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

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

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ 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 int vcpu_load(struct kvm_vcpu *vcpu)
155 {
156         int cpu;
157 
158         if (mutex_lock_killable(&vcpu->mutex))
159                 return -EINTR;
160         cpu = get_cpu();
161         preempt_notifier_register(&vcpu->preempt_notifier);
162         kvm_arch_vcpu_load(vcpu, cpu);
163         put_cpu();
164         return 0;
165 }
166 EXPORT_SYMBOL_GPL(vcpu_load);
167 
168 void vcpu_put(struct kvm_vcpu *vcpu)
169 {
170         preempt_disable();
171         kvm_arch_vcpu_put(vcpu);
172         preempt_notifier_unregister(&vcpu->preempt_notifier);
173         preempt_enable();
174         mutex_unlock(&vcpu->mutex);
175 }
176 EXPORT_SYMBOL_GPL(vcpu_put);
177 
178 /* TODO: merge with kvm_arch_vcpu_should_kick */
179 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
180 {
181         int mode = kvm_vcpu_exiting_guest_mode(vcpu);
182 
183         /*
184          * We need to wait for the VCPU to reenable interrupts and get out of
185          * READING_SHADOW_PAGE_TABLES mode.
186          */
187         if (req & KVM_REQUEST_WAIT)
188                 return mode != OUTSIDE_GUEST_MODE;
189 
190         /*
191          * Need to kick a running VCPU, but otherwise there is nothing to do.
192          */
193         return mode == IN_GUEST_MODE;
194 }
195 
196 static void ack_flush(void *_completed)
197 {
198 }
199 
200 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
201 {
202         if (unlikely(!cpus))
203                 cpus = cpu_online_mask;
204 
205         if (cpumask_empty(cpus))
206                 return false;
207 
208         smp_call_function_many(cpus, ack_flush, NULL, wait);
209         return true;
210 }
211 
212 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
213 {
214         int i, cpu, me;
215         cpumask_var_t cpus;
216         bool called;
217         struct kvm_vcpu *vcpu;
218 
219         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
220 
221         me = get_cpu();
222         kvm_for_each_vcpu(i, vcpu, kvm) {
223                 kvm_make_request(req, vcpu);
224                 cpu = vcpu->cpu;
225 
226                 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
227                         continue;
228 
229                 if (cpus != NULL && cpu != -1 && cpu != me &&
230                     kvm_request_needs_ipi(vcpu, req))
231                         __cpumask_set_cpu(cpu, cpus);
232         }
233         called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
234         put_cpu();
235         free_cpumask_var(cpus);
236         return called;
237 }
238 
239 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
240 void kvm_flush_remote_tlbs(struct kvm *kvm)
241 {
242         /*
243          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
244          * kvm_make_all_cpus_request.
245          */
246         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
247 
248         /*
249          * We want to publish modifications to the page tables before reading
250          * mode. Pairs with a memory barrier in arch-specific code.
251          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
252          * and smp_mb in walk_shadow_page_lockless_begin/end.
253          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
254          *
255          * There is already an smp_mb__after_atomic() before
256          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
257          * barrier here.
258          */
259         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
260                 ++kvm->stat.remote_tlb_flush;
261         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
262 }
263 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
264 #endif
265 
266 void kvm_reload_remote_mmus(struct kvm *kvm)
267 {
268         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
269 }
270 
271 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
272 {
273         struct page *page;
274         int r;
275 
276         mutex_init(&vcpu->mutex);
277         vcpu->cpu = -1;
278         vcpu->kvm = kvm;
279         vcpu->vcpu_id = id;
280         vcpu->pid = NULL;
281         init_swait_queue_head(&vcpu->wq);
282         kvm_async_pf_vcpu_init(vcpu);
283 
284         vcpu->pre_pcpu = -1;
285         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
286 
287         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
288         if (!page) {
289                 r = -ENOMEM;
290                 goto fail;
291         }
292         vcpu->run = page_address(page);
293 
294         kvm_vcpu_set_in_spin_loop(vcpu, false);
295         kvm_vcpu_set_dy_eligible(vcpu, false);
296         vcpu->preempted = false;
297 
298         r = kvm_arch_vcpu_init(vcpu);
299         if (r < 0)
300                 goto fail_free_run;
301         return 0;
302 
303 fail_free_run:
304         free_page((unsigned long)vcpu->run);
305 fail:
306         return r;
307 }
308 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
309 
310 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
311 {
312         /*
313          * no need for rcu_read_lock as VCPU_RUN is the only place that
314          * will change the vcpu->pid pointer and on uninit all file
315          * descriptors are already gone.
316          */
317         put_pid(rcu_dereference_protected(vcpu->pid, 1));
318         kvm_arch_vcpu_uninit(vcpu);
319         free_page((unsigned long)vcpu->run);
320 }
321 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
322 
323 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
324 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
325 {
326         return container_of(mn, struct kvm, mmu_notifier);
327 }
328 
329 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
330                                         struct mm_struct *mm,
331                                         unsigned long address,
332                                         pte_t pte)
333 {
334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
335         int idx;
336 
337         idx = srcu_read_lock(&kvm->srcu);
338         spin_lock(&kvm->mmu_lock);
339         kvm->mmu_notifier_seq++;
340         kvm_set_spte_hva(kvm, address, pte);
341         spin_unlock(&kvm->mmu_lock);
342         srcu_read_unlock(&kvm->srcu, idx);
343 }
344 
345 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
346                                                     struct mm_struct *mm,
347                                                     unsigned long start,
348                                                     unsigned long end)
349 {
350         struct kvm *kvm = mmu_notifier_to_kvm(mn);
351         int need_tlb_flush = 0, idx;
352 
353         idx = srcu_read_lock(&kvm->srcu);
354         spin_lock(&kvm->mmu_lock);
355         /*
356          * The count increase must become visible at unlock time as no
357          * spte can be established without taking the mmu_lock and
358          * count is also read inside the mmu_lock critical section.
359          */
360         kvm->mmu_notifier_count++;
361         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
362         need_tlb_flush |= kvm->tlbs_dirty;
363         /* we've to flush the tlb before the pages can be freed */
364         if (need_tlb_flush)
365                 kvm_flush_remote_tlbs(kvm);
366 
367         spin_unlock(&kvm->mmu_lock);
368 
369         kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
370 
371         srcu_read_unlock(&kvm->srcu, idx);
372 }
373 
374 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
375                                                   struct mm_struct *mm,
376                                                   unsigned long start,
377                                                   unsigned long end)
378 {
379         struct kvm *kvm = mmu_notifier_to_kvm(mn);
380 
381         spin_lock(&kvm->mmu_lock);
382         /*
383          * This sequence increase will notify the kvm page fault that
384          * the page that is going to be mapped in the spte could have
385          * been freed.
386          */
387         kvm->mmu_notifier_seq++;
388         smp_wmb();
389         /*
390          * The above sequence increase must be visible before the
391          * below count decrease, which is ensured by the smp_wmb above
392          * in conjunction with the smp_rmb in mmu_notifier_retry().
393          */
394         kvm->mmu_notifier_count--;
395         spin_unlock(&kvm->mmu_lock);
396 
397         BUG_ON(kvm->mmu_notifier_count < 0);
398 }
399 
400 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
401                                               struct mm_struct *mm,
402                                               unsigned long start,
403                                               unsigned long end)
404 {
405         struct kvm *kvm = mmu_notifier_to_kvm(mn);
406         int young, idx;
407 
408         idx = srcu_read_lock(&kvm->srcu);
409         spin_lock(&kvm->mmu_lock);
410 
411         young = kvm_age_hva(kvm, start, end);
412         if (young)
413                 kvm_flush_remote_tlbs(kvm);
414 
415         spin_unlock(&kvm->mmu_lock);
416         srcu_read_unlock(&kvm->srcu, idx);
417 
418         return young;
419 }
420 
421 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
422                                         struct mm_struct *mm,
423                                         unsigned long start,
424                                         unsigned long end)
425 {
426         struct kvm *kvm = mmu_notifier_to_kvm(mn);
427         int young, idx;
428 
429         idx = srcu_read_lock(&kvm->srcu);
430         spin_lock(&kvm->mmu_lock);
431         /*
432          * Even though we do not flush TLB, this will still adversely
433          * affect performance on pre-Haswell Intel EPT, where there is
434          * no EPT Access Bit to clear so that we have to tear down EPT
435          * tables instead. If we find this unacceptable, we can always
436          * add a parameter to kvm_age_hva so that it effectively doesn't
437          * do anything on clear_young.
438          *
439          * Also note that currently we never issue secondary TLB flushes
440          * from clear_young, leaving this job up to the regular system
441          * cadence. If we find this inaccurate, we might come up with a
442          * more sophisticated heuristic later.
443          */
444         young = kvm_age_hva(kvm, start, end);
445         spin_unlock(&kvm->mmu_lock);
446         srcu_read_unlock(&kvm->srcu, idx);
447 
448         return young;
449 }
450 
451 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
452                                        struct mm_struct *mm,
453                                        unsigned long address)
454 {
455         struct kvm *kvm = mmu_notifier_to_kvm(mn);
456         int young, idx;
457 
458         idx = srcu_read_lock(&kvm->srcu);
459         spin_lock(&kvm->mmu_lock);
460         young = kvm_test_age_hva(kvm, address);
461         spin_unlock(&kvm->mmu_lock);
462         srcu_read_unlock(&kvm->srcu, idx);
463 
464         return young;
465 }
466 
467 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
468                                      struct mm_struct *mm)
469 {
470         struct kvm *kvm = mmu_notifier_to_kvm(mn);
471         int idx;
472 
473         idx = srcu_read_lock(&kvm->srcu);
474         kvm_arch_flush_shadow_all(kvm);
475         srcu_read_unlock(&kvm->srcu, idx);
476 }
477 
478 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
479         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
480         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
481         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
482         .clear_young            = kvm_mmu_notifier_clear_young,
483         .test_young             = kvm_mmu_notifier_test_young,
484         .change_pte             = kvm_mmu_notifier_change_pte,
485         .release                = kvm_mmu_notifier_release,
486 };
487 
488 static int kvm_init_mmu_notifier(struct kvm *kvm)
489 {
490         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
491         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
492 }
493 
494 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
495 
496 static int kvm_init_mmu_notifier(struct kvm *kvm)
497 {
498         return 0;
499 }
500 
501 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
502 
503 static struct kvm_memslots *kvm_alloc_memslots(void)
504 {
505         int i;
506         struct kvm_memslots *slots;
507 
508         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
509         if (!slots)
510                 return NULL;
511 
512         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
513                 slots->id_to_index[i] = slots->memslots[i].id = i;
514 
515         return slots;
516 }
517 
518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
519 {
520         if (!memslot->dirty_bitmap)
521                 return;
522 
523         kvfree(memslot->dirty_bitmap);
524         memslot->dirty_bitmap = NULL;
525 }
526 
527 /*
528  * Free any memory in @free but not in @dont.
