         if (is_error_noslot_pfn(pfn))
1614                 return KVM_ERR_PTR_BAD_PAGE;
1615 
1616         if (kvm_is_reserved_pfn(pfn)) {
1617                 WARN_ON(1);
1618                 return KVM_ERR_PTR_BAD_PAGE;
1619         }
1620 
1621         return pfn_to_page(pfn);
1622 }
1623 
1624 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1625 {
1626         kvm_pfn_t pfn;
1627 
1628         pfn = gfn_to_pfn(kvm, gfn);
1629 
1630         return kvm_pfn_to_page(pfn);
1631 }
1632 EXPORT_SYMBOL_GPL(gfn_to_page);
1633 
1634 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1635 {
1636         kvm_pfn_t pfn;
1637 
1638         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1639 
1640         return kvm_pfn_to_page(pfn);
1641 }
1642 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1643 
1644 void kvm_release_page_clean(struct page *page)
1645 {
1646         WARN_ON(is_error_page(page));
1647 
1648         kvm_release_pfn_clean(page_to_pfn(page));
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1651 
1652 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1653 {
1654         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1655                 put_page(pfn_to_page(pfn));
1656 }
1657 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1658 
1659 void kvm_release_page_dirty(struct page *page)
1660 {
1661         WARN_ON(is_error_page(page));
1662 
1663         kvm_release_pfn_dirty(page_to_pfn(page));
1664 }
1665 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1666 
1667 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1668 {
1669         kvm_set_pfn_dirty(pfn);
1670         kvm_release_pfn_clean(pfn);
1671 }
1672 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1673 
1674 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1675 {
1676         if (!kvm_is_reserved_pfn(pfn)) {
1677                 struct page *page = pfn_to_page(pfn);
1678 
1679                 if (!PageReserved(page))
1680                         SetPageDirty(page);
1681         }
1682 }
1683 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1684 
1685 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1686 {
1687         if (!kvm_is_reserved_pfn(pfn))
1688                 mark_page_accessed(pfn_to_page(pfn));
1689 }
1690 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1691 
1692 void kvm_get_pfn(kvm_pfn_t pfn)
1693 {
1694         if (!kvm_is_reserved_pfn(pfn))
1695                 get_page(pfn_to_page(pfn));
1696 }
1697 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1698 
1699 static int next_segment(unsigned long len, int offset)
1700 {
1701         if (len > PAGE_SIZE - offset)
1702                 return PAGE_SIZE - offset;
1703         else
1704                 return len;
1705 }
1706 
1707 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1708                                  void *data, int offset, int len)
1709 {
1710         int r;
1711         unsigned long addr;
1712 
1713         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1714         if (kvm_is_error_hva(addr))
1715                 return -EFAULT;
1716         r = __copy_from_user(data, (void __user *)addr + offset, len);
1717         if (r)
1718                 return -EFAULT;
1719         return 0;
1720 }
1721 
1722 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1723                         int len)
1724 {
1725         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1726 
1727         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1728 }
1729 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1730 
1731 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1732                              int offset, int len)
1733 {
1734         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1735 
1736         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1737 }
1738 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1739 
1740 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1741 {
1742         gfn_t gfn = gpa >> PAGE_SHIFT;
1743         int seg;
1744         int offset = offset_in_page(gpa);
1745         int ret;
1746 
1747         while ((seg = next_segment(len, offset)) != 0) {
1748                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1749                 if (ret < 0)
1750                         return ret;
1751                 offset = 0;
1752                 len -= seg;
1753                 data += seg;
1754                 ++gfn;
1755         }
1756         return 0;
1757 }
1758 EXPORT_SYMBOL_GPL(kvm_read_guest);
1759 
1760 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1761 {
1762         gfn_t gfn = gpa >> PAGE_SHIFT;
1763         int seg;
1764         int offset = offset_in_page(gpa);
1765         int ret;
1766 
1767         while ((seg = next_segment(len, offset)) != 0) {
1768                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1769                 if (ret < 0)
1770                         return ret;
1771                 offset = 0;
1772                 len -= seg;
1773                 data += seg;
1774                 ++gfn;
1775         }
1776         return 0;
1777 }
1778 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1779 
1780 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1781                                    void *data, int offset, unsigned long len)
1782 {
1783         int r;
1784         unsigned long addr;
1785 
1786         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1787         if (kvm_is_error_hva(addr))
1788                 return -EFAULT;
1789         pagefault_disable();
1790         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1791         pagefault_enable();
1792         if (r)
1793                 return -EFAULT;
1794         return 0;
1795 }
1796 
1797 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1798                           unsigned long len)
1799 {
1800         gfn_t gfn = gpa >> PAGE_SHIFT;
1801         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1802         int offset = offset_in_page(gpa);
1803 
1804         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1805 }
1806 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1807 
1808 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1809                                void *data, unsigned long len)
1810 {
1811         gfn_t gfn = gpa >> PAGE_SHIFT;
1812         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1813         int offset = offset_in_page(gpa);
1814 
1815         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1816 }
1817 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1818 
1819 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1820                                   const void *data, int offset, int len)
1821 {
1822         int r;
1823         unsigned long addr;
1824 
1825         addr = gfn_to_hva_memslot(memslot, gfn);
1826         if (kvm_is_error_hva(addr))
1827                 return -EFAULT;
1828         r = __copy_to_user((void __user *)addr + offset, data, len);
1829         if (r)
1830                 return -EFAULT;
1831         mark_page_dirty_in_slot(memslot, gfn);
1832         return 0;
1833 }
1834 
1835 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1836                          const void *data, int offset, int len)
1837 {
1838         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1839 
1840         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1841 }
1842 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1843 
1844 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1845                               const void *data, int offset, int len)
1846 {
1847         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1848 
1849         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1850 }
1851 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1852 
1853 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1854                     unsigned long len)
1855 {
1856         gfn_t gfn = gpa >> PAGE_SHIFT;
1857         int seg;
1858         int offset = offset_in_page(gpa);
1859         int ret;
1860 
1861         while ((seg = next_segment(len, offset)) != 0) {
1862                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1863                 if (ret < 0)
1864                         return ret;
1865                 offset = 0;
1866                 len -= seg;
1867                 data += seg;
1868                 ++gfn;
1869         }
1870         return 0;
1871 }
1872 EXPORT_SYMBOL_GPL(kvm_write_guest);
1873 
1874 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1875                          unsigned long len)
1876 {
1877         gfn_t gfn = gpa >> PAGE_SHIFT;
1878         int seg;
1879         int offset = offset_in_page(gpa);
1880         int ret;
1881 
1882         while ((seg = next_segment(len, offset)) != 0) {
1883                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1884                 if (ret < 0)
1885                         return ret;
1886                 offset = 0;
1887                 len -= seg;
1888                 data += seg;
1889                 ++gfn;
1890         }
1891         return 0;
1892 }
1893 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1894 
1895 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1896                                        struct gfn_to_hva_cache *ghc,
1897                                        gpa_t gpa, unsigned long len)
1898 {
1899         int offset = offset_in_page(gpa);
1900         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1901         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1902         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1903         gfn_t nr_pages_avail;
1904 
1905         ghc->gpa = gpa;
1906         ghc->generation = slots->generation;
1907         ghc->len = len;
1908         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1909         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1910         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1911                 ghc->hva += offset;
1912         } else {
1913                 /*
1914                  * If the requested region crosses two memslots, we still
1915                  * verify that the entire region is valid here.
1916                  */
1917                 while (start_gfn <= end_gfn) {
1918                         nr_pages_avail = 0;
1919                         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1920                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1921                                                    &nr_pages_avail);
1922                         if (kvm_is_error_hva(ghc->hva))
1923                                 return -EFAULT;
1924                         start_gfn += nr_pages_avail;
1925                 }
1926                 /* Use the slow path for cross page reads and writes. */
1927                 ghc->memslot = NULL;
1928         }
1929         return 0;
1930 }
1931 
1932 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1933                               gpa_t gpa, unsigned long len)
1934 {
1935         struct kvm_memslots *slots = kvm_memslots(kvm);
1936         return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1937 }
1938 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1939 
1940 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1941                            void *data, int offset, unsigned long len)
1942 {
1943         struct kvm_memslots *slots = kvm_memslots(kvm);
1944         int r;
1945         gpa_t gpa = ghc->gpa + offset;
1946 
1947         BUG_ON(len + offset > ghc->len);
1948 
1949         if (slots->generation != ghc->generation)
1950                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1951 
1952         if (unlikely(!ghc->memslot))
1953                 return kvm_write_guest(kvm, gpa, data, len);
1954 
1955         if (kvm_is_error_hva(ghc->hva))
1956                 return -EFAULT;
1957 
1958         r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1959         if (r)
1960                 return -EFAULT;
1961         mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1962 
1963         return 0;
1964 }
1965 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1966 
1967 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1968                            void *data, unsigned long len)
1969 {
1970         return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1971 }
1972 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1973 
1974 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1975                            void *data, unsigned long len)
1976 {
1977         struct kvm_memslots *slots = kvm_memslots(kvm);
1978         int r;
1979 
1980         BUG_ON(len > ghc->len);
1981 
1982         if (slots->generation != ghc->generation)
1983                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1984 
1985         if (unlikely(!ghc->memslot))
1986                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1987 
1988         if (kvm_is_error_hva(ghc->hva))
1989                 return -EFAULT;
1990 
1991         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1992         if (r)
1993                 return -EFAULT;
1994 
1995         return 0;
1996 }
1997 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1998 
1999 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2000 {
2001         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2002 
2003         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2004 }
2005 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2006 
2007 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2008 {
2009         gfn_t gfn = gpa >> PAGE_SHIFT;
2010         int seg;
2011         int offset = offset_in_page(gpa);
2012         int ret;
2013 
2014         while ((seg = next_segment(len, offset)) != 0) {
2015                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2016                 if (ret < 0)
2017                         return ret;
2018                 offset = 0;
2019                 len -= seg;
2020                 ++gfn;
2021         }
2022         return 0;
2023 }
2024 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2025 
2026 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2027                                     gfn_t gfn)
2028 {
2029         if (memslot && memslot->dirty_bitmap) {
2030                 unsigned long rel_gfn = gfn - memslot->base_gfn;
2031 
2032                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2033         }
2034 }
2035 
2036 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2037 {
2038         struct kvm_memory_slot *memslot;
2039 
2040         memslot = gfn_to_memslot(kvm, gfn);
2041         mark_page_dirty_in_slot(memslot, gfn);
2042 }
2043 EXPORT_SYMBOL_GPL(mark_page_dirty);
2044 
2045 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2046 {
2047         struct kvm_memory_slot *memslot;
2048 
2049         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2050         mark_page_dirty_in_slot(memslot, gfn);
2051 }
2052 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2053 
2054 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2055 {
2056         if (!vcpu->sigset_active)
2057                 return;
2058 
2059         /*
2060          * This does a lockless modification of ->real_blocked, which is fine
2061          * because, only current can change ->real_blocked and all readers of
2062          * ->real_blocked don't care as long ->real_blocked is always a subset
2063          * of ->blocked.
2064          */
2065         sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
2066 }
2067 
2068 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2069 {
2070         if (!vcpu->sigset_active)
2071                 return;
2072 
2073         sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
2074         sigemptyset(&current->real_blocked);
2075 }
2076 
2077 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2078 {
2079         unsigned int old, val, grow;
2080 
2081         old = val = vcpu->halt_poll_ns;
2082         grow = READ_ONCE(halt_poll_ns_grow);
2083         /* 10us base */
2084         if (val == 0 && grow)
2085                 val = 10000;
2086         else
2087                 val *= grow;
2088 
2089         if (val > halt_poll_ns)
2090                 val = halt_poll_ns;
2091 
2092         vcpu->halt_poll_ns = val;
2093         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2094 }
2095 
2096 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2097 {
2098         unsigned int old, val, shrink;
2099 
2100         old = val = vcpu->halt_poll_ns;
2101         shrink = READ_ONCE(halt_poll_ns_shrink);
2102         if (shrink == 0)
2103                 val = 0;
2104         else
2105                 val /= shrink;
2106 
2107         vcpu->halt_poll_ns = val;
2108         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2109 }
2110 
2111 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2112 {
2113         if (kvm_arch_vcpu_runnable(vcpu)) {
2114                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2115                 return -EINTR;
2116         }
2117         if (kvm_cpu_has_pending_timer(vcpu))
2118                 return -EINTR;
2119         if (signal_pending(current))
2120                 return -EINTR;
2121 
2122         return 0;
2123 }
2124 
2125 /*
2126  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2127  */
2128 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2129 {
2130         ktime_t start, cur;
2131         DECLARE_SWAITQUEUE(wait);
2132         bool waited = false;
2133         u64 block_ns;
2134 
2135         start = cur = ktime_get();
2136         if (vcpu->halt_poll_ns) {
2137                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2138 
2139                 ++vcpu->stat.halt_attempted_poll;
2140                 do {
2141                         /*
2142                          * This sets KVM_REQ_UNHALT if an interrupt
2143                          * arrives.
2144                          */
2145                         if (kvm_vcpu_check_block(vcpu) < 0) {
2146                                 ++vcpu->stat.halt_successful_poll;
2147                                 if (!vcpu_valid_wakeup(vcpu))
2148                                         ++vcpu->stat.halt_poll_invalid;
2149                                 goto out;
2150                         }
2151                         cur = ktime_get();
2152                 } while (single_task_running() && ktime_before(cur, stop));
2153         }
2154 
2155         kvm_arch_vcpu_blocking(vcpu);
2156 
2157         for (;;) {
2158                 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2159 
2160                 if (kvm_vcpu_check_block(vcpu) < 0)
2161                         break;
2162 
2163                 waited = true;
2164                 schedule();
2165         }
2166 
2167         finish_swait(&vcpu->wq, &wait);
2168         cur = ktime_get();
2169 
2170         kvm_arch_vcpu_unblocking(vcpu);
2171 out:
2172         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2173 
2174         if (!vcpu_valid_wakeup(vcpu))
2175                 shrink_halt_poll_ns(vcpu);
2176         else if (halt_poll_ns) {
2177                 if (block_ns <= vcpu->halt_poll_ns)
2178                         ;
2179                 /* we had a long block, shrink polling */
2180                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2181                         shrink_halt_poll_ns(vcpu);
2182                 /* we had a short halt and our poll time is too small */
2183                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2184                         block_ns < halt_poll_ns)
2185                         grow_halt_poll_ns(vcpu);
2186         } else
2187                 vcpu->halt_poll_ns = 0;
2188 
2189         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2190         kvm_arch_vcpu_block_finish(vcpu);
2191 }
2192 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2193 
2194 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2195 {
2196         struct swait_queue_head *wqp;
2197 
2198         wqp = kvm_arch_vcpu_wq(vcpu);
2199         if (swq_has_sleeper(wqp)) {
2200                 swake_up(wqp);
2201                 ++vcpu->stat.halt_wakeup;
2202                 return true;
2203         }
2204 
2205         return false;
2206 }
2207 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2208 
2209 #ifndef CONFIG_S390
2210 /*
2211  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2212  */
2213 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2214 {
2215         int me;
2216         int cpu = vcpu->cpu;
2217 
2218         if (kvm_vcpu_wake_up(vcpu))
2219                 return;
2220 
2221         me = get_cpu();
2222         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2223                 if (kvm_arch_vcpu_should_kick(vcpu))
2224                         smp_send_reschedule(cpu);
2225         put_cpu();
2226 }
2227 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2228 #endif /* !CONFIG_S390 */
2229 
2230 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2231 {
2232         struct pid *pid;
2233         struct task_struct *task = NULL;
2234         int ret = 0;
2235 
2236         rcu_read_lock();
2237         pid = rcu_dereference(target->pid);
2238         if (pid)
2239                 task = get_pid_task(pid, PIDTYPE_PID);
2240         rcu_read_unlock();
2241         if (!task)
2242                 return ret;
2243         ret = yield_to(task, 1);
2244         put_task_struct(task);
2245 
2246         return ret;
2247 }
2248 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2249 
2250 /*
2251  * Helper that checks whether a VCPU is eligible for directed yield.
2252  * Most eligible candidate to yield is decided by following heuristics:
2253  *
2254  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2255  *  (preempted lock holder), indicated by @in_spin_loop.
2256  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2257  *
2258  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2259  *  chance last time (mostly it has become eligible now since we have probably
2260  *  yielded to lockholder in last iteration. This is done by toggling
2261  *  @dy_eligible each time a VCPU checked for eligibility.)
2262  *
2263  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2264  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2265  *  burning. Giving priority for a potential lock-holder increases lock
2266  *  progress.
2267  *
2268  *  Since algorithm is based on heuristics, accessing another VCPU data without
2269  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2270  *  and continue with next VCPU and so on.
2271  */
2272 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2273 {
2274 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2275         bool eligible;
2276 
2277         eligible = !vcpu->spin_loop.in_spin_loop ||
2278                     vcpu->spin_loop.dy_eligible;
2279 
2280         if (vcpu->spin_loop.in_spin_loop)
2281                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2282 
2283         return eligible;
2284 #else
2285         return true;
2286 #endif
2287 }
2288 
2289 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2290 {
2291         struct kvm *kvm = me->kvm;
2292         struct kvm_vcpu *vcpu;
2293         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2294         int yielded = 0;
2295         int try = 3;
2296         int pass;
2297         int i;
2298 
2299         kvm_vcpu_set_in_spin_loop(me, true);
2300         /*
2301          * We boost the priority of a VCPU that is runnable but not
2302          * currently running, because it got preempted by something
2303          * else and called schedule in __vcpu_run.  Hopefully that
2304          * VCPU is holding the lock that we need and will release it.