529  */
530 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
531                               struct kvm_memory_slot *dont)
532 {
533         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
534                 kvm_destroy_dirty_bitmap(free);
535 
536         kvm_arch_free_memslot(kvm, free, dont);
537 
538         free->npages = 0;
539 }
540 
541 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
542 {
543         struct kvm_memory_slot *memslot;
544 
545         if (!slots)
546                 return;
547 
548         kvm_for_each_memslot(memslot, slots)
549                 kvm_free_memslot(kvm, memslot, NULL);
550 
551         kvfree(slots);
552 }
553 
554 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
555 {
556         int i;
557 
558         if (!kvm->debugfs_dentry)
559                 return;
560 
561         debugfs_remove_recursive(kvm->debugfs_dentry);
562 
563         if (kvm->debugfs_stat_data) {
564                 for (i = 0; i < kvm_debugfs_num_entries; i++)
565                         kfree(kvm->debugfs_stat_data[i]);
566                 kfree(kvm->debugfs_stat_data);
567         }
568 }
569 
570 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
571 {
572         char dir_name[ITOA_MAX_LEN * 2];
573         struct kvm_stat_data *stat_data;
574         struct kvm_stats_debugfs_item *p;
575 
576         if (!debugfs_initialized())
577                 return 0;
578 
579         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
580         kvm->debugfs_dentry = debugfs_create_dir(dir_name,
581                                                  kvm_debugfs_dir);
582         if (!kvm->debugfs_dentry)
583                 return -ENOMEM;
584 
585         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
586                                          sizeof(*kvm->debugfs_stat_data),
587                                          GFP_KERNEL);
588         if (!kvm->debugfs_stat_data)
589                 return -ENOMEM;
590 
591         for (p = debugfs_entries; p->name; p++) {
592                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
593                 if (!stat_data)
594                         return -ENOMEM;
595 
596                 stat_data->kvm = kvm;
597                 stat_data->offset = p->offset;
598                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
599                 if (!debugfs_create_file(p->name, 0644,
600                                          kvm->debugfs_dentry,
601                                          stat_data,
602                                          stat_fops_per_vm[p->kind]))
603                         return -ENOMEM;
604         }
605         return 0;
606 }
607 
608 static struct kvm *kvm_create_vm(unsigned long type)
609 {
610         int r, i;
611         struct kvm *kvm = kvm_arch_alloc_vm();
612 
613         if (!kvm)
614                 return ERR_PTR(-ENOMEM);
615 
616         spin_lock_init(&kvm->mmu_lock);
617         mmgrab(current->mm);
618         kvm->mm = current->mm;
619         kvm_eventfd_init(kvm);
620         mutex_init(&kvm->lock);
621         mutex_init(&kvm->irq_lock);
622         mutex_init(&kvm->slots_lock);
623         refcount_set(&kvm->users_count, 1);
624         INIT_LIST_HEAD(&kvm->devices);
625 
626         r = kvm_arch_init_vm(kvm, type);
627         if (r)
628                 goto out_err_no_disable;
629 
630         r = hardware_enable_all();
631         if (r)
632                 goto out_err_no_disable;
633 
634 #ifdef CONFIG_HAVE_KVM_IRQFD
635         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
636 #endif
637 
638         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
639 
640         r = -ENOMEM;
641         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
642                 struct kvm_memslots *slots = kvm_alloc_memslots();
643                 if (!slots)
644                         goto out_err_no_srcu;
645                 /*
646                  * Generations must be different for each address space.
647                  * Init kvm generation close to the maximum to easily test the
648                  * code of handling generation number wrap-around.
649                  */
650                 slots->generation = i * 2 - 150;
651                 rcu_assign_pointer(kvm->memslots[i], slots);
652         }
653 
654         if (init_srcu_struct(&kvm->srcu))
655                 goto out_err_no_srcu;
656         if (init_srcu_struct(&kvm->irq_srcu))
657                 goto out_err_no_irq_srcu;
658         for (i = 0; i < KVM_NR_BUSES; i++) {
659                 rcu_assign_pointer(kvm->buses[i],
660                         kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
661                 if (!kvm->buses[i])
662                         goto out_err;
663         }
664 
665         r = kvm_init_mmu_notifier(kvm);
666         if (r)
667                 goto out_err;
668 
669         spin_lock(&kvm_lock);
670         list_add(&kvm->vm_list, &vm_list);
671         spin_unlock(&kvm_lock);
672 
673         preempt_notifier_inc();
674 
675         return kvm;
676 
677 out_err:
678         cleanup_srcu_struct(&kvm->irq_srcu);
679 out_err_no_irq_srcu:
680         cleanup_srcu_struct(&kvm->srcu);
681 out_err_no_srcu:
682         hardware_disable_all();
683 out_err_no_disable:
684         refcount_set(&kvm->users_count, 0);
685         for (i = 0; i < KVM_NR_BUSES; i++)
686                 kfree(kvm_get_bus(kvm, i));
687         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
688                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
689         kvm_arch_free_vm(kvm);
690         mmdrop(current->mm);
691         return ERR_PTR(r);
692 }
693 
694 static void kvm_destroy_devices(struct kvm *kvm)
695 {
696         struct kvm_device *dev, *tmp;
697 
698         /*
699          * We do not need to take the kvm->lock here, because nobody else
700          * has a reference to the struct kvm at this point and therefore
701          * cannot access the devices list anyhow.
702          */
703         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
704                 list_del(&dev->vm_node);
705                 dev->ops->destroy(dev);
706         }
707 }
708 
709 static void kvm_destroy_vm(struct kvm *kvm)
710 {
711         int i;
712         struct mm_struct *mm = kvm->mm;
713 
714         kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
715         kvm_destroy_vm_debugfs(kvm);
716         kvm_arch_sync_events(kvm);
717         spin_lock(&kvm_lock);
718         list_del(&kvm->vm_list);
719         spin_unlock(&kvm_lock);
720         kvm_free_irq_routing(kvm);
721         for (i = 0; i < KVM_NR_BUSES; i++) {
722                 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
723 
724                 if (bus)
725                         kvm_io_bus_destroy(bus);
726                 kvm->buses[i] = NULL;
727         }
728         kvm_coalesced_mmio_free(kvm);
729 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
730         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
731 #else
732         kvm_arch_flush_shadow_all(kvm);
733 #endif
734         kvm_arch_destroy_vm(kvm);
735         kvm_destroy_devices(kvm);
736         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
737                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
738         cleanup_srcu_struct(&kvm->irq_srcu);
739         cleanup_srcu_struct(&kvm->srcu);
740         kvm_arch_free_vm(kvm);
741         preempt_notifier_dec();
742         hardware_disable_all();
743         mmdrop(mm);
744 }
745 
746 void kvm_get_kvm(struct kvm *kvm)
747 {
748         refcount_inc(&kvm->users_count);
749 }
750 EXPORT_SYMBOL_GPL(kvm_get_kvm);
751 
752 void kvm_put_kvm(struct kvm *kvm)
753 {
754         if (refcount_dec_and_test(&kvm->users_count))
755                 kvm_destroy_vm(kvm);
756 }
757 EXPORT_SYMBOL_GPL(kvm_put_kvm);
758 
759 
760 static int kvm_vm_release(struct inode *inode, struct file *filp)
761 {
762         struct kvm *kvm = filp->private_data;
763 
764         kvm_irqfd_release(kvm);
765 
766         kvm_put_kvm(kvm);
767         return 0;
768 }
769 
770 /*
771  * Allocation size is twice as large as the actual dirty bitmap size.
772  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
773  */
774 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
775 {
776         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
777 
778         memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
779         if (!memslot->dirty_bitmap)
780                 return -ENOMEM;
781 
782         return 0;
783 }
784 
785 /*
786  * Insert memslot and re-sort memslots based on their GFN,
787  * so binary search could be used to lookup GFN.
788  * Sorting algorithm takes advantage of having initially
789  * sorted array and known changed memslot position.
790  */
791 static void update_memslots(struct kvm_memslots *slots,
792                             struct kvm_memory_slot *new)
793 {
794         int id = new->id;
795         int i = slots->id_to_index[id];
796         struct kvm_memory_slot *mslots = slots->memslots;
797 
798         WARN_ON(mslots[i].id != id);
799         if (!new->npages) {
800                 WARN_ON(!mslots[i].npages);
801                 if (mslots[i].npages)
802                         slots->used_slots--;
803         } else {
804                 if (!mslots[i].npages)
805                         slots->used_slots++;
806         }
807 
808         while (i < KVM_MEM_SLOTS_NUM - 1 &&
809                new->base_gfn <= mslots[i + 1].base_gfn) {
810                 if (!mslots[i + 1].npages)
811                         break;
812                 mslots[i] = mslots[i + 1];
813                 slots->id_to_index[mslots[i].id] = i;
814                 i++;
815         }
816 
817         /*
818          * The ">=" is needed when creating a slot with base_gfn == 0,
819          * so that it moves before all those with base_gfn == npages == 0.
820          *
821          * On the other hand, if new->npages is zero, the above loop has
822          * already left i pointing to the beginning of the empty part of
823          * mslots, and the ">=" would move the hole backwards in this
824          * case---which is wrong.  So skip the loop when deleting a slot.
825          */
826         if (new->npages) {
827                 while (i > 0 &&
828                        new->base_gfn >= mslots[i - 1].base_gfn) {
829                         mslots[i] = mslots[i - 1];
830                         slots->id_to_index[mslots[i].id] = i;
831                         i--;
832                 }
833         } else
834                 WARN_ON_ONCE(i != slots->used_slots);
835 
836         mslots[i] = *new;
837         slots->id_to_index[mslots[i].id] = i;
838 }
839 
840 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
841 {
842         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
843 
844 #ifdef __KVM_HAVE_READONLY_MEM
845         valid_flags |= KVM_MEM_READONLY;
846 #endif
847 
848         if (mem->flags & ~valid_flags)
849                 return -EINVAL;
850 
851         return 0;
852 }
853 
854 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
855                 int as_id, struct kvm_memslots *slots)
856 {
857         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
858 
859         /*
860          * Set the low bit in the generation, which disables SPTE caching
861          * until the end of synchronize_srcu_expedited.
862          */
863         WARN_ON(old_memslots->generation & 1);
864         slots->generation = old_memslots->generation + 1;
865 
866         rcu_assign_pointer(kvm->memslots[as_id], slots);
867         synchronize_srcu_expedited(&kvm->srcu);
868 
869         /*
870          * Increment the new memslot generation a second time. This prevents
871          * vm exits that race with memslot updates from caching a memslot
872          * generation that will (potentially) be valid forever.
873          *
874          * Generations must be unique even across address spaces.  We do not need
875          * a global counter for that, instead the generation space is evenly split
876          * across address spaces.  For example, with two address spaces, address
877          * space 0 will use generations 0, 4, 8, ... while * address space 1 will
878          * use generations 2, 6, 10, 14, ...
879          */
880         slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
881 
882         kvm_arch_memslots_updated(kvm, slots);
883 
884         return old_memslots;
885 }
886 
887 /*
888  * Allocate some memory and give it an address in the guest physical address
889  * space.
890  *
891  * Discontiguous memory is allowed, mostly for framebuffers.
892  *
893  * Must be called holding kvm->slots_lock for write.