2305          * We approximate round-robin by starting at the last boosted VCPU.
2306          */
2307         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2308                 kvm_for_each_vcpu(i, vcpu, kvm) {
2309                         if (!pass && i <= last_boosted_vcpu) {
2310                                 i = last_boosted_vcpu;
2311                                 continue;
2312                         } else if (pass && i > last_boosted_vcpu)
2313                                 break;
2314                         if (!READ_ONCE(vcpu->preempted))
2315                                 continue;
2316                         if (vcpu == me)
2317                                 continue;
2318                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2319                                 continue;
2320                         if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2321                                 continue;
2322                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2323                                 continue;
2324 
2325                         yielded = kvm_vcpu_yield_to(vcpu);
2326                         if (yielded > 0) {
2327                                 kvm->last_boosted_vcpu = i;
2328                                 break;
2329                         } else if (yielded < 0) {
2330                                 try--;
2331                                 if (!try)
2332                                         break;
2333                         }
2334                 }
2335         }
2336         kvm_vcpu_set_in_spin_loop(me, false);
2337 
2338         /* Ensure vcpu is not eligible during next spinloop */
2339         kvm_vcpu_set_dy_eligible(me, false);
2340 }
2341 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2342 
2343 static int kvm_vcpu_fault(struct vm_fault *vmf)
2344 {
2345         struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2346         struct page *page;
2347 
2348         if (vmf->pgoff == 0)
2349                 page = virt_to_page(vcpu->run);
2350 #ifdef CONFIG_X86
2351         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2352                 page = virt_to_page(vcpu->arch.pio_data);
2353 #endif
2354 #ifdef CONFIG_KVM_MMIO
2355         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2356                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2357 #endif
2358         else
2359                 return kvm_arch_vcpu_fault(vcpu, vmf);
2360         get_page(page);
2361         vmf->page = page;
2362         return 0;
2363 }
2364 
2365 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2366         .fault = kvm_vcpu_fault,
2367 };
2368 
2369 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2370 {
2371         vma->vm_ops = &kvm_vcpu_vm_ops;
2372         return 0;
2373 }
2374 
2375 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2376 {
2377         struct kvm_vcpu *vcpu = filp->private_data;
2378 
2379         debugfs_remove_recursive(vcpu->debugfs_dentry);
2380         kvm_put_kvm(vcpu->kvm);
2381         return 0;
2382 }
2383 
2384 static struct file_operations kvm_vcpu_fops = {
2385         .release        = kvm_vcpu_release,
2386         .unlocked_ioctl = kvm_vcpu_ioctl,
2387 #ifdef CONFIG_KVM_COMPAT
2388         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2389 #endif
2390         .mmap           = kvm_vcpu_mmap,
2391         .llseek         = noop_llseek,
2392 };
2393 
2394 /*
2395  * Allocates an inode for the vcpu.
2396  */
2397 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2398 {
2399         char name[8 + 1 + ITOA_MAX_LEN + 1];
2400 
2401         snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2402         return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2403 }
2404 
2405 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2406 {
2407         char dir_name[ITOA_MAX_LEN * 2];
2408         int ret;
2409 
2410         if (!kvm_arch_has_vcpu_debugfs())
2411                 return 0;
2412 
2413         if (!debugfs_initialized())
2414                 return 0;
2415 
2416         snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2417         vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2418                                                                 vcpu->kvm->debugfs_dentry);
2419         if (!vcpu->debugfs_dentry)
2420                 return -ENOMEM;
2421 
2422         ret = kvm_arch_create_vcpu_debugfs(vcpu);
2423         if (ret < 0) {
2424                 debugfs_remove_recursive(vcpu->debugfs_dentry);
2425                 return ret;
2426         }
2427 
2428         return 0;
2429 }
2430 
2431 /*
2432  * Creates some virtual cpus.  Good luck creating more than one.
2433  */
2434 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2435 {
2436         int r;
2437         struct kvm_vcpu *vcpu;
2438 
2439         if (id >= KVM_MAX_VCPU_ID)
2440                 return -EINVAL;
2441 
2442         mutex_lock(&kvm->lock);
2443         if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2444                 mutex_unlock(&kvm->lock);
2445                 return -EINVAL;
2446         }
2447 
2448         kvm->created_vcpus++;
2449         mutex_unlock(&kvm->lock);
2450 
2451         vcpu = kvm_arch_vcpu_create(kvm, id);
2452         if (IS_ERR(vcpu)) {
2453                 r = PTR_ERR(vcpu);
2454                 goto vcpu_decrement;
2455         }
2456 
2457         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2458 
2459         r = kvm_arch_vcpu_setup(vcpu);
2460         if (r)
2461                 goto vcpu_destroy;
2462 
2463         r = kvm_create_vcpu_debugfs(vcpu);
2464         if (r)
2465                 goto vcpu_destroy;
2466 
2467         mutex_lock(&kvm->lock);
2468         if (kvm_get_vcpu_by_id(kvm, id)) {
2469                 r = -EEXIST;
2470                 goto unlock_vcpu_destroy;
2471         }
2472 
2473         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2474 
2475         /* Now it's all set up, let userspace reach it */
2476         kvm_get_kvm(kvm);
2477         r = create_vcpu_fd(vcpu);
2478         if (r < 0) {
2479                 kvm_put_kvm(kvm);
2480                 goto unlock_vcpu_destroy;
2481         }
2482 
2483         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2484 
2485         /*
2486          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2487          * before kvm->online_vcpu's incremented value.
2488          */
2489         smp_wmb();
2490         atomic_inc(&kvm->online_vcpus);
2491 
2492         mutex_unlock(&kvm->lock);
2493         kvm_arch_vcpu_postcreate(vcpu);
2494         return r;
2495 
2496 unlock_vcpu_destroy:
2497         mutex_unlock(&kvm->lock);
2498         debugfs_remove_recursive(vcpu->debugfs_dentry);
2499 vcpu_destroy:
2500         kvm_arch_vcpu_destroy(vcpu);
2501 vcpu_decrement:
2502         mutex_lock(&kvm->lock);
2503         kvm->created_vcpus--;
2504         mutex_unlock(&kvm->lock);
2505         return r;
2506 }
2507 
2508 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2509 {
2510         if (sigset) {
2511                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2512                 vcpu->sigset_active = 1;
2513                 vcpu->sigset = *sigset;
2514         } else
2515                 vcpu->sigset_active = 0;
2516         return 0;
2517 }
2518 
2519 static long kvm_vcpu_ioctl(struct file *filp,
2520                            unsigned int ioctl, unsigned long arg)
2521 {
2522         struct kvm_vcpu *vcpu = filp->private_data;
2523         void __user *argp = (void __user *)arg;
2524         int r;
2525         struct kvm_fpu *fpu = NULL;
2526         struct kvm_sregs *kvm_sregs = NULL;
2527 
2528         if (vcpu->kvm->mm != current->mm)
2529                 return -EIO;
2530 
2531         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2532                 return -EINVAL;
2533 
2534         /*
2535          * Some architectures have vcpu ioctls that are asynchronous to vcpu
2536          * execution; mutex_lock() would break them.
2537          */
2538         r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2539         if (r != -ENOIOCTLCMD)
2540                 return r;
2541 
2542         if (mutex_lock_killable(&vcpu->mutex))
2543                 return -EINTR;
2544         switch (ioctl) {
2545         case KVM_RUN: {
2546                 struct pid *oldpid;
2547                 r = -EINVAL;
2548                 if (arg)
2549                         goto out;
2550                 oldpid = rcu_access_pointer(vcpu->pid);
2551                 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2552                         /* The thread running this VCPU changed. */
2553                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2554 
2555                         rcu_assign_pointer(vcpu->pid, newpid);
2556                         if (oldpid)
2557                                 synchronize_rcu();
2558                         put_pid(oldpid);
2559                 }
2560                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2561                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2562                 break;
2563         }
2564         case KVM_GET_REGS: {
2565                 struct kvm_regs *kvm_regs;
2566 
2567                 r = -ENOMEM;
2568                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2569                 if (!kvm_regs)
2570                         goto out;
2571                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2572                 if (r)
2573                         goto out_free1;
2574                 r = -EFAULT;
2575                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2576                         goto out_free1;
2577                 r = 0;
2578 out_free1:
2579                 kfree(kvm_regs);
2580                 break;
2581         }
2582         case KVM_SET_REGS: {
2583                 struct kvm_regs *kvm_regs;
2584 
2585                 r = -ENOMEM;
2586                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2587                 if (IS_ERR(kvm_regs)) {
2588                         r = PTR_ERR(kvm_regs);
2589                         goto out;
2590                 }
2591                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2592                 kfree(kvm_regs);
2593                 break;
2594         }
2595         case KVM_GET_SREGS: {
2596                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2597                 r = -ENOMEM;
2598                 if (!kvm_sregs)
2599                         goto out;
2600                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2601                 if (r)
2602                         goto out;
2603                 r = -EFAULT;
2604                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2605                         goto out;
2606                 r = 0;
2607                 break;
2608         }
2609         case KVM_SET_SREGS: {
2610                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2611                 if (IS_ERR(kvm_sregs)) {
2612                         r = PTR_ERR(kvm_sregs);
2613                         kvm_sregs = NULL;
2614                         goto out;
2615                 }
2616                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2617                 break;
2618         }
2619         case KVM_GET_MP_STATE: {
2620                 struct kvm_mp_state mp_state;
2621 
2622                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2623                 if (r)
2624                         goto out;
2625                 r = -EFAULT;
2626                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2627                         goto out;
2628                 r = 0;
2629                 break;
2630         }
2631         case KVM_SET_MP_STATE: {
2632                 struct kvm_mp_state mp_state;
2633 
2634                 r = -EFAULT;
2635                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2636                         goto out;
2637                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2638                 break;
2639         }
2640         case KVM_TRANSLATE: {
2641                 struct kvm_translation tr;
2642 
2643                 r = -EFAULT;
2644                 if (copy_from_user(&tr, argp, sizeof(tr)))
2645                         goto out;
2646                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2647                 if (r)
2648                         goto out;
2649                 r = -EFAULT;
2650                 if (copy_to_user(argp, &tr, sizeof(tr)))
2651                         goto out;
2652                 r = 0;
2653                 break;
2654         }
2655         case KVM_SET_GUEST_DEBUG: {
2656                 struct kvm_guest_debug dbg;
2657 
2658                 r = -EFAULT;
2659                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2660                         goto out;
2661                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2662                 break;
2663         }
2664         case KVM_SET_SIGNAL_MASK: {
2665                 struct kvm_signal_mask __user *sigmask_arg = argp;
2666                 struct kvm_signal_mask kvm_sigmask;
2667                 sigset_t sigset, *p;
2668 
2669                 p = NULL;
2670                 if (argp) {
2671                         r = -EFAULT;
2672                         if (copy_from_user(&kvm_sigmask, argp,
2673                                            sizeof(kvm_sigmask)))
2674                                 goto out;
2675                         r = -EINVAL;
2676                         if (kvm_sigmask.len != sizeof(sigset))
2677                                 goto out;
2678                         r = -EFAULT;
2679                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2680                                            sizeof(sigset)))
2681                                 goto out;
2682                         p = &sigset;
2683                 }
2684                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2685                 break;
2686         }
2687         case KVM_GET_FPU: {
2688                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2689                 r = -ENOMEM;
2690                 if (!fpu)
2691                         goto out;
2692                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2693                 if (r)
2694                         goto out;
2695                 r = -EFAULT;
2696                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2697                         goto out;
2698                 r = 0;
2699                 break;
2700         }
2701         case KVM_SET_FPU: {
2702                 fpu = memdup_user(argp, sizeof(*fpu));
2703                 if (IS_ERR(fpu)) {
2704                         r = PTR_ERR(fpu);
2705                         fpu = NULL;
2706                         goto out;
2707                 }
2708                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2709                 break;
2710         }
2711         default:
2712                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2713         }
2714 out:
2715         mutex_unlock(&vcpu->mutex);
2716         kfree(fpu);
2717         kfree(kvm_sregs);
2718         return r;
2719 }
2720 
2721 #ifdef CONFIG_KVM_COMPAT
2722 static long kvm_vcpu_compat_ioctl(struct file *filp,
2723                                   unsigned int ioctl, unsigned long arg)
2724 {
2725         struct kvm_vcpu *vcpu = filp->private_data;
2726         void __user *argp = compat_ptr(arg);
2727         int r;
2728 
2729         if (vcpu->kvm->mm != current->mm)
2730                 return -EIO;
2731 
2732         switch (ioctl) {
2733         case KVM_SET_SIGNAL_MASK: {
2734                 struct kvm_signal_mask __user *sigmask_arg = argp;
2735                 struct kvm_signal_mask kvm_sigmask;
2736                 sigset_t sigset;
2737 
2738                 if (argp) {
2739                         r = -EFAULT;
2740                         if (copy_from_user(&kvm_sigmask, argp,
2741                                            sizeof(kvm_sigmask)))
2742                                 goto out;
2743                         r = -EINVAL;
2744                         if (kvm_sigmask.len != sizeof(compat_sigset_t))
2745                                 goto out;
2746                         r = -EFAULT;
2747                         if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2748                                 goto out;
2749                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2750                 } else
2751                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2752                 break;
2753         }
2754         default:
2755                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2756         }
2757 
2758 out:
2759         return r;
2760 }
2761 #endif
2762 
2763 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2764                                  int (*accessor)(struct kvm_device *dev,
2765                                                  struct kvm_device_attr *attr),
2766                                  unsigned long arg)
2767 {
2768         struct kvm_device_attr attr;
2769 
2770         if (!accessor)
2771                 return -EPERM;
2772 
2773         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2774                 return -EFAULT;
2775 
2776         return accessor(dev, &attr);
2777 }
2778 
2779 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2780                              unsigned long arg)
2781 {
2782         struct kvm_device *dev = filp->private_data;
2783 
2784         switch (ioctl) {
2785         case KVM_SET_DEVICE_ATTR:
2786                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2787         case KVM_GET_DEVICE_ATTR:
2788                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2789         case KVM_HAS_DEVICE_ATTR:
2790                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2791         default:
2792                 if (dev->ops->ioctl)
2793                         return dev->ops->ioctl(dev, ioctl, arg);
2794 
2795                 return -ENOTTY;
2796         }
2797 }
2798 
2799 static int kvm_device_release(struct inode *inode, struct file *filp)
2800 {
2801         struct kvm_device *dev = filp->private_data;
2802         struct kvm *kvm = dev->kvm;
2803 
2804         kvm_put_kvm(kvm);
2805         return 0;
2806 }
2807 
2808 static const struct file_operations kvm_device_fops = {
2809         .unlocked_ioctl = kvm_device_ioctl,
2810 #ifdef CONFIG_KVM_COMPAT
2811         .compat_ioctl = kvm_device_ioctl,
2812 #endif
2813         .release = kvm_device_release,
2814 };
2815 
2816 struct kvm_device *kvm_device_from_filp(struct file *filp)
2817 {
2818         if (filp->f_op != &kvm_device_fops)
2819                 return NULL;
2820 
2821         return filp->private_data;
2822 }
2823 
2824 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2825 #ifdef CONFIG_KVM_MPIC
2826         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2827         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2828 #endif
2829 };
2830 
2831 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2832 {
2833         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2834                 return -ENOSPC;
2835 
2836         if (kvm_device_ops_table[type] != NULL)
2837                 return -EEXIST;
2838 
2839         kvm_device_ops_table[type] = ops;
2840         return 0;
2841 }
2842 
2843 void kvm_unregister_device_ops(u32 type)
2844 {
2845         if (kvm_device_ops_table[type] != NULL)
2846                 kvm_device_ops_table[type] = NULL;
2847 }
2848 
2849 static int kvm_ioctl_create_device(struct kvm *kvm,
2850                                    struct kvm_create_device *cd)
2851 {
2852         struct kvm_device_ops *ops = NULL;
2853         struct kvm_device *dev;