894  */
895 int __kvm_set_memory_region(struct kvm *kvm,
896                             const struct kvm_userspace_memory_region *mem)
897 {
898         int r;
899         gfn_t base_gfn;
900         unsigned long npages;
901         struct kvm_memory_slot *slot;
902         struct kvm_memory_slot old, new;
903         struct kvm_memslots *slots = NULL, *old_memslots;
904         int as_id, id;
905         enum kvm_mr_change change;
906 
907         r = check_memory_region_flags(mem);
908         if (r)
909                 goto out;
910 
911         r = -EINVAL;
912         as_id = mem->slot >> 16;
913         id = (u16)mem->slot;
914 
915         /* General sanity checks */
916         if (mem->memory_size & (PAGE_SIZE - 1))
917                 goto out;
918         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
919                 goto out;
920         /* We can read the guest memory with __xxx_user() later on. */
921         if ((id < KVM_USER_MEM_SLOTS) &&
922             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
923              !access_ok(VERIFY_WRITE,
924                         (void __user *)(unsigned long)mem->userspace_addr,
925                         mem->memory_size)))
926                 goto out;
927         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
928                 goto out;
929         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
930                 goto out;
931 
932         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
933         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
934         npages = mem->memory_size >> PAGE_SHIFT;
935 
936         if (npages > KVM_MEM_MAX_NR_PAGES)
937                 goto out;
938 
939         new = old = *slot;
940 
941         new.id = id;
942         new.base_gfn = base_gfn;
943         new.npages = npages;
944         new.flags = mem->flags;
945 
946         if (npages) {
947                 if (!old.npages)
948                         change = KVM_MR_CREATE;
949                 else { /* Modify an existing slot. */
950                         if ((mem->userspace_addr != old.userspace_addr) ||
951                             (npages != old.npages) ||
952                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
953                                 goto out;
954 
955                         if (base_gfn != old.base_gfn)
956                                 change = KVM_MR_MOVE;
957                         else if (new.flags != old.flags)
958                                 change = KVM_MR_FLAGS_ONLY;
959                         else { /* Nothing to change. */
960                                 r = 0;
961                                 goto out;
962                         }
963                 }
964         } else {
965                 if (!old.npages)
966                         goto out;
967 
968                 change = KVM_MR_DELETE;
969                 new.base_gfn = 0;
970                 new.flags = 0;
971         }
972 
973         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
974                 /* Check for overlaps */
975                 r = -EEXIST;
976                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
977                         if (slot->id == id)
978                                 continue;
979                         if (!((base_gfn + npages <= slot->base_gfn) ||
980                               (base_gfn >= slot->base_gfn + slot->npages)))
981                                 goto out;
982                 }
983         }
984 
985         /* Free page dirty bitmap if unneeded */
986         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
987                 new.dirty_bitmap = NULL;
988 
989         r = -ENOMEM;
990         if (change == KVM_MR_CREATE) {
991                 new.userspace_addr = mem->userspace_addr;
992 
993                 if (kvm_arch_create_memslot(kvm, &new, npages))
994                         goto out_free;
995         }
996 
997         /* Allocate page dirty bitmap if needed */
998         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
999                 if (kvm_create_dirty_bitmap(&new) < 0)
1000                         goto out_free;
1001         }
1002 
1003         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1004         if (!slots)
1005                 goto out_free;
1006         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1007 
1008         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1009                 slot = id_to_memslot(slots, id);
1010                 slot->flags |= KVM_MEMSLOT_INVALID;
1011 
1012                 old_memslots = install_new_memslots(kvm, as_id, slots);
1013 
1014                 /* From this point no new shadow pages pointing to a deleted,
1015                  * or moved, memslot will be created.
1016                  *
1017                  * validation of sp->gfn happens in:
1018                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1019                  *      - kvm_is_visible_gfn (mmu_check_roots)
1020                  */
1021                 kvm_arch_flush_shadow_memslot(kvm, slot);
1022 
1023                 /*
1024                  * We can re-use the old_memslots from above, the only difference
1025                  * from the currently installed memslots is the invalid flag.  This
1026                  * will get overwritten by update_memslots anyway.
1027                  */
1028                 slots = old_memslots;
1029         }
1030 
1031         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1032         if (r)
1033                 goto out_slots;
1034 
1035         /* actual memory is freed via old in kvm_free_memslot below */
1036         if (change == KVM_MR_DELETE) {
1037                 new.dirty_bitmap = NULL;
1038                 memset(&new.arch, 0, sizeof(new.arch));
1039         }
1040 
1041         update_memslots(slots, &new);
1042         old_memslots = install_new_memslots(kvm, as_id, slots);
1043 
1044         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1045 
1046         kvm_free_memslot(kvm, &old, &new);
1047         kvfree(old_memslots);
1048         return 0;
1049 
1050 out_slots:
1051         kvfree(slots);
1052 out_free:
1053         kvm_free_memslot(kvm, &new, &old);
1054 out:
1055         return r;
1056 }
1057 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1058 
1059 int kvm_set_memory_region(struct kvm *kvm,
1060                           const struct kvm_userspace_memory_region *mem)
1061 {
1062         int r;
1063 
1064         mutex_lock(&kvm->slots_lock);
1065         r = __kvm_set_memory_region(kvm, mem);
1066         mutex_unlock(&kvm->slots_lock);
1067         return r;
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1070 
1071 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1072                                           struct kvm_userspace_memory_region *mem)
1073 {
1074         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1075                 return -EINVAL;
1076 
1077         return kvm_set_memory_region(kvm, mem);
1078 }
1079 
1080 int kvm_get_dirty_log(struct kvm *kvm,
1081                         struct kvm_dirty_log *log, int *is_dirty)
1082 {
1083         struct kvm_memslots *slots;
1084         struct kvm_memory_slot *memslot;
1085         int i, as_id, id;
1086         unsigned long n;
1087         unsigned long any = 0;
1088 
1089         as_id = log->slot >> 16;
1090         id = (u16)log->slot;
1091         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1092                 return -EINVAL;
1093 
1094         slots = __kvm_memslots(kvm, as_id);
1095         memslot = id_to_memslot(slots, id);
1096         if (!memslot->dirty_bitmap)
1097                 return -ENOENT;
1098 
1099         n = kvm_dirty_bitmap_bytes(memslot);
1100 
1101         for (i = 0; !any && i < n/sizeof(long); ++i)
1102                 any = memslot->dirty_bitmap[i];
1103 
1104         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1105                 return -EFAULT;
1106 
1107         if (any)
1108                 *is_dirty = 1;
1109         return 0;
1110 }
1111 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1112 
1113 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1114 /**
1115  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1116  *      are dirty write protect them for next write.
1117  * @kvm:        pointer to kvm instance
1118  * @log:        slot id and address to which we copy the log
1119  * @is_dirty:   flag set if any page is dirty
1120  *
1121  * We need to keep it in mind that VCPU threads can write to the bitmap
1122  * concurrently. So, to avoid losing track of dirty pages we keep the
1123  * following order:
1124  *
1125  *    1. Take a snapshot of the bit and clear it if needed.
1126  *    2. Write protect the corresponding page.
1127  *    3. Copy the snapshot to the userspace.
1128  *    4. Upon return caller flushes TLB's if needed.
1129  *
1130  * Between 2 and 4, the guest may write to the page using the remaining TLB
1131  * entry.  This is not a problem because the page is reported dirty using
1132  * the snapshot taken before and step 4 ensures that writes done after
1133  * exiting to userspace will be logged for the next call.
1134  *
1135  */
1136 int kvm_get_dirty_log_protect(struct kvm *kvm,
1137                         struct kvm_dirty_log *log, bool *is_dirty)
1138 {
1139         struct kvm_memslots *slots;
1140         struct kvm_memory_slot *memslot;
1141         int i, as_id, id;
1142         unsigned long n;
1143         unsigned long *dirty_bitmap;
1144         unsigned long *dirty_bitmap_buffer;
1145 
1146         as_id = log->slot >> 16;
1147         id = (u16)log->slot;
1148         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1149                 return -EINVAL;
1150 
1151         slots = __kvm_memslots(kvm, as_id);
1152         memslot = id_to_memslot(slots, id);
1153 
1154         dirty_bitmap = memslot->dirty_bitmap;
1155         if (!dirty_bitmap)
1156                 return -ENOENT;
1157 
1158         n = kvm_dirty_bitmap_bytes(memslot);
1159 
1160         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1161         memset(dirty_bitmap_buffer, 0, n);
1162 
1163         spin_lock(&kvm->mmu_lock);
1164         *is_dirty = false;
1165         for (i = 0; i < n / sizeof(long); i++) {
1166                 unsigned long mask;
1167                 gfn_t offset;
1168 
1169                 if (!dirty_bitmap[i])
1170                         continue;
1171 
1172                 *is_dirty = true;
1173 
1174                 mask = xchg(&dirty_bitmap[i], 0);
1175                 dirty_bitmap_buffer[i] = mask;
1176 
1177                 if (mask) {
1178                         offset = i * BITS_PER_LONG;
1179                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1180                                                                 offset, mask);
1181                 }
1182         }
1183 
1184         spin_unlock(&kvm->mmu_lock);
1185         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1186                 return -EFAULT;
1187         return 0;
1188 }
1189 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1190 #endif
1191 
1192 bool kvm_largepages_enabled(void)
1193 {
1194         return largepages_enabled;
1195 }
1196 
1197 void kvm_disable_largepages(void)
1198 {
1199         largepages_enabled = false;
1200 }
1201 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1202 
1203 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1204 {
1205         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1206 }
1207 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1208 
1209 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1210 {
1211         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1212 }
1213 
1214 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1215 {
1216         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1217 
1218         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1219               memslot->flags & KVM_MEMSLOT_INVALID)
1220                 return false;
1221 
1222         return true;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1225 
1226 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1227 {
1228         struct vm_area_struct *vma;
1229         unsigned long addr, size;
1230 
1231         size = PAGE_SIZE;
1232 
1233         addr = gfn_to_hva(kvm, gfn);
1234         if (kvm_is_error_hva(addr))
1235                 return PAGE_SIZE;
1236 
1237         down_read(&current->mm->mmap_sem);
1238         vma = find_vma(current->mm, addr);
1239         if (!vma)
1240                 goto out;
1241 
1242         size = vma_kernel_pagesize(vma);
1243 
1244 out:
1245         up_read(&current->mm->mmap_sem);
1246 
1247         return size;
1248 }
1249 
1250 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1251 {
1252         return slot->flags & KVM_MEM_READONLY;
1253 }
1254 
1255 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1256                                        gfn_t *nr_pages, bool write)
1257 {
1258         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1259                 return KVM_HVA_ERR_BAD;
1260 
1261         if (memslot_is_readonly(slot) && write)
1262                 return KVM_HVA_ERR_RO_BAD;
1263 
1264         if (nr_pages)
1265                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1266 
1267         return __gfn_to_hva_memslot(slot, gfn);
1268 }
1269 
1270 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1271                                      gfn_t *nr_pages)
1272 {
1273         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1274 }
1275 
1276 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1277                                         gfn_t gfn)
1278 {
1279         return gfn_to_hva_many(slot, gfn, NULL);
1280 }
1281 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1282 
1283 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1284 {
1285         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1286 }
1287 EXPORT_SYMBOL_GPL(gfn_to_hva);
1288 
1289 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1290 {
1291         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1292 }
1293 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1294 
1295 /*
1296  * If writable is set to false, the hva returned by this function is only
1297  * allowed to be read.
1298  */
1299 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1300                                       gfn_t gfn, bool *writable)
1301 {
1302         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1303 
1304         if (!kvm_is_error_hva(hva) && writable)
1305                 *writable = !memslot_is_readonly(slot);
1306 
1307         return hva;
1308 }
1309 
1310 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1311 {
1312         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1313 
1314         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1315 }
1316 
1317 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1318 {
1319         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1320 
1321         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1322 }
1323 
1324 static int get_user_page_nowait(unsigned long start, int write,
1325                 struct page **page)
1326 {
1327         int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1328 
1329         if (write)
1330                 flags |= FOLL_WRITE;
1331 
1332         return get_user_pages(start, 1, flags, page, NULL);
1333 }
1334 
1335 static inline int check_user_page_hwpoison(unsigned long addr)
1336 {
1337         int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1338 
1339         rc = get_user_pages(addr, 1, flags, NULL, NULL);
1340         return rc == -EHWPOISON;
1341 }
1342 
1343 /*
1344  * The atomic path to get the writable pfn which will be stored in @pfn,
1345  * true indicates success, otherwise false is returned.