2854         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2855         int ret;
2856 
2857         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2858                 return -ENODEV;
2859 
2860         ops = kvm_device_ops_table[cd->type];
2861         if (ops == NULL)
2862                 return -ENODEV;
2863 
2864         if (test)
2865                 return 0;
2866 
2867         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2868         if (!dev)
2869                 return -ENOMEM;
2870 
2871         dev->ops = ops;
2872         dev->kvm = kvm;
2873 
2874         mutex_lock(&kvm->lock);
2875         ret = ops->create(dev, cd->type);
2876         if (ret < 0) {
2877                 mutex_unlock(&kvm->lock);
2878                 kfree(dev);
2879                 return ret;
2880         }
2881         list_add(&dev->vm_node, &kvm->devices);
2882         mutex_unlock(&kvm->lock);
2883 
2884         if (ops->init)
2885                 ops->init(dev);
2886 
2887         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2888         if (ret < 0) {
2889                 mutex_lock(&kvm->lock);
2890                 list_del(&dev->vm_node);
2891                 mutex_unlock(&kvm->lock);
2892                 ops->destroy(dev);
2893                 return ret;
2894         }
2895 
2896         kvm_get_kvm(kvm);
2897         cd->fd = ret;
2898         return 0;
2899 }
2900 
2901 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2902 {
2903         switch (arg) {
2904         case KVM_CAP_USER_MEMORY:
2905         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2906         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2907         case KVM_CAP_INTERNAL_ERROR_DATA:
2908 #ifdef CONFIG_HAVE_KVM_MSI
2909         case KVM_CAP_SIGNAL_MSI:
2910 #endif
2911 #ifdef CONFIG_HAVE_KVM_IRQFD
2912         case KVM_CAP_IRQFD:
2913         case KVM_CAP_IRQFD_RESAMPLE:
2914 #endif
2915         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2916         case KVM_CAP_CHECK_EXTENSION_VM:
2917                 return 1;
2918 #ifdef CONFIG_KVM_MMIO
2919         case KVM_CAP_COALESCED_MMIO:
2920                 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2921 #endif
2922 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2923         case KVM_CAP_IRQ_ROUTING:
2924                 return KVM_MAX_IRQ_ROUTES;
2925 #endif
2926 #if KVM_ADDRESS_SPACE_NUM > 1
2927         case KVM_CAP_MULTI_ADDRESS_SPACE:
2928                 return KVM_ADDRESS_SPACE_NUM;
2929 #endif
2930         case KVM_CAP_MAX_VCPU_ID:
2931                 return KVM_MAX_VCPU_ID;
2932         default:
2933                 break;
2934         }
2935         return kvm_vm_ioctl_check_extension(kvm, arg);
2936 }
2937 
2938 static long kvm_vm_ioctl(struct file *filp,
2939                            unsigned int ioctl, unsigned long arg)
2940 {
2941         struct kvm *kvm = filp->private_data;
2942         void __user *argp = (void __user *)arg;
2943         int r;
2944 
2945         if (kvm->mm != current->mm)
2946                 return -EIO;
2947         switch (ioctl) {
2948         case KVM_CREATE_VCPU:
2949                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2950                 break;
2951         case KVM_SET_USER_MEMORY_REGION: {
2952                 struct kvm_userspace_memory_region kvm_userspace_mem;
2953 
2954                 r = -EFAULT;
2955                 if (copy_from_user(&kvm_userspace_mem, argp,
2956                                                 sizeof(kvm_userspace_mem)))
2957                         goto out;
2958 
2959                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2960                 break;
2961         }
2962         case KVM_GET_DIRTY_LOG: {
2963                 struct kvm_dirty_log log;
2964 
2965                 r = -EFAULT;
2966                 if (copy_from_user(&log, argp, sizeof(log)))
2967                         goto out;
2968                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2969                 break;
2970         }
2971 #ifdef CONFIG_KVM_MMIO
2972         case KVM_REGISTER_COALESCED_MMIO: {
2973                 struct kvm_coalesced_mmio_zone zone;
2974 
2975                 r = -EFAULT;
2976                 if (copy_from_user(&zone, argp, sizeof(zone)))
2977                         goto out;
2978                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2979                 break;
2980         }
2981         case KVM_UNREGISTER_COALESCED_MMIO: {
2982                 struct kvm_coalesced_mmio_zone zone;
2983 
2984                 r = -EFAULT;
2985                 if (copy_from_user(&zone, argp, sizeof(zone)))
2986                         goto out;
2987                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2988                 break;
2989         }
2990 #endif
2991         case KVM_IRQFD: {
2992                 struct kvm_irqfd data;
2993 
2994                 r = -EFAULT;
2995                 if (copy_from_user(&data, argp, sizeof(data)))
2996                         goto out;
2997                 r = kvm_irqfd(kvm, &data);
2998                 break;
2999         }
3000         case KVM_IOEVENTFD: {
3001                 struct kvm_ioeventfd data;
3002 
3003                 r = -EFAULT;
3004                 if (copy_from_user(&data, argp, sizeof(data)))
3005                         goto out;
3006                 r = kvm_ioeventfd(kvm, &data);
3007                 break;
3008         }
3009 #ifdef CONFIG_HAVE_KVM_MSI
3010         case KVM_SIGNAL_MSI: {
3011                 struct kvm_msi msi;
3012 
3013                 r = -EFAULT;
3014                 if (copy_from_user(&msi, argp, sizeof(msi)))
3015                         goto out;
3016                 r = kvm_send_userspace_msi(kvm, &msi);
3017                 break;
3018         }
3019 #endif
3020 #ifdef __KVM_HAVE_IRQ_LINE
3021         case KVM_IRQ_LINE_STATUS:
3022         case KVM_IRQ_LINE: {
3023                 struct kvm_irq_level irq_event;
3024 
3025                 r = -EFAULT;
3026                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3027                         goto out;
3028 
3029                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3030                                         ioctl == KVM_IRQ_LINE_STATUS);
3031                 if (r)
3032                         goto out;
3033 
3034                 r = -EFAULT;
3035                 if (ioctl == KVM_IRQ_LINE_STATUS) {
3036                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3037                                 goto out;
3038                 }
3039 
3040                 r = 0;
3041                 break;
3042         }
3043 #endif
3044 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3045         case KVM_SET_GSI_ROUTING: {
3046                 struct kvm_irq_routing routing;
3047                 struct kvm_irq_routing __user *urouting;
3048                 struct kvm_irq_routing_entry *entries = NULL;
3049 
3050                 r = -EFAULT;
3051                 if (copy_from_user(&routing, argp, sizeof(routing)))
3052                         goto out;
3053                 r = -EINVAL;
3054                 if (!kvm_arch_can_set_irq_routing(kvm))
3055                         goto out;
3056                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3057                         goto out;
3058                 if (routing.flags)
3059                         goto out;
3060                 if (routing.nr) {
3061                         r = -ENOMEM;
3062                         entries = vmalloc(routing.nr * sizeof(*entries));
3063                         if (!entries)
3064                                 goto out;
3065                         r = -EFAULT;
3066                         urouting = argp;
3067                         if (copy_from_user(entries, urouting->entries,
3068                                            routing.nr * sizeof(*entries)))
3069                                 goto out_free_irq_routing;
3070                 }
3071                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3072                                         routing.flags);
3073 out_free_irq_routing:
3074                 vfree(entries);
3075                 break;
3076         }
3077 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3078         case KVM_CREATE_DEVICE: {
3079                 struct kvm_create_device cd;
3080 
3081                 r = -EFAULT;
3082                 if (copy_from_user(&cd, argp, sizeof(cd)))
3083                         goto out;
3084 
3085                 r = kvm_ioctl_create_device(kvm, &cd);
3086                 if (r)
3087                         goto out;
3088 
3089                 r = -EFAULT;
3090                 if (copy_to_user(argp, &cd, sizeof(cd)))
3091                         goto out;
3092 
3093                 r = 0;
3094                 break;
3095         }
3096         case KVM_CHECK_EXTENSION:
3097                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3098                 break;
3099         default:
3100                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3101         }
3102 out:
3103         return r;
3104 }
3105 
3106 #ifdef CONFIG_KVM_COMPAT
3107 struct compat_kvm_dirty_log {
3108         __u32 slot;
3109         __u32 padding1;
3110         union {
3111                 compat_uptr_t dirty_bitmap; /* one bit per page */
3112                 __u64 padding2;
3113         };
3114 };
3115 
3116 static long kvm_vm_compat_ioctl(struct file *filp,
3117                            unsigned int ioctl, unsigned long arg)
3118 {
3119         struct kvm *kvm = filp->private_data;
3120         int r;
3121 
3122         if (kvm->mm != current->mm)
3123                 return -EIO;
3124         switch (ioctl) {
3125         case KVM_GET_DIRTY_LOG: {
3126                 struct compat_kvm_dirty_log compat_log;
3127                 struct kvm_dirty_log log;
3128 
3129                 if (copy_from_user(&compat_log, (void __user *)arg,
3130                                    sizeof(compat_log)))
3131                         return -EFAULT;
3132                 log.slot         = compat_log.slot;
3133                 log.padding1     = compat_log.padding1;
3134                 log.padding2     = compat_log.padding2;
3135                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3136 
3137                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3138                 break;
3139         }
3140         default:
3141                 r = kvm_vm_ioctl(filp, ioctl, arg);
3142         }
3143         return r;
3144 }
3145 #endif
3146 
3147 static struct file_operations kvm_vm_fops = {
3148         .release        = kvm_vm_release,
3149         .unlocked_ioctl = kvm_vm_ioctl,
3150 #ifdef CONFIG_KVM_COMPAT
3151         .compat_ioctl   = kvm_vm_compat_ioctl,
3152 #endif
3153         .llseek         = noop_llseek,
3154 };
3155 
3156 static int kvm_dev_ioctl_create_vm(unsigned long type)
3157 {
3158         int r;
3159         struct kvm *kvm;
3160         struct file *file;
3161 
3162         kvm = kvm_create_vm(type);
3163         if (IS_ERR(kvm))
3164                 return PTR_ERR(kvm);
3165 #ifdef CONFIG_KVM_MMIO
3166         r = kvm_coalesced_mmio_init(kvm);
3167         if (r < 0)
3168                 goto put_kvm;
3169 #endif
3170         r = get_unused_fd_flags(O_CLOEXEC);
3171         if (r < 0)
3172                 goto put_kvm;
3173 
3174         file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3175         if (IS_ERR(file)) {
3176                 put_unused_fd(r);
3177                 r = PTR_ERR(file);
3178                 goto put_kvm;
3179         }
3180 
3181         /*
3182          * Don't call kvm_put_kvm anymore at this point; file->f_op is
3183          * already set, with ->release() being kvm_vm_release().  In error
3184          * cases it will be called by the final fput(file) and will take
3185          * care of doing kvm_put_kvm(kvm).