1346  */
1347 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1348                             bool write_fault, bool *writable, kvm_pfn_t *pfn)
1349 {
1350         struct page *page[1];
1351         int npages;
1352 
1353         if (!(async || atomic))
1354                 return false;
1355 
1356         /*
1357          * Fast pin a writable pfn only if it is a write fault request
1358          * or the caller allows to map a writable pfn for a read fault
1359          * request.
1360          */
1361         if (!(write_fault || writable))
1362                 return false;
1363 
1364         npages = __get_user_pages_fast(addr, 1, 1, page);
1365         if (npages == 1) {
1366                 *pfn = page_to_pfn(page[0]);
1367 
1368                 if (writable)
1369                         *writable = true;
1370                 return true;
1371         }
1372 
1373         return false;
1374 }
1375 
1376 /*
1377  * The slow path to get the pfn of the specified host virtual address,
1378  * 1 indicates success, -errno is returned if error is detected.
1379  */
1380 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1381                            bool *writable, kvm_pfn_t *pfn)
1382 {
1383         struct page *page[1];
1384         int npages = 0;
1385 
1386         might_sleep();
1387 
1388         if (writable)
1389                 *writable = write_fault;
1390 
1391         if (async) {
1392                 down_read(&current->mm->mmap_sem);
1393                 npages = get_user_page_nowait(addr, write_fault, page);
1394                 up_read(&current->mm->mmap_sem);
1395         } else {
1396                 unsigned int flags = FOLL_HWPOISON;
1397 
1398                 if (write_fault)
1399                         flags |= FOLL_WRITE;
1400 
1401                 npages = get_user_pages_unlocked(addr, 1, page, flags);
1402         }
1403         if (npages != 1)
1404                 return npages;
1405 
1406         /* map read fault as writable if possible */
1407         if (unlikely(!write_fault) && writable) {
1408                 struct page *wpage[1];
1409 
1410                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1411                 if (npages == 1) {
1412                         *writable = true;
1413                         put_page(page[0]);
1414                         page[0] = wpage[0];
1415                 }
1416 
1417                 npages = 1;
1418         }
1419         *pfn = page_to_pfn(page[0]);
1420         return npages;
1421 }
1422 
1423 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1424 {
1425         if (unlikely(!(vma->vm_flags & VM_READ)))
1426                 return false;
1427 
1428         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1429                 return false;
1430 
1431         return true;
1432 }
1433 
1434 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1435                                unsigned long addr, bool *async,
1436                                bool write_fault, kvm_pfn_t *p_pfn)
1437 {
1438         unsigned long pfn;
1439         int r;
1440 
1441         r = follow_pfn(vma, addr, &pfn);
1442         if (r) {
1443                 /*
1444                  * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1445                  * not call the fault handler, so do it here.
1446                  */
1447                 bool unlocked = false;
1448                 r = fixup_user_fault(current, current->mm, addr,
1449                                      (write_fault ? FAULT_FLAG_WRITE : 0),
1450                                      &unlocked);
1451                 if (unlocked)
1452                         return -EAGAIN;
1453                 if (r)
1454                         return r;
1455 
1456                 r = follow_pfn(vma, addr, &pfn);
1457                 if (r)
1458                         return r;
1459 
1460         }
1461 
1462 
1463         /*
1464          * Get a reference here because callers of *hva_to_pfn* and
1465          * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1466          * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1467          * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1468          * simply do nothing for reserved pfns.
1469          *
1470          * Whoever called remap_pfn_range is also going to call e.g.
1471          * unmap_mapping_range before the underlying pages are freed,
1472          * causing a call to our MMU notifier.
1473          */ 
1474         kvm_get_pfn(pfn);
1475 
1476         *p_pfn = pfn;
1477         return 0;
1478 }
1479 
1480 /*
1481  * Pin guest page in memory and return its pfn.
1482  * @addr: host virtual address which maps memory to the guest
1483  * @atomic: whether this function can sleep
1484  * @async: whether this function need to wait IO complete if the
1485  *         host page is not in the memory
1486  * @write_fault: whether we should get a writable host page
1487  * @writable: whether it allows to map a writable host page for !@write_fault
1488  *
1489  * The function will map a writable host page for these two cases:
1490  * 1): @write_fault = true
1491  * 2): @write_fault = false && @writable, @writable will tell the caller
1492  *     whether the mapping is writable.
1493  */
1494 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1495                         bool write_fault, bool *writable)
1496 {
1497         struct vm_area_struct *vma;
1498         kvm_pfn_t pfn = 0;
1499         int npages, r;
1500 
1501         /* we can do it either atomically or asynchronously, not both */
1502         BUG_ON(atomic && async);
1503 
1504         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1505                 return pfn;
1506 
1507         if (atomic)
1508                 return KVM_PFN_ERR_FAULT;
1509 
1510         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1511         if (npages == 1)
1512                 return pfn;
1513 
1514         down_read(&current->mm->mmap_sem);
1515         if (npages == -EHWPOISON ||
1516               (!async && check_user_page_hwpoison(addr))) {
1517                 pfn = KVM_PFN_ERR_HWPOISON;
1518                 goto exit;
1519         }
1520 
1521 retry:
1522         vma = find_vma_intersection(current->mm, addr, addr + 1);
1523 
1524         if (vma == NULL)
1525                 pfn = KVM_PFN_ERR_FAULT;
1526         else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1527                 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1528                 if (r == -EAGAIN)
1529                         goto retry;
1530                 if (r < 0)
1531                         pfn = KVM_PFN_ERR_FAULT;
1532         } else {
1533                 if (async && vma_is_valid(vma, write_fault))
1534                         *async = true;
1535                 pfn = KVM_PFN_ERR_FAULT;
1536         }
1537 exit:
1538         up_read(&current->mm->mmap_sem);
1539         return pfn;
1540 }
1541 
1542 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1543                                bool atomic, bool *async, bool write_fault,
1544                                bool *writable)
1545 {
1546         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1547 
1548         if (addr == KVM_HVA_ERR_RO_BAD) {
1549                 if (writable)
1550                         *writable = false;
1551                 return KVM_PFN_ERR_RO_FAULT;
1552         }
1553 
1554         if (kvm_is_error_hva(addr)) {
1555                 if (writable)
1556                         *writable = false;
1557                 return KVM_PFN_NOSLOT;
1558         }
1559 
1560         /* Do not map writable pfn in the readonly memslot. */
1561         if (writable && memslot_is_readonly(slot)) {
1562                 *writable = false;
1563                 writable = NULL;
1564         }
1565 
1566         return hva_to_pfn(addr, atomic, async, write_fault,
1567                           writable);
1568 }
1569 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1570 
1571 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1572                       bool *writable)
1573 {
1574         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1575                                     write_fault, writable);
1576 }
1577 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1578 
1579 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1580 {
1581         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1582 }
1583 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1584 
1585 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1586 {
1587         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1588 }
1589 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1590 
1591 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1592 {
1593         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1594 }
1595 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1596 
1597 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1598 {
1599         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1600 }
1601 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1602 
1603 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1604 {
1605         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1606 }
1607 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1608 
1609 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1610 {
1611         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1612 }
1613 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1614 
1615 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1616                             struct page **pages, int nr_pages)
1617 {
1618         unsigned long addr;
1619         gfn_t entry = 0;
1620 
1621         addr = gfn_to_hva_many(slot, gfn, &entry);
1622         if (kvm_is_error_hva(addr))
1623                 return -1;
1624 
1625         if (entry < nr_pages)
1626                 return 0;
1627 
1628         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1629 }
1630 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1631 
1632 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1633 {
1634         if (is_error_noslot_pfn(pfn))
1635                 return KVM_ERR_PTR_BAD_PAGE;
1636 
1637         if (kvm_is_reserved_pfn(pfn)) {
1638                 WARN_ON(1);
1639                 return KVM_ERR_PTR_BAD_PAGE;
1640         }
1641 
1642         return pfn_to_page(pfn);
1643 }
1644 
1645 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1646 {
1647         kvm_pfn_t pfn;
1648 
1649         pfn = gfn_to_pfn(kvm, gfn);
1650 
1651         return kvm_pfn_to_page(pfn);
1652 }
1653 EXPORT_SYMBOL_GPL(gfn_to_page);
1654 
1655 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1656 {
1657         kvm_pfn_t pfn;
1658 
1659         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1660 
1661         return kvm_pfn_to_page(pfn);
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1664 
1665 void kvm_release_page_clean(struct page *page)
1666 {
1667         WARN_ON(is_error_page(page));
1668 
1669         kvm_release_pfn_clean(page_to_pfn(page));
1670 }
1671 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1672 
1673 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1674 {
1675         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1676                 put_page(pfn_to_page(pfn));
1677 }
1678 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1679 
1680 void kvm_release_page_dirty(struct page *page)
1681 {
1682         WARN_ON(is_error_page(page));
1683 
1684         kvm_release_pfn_dirty(page_to_pfn(page));
1685 }
1686 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1687 
1688 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1689 {
1690         kvm_set_pfn_dirty(pfn);
1691         kvm_release_pfn_clean(pfn);
1692 }
1693 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1694 
1695 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1696 {
1697         if (!kvm_is_reserved_pfn(pfn)) {
1698                 struct page *page = pfn_to_page(pfn);
1699 
1700                 if (!PageReserved(page))
1701                         SetPageDirty(page);
1702         }
1703 }
1704 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1705 
1706 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1707 {
1708         if (!kvm_is_reserved_pfn(pfn))
1709                 mark_page_accessed(pfn_to_page(pfn));
1710 }
1711 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1712 
1713 void kvm_get_pfn(kvm_pfn_t pfn)
1714 {
1715         if (!kvm_is_reserved_pfn(pfn))
1716                 get_page(pfn_to_page(pfn));
1717 }
1718 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1719 
1720 static int next_segment(unsigned long len, int offset)
1721 {
1722         if (len > PAGE_SIZE - offset)
1723                 return PAGE_SIZE - offset;
1724         else
1725                 return len;
1726 }
1727 
1728 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1729                                  void *data, int offset, int len)
1730 {
1731         int r;
1732         unsigned long addr;
1733 
1734         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1735         if (kvm_is_error_hva(addr))
1736                 return -EFAULT;
1737         r = __copy_from_user(data, (void __user *)addr + offset, len);
1738         if (r)
1739                 return -EFAULT;
1740         return 0;
1741 }
1742 
1743 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1744                         int len)
1745 {
1746         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1747 
1748         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1749 }
1750 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1751 
1752 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1753                              int offset, int len)
1754 {
1755         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1756 
1757         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1758 }
1759 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1760 
1761 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1762 {
1763         gfn_t gfn = gpa >> PAGE_SHIFT;
1764         int seg;
1765         int offset = offset_in_page(gpa);
1766         int ret;
1767 
1768         while ((seg = next_segment(len, offset)) != 0) {
1769                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1770                 if (ret < 0)
1771                         return ret;
1772                 offset = 0;
1773                 len -= seg;
1774                 data += seg;
1775                 ++gfn;
1776         }
1777         return 0;
1778 }
1779 EXPORT_SYMBOL_GPL(kvm_read_guest);
1780 
1781 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1782 {
1783         gfn_t gfn = gpa >> PAGE_SHIFT;
1784         int seg;
1785         int offset = offset_in_page(gpa);
1786         int ret;
1787 
1788         while ((seg = next_segment(len, offset)) != 0) {
1789                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1790                 if (ret < 0)
1791                         return ret;
1792                 offset = 0;
1793                 len -= seg;
1794                 data += seg;
1795                 ++gfn;
1796         }
1797         return 0;
1798 }
1799 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1800 
1801 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1802                                    void *data, int offset, unsigned long len)
1803 {
1804         int r;
1805         unsigned long addr;
1806 
1807         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1808         if (kvm_is_error_hva(addr))
1809                 return -EFAULT;
1810         pagefault_disable();
1811         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1812         pagefault_enable();
1813         if (r)
1814                 return -EFAULT;
1815         return 0;
1816 }
1817 
1818 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1819                           unsigned long len)
1820 {
1821         gfn_t gfn = gpa >> PAGE_SHIFT;
1822         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1823         int offset = offset_in_page(gpa);
1824 
1825         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1826 }
1827 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1828 
1829 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1830                                void *data, unsigned long len)
1831 {
1832         gfn_t gfn = gpa >> PAGE_SHIFT;
1833         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1834         int offset = offset_in_page(gpa);
1835 
1836         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1837 }
1838 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1839 
1840 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1841                                   const void *data, int offset, int len)
1842 {
1843         int r;
1844         unsigned long addr;
1845 
1846         addr = gfn_to_hva_memslot(memslot, gfn);
1847         if (kvm_is_error_hva(addr))
1848                 return -EFAULT;
1849         r = __copy_to_user((void __user *)addr + offset, data, len);
1850         if (r)
1851                 return -EFAULT;
1852         mark_page_dirty_in_slot(memslot, gfn);
1853         return 0;
1854 }
1855 
1856 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1857                          const void *data, int offset, int len)
1858 {
1859         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1860 
1861         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1862 }
1863 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1864 
1865 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1866                               const void *data, int offset, int len)
1867 {
1868         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1869 
1870         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1871 }
1872 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1873 
1874 int kvm_write_guest(struct kvm *kvm, 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_write_guest_page(kvm, 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_write_guest);
1894 
1895 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1896                          unsigned long len)
1897 {
1898         gfn_t gfn = gpa >> PAGE_SHIFT;
1899         int seg;
1900         int offset = offset_in_page(gpa);
1901         int ret;
1902 
1903         while ((seg = next_segment(len, offset)) != 0) {
1904                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1905                 if (ret < 0)
1906                         return ret;
1907                 offset = 0;
1908                 len -= seg;
1909                 data += seg;
1910                 ++gfn;
1911         }
1912         return 0;
1913 }
1914 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1915 
1916 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1917                                        struct gfn_to_hva_cache *ghc,
1918                                        gpa_t gpa, unsigned long len)
1919 {
1920         int offset = offset_in_page(gpa);
1921         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1922         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1923         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1924         gfn_t nr_pages_avail;
1925 
1926         ghc->gpa = gpa;
1927         ghc->generation = slots->generation;
1928         ghc->len = len;
1929         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1930         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1931         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1932                 ghc->hva += offset;
1933         } else {
1934                 /*
1935                  * If the requested region crosses two memslots, we still
1936                  * verify that the entire region is valid here.
1937                  */
1938                 while (start_gfn <= end_gfn) {
1939                         nr_pages_avail = 0;
1940                         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1941                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1942                                                    &nr_pages_avail);
1943                         if (kvm_is_error_hva(ghc->hva))
1944                                 return -EFAULT;
1945                         start_gfn += nr_pages_avail;
1946                 }
1947                 /* Use the slow path for cross page reads and writes. */
1948                 ghc->memslot = NULL;
1949         }
1950         return 0;
1951 }
1952 
1953 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1954                               gpa_t gpa, unsigned long len)
1955 {
1956         struct kvm_memslots *slots = kvm_memslots(kvm);
1957         return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1958 }
1959 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1960 
1961 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1962                            void *data, int offset, unsigned long len)
1963 {
1964         struct kvm_memslots *slots = kvm_memslots(kvm);
1965         int r;
1966         gpa_t gpa = ghc->gpa + offset;
1967 
1968         BUG_ON(len + offset > ghc->len);
1969 
1970         if (slots->generation != ghc->generation)
1971                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1972 
1973         if (unlikely(!ghc->memslot))
1974                 return kvm_write_guest(kvm, gpa, data, len);
1975 
1976         if (kvm_is_error_hva(ghc->hva))
1977                 return -EFAULT;
1978 
1979         r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1980         if (r)
1981                 return -EFAULT;
1982         mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1983 
1984         return 0;
1985 }
1986 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1987 
1988 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1989                            void *data, unsigned long len)
1990 {
1991         return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1992 }
1993 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1994 
1995 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1996                            void *data, unsigned long len)
1997 {
1998         struct kvm_memslots *slots = kvm_memslots(kvm);
1999         int r;
2000 
2001         BUG_ON(len > ghc->len);
2002 
2003         if (slots->generation != ghc->generation)
2004                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2005 
2006         if (unlikely(!ghc->memslot))
2007                 return kvm_read_guest(kvm, ghc->gpa, data, len);
2008 
2009         if (kvm_is_error_hva(ghc->hva))
2010                 return -EFAULT;
2011 
2012         r = __copy_from_user(data, (void __user *)ghc->hva, len);
2013         if (r)
2014                 return -EFAULT;
2015 
2016         return 0;
2017 }
2018 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2019 
2020 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2021 {
2022         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2023 
2024         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2025 }
2026 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2027 
2028 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2029 {
2030         gfn_t gfn = gpa >> PAGE_SHIFT;
2031         int seg;
2032         int offset = offset_in_page(gpa);
2033         int ret;
2034 
2035         while ((seg = next_segment(len, offset)) != 0) {
2036                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2037                 if (ret < 0)
2038                         return ret;
2039                 offset = 0;
2040                 len -= seg;
2041                 ++gfn;
2042         }
2043         return 0;
2044 }
2045 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2046 
2047 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2048                                     gfn_t gfn)
2049 {
2050         if (memslot && memslot->dirty_bitmap) {
2051                 unsigned long rel_gfn = gfn - memslot->base_gfn;
2052 
2053                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2054         }
2055 }
2056 
2057 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2058 {
2059         struct kvm_memory_slot *memslot;
2060 
2061         memslot = gfn_to_memslot(kvm, gfn);
2062         mark_page_dirty_in_slot(memslot, gfn);
2063 }
2064 EXPORT_SYMBOL_GPL(mark_page_dirty);
2065 
2066 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2067 {
2068         struct kvm_memory_slot *memslot;
2069 
2070         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2071         mark_page_dirty_in_slot(memslot, gfn);
2072 }
2073 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2074 
2075 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2076 {
2077         if (!vcpu->sigset_active)
2078                 return;
2079 
2080         /*
2081          * This does a lockless modification of ->real_blocked, which is fine
2082          * because, only current can change ->real_blocked and all readers of
2083          * ->real_blocked don't care as long ->real_blocked is always a subset
2084          * of ->blocked.
2085          */
2086         sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
2087 }
2088 
2089 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2090 {
2091         if (!vcpu->sigset_active)
2092                 return;
2093 
2094         sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
2095         sigemptyset(&current->real_blocked);
2096 }
2097 
2098 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2099 {
2100         unsigned int old, val, grow;
2101 
2102         old = val = vcpu->halt_poll_ns;
2103         grow = READ_ONCE(halt_poll_ns_grow);
2104         /* 10us base */
2105         if (val == 0 && grow)
2106                 val = 10000;
2107         else
2108                 val *= grow;
2109 
2110         if (val > halt_poll_ns)
2111                 val = halt_poll_ns;
2112 
2113         vcpu->halt_poll_ns = val;
2114         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2115 }
2116 
2117 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2118 {
2119         unsigned int old, val, shrink;
2120 
2121         old = val = vcpu->halt_poll_ns;
2122         shrink = READ_ONCE(halt_poll_ns_shrink);
2123         if (shrink == 0)
2124                 val = 0;
2125         else
2126                 val /= shrink;
2127 
2128         vcpu->halt_poll_ns = val;
2129         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2130 }
2131 
2132 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2133 {
2134         if (kvm_arch_vcpu_runnable(vcpu)) {
2135                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2136                 return -EINTR;
2137         }
2138         if (kvm_cpu_has_pending_timer(vcpu))
2139                 return -EINTR;
2140         if (signal_pending(current))
2141                 return -EINTR;
2142 
2143         return 0;
2144 }
2145 
2146 /*
2147  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2148  */
2149 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2150 {
2151         ktime_t start, cur;
2152         DECLARE_SWAITQUEUE(wait);
2153         bool waited = false;
2154         u64 block_ns;
2155 
2156         start = cur = ktime_get();
2157         if (vcpu->halt_poll_ns) {
2158                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2159 
2160                 ++vcpu->stat.halt_attempted_poll;
2161                 do {
2162                         /*
2163                          * This sets KVM_REQ_UNHALT if an interrupt
2164                          * arrives.
2165                          */
2166                         if (kvm_vcpu_check_block(vcpu) < 0) {
2167                                 ++vcpu->stat.halt_successful_poll;
2168                                 if (!vcpu_valid_wakeup(vcpu))
2169                                         ++vcpu->stat.halt_poll_invalid;
2170                                 goto out;
2171                         }
2172                         cur = ktime_get();
2173                 } while (single_task_running() && ktime_before(cur, stop));
2174         }
2175 
2176         kvm_arch_vcpu_blocking(vcpu);
2177 
2178         for (;;) {
2179                 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2180 
2181                 if (kvm_vcpu_check_block(vcpu) < 0)
2182                         break;
2183 
2184                 waited = true;
2185                 schedule();
2186         }
2187 
2188         finish_swait(&vcpu->wq, &wait);
2189         cur = ktime_get();
2190 
2191         kvm_arch_vcpu_unblocking(vcpu);
2192 out:
2193         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2194 
2195         if (!vcpu_valid_wakeup(vcpu))
2196                 shrink_halt_poll_ns(vcpu);
2197         else if (halt_poll_ns) {
2198                 if (block_ns <= vcpu->halt_poll_ns)
2199                         ;
2200                 /* we had a long block, shrink polling */
2201                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2202                         shrink_halt_poll_ns(vcpu);
2203                 /* we had a short halt and our poll time is too small */
2204                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2205                         block_ns < halt_poll_ns)
2206                         grow_halt_poll_ns(vcpu);
2207         } else
2208                 vcpu->halt_poll_ns = 0;
2209 
2210         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2211         kvm_arch_vcpu_block_finish(vcpu);
2212 }
2213 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2214 
2215 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2216 {
2217         struct swait_queue_head *wqp;
2218 
2219         wqp = kvm_arch_vcpu_wq(vcpu);
2220         if (swq_has_sleeper(wqp)) {
2221                 swake_up(wqp);
2222                 ++vcpu->stat.halt_wakeup;
2223                 return true;
2224         }
2225 
2226         return false;
2227 }
2228 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2229 
2230 #ifndef CONFIG_S390
2231 /*
2232  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2233  */
2234 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2235 {
2236         int me;
2237         int cpu = vcpu->cpu;
2238 
2239         if (kvm_vcpu_wake_up(vcpu))
2240                 return;
2241 
2242         me = get_cpu();
2243         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2244                 if (kvm_arch_vcpu_should_kick(vcpu))
2245                         smp_send_reschedule(cpu);
2246         put_cpu();
2247 }
2248 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2249 #endif /* !CONFIG_S390 */
2250 
2251 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2252 {
2253         struct pid *pid;
2254         struct task_struct *task = NULL;
2255         int ret = 0;
2256 
2257         rcu_read_lock();
2258         pid = rcu_dereference(target->pid);
2259         if (pid)
2260                 task = get_pid_task(pid, PIDTYPE_PID);
2261         rcu_read_unlock();
2262         if (!task)
2263                 return ret;
2264         ret = yield_to(task, 1);
2265         put_task_struct(task);
2266 
2267         return ret;
2268 }
2269 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2270 
2271 /*
2272  * Helper that checks whether a VCPU is eligible for directed yield.