3186          */
3187         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3188                 put_unused_fd(r);
3189                 fput(file);
3190                 return -ENOMEM;
3191         }
3192         kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3193 
3194         fd_install(r, file);
3195         return r;
3196 
3197 put_kvm:
3198         kvm_put_kvm(kvm);
3199         return r;
3200 }
3201 
3202 static long kvm_dev_ioctl(struct file *filp,
3203                           unsigned int ioctl, unsigned long arg)
3204 {
3205         long r = -EINVAL;
3206 
3207         switch (ioctl) {
3208         case KVM_GET_API_VERSION:
3209                 if (arg)
3210                         goto out;
3211                 r = KVM_API_VERSION;
3212                 break;
3213         case KVM_CREATE_VM:
3214                 r = kvm_dev_ioctl_create_vm(arg);
3215                 break;
3216         case KVM_CHECK_EXTENSION:
3217                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3218                 break;
3219         case KVM_GET_VCPU_MMAP_SIZE:
3220                 if (arg)
3221                         goto out;
3222                 r = PAGE_SIZE;     /* struct kvm_run */
3223 #ifdef CONFIG_X86
3224                 r += PAGE_SIZE;    /* pio data page */
3225 #endif
3226 #ifdef CONFIG_KVM_MMIO
3227                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3228 #endif
3229                 break;
3230         case KVM_TRACE_ENABLE:
3231         case KVM_TRACE_PAUSE:
3232         case KVM_TRACE_DISABLE:
3233                 r = -EOPNOTSUPP;
3234                 break;
3235         default:
3236                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3237         }
3238 out:
3239         return r;
3240 }
3241 
3242 static struct file_operations kvm_chardev_ops = {
3243         .unlocked_ioctl = kvm_dev_ioctl,
3244         .compat_ioctl   = kvm_dev_ioctl,
3245         .llseek         = noop_llseek,
3246 };
3247 
3248 static struct miscdevice kvm_dev = {
3249         KVM_MINOR,
3250         "kvm",
3251         &kvm_chardev_ops,
3252 };
3253 
3254 static void hardware_enable_nolock(void *junk)
3255 {
3256         int cpu = raw_smp_processor_id();
3257         int r;
3258 
3259         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3260                 return;
3261 
3262         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3263 
3264         r = kvm_arch_hardware_enable();
3265 
3266         if (r) {
3267                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3268                 atomic_inc(&hardware_enable_failed);
3269                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3270         }
3271 }
3272 
3273 static int kvm_starting_cpu(unsigned int cpu)
3274 {
3275         raw_spin_lock(&kvm_count_lock);
3276         if (kvm_usage_count)
3277                 hardware_enable_nolock(NULL);
3278         raw_spin_unlock(&kvm_count_lock);
3279         return 0;
3280 }
3281 
3282 static void hardware_disable_nolock(void *junk)
3283 {
3284         int cpu = raw_smp_processor_id();
3285 
3286         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3287                 return;
3288         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3289         kvm_arch_hardware_disable();
3290 }
3291 
3292 static int kvm_dying_cpu(unsigned int cpu)
3293 {
3294         raw_spin_lock(&kvm_count_lock);
3295         if (kvm_usage_count)
3296                 hardware_disable_nolock(NULL);
3297         raw_spin_unlock(&kvm_count_lock);
3298         return 0;
3299 }
3300 
3301 static void hardware_disable_all_nolock(void)
3302 {
3303         BUG_ON(!kvm_usage_count);
3304 
3305         kvm_usage_count--;
3306         if (!kvm_usage_count)
3307                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3308 }
3309 
3310 static void hardware_disable_all(void)
3311 {
3312         raw_spin_lock(&kvm_count_lock);
3313         hardware_disable_all_nolock();
3314         raw_spin_unlock(&kvm_count_lock);
3315 }
3316 
3317 static int hardware_enable_all(void)
3318 {
3319         int r = 0;
3320 
3321         raw_spin_lock(&kvm_count_lock);
3322 
3323         kvm_usage_count++;
3324         if (kvm_usage_count == 1) {
3325                 atomic_set(&hardware_enable_failed, 0);
3326                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3327 
3328                 if (atomic_read(&hardware_enable_failed)) {
3329                         hardware_disable_all_nolock();
3330                         r = -EBUSY;
3331                 }
3332         }
3333 
3334         raw_spin_unlock(&kvm_count_lock);
3335 
3336         return r;
3337 }
3338 
3339 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3340                       void *v)
3341 {
3342         /*
3343          * Some (well, at least mine) BIOSes hang on reboot if
3344          * in vmx root mode.
3345          *
3346          * And Intel TXT required VMX off for all cpu when system shutdown.
3347          */
3348         pr_info("kvm: exiting hardware virtualization\n");
3349         kvm_rebooting = true;
3350         on_each_cpu(hardware_disable_nolock, NULL, 1);
3351         return NOTIFY_OK;
3352 }
3353 
3354 static struct notifier_block kvm_reboot_notifier = {
3355         .notifier_call = kvm_reboot,
3356         .priority = 0,
3357 };
3358 
3359 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3360 {
3361         int i;
3362 
3363         for (i = 0; i < bus->dev_count; i++) {
3364                 struct kvm_io_device *pos = bus->range[i].dev;
3365 
3366                 kvm_iodevice_destructor(pos);
3367         }
3368         kfree(bus);
3369 }
3370 
3371 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3372                                  const struct kvm_io_range *r2)
3373 {
3374         gpa_t addr1 = r1->addr;
3375         gpa_t addr2 = r2->addr;
3376 
3377         if (addr1 < addr2)
3378                 return -1;
3379 
3380         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3381          * accept any overlapping write.  Any order is acceptable for
3382          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3383          * we process all of them.
3384          */
3385         if (r2->len) {
3386                 addr1 += r1->len;
3387                 addr2 += r2->len;
3388         }
3389 
3390         if (addr1 > addr2)
3391                 return 1;
3392 
3393         return 0;
3394 }
3395 
3396 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3397 {
3398         return kvm_io_bus_cmp(p1, p2);
3399 }
3400 
3401 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3402                           gpa_t addr, int len)
3403 {
3404         bus->range[bus->dev_count++] = (struct kvm_io_range) {
3405                 .addr = addr,
3406                 .len = len,
3407                 .dev = dev,
3408         };
3409 
3410         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3411                 kvm_io_bus_sort_cmp, NULL);
3412 
3413         return 0;
3414 }
3415 
3416 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3417                              gpa_t addr, int len)
3418 {
3419         struct kvm_io_range *range, key;
3420         int off;
3421 
3422         key = (struct kvm_io_range) {
3423                 .addr = addr,
3424                 .len = len,
3425         };
3426 
3427         range = bsearch(&key, bus->range, bus->dev_count,
3428                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3429         if (range == NULL)
3430                 return -ENOENT;
3431 
3432         off = range - bus->range;
3433 
3434         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3435                 off--;
3436 
3437         return off;
3438 }
3439 
3440 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3441                               struct kvm_io_range *range, const void *val)
3442 {
3443         int idx;
3444 
3445         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3446         if (idx < 0)
3447                 return -EOPNOTSUPP;
3448 
3449         while (idx < bus->dev_count &&
3450                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3451                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3452                                         range->len, val))
3453                         return idx;
3454                 idx++;
3455         }
3456 
3457         return -EOPNOTSUPP;
3458 }
3459 
3460 /* kvm_io_bus_write - called under kvm->slots_lock */
3461 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3462                      int len, const void *val)
3463 {
3464         struct kvm_io_bus *bus;
3465         struct kvm_io_range range;
3466         int r;
3467 
3468         range = (struct kvm_io_range) {
3469                 .addr = addr,
3470                 .len = len,
3471         };
3472 
3473         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3474         if (!bus)
3475                 return -ENOMEM;
3476         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3477         return r < 0 ? r : 0;
3478 }
3479 
3480 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3481 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3482                             gpa_t addr, int len, const void *val, long cookie)
3483 {
3484         struct kvm_io_bus *bus;
3485         struct kvm_io_range range;
3486 
3487         range = (struct kvm_io_range) {
3488                 .addr = addr,
3489                 .len = len,
3490         };
3491 
3492         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3493         if (!bus)
3494                 return -ENOMEM;
3495 
3496         /* First try the device referenced by cookie. */
3497         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3498             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3499                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3500                                         val))
3501                         return cookie;
3502 
3503         /*
3504          * cookie contained garbage; fall back to search and return the
3505          * correct cookie value.