2273  * Most eligible candidate to yield is decided by following heuristics:
2274  *
2275  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2276  *  (preempted lock holder), indicated by @in_spin_loop.
2277  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2278  *
2279  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2280  *  chance last time (mostly it has become eligible now since we have probably
2281  *  yielded to lockholder in last iteration. This is done by toggling
2282  *  @dy_eligible each time a VCPU checked for eligibility.)
2283  *
2284  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2285  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2286  *  burning. Giving priority for a potential lock-holder increases lock
2287  *  progress.
2288  *
2289  *  Since algorithm is based on heuristics, accessing another VCPU data without
2290  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2291  *  and continue with next VCPU and so on.
2292  */
2293 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2294 {
2295 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2296         bool eligible;
2297 
2298         eligible = !vcpu->spin_loop.in_spin_loop ||
2299                     vcpu->spin_loop.dy_eligible;
2300 
2301         if (vcpu->spin_loop.in_spin_loop)
2302                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2303 
2304         return eligible;
2305 #else
2306         return true;
2307 #endif
2308 }
2309 
2310 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2311 {
2312         struct kvm *kvm = me->kvm;
2313         struct kvm_vcpu *vcpu;
2314         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2315         int yielded = 0;
2316         int try = 3;
2317         int pass;
2318         int i;
2319 
2320         kvm_vcpu_set_in_spin_loop(me, true);
2321         /*
2322          * We boost the priority of a VCPU that is runnable but not
2323          * currently running, because it got preempted by something
2324          * else and called schedule in __vcpu_run.  Hopefully that
2325          * VCPU is holding the lock that we need and will release it.
2326          * We approximate round-robin by starting at the last boosted VCPU.
2327          */
2328         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2329                 kvm_for_each_vcpu(i, vcpu, kvm) {
2330                         if (!pass && i <= last_boosted_vcpu) {
2331                                 i = last_boosted_vcpu;
2332                                 continue;
2333                         } else if (pass && i > last_boosted_vcpu)
2334                                 break;
2335                         if (!READ_ONCE(vcpu->preempted))
2336                                 continue;
2337                         if (vcpu == me)
2338                                 continue;
2339                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2340                                 continue;
2341                         if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2342                                 continue;
2343                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2344                                 continue;
2345 
2346                         yielded = kvm_vcpu_yield_to(vcpu);
2347                         if (yielded > 0) {
2348                                 kvm->last_boosted_vcpu = i;
2349                                 break;
2350                         } else if (yielded < 0) {
2351                                 try--;
2352                                 if (!try)
2353                                         break;
2354                         }
2355                 }
2356         }
2357         kvm_vcpu_set_in_spin_loop(me, false);
2358 
2359         /* Ensure vcpu is not eligible during next spinloop */
2360         kvm_vcpu_set_dy_eligible(me, false);
2361 }
2362 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2363 
2364 static int kvm_vcpu_fault(struct vm_fault *vmf)
2365 {
2366         struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2367         struct page *page;
2368 
2369         if (vmf->pgoff == 0)
2370                 page = virt_to_page(vcpu->run);
2371 #ifdef CONFIG_X86
2372         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2373                 page = virt_to_page(vcpu->arch.pio_data);
2374 #endif
2375 #ifdef CONFIG_KVM_MMIO
2376         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2377                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2378 #endif
2379         else
2380                 return kvm_arch_vcpu_fault(vcpu, vmf);
2381         get_page(page);
2382         vmf->page = page;
2383         return 0;
2384 }
2385 
2386 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2387         .fault = kvm_vcpu_fault,
2388 };
2389 
2390 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2391 {
2392         vma->vm_ops = &kvm_vcpu_vm_ops;
2393         return 0;
2394 }
2395 
2396 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2397 {
2398         struct kvm_vcpu *vcpu = filp->private_data;
2399 
2400         debugfs_remove_recursive(vcpu->debugfs_dentry);
2401         kvm_put_kvm(vcpu->kvm);
2402         return 0;
2403 }
2404 
2405 static struct file_operations kvm_vcpu_fops = {
2406         .release        = kvm_vcpu_release,
2407         .unlocked_ioctl = kvm_vcpu_ioctl,
2408 #ifdef CONFIG_KVM_COMPAT
2409         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2410 #endif
2411         .mmap           = kvm_vcpu_mmap,
2412         .llseek         = noop_llseek,
2413 };
2414 
2415 /*
2416  * Allocates an inode for the vcpu.
2417  */
2418 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2419 {
2420         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2421 }
2422 
2423 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2424 {
2425         char dir_name[ITOA_MAX_LEN * 2];
2426         int ret;
2427 
2428         if (!kvm_arch_has_vcpu_debugfs())
2429                 return 0;
2430 
2431         if (!debugfs_initialized())
2432                 return 0;
2433 
2434         snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2435         vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2436                                                                 vcpu->kvm->debugfs_dentry);
2437         if (!vcpu->debugfs_dentry)
2438                 return -ENOMEM;
2439 
2440         ret = kvm_arch_create_vcpu_debugfs(vcpu);
2441         if (ret < 0) {
2442                 debugfs_remove_recursive(vcpu->debugfs_dentry);
2443                 return ret;
2444         }
2445 
2446         return 0;
2447 }
2448 
2449 /*
2450  * Creates some virtual cpus.  Good luck creating more than one.
2451  */
2452 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2453 {
2454         int r;
2455         struct kvm_vcpu *vcpu;
2456 
2457         if (id >= KVM_MAX_VCPU_ID)
2458                 return -EINVAL;
2459 
2460         mutex_lock(&kvm->lock);
2461         if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2462                 mutex_unlock(&kvm->lock);
2463                 return -EINVAL;
2464         }
2465 
2466         kvm->created_vcpus++;
2467         mutex_unlock(&kvm->lock);
2468 
2469         vcpu = kvm_arch_vcpu_create(kvm, id);
2470         if (IS_ERR(vcpu)) {
2471                 r = PTR_ERR(vcpu);
2472                 goto vcpu_decrement;
2473         }
2474 
2475         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2476 
2477         r = kvm_arch_vcpu_setup(vcpu);
2478         if (r)
2479                 goto vcpu_destroy;
2480 
2481         r = kvm_create_vcpu_debugfs(vcpu);
2482         if (r)
2483                 goto vcpu_destroy;
2484 
2485         mutex_lock(&kvm->lock);
2486         if (kvm_get_vcpu_by_id(kvm, id)) {
2487                 r = -EEXIST;
2488                 goto unlock_vcpu_destroy;
2489         }
2490 
2491         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2492 
2493         /* Now it's all set up, let userspace reach it */
2494         kvm_get_kvm(kvm);
2495         r = create_vcpu_fd(vcpu);
2496         if (r < 0) {
2497                 kvm_put_kvm(kvm);
2498                 goto unlock_vcpu_destroy;
2499         }
2500 
2501         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2502 
2503         /*
2504          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2505          * before kvm->online_vcpu's incremented value.
2506          */
2507         smp_wmb();
2508         atomic_inc(&kvm->online_vcpus);
2509 
2510         mutex_unlock(&kvm->lock);
2511         kvm_arch_vcpu_postcreate(vcpu);
2512         return r;
2513 
2514 unlock_vcpu_destroy:
2515         mutex_unlock(&kvm->lock);
2516         debugfs_remove_recursive(vcpu->debugfs_dentry);
2517 vcpu_destroy:
2518         kvm_arch_vcpu_destroy(vcpu);
2519 vcpu_decrement:
2520         mutex_lock(&kvm->lock);
2521         kvm->created_vcpus--;
2522         mutex_unlock(&kvm->lock);
2523         return r;
2524 }
2525 
2526 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2527 {
2528         if (sigset) {
2529                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2530                 vcpu->sigset_active = 1;
2531                 vcpu->sigset = *sigset;
2532         } else
2533                 vcpu->sigset_active = 0;
2534         return 0;
2535 }
2536 
2537 static long kvm_vcpu_ioctl(struct file *filp,
2538                            unsigned int ioctl, unsigned long arg)
2539 {
2540         struct kvm_vcpu *vcpu = filp->private_data;
2541         void __user *argp = (void __user *)arg;
2542         int r;
2543         struct kvm_fpu *fpu = NULL;
2544         struct kvm_sregs *kvm_sregs = NULL;
2545 
2546         if (vcpu->kvm->mm != current->mm)
2547                 return -EIO;
2548 
2549         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2550                 return -EINVAL;
2551 
2552 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2553         /*
2554          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2555          * so vcpu_load() would break it.
2556          */
2557         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2558                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2559 #endif
2560 
2561 
2562         r = vcpu_load(vcpu);
2563         if (r)
2564                 return r;
2565         switch (ioctl) {
2566         case KVM_RUN: {
2567                 struct pid *oldpid;
2568                 r = -EINVAL;
2569                 if (arg)
2570                         goto out;
2571                 oldpid = rcu_access_pointer(vcpu->pid);
2572                 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2573                         /* The thread running this VCPU changed. */
2574                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2575 
2576                         rcu_assign_pointer(vcpu->pid, newpid);
2577                         if (oldpid)
2578                                 synchronize_rcu();
2579                         put_pid(oldpid);
2580                 }
2581                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2582                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2583                 break;
2584         }
2585         case KVM_GET_REGS: {
2586                 struct kvm_regs *kvm_regs;
2587 
2588                 r = -ENOMEM;
2589                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2590                 if (!kvm_regs)
2591                         goto out;
2592                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2593                 if (r)
2594                         goto out_free1;
2595                 r = -EFAULT;
2596                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2597                         goto out_free1;
2598                 r = 0;
2599 out_free1:
2600                 kfree(kvm_regs);
2601                 break;
2602         }
2603         case KVM_SET_REGS: {
2604                 struct kvm_regs *kvm_regs;
2605 
2606                 r = -ENOMEM;
2607                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2608                 if (IS_ERR(kvm_regs)) {
2609                         r = PTR_ERR(kvm_regs);
2610                         goto out;
2611                 }
2612                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2613                 kfree(kvm_regs);
2614                 break;
2615         }
2616         case KVM_GET_SREGS: {
2617                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2618                 r = -ENOMEM;
2619                 if (!kvm_sregs)
2620                         goto out;
2621                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2622                 if (r)
2623                         goto out;
2624                 r = -EFAULT;
2625                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2626                         goto out;
2627                 r = 0;
2628                 break;
2629         }
2630         case KVM_SET_SREGS: {
2631                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2632                 if (IS_ERR(kvm_sregs)) {
2633                         r = PTR_ERR(kvm_sregs);
2634                         kvm_sregs = NULL;
2635                         goto out;
2636                 }
2637                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2638                 break;
2639         }
2640         case KVM_GET_MP_STATE: {
2641                 struct kvm_mp_state mp_state;
2642 
2643                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2644                 if (r)
2645                         goto out;
2646                 r = -EFAULT;
2647                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2648                         goto out;
2649                 r = 0;
2650                 break;
2651         }
2652         case KVM_SET_MP_STATE: {
2653                 struct kvm_mp_state mp_state;
2654 
2655                 r = -EFAULT;
2656                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2657                         goto out;
2658                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2659                 break;
2660         }
2661         case KVM_TRANSLATE: {
2662                 struct kvm_translation tr;
2663 
2664                 r = -EFAULT;
2665                 if (copy_from_user(&tr, argp, sizeof(tr)))
2666                         goto out;
2667                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2668                 if (r)
2669                         goto out;
2670                 r = -EFAULT;
2671                 if (copy_to_user(argp, &tr, sizeof(tr)))
2672                         goto out;
2673                 r = 0;
2674                 break;
2675         }
2676         case KVM_SET_GUEST_DEBUG: {
2677                 struct kvm_guest_debug dbg;
2678 
2679                 r = -EFAULT;
2680                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2681                         goto out;
2682                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2683                 break;
2684         }
2685         case KVM_SET_SIGNAL_MASK: {
2686                 struct kvm_signal_mask __user *sigmask_arg = argp;
2687                 struct kvm_signal_mask kvm_sigmask;
2688                 sigset_t sigset, *p;
2689 
2690                 p = NULL;
2691                 if (argp) {
2692                         r = -EFAULT;
2693                         if (copy_from_user(&kvm_sigmask, argp,
2694                                            sizeof(kvm_sigmask)))
2695                                 goto out;
2696                         r = -EINVAL;
2697                         if (kvm_sigmask.len != sizeof(sigset))
2698                                 goto out;
2699                         r = -EFAULT;
2700                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2701                                            sizeof(sigset)))
2702                                 goto out;
2703                         p = &sigset;
2704                 }
2705                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2706                 break;
2707         }
2708         case KVM_GET_FPU: {
2709                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2710                 r = -ENOMEM;
2711                 if (!fpu)
2712                         goto out;
2713                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2714                 if (r)
2715                         goto out;
2716                 r = -EFAULT;
2717                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2718                         goto out;
2719                 r = 0;
2720                 break;
2721         }
2722         case KVM_SET_FPU: {
2723                 fpu = memdup_user(argp, sizeof(*fpu));
2724                 if (IS_ERR(fpu)) {
2725                         r = PTR_ERR(fpu);
2726                         fpu = NULL;
2727                         goto out;
2728                 }
2729                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2730                 break;
2731         }
2732         default:
2733                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2734         }
2735 out:
2736         vcpu_put(vcpu);
2737         kfree(fpu);
2738         kfree(kvm_sregs);
2739         return r;
2740 }
2741 
2742 #ifdef CONFIG_KVM_COMPAT
2743 static long kvm_vcpu_compat_ioctl(struct file *filp,
2744                                   unsigned int ioctl, unsigned long arg)
2745 {
2746         struct kvm_vcpu *vcpu = filp->private_data;
2747         void __user *argp = compat_ptr(arg);
2748         int r;
2749 
2750         if (vcpu->kvm->mm != current->mm)
2751                 return -EIO;
2752 
2753         switch (ioctl) {
2754         case KVM_SET_SIGNAL_MASK: {
2755                 struct kvm_signal_mask __user *sigmask_arg = argp;
2756                 struct kvm_signal_mask kvm_sigmask;
2757                 sigset_t sigset;
2758 
2759                 if (argp) {
2760                         r = -EFAULT;
2761                         if (copy_from_user(&kvm_sigmask, argp,
2762                                            sizeof(kvm_sigmask)))
2763                                 goto out;
2764                         r = -EINVAL;
2765                         if (kvm_sigmask.len != sizeof(compat_sigset_t))
2766                                 goto out;
2767                         r = -EFAULT;
2768                         if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2769                                 goto out;
2770                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2771                 } else
2772                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2773                 break;
2774         }
2775         default:
2776                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2777         }
2778 
2779 out:
2780         return r;
2781 }
2782 #endif
2783 
2784 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2785                                  int (*accessor)(struct kvm_device *dev,
2786                                                  struct kvm_device_attr *attr),
2787                                  unsigned long arg)
2788 {
2789         struct kvm_device_attr attr;
2790 
2791         if (!accessor)
2792                 return -EPERM;
2793 
2794         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2795                 return -EFAULT;
2796 
2797         return accessor(dev, &attr);
2798 }
2799 
2800 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2801                              unsigned long arg)
2802 {
2803         struct kvm_device *dev = filp->private_data;
2804 
2805         switch (ioctl) {
2806         case KVM_SET_DEVICE_ATTR:
2807                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2808         case KVM_GET_DEVICE_ATTR:
2809                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2810         case KVM_HAS_DEVICE_ATTR:
2811                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2812         default:
2813                 if (dev->ops->ioctl)
2814                         return dev->ops->ioctl(dev, ioctl, arg);
2815 
2816                 return -ENOTTY;
2817         }
2818 }
2819 
2820 static int kvm_device_release(struct inode *inode, struct file *filp)
2821 {
2822         struct kvm_device *dev = filp->private_data;
2823         struct kvm *kvm = dev->kvm;
2824 
2825         kvm_put_kvm(kvm);
2826         return 0;
2827 }
2828 
2829 static const struct file_operations kvm_device_fops = {
2830         .unlocked_ioctl = kvm_device_ioctl,
2831 #ifdef CONFIG_KVM_COMPAT
2832         .compat_ioctl = kvm_device_ioctl,
2833 #endif
2834         .release = kvm_device_release,
2835 };
2836 
2837 struct kvm_device *kvm_device_from_filp(struct file *filp)
2838 {
2839         if (filp->f_op != &kvm_device_fops)
2840                 return NULL;
2841 
2842         return filp->private_data;
2843 }
2844 
2845 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2846 #ifdef CONFIG_KVM_MPIC
2847         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2848         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2849 #endif
2850 };
2851 
2852 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2853 {
2854         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2855                 return -ENOSPC;
2856 
2857         if (kvm_device_ops_table[type] != NULL)
2858                 return -EEXIST;
2859 
2860         kvm_device_ops_table[type] = ops;
2861         return 0;
2862 }
2863 
2864 void kvm_unregister_device_ops(u32 type)
2865 {
2866         if (kvm_device_ops_table[type] != NULL)
2867                 kvm_device_ops_table[type] = NULL;
2868 }
2869 
2870 static int kvm_ioctl_create_device(struct kvm *kvm,
2871                                    struct kvm_create_device *cd)
2872 {
2873         struct kvm_device_ops *ops = NULL;
2874         struct kvm_device *dev;
2875         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2876         int ret;
2877 
2878         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2879                 return -ENODEV;
2880 
2881         ops = kvm_device_ops_table[cd->type];
2882         if (ops == NULL)
2883                 return -ENODEV;
2884 
2885         if (test)
2886                 return 0;
2887 
2888         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2889         if (!dev)
2890                 return -ENOMEM;
2891 
2892         dev->ops = ops;
2893         dev->kvm = kvm;
2894 
2895         mutex_lock(&kvm->lock);
2896         ret = ops->create(dev, cd->type);
2897         if (ret < 0) {
2898                 mutex_unlock(&kvm->lock);
2899                 kfree(dev);
2900                 return ret;
2901         }
2902         list_add(&dev->vm_node, &kvm->devices);
2903         mutex_unlock(&kvm->lock);
2904 
2905         if (ops->init)
2906                 ops->init(dev);
2907 
2908         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2909         if (ret < 0) {
2910                 mutex_lock(&kvm->lock);
2911                 list_del(&dev->vm_node);
2912                 mutex_unlock(&kvm->lock);
2913                 ops->destroy(dev);
2914                 return ret;
2915         }
2916 
2917         kvm_get_kvm(kvm);
2918         cd->fd = ret;
2919         return 0;
2920 }
2921 
2922 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2923 {
2924         switch (arg) {
2925         case KVM_CAP_USER_MEMORY:
2926         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2927         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2928         case KVM_CAP_INTERNAL_ERROR_DATA:
2929 #ifdef CONFIG_HAVE_KVM_MSI
2930         case KVM_CAP_SIGNAL_MSI:
2931 #endif
2932 #ifdef CONFIG_HAVE_KVM_IRQFD
2933         case KVM_CAP_IRQFD:
2934         case KVM_CAP_IRQFD_RESAMPLE:
2935 #endif
2936         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2937         case KVM_CAP_CHECK_EXTENSION_VM:
2938                 return 1;
2939 #ifdef CONFIG_KVM_MMIO
2940         case KVM_CAP_COALESCED_MMIO:
2941                 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2942 #endif
2943 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2944         case KVM_CAP_IRQ_ROUTING:
2945                 return KVM_MAX_IRQ_ROUTES;
2946 #endif
2947 #if KVM_ADDRESS_SPACE_NUM > 1
2948         case KVM_CAP_MULTI_ADDRESS_SPACE:
2949                 return KVM_ADDRESS_SPACE_NUM;
2950 #endif
2951         case KVM_CAP_MAX_VCPU_ID:
2952                 return KVM_MAX_VCPU_ID;
2953         default:
2954                 break;
2955         }
2956         return kvm_vm_ioctl_check_extension(kvm, arg);
2957 }
2958 
2959 static long kvm_vm_ioctl(struct file *filp,
2960                            unsigned int ioctl, unsigned long arg)
2961 {
2962         struct kvm *kvm = filp->private_data;
2963         void __user *argp = (void __user *)arg;
2964         int r;
2965 
2966         if (kvm->mm != current->mm)
2967                 return -EIO;
2968         switch (ioctl) {
2969         case KVM_CREATE_VCPU:
2970                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2971                 break;