3506          */
3507         return __kvm_io_bus_write(vcpu, bus, &range, val);
3508 }
3509 
3510 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3511                              struct kvm_io_range *range, void *val)
3512 {
3513         int idx;
3514 
3515         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3516         if (idx < 0)
3517                 return -EOPNOTSUPP;
3518 
3519         while (idx < bus->dev_count &&
3520                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3521                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3522                                        range->len, val))
3523                         return idx;
3524                 idx++;
3525         }
3526 
3527         return -EOPNOTSUPP;
3528 }
3529 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3530 
3531 /* kvm_io_bus_read - called under kvm->slots_lock */
3532 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3533                     int len, void *val)
3534 {
3535         struct kvm_io_bus *bus;
3536         struct kvm_io_range range;
3537         int r;
3538 
3539         range = (struct kvm_io_range) {
3540                 .addr = addr,
3541                 .len = len,
3542         };
3543 
3544         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3545         if (!bus)
3546                 return -ENOMEM;
3547         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3548         return r < 0 ? r : 0;
3549 }
3550 
3551 
3552 /* Caller must hold slots_lock. */
3553 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3554                             int len, struct kvm_io_device *dev)
3555 {
3556         struct kvm_io_bus *new_bus, *bus;
3557 
3558         bus = kvm_get_bus(kvm, bus_idx);
3559         if (!bus)
3560                 return -ENOMEM;
3561 
3562         /* exclude ioeventfd which is limited by maximum fd */
3563         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3564                 return -ENOSPC;
3565 
3566         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3567                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3568         if (!new_bus)
3569                 return -ENOMEM;
3570         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3571                sizeof(struct kvm_io_range)));
3572         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3573         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3574         synchronize_srcu_expedited(&kvm->srcu);
3575         kfree(bus);
3576 
3577         return 0;
3578 }
3579 
3580 /* Caller must hold slots_lock. */
3581 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3582                                struct kvm_io_device *dev)
3583 {
3584         int i;
3585         struct kvm_io_bus *new_bus, *bus;
3586 
3587         bus = kvm_get_bus(kvm, bus_idx);
3588         if (!bus)
3589                 return;
3590 
3591         for (i = 0; i < bus->dev_count; i++)
3592                 if (bus->range[i].dev == dev) {
3593                         break;
3594                 }
3595 
3596         if (i == bus->dev_count)
3597                 return;
3598 
3599         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3600                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3601         if (!new_bus)  {
3602                 pr_err("kvm: failed to shrink bus, removing it completely\n");
3603                 goto broken;
3604         }
3605 
3606         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3607         new_bus->dev_count--;
3608         memcpy(new_bus->range + i, bus->range + i + 1,
3609                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3610 
3611 broken:
3612         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3613         synchronize_srcu_expedited(&kvm->srcu);
3614         kfree(bus);
3615         return;
3616 }
3617 
3618 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3619                                          gpa_t addr)
3620 {
3621         struct kvm_io_bus *bus;
3622         int dev_idx, srcu_idx;
3623         struct kvm_io_device *iodev = NULL;
3624 
3625         srcu_idx = srcu_read_lock(&kvm->srcu);
3626 
3627         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3628         if (!bus)
3629                 goto out_unlock;
3630 
3631         dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3632         if (dev_idx < 0)
3633                 goto out_unlock;
3634 
3635         iodev = bus->range[dev_idx].dev;
3636 
3637 out_unlock:
3638         srcu_read_unlock(&kvm->srcu, srcu_idx);
3639 
3640         return iodev;
3641 }
3642 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3643 
3644 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3645                            int (*get)(void *, u64 *), int (*set)(void *, u64),
3646                            const char *fmt)
3647 {
3648         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3649                                           inode->i_private;
3650 
3651         /* The debugfs files are a reference to the kvm struct which
3652          * is still valid when kvm_destroy_vm is called.
3653          * To avoid the race between open and the removal of the debugfs
3654          * directory we test against the users count.
3655          */
3656         if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3657                 return -ENOENT;
3658 
3659         if (simple_attr_open(inode, file, get, set, fmt)) {
3660                 kvm_put_kvm(stat_data->kvm);
3661                 return -ENOMEM;
3662         }
3663 
3664         return 0;
3665 }
3666 
3667 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3668 {
3669         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3670                                           inode->i_private;
3671 
3672         simple_attr_release(inode, file);
3673         kvm_put_kvm(stat_data->kvm);
3674 
3675         return 0;
3676 }
3677 
3678 static int vm_stat_get_per_vm(void *data, u64 *val)
3679 {
3680         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3681 
3682         *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3683 
3684         return 0;
3685 }
3686 
3687 static int vm_stat_clear_per_vm(void *data, u64 val)
3688 {
3689         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3690 
3691         if (val)
3692                 return -EINVAL;
3693 
3694         *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3695 
3696         return 0;
3697 }
3698 
3699 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3700 {
3701         __simple_attr_check_format("%llu\n", 0ull);
3702         return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3703                                 vm_stat_clear_per_vm, "%llu\n");
3704 }
3705 
3706 static const struct file_operations vm_stat_get_per_vm_fops = {
3707         .owner   = THIS_MODULE,
3708         .open    = vm_stat_get_per_vm_open,
3709         .release = kvm_debugfs_release,
3710         .read    = simple_attr_read,
3711         .write   = simple_attr_write,
3712         .llseek  = no_llseek,
3713 };
3714 
3715 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3716 {
3717         int i;
3718         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3719         struct kvm_vcpu *vcpu;
3720 
3721         *val = 0;
3722 
3723         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3724                 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3725 
3726         return 0;
3727 }
3728 
3729 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3730 {
3731         int i;
3732         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3733         struct kvm_vcpu *vcpu;
3734 
3735         if (val)
3736                 return -EINVAL;
3737 
3738         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3739                 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3740 
3741         return 0;
3742 }
3743 
3744 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3745 {
3746         __simple_attr_check_format("%llu\n", 0ull);
3747         return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3748                                  vcpu_stat_clear_per_vm, "%llu\n");
3749 }
3750 
3751 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3752         .owner   = THIS_MODULE,
3753         .open    = vcpu_stat_get_per_vm_open,
3754         .release = kvm_debugfs_release,
3755         .read    = simple_attr_read,
3756         .write   = simple_attr_write,
3757         .llseek  = no_llseek,
3758 };
3759 
3760 static const struct file_operations *stat_fops_per_vm[] = {
3761         [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3762         [KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
3763 };
3764 
3765 static int vm_stat_get(void *_offset, u64 *val)
3766 {
3767         unsigned offset = (long)_offset;
3768         struct kvm *kvm;
3769         struct kvm_stat_data stat_tmp = {.offset = offset};
3770         u64 tmp_val;
3771 
3772         *val = 0;
3773         spin_lock(&kvm_lock);
3774         list_for_each_entry(kvm, &vm_list, vm_list) {
3775                 stat_tmp.kvm = kvm;
3776                 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3777                 *val += tmp_val;
3778         }
3779         spin_unlock(&kvm_lock);
3780         return 0;
3781 }
3782 
3783 static int vm_stat_clear(void *_offset, u64 val)
3784 {
3785         unsigned offset = (long)_offset;
3786         struct kvm *kvm;
3787         struct kvm_stat_data stat_tmp = {.offset = offset};
3788 
3789         if (val)
3790                 return -EINVAL;
3791 
3792         spin_lock(&kvm_lock);
3793         list_for_each_entry(kvm, &vm_list, vm_list) {
3794                 stat_tmp.kvm = kvm;
3795                 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3796         }
3797         spin_unlock(&kvm_lock);
3798 
3799         return 0;
3800 }
3801 
3802 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3803 
3804 static int vcpu_stat_get(void *_offset, u64 *val)
3805 {
3806         unsigned offset = (long)_offset;
3807         struct kvm *kvm;
3808         struct kvm_stat_data stat_tmp = {.offset = offset};
3809         u64 tmp_val;
3810 
3811         *val = 0;
3812         spin_lock(&kvm_lock);
3813         list_for_each_entry(kvm, &vm_list, vm_list) {
3814                 stat_tmp.kvm = kvm;
3815                 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3816                 *val += tmp_val;
3817         }
3818         spin_unlock(&kvm_lock);
3819         return 0;
3820 }
3821 
3822 static int vcpu_stat_clear(void *_offset, u64 val)
3823 {
3824         unsigned offset = (long)_offset;
3825         struct kvm *kvm;
3826         struct kvm_stat_data stat_tmp = {.offset = offset};
3827 
3828         if (val)
3829                 return -EINVAL;
3830 
3831         spin_lock(&kvm_lock);
3832         list_for_each_entry(kvm, &vm_list, vm_list) {
3833                 stat_tmp.kvm = kvm;
3834                 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3835         }
3836         spin_unlock(&kvm_lock);
3837 
3838         return 0;
3839 }
3840 
3841 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3842                         "%llu\n");
3843 
3844 static const struct file_operations *stat_fops[] = {
3845         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3846         [KVM_STAT_VM]   = &vm_stat_fops,
3847 };
3848 
3849 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3850 {
3851         struct kobj_uevent_env *env;
3852         unsigned long long created, active;
3853 
3854         if (!kvm_dev.this_device || !kvm)
3855                 return;
3856 
3857         spin_lock(&kvm_lock);
3858         if (type == KVM_EVENT_CREATE_VM) {
3859                 kvm_createvm_count++;
3860                 kvm_active_vms++;
3861         } else if (type == KVM_EVENT_DESTROY_VM) {
3862                 kvm_active_vms--;
3863         }
3864         created = kvm_createvm_count;
3865         active = kvm_active_vms;
3866         spin_unlock(&kvm_lock);
3867 
3868         env = kzalloc(sizeof(*env), GFP_KERNEL);
3869         if (!env)
3870                 return;
3871 
3872         add_uevent_var(env, "CREATED=%llu", created);
3873         add_uevent_var(env, "COUNT=%llu", active);
3874 
3875         if (type == KVM_EVENT_CREATE_VM) {
3876                 add_uevent_var(env, "EVENT=create");
3877                 kvm->userspace_pid = task_pid_nr(current);
3878         } else if (type == KVM_EVENT_DESTROY_VM) {
3879                 add_uevent_var(env, "EVENT=destroy");
3880         }
3881         add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3882 
3883         if (kvm->debugfs_dentry) {
3884                 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3885 
3886                 if (p) {
3887                         tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3888                         if (!IS_ERR(tmp))
3889                                 add_uevent_var(env, "STATS_PATH=%s", tmp);
3890                         kfree(p);
3891                 }
3892         }
3893         /* no need for checks, since we are adding at most only 5 keys */
3894         env->envp[env->envp_idx++] = NULL;
3895         kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3896         kfree(env);
3897 }
3898 
3899 static int kvm_init_debug(void)
3900 {
3901         int r = -EEXIST;
3902         struct kvm_stats_debugfs_item *p;
3903 
3904         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3905         if (kvm_debugfs_dir == NULL)
3906                 goto out;
3907 
3908         kvm_debugfs_num_entries = 0;
3909         for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3910                 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3911                                          (void *)(long)p->offset,
3912                                          stat_fops[p->kind]))
3913                         goto out_dir;
3914         }
3915 
3916         return 0;
3917 
3918 out_dir:
3919         debugfs_remove_recursive(kvm_debugfs_dir);
3920 out:
3921         return r;
3922 }
3923 
3924 static int kvm_suspend(void)
3925 {
3926         if (kvm_usage_count)
3927                 hardware_disable_nolock(NULL);
3928         return 0;
3929 }
3930 
3931 static void kvm_resume(void)
3932 {
3933         if (kvm_usage_count) {
3934                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3935                 hardware_enable_nolock(NULL);
3936         }
3937 }
3938 
3939 static struct syscore_ops kvm_syscore_ops = {
3940         .suspend = kvm_suspend,
3941         .resume = kvm_resume,
3942 };
3943 
3944 static inline
3945 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3946 {
3947         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3948 }
3949 
3950 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3951 {
3952         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3953 
3954         if (vcpu->preempted)
3955                 vcpu->preempted = false;
3956 
3957         kvm_arch_sched_in(vcpu, cpu);
3958 
3959         kvm_arch_vcpu_load(vcpu, cpu);
3960 }
3961 
3962 static void kvm_sched_out(struct preempt_notifier *pn,
3963                           struct task_struct *next)
3964 {
3965         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3966 
3967         if (current->state == TASK_RUNNING)
3968                 vcpu->preempted = true;
3969         kvm_arch_vcpu_put(vcpu);
3970 }
3971 
3972 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3973                   struct module *module)
3974 {
3975         int r;
3976         int cpu;
3977 
3978         r = kvm_arch_init(opaque);
3979         if (r)
3980                 goto out_fail;
3981 
3982         /*
3983          * kvm_arch_init makes sure there's at most one caller
3984          * for architectures that support multiple implementations,
3985          * like intel and amd on x86.
3986          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3987          * conflicts in case kvm is already setup for another implementation.
3988          */
3989         r = kvm_irqfd_init();
3990         if (r)
3991                 goto out_irqfd;
3992 
3993         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3994                 r = -ENOMEM;
3995                 goto out_free_0;
3996         }
3997 
3998         r = kvm_arch_hardware_setup();
3999         if (r < 0)
4000                 goto out_free_0a;
4001 
4002         for_each_online_cpu(cpu) {
4003                 smp_call_function_single(cpu,
4004                                 kvm_arch_check_processor_compat,
4005                                 &r, 1);
4006                 if (r < 0)
4007                         goto out_free_1;
4008         }
4009 
4010         r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4011                                       kvm_starting_cpu, kvm_dying_cpu);
4012         if (r)
4013                 goto out_free_2;
4014         register_reboot_notifier(&kvm_reboot_notifier);
4015 
4016         /* A kmem cache lets us meet the alignment requirements of fx_save. */
4017         if (!vcpu_align)
4018                 vcpu_align = __alignof__(struct kvm_vcpu);
4019         kvm_vcpu_cache =
4020                 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4021                                            SLAB_ACCOUNT,
4022                                            offsetof(struct kvm_vcpu, arch),
4023                                            sizeof_field(struct kvm_vcpu, arch),
4024                                            NULL);
4025         if (!kvm_vcpu_cache) {
4026                 r = -ENOMEM;
4027                 goto out_free_3;
4028         }
4029 
4030         r = kvm_async_pf_init();
4031         if (r)
4032                 goto out_free;
4033 
4034         kvm_chardev_ops.owner = module;
4035         kvm_vm_fops.owner = module;
4036         kvm_vcpu_fops.owner = module;
4037 
4038         r = misc_register(&kvm_dev);
4039         if (r) {
4040                 pr_err("kvm: misc device register failed\n");
4041                 goto out_unreg;
4042         }
4043 
4044         register_syscore_ops(&kvm_syscore_ops);
4045 
4046         kvm_preempt_ops.sched_in = kvm_sched_in;
4047         kvm_preempt_ops.sched_out = kvm_sched_out;
4048 
4049         r = kvm_init_debug();
4050         if (r) {
4051                 pr_err("kvm: create debugfs files failed\n");
4052                 goto out_undebugfs;
4053         }
4054 
4055         r = kvm_vfio_ops_init();
4056         WARN_ON(r);
4057 
4058         return 0;
4059 
4060 out_undebugfs:
4061         unregister_syscore_ops(&kvm_syscore_ops);
4062         misc_deregister(&kvm_dev);
4063 out_unreg:
4064         kvm_async_pf_deinit();
4065 out_free:
4066         kmem_cache_destroy(kvm_vcpu_cache);
4067 out_free_3:
4068         unregister_reboot_notifier(&kvm_reboot_notifier);
4069         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4070 out_free_2:
4071 out_free_1:
4072         kvm_arch_hardware_unsetup();
4073 out_free_0a:
4074         free_cpumask_var(cpus_hardware_enabled);
4075 out_free_0:
4076         kvm_irqfd_exit();
4077 out_irqfd:
4078         kvm_arch_exit();
4079 out_fail:
4080         return r;
4081 }
4082 EXPORT_SYMBOL_GPL(kvm_init);
4083 
4084 void kvm_exit(void)
4085 {
4086         debugfs_remove_recursive(kvm_debugfs_dir);
4087         misc_deregister(&kvm_dev);
4088         kmem_cache_destroy(kvm_vcpu_cache);
4089         kvm_async_pf_deinit();
4090         unregister_syscore_ops(&kvm_syscore_ops);
4091         unregister_reboot_notifier(&kvm_reboot_notifier);
4092         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4093         on_each_cpu(hardware_disable_nolock, NULL, 1);
4094         kvm_arch_hardware_unsetup();
4095         kvm_arch_exit();
4096         kvm_irqfd_exit();
4097         free_cpumask_var(cpus_hardware_enabled);
4098         kvm_vfio_ops_exit();
4099 }
4100 EXPORT_SYMBOL_GPL(kvm_exit);
4101 

~ [ 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