2972         case KVM_SET_USER_MEMORY_REGION: {
2973                 struct kvm_userspace_memory_region kvm_userspace_mem;
2974 
2975                 r = -EFAULT;
2976                 if (copy_from_user(&kvm_userspace_mem, argp,
2977                                                 sizeof(kvm_userspace_mem)))
2978                         goto out;
2979 
2980                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2981                 break;
2982         }
2983         case KVM_GET_DIRTY_LOG: {
2984                 struct kvm_dirty_log log;
2985 
2986                 r = -EFAULT;
2987                 if (copy_from_user(&log, argp, sizeof(log)))
2988                         goto out;
2989                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2990                 break;
2991         }
2992 #ifdef CONFIG_KVM_MMIO
2993         case KVM_REGISTER_COALESCED_MMIO: {
2994                 struct kvm_coalesced_mmio_zone zone;
2995 
2996                 r = -EFAULT;
2997                 if (copy_from_user(&zone, argp, sizeof(zone)))
2998                         goto out;
2999                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3000                 break;
3001         }
3002         case KVM_UNREGISTER_COALESCED_MMIO: {
3003                 struct kvm_coalesced_mmio_zone zone;
3004 
3005                 r = -EFAULT;
3006                 if (copy_from_user(&zone, argp, sizeof(zone)))
3007                         goto out;
3008                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3009                 break;
3010         }
3011 #endif
3012         case KVM_IRQFD: {
3013                 struct kvm_irqfd data;
3014 
3015                 r = -EFAULT;
3016                 if (copy_from_user(&data, argp, sizeof(data)))
3017                         goto out;
3018                 r = kvm_irqfd(kvm, &data);
3019                 break;
3020         }
3021         case KVM_IOEVENTFD: {
3022                 struct kvm_ioeventfd data;
3023 
3024                 r = -EFAULT;
3025                 if (copy_from_user(&data, argp, sizeof(data)))
3026                         goto out;
3027                 r = kvm_ioeventfd(kvm, &data);
3028                 break;
3029         }
3030 #ifdef CONFIG_HAVE_KVM_MSI
3031         case KVM_SIGNAL_MSI: {
3032                 struct kvm_msi msi;
3033 
3034                 r = -EFAULT;
3035                 if (copy_from_user(&msi, argp, sizeof(msi)))
3036                         goto out;
3037                 r = kvm_send_userspace_msi(kvm, &msi);
3038                 break;
3039         }
3040 #endif
3041 #ifdef __KVM_HAVE_IRQ_LINE
3042         case KVM_IRQ_LINE_STATUS:
3043         case KVM_IRQ_LINE: {
3044                 struct kvm_irq_level irq_event;
3045 
3046                 r = -EFAULT;
3047                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3048                         goto out;
3049 
3050                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3051                                         ioctl == KVM_IRQ_LINE_STATUS);
3052                 if (r)
3053                         goto out;
3054 
3055                 r = -EFAULT;
3056                 if (ioctl == KVM_IRQ_LINE_STATUS) {
3057                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3058                                 goto out;
3059                 }
3060 
3061                 r = 0;
3062                 break;
3063         }
3064 #endif
3065 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3066         case KVM_SET_GSI_ROUTING: {
3067                 struct kvm_irq_routing routing;
3068                 struct kvm_irq_routing __user *urouting;
3069                 struct kvm_irq_routing_entry *entries = NULL;
3070 
3071                 r = -EFAULT;
3072                 if (copy_from_user(&routing, argp, sizeof(routing)))
3073                         goto out;
3074                 r = -EINVAL;
3075                 if (!kvm_arch_can_set_irq_routing(kvm))
3076                         goto out;
3077                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3078                         goto out;
3079                 if (routing.flags)
3080                         goto out;
3081                 if (routing.nr) {
3082                         r = -ENOMEM;
3083                         entries = vmalloc(routing.nr * sizeof(*entries));
3084                         if (!entries)
3085                                 goto out;
3086                         r = -EFAULT;
3087                         urouting = argp;
3088                         if (copy_from_user(entries, urouting->entries,
3089                                            routing.nr * sizeof(*entries)))
3090                                 goto out_free_irq_routing;
3091                 }
3092                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3093                                         routing.flags);
3094 out_free_irq_routing:
3095                 vfree(entries);
3096                 break;
3097         }
3098 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3099         case KVM_CREATE_DEVICE: {
3100                 struct kvm_create_device cd;
3101 
3102                 r = -EFAULT;
3103                 if (copy_from_user(&cd, argp, sizeof(cd)))
3104                         goto out;
3105 
3106                 r = kvm_ioctl_create_device(kvm, &cd);
3107                 if (r)
3108                         goto out;
3109 
3110                 r = -EFAULT;
3111                 if (copy_to_user(argp, &cd, sizeof(cd)))
3112                         goto out;
3113 
3114                 r = 0;
3115                 break;
3116         }
3117         case KVM_CHECK_EXTENSION:
3118                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3119                 break;
3120         default:
3121                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3122         }
3123 out:
3124         return r;
3125 }
3126 
3127 #ifdef CONFIG_KVM_COMPAT
3128 struct compat_kvm_dirty_log {
3129         __u32 slot;
3130         __u32 padding1;
3131         union {
3132                 compat_uptr_t dirty_bitmap; /* one bit per page */
3133                 __u64 padding2;
3134         };
3135 };
3136 
3137 static long kvm_vm_compat_ioctl(struct file *filp,
3138                            unsigned int ioctl, unsigned long arg)
3139 {
3140         struct kvm *kvm = filp->private_data;
3141         int r;
3142 
3143         if (kvm->mm != current->mm)
3144                 return -EIO;
3145         switch (ioctl) {
3146         case KVM_GET_DIRTY_LOG: {
3147                 struct compat_kvm_dirty_log compat_log;
3148                 struct kvm_dirty_log log;
3149 
3150                 if (copy_from_user(&compat_log, (void __user *)arg,
3151                                    sizeof(compat_log)))
3152                         return -EFAULT;
3153                 log.slot         = compat_log.slot;
3154                 log.padding1     = compat_log.padding1;
3155                 log.padding2     = compat_log.padding2;
3156                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3157 
3158                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3159                 break;
3160         }
3161         default:
3162                 r = kvm_vm_ioctl(filp, ioctl, arg);
3163         }
3164         return r;
3165 }
3166 #endif
3167 
3168 static struct file_operations kvm_vm_fops = {
3169         .release        = kvm_vm_release,
3170         .unlocked_ioctl = kvm_vm_ioctl,
3171 #ifdef CONFIG_KVM_COMPAT
3172         .compat_ioctl   = kvm_vm_compat_ioctl,
3173 #endif
3174         .llseek         = noop_llseek,
3175 };
3176 
3177 static int kvm_dev_ioctl_create_vm(unsigned long type)
3178 {
3179         int r;
3180         struct kvm *kvm;
3181         struct file *file;
3182 
3183         kvm = kvm_create_vm(type);
3184         if (IS_ERR(kvm))
3185                 return PTR_ERR(kvm);
3186 #ifdef CONFIG_KVM_MMIO
3187         r = kvm_coalesced_mmio_init(kvm);
3188         if (r < 0) {
3189                 kvm_put_kvm(kvm);
3190                 return r;
3191         }
3192 #endif
3193         r = get_unused_fd_flags(O_CLOEXEC);
3194         if (r < 0) {
3195                 kvm_put_kvm(kvm);
3196                 return r;
3197         }
3198         file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3199         if (IS_ERR(file)) {
3200                 put_unused_fd(r);
3201                 kvm_put_kvm(kvm);
3202                 return PTR_ERR(file);
3203         }
3204 
3205         /*
3206          * Don't call kvm_put_kvm anymore at this point; file->f_op is
3207          * already set, with ->release() being kvm_vm_release().  In error
3208          * cases it will be called by the final fput(file) and will take
3209          * care of doing kvm_put_kvm(kvm).
3210          */
3211         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3212                 put_unused_fd(r);
3213                 fput(file);
3214                 return -ENOMEM;
3215         }
3216         kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3217 
3218         fd_install(r, file);
3219         return r;
3220 }
3221 
3222 static long kvm_dev_ioctl(struct file *filp,
3223                           unsigned int ioctl, unsigned long arg)
3224 {
3225         long r = -EINVAL;
3226 
3227         switch (ioctl) {
3228         case KVM_GET_API_VERSION:
3229                 if (arg)
3230                         goto out;
3231                 r = KVM_API_VERSION;
3232                 break;
3233         case KVM_CREATE_VM:
3234                 r = kvm_dev_ioctl_create_vm(arg);
3235                 break;
3236         case KVM_CHECK_EXTENSION:
3237                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3238                 break;
3239         case KVM_GET_VCPU_MMAP_SIZE:
3240                 if (arg)
3241                         goto out;
3242                 r = PAGE_SIZE;     /* struct kvm_run */
3243 #ifdef CONFIG_X86
3244                 r += PAGE_SIZE;    /* pio data page */
3245 #endif
3246 #ifdef CONFIG_KVM_MMIO
3247                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3248 #endif
3249                 break;
3250         case KVM_TRACE_ENABLE:
3251         case KVM_TRACE_PAUSE:
3252         case KVM_TRACE_DISABLE:
3253                 r = -EOPNOTSUPP;
3254                 break;
3255         default:
3256                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3257         }
3258 out:
3259         return r;
3260 }
3261 
3262 static struct file_operations kvm_chardev_ops = {
3263         .unlocked_ioctl = kvm_dev_ioctl,
3264         .compat_ioctl   = kvm_dev_ioctl,
3265         .llseek         = noop_llseek,
3266 };
3267 
3268 static struct miscdevice kvm_dev = {
3269         KVM_MINOR,
3270         "kvm",
3271         &kvm_chardev_ops,
3272 };
3273 
3274 static void hardware_enable_nolock(void *junk)
3275 {
3276         int cpu = raw_smp_processor_id();
3277         int r;
3278 
3279         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3280                 return;
3281 
3282         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3283 
3284         r = kvm_arch_hardware_enable();
3285 
3286         if (r) {
3287                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3288                 atomic_inc(&hardware_enable_failed);
3289                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3290         }
3291 }
3292 
3293 static int kvm_starting_cpu(unsigned int cpu)
3294 {
3295         raw_spin_lock(&kvm_count_lock);
3296         if (kvm_usage_count)
3297                 hardware_enable_nolock(NULL);
3298         raw_spin_unlock(&kvm_count_lock);
3299         return 0;
3300 }
3301 
3302 static void hardware_disable_nolock(void *junk)
3303 {
3304         int cpu = raw_smp_processor_id();
3305 
3306         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3307                 return;
3308         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3309         kvm_arch_hardware_disable();
3310 }
3311 
3312 static int kvm_dying_cpu(unsigned int cpu)
3313 {
3314         raw_spin_lock(&kvm_count_lock);
3315         if (kvm_usage_count)
3316                 hardware_disable_nolock(NULL);
3317         raw_spin_unlock(&kvm_count_lock);
3318         return 0;
3319 }
3320 
3321 static void hardware_disable_all_nolock(void)
3322 {
3323         BUG_ON(!kvm_usage_count);
3324 
3325         kvm_usage_count--;
3326         if (!kvm_usage_count)
3327                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3328 }
3329 
3330 static void hardware_disable_all(void)
3331 {
3332         raw_spin_lock(&kvm_count_lock);
3333         hardware_disable_all_nolock();
3334         raw_spin_unlock(&kvm_count_lock);
3335 }
3336 
3337 static int hardware_enable_all(void)
3338 {
3339         int r = 0;
3340 
3341         raw_spin_lock(&kvm_count_lock);
3342 
3343         kvm_usage_count++;
3344         if (kvm_usage_count == 1) {
3345                 atomic_set(&hardware_enable_failed, 0);
3346                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3347 
3348                 if (atomic_read(&hardware_enable_failed)) {
3349                         hardware_disable_all_nolock();
3350                         r = -EBUSY;
3351                 }
3352         }
3353 
3354         raw_spin_unlock(&kvm_count_lock);
3355 
3356         return r;
3357 }
3358 
3359 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3360                       void *v)
3361 {
3362         /*
3363          * Some (well, at least mine) BIOSes hang on reboot if
3364          * in vmx root mode.
3365          *
3366          * And Intel TXT required VMX off for all cpu when system shutdown.
3367          */
3368         pr_info("kvm: exiting hardware virtualization\n");
3369         kvm_rebooting = true;
3370         on_each_cpu(hardware_disable_nolock, NULL, 1);
3371         return NOTIFY_OK;
3372 }
3373 
3374 static struct notifier_block kvm_reboot_notifier = {
3375         .notifier_call = kvm_reboot,
3376         .priority = 0,
3377 };
3378 
3379 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3380 {
3381         int i;
3382 
3383         for (i = 0; i < bus->dev_count; i++) {
3384                 struct kvm_io_device *pos = bus->range[i].dev;
3385 
3386                 kvm_iodevice_destructor(pos);
3387         }
3388         kfree(bus);
3389 }
3390 
3391 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3392                                  const struct kvm_io_range *r2)
3393 {
3394         gpa_t addr1 = r1->addr;
3395         gpa_t addr2 = r2->addr;
3396 
3397         if (addr1 < addr2)
3398                 return -1;
3399 
3400         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3401          * accept any overlapping write.  Any order is acceptable for
3402          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3403          * we process all of them.
3404          */
3405         if (r2->len) {
3406                 addr1 += r1->len;
3407                 addr2 += r2->len;
3408         }
3409 
3410         if (addr1 > addr2)
3411                 return 1;
3412 
3413         return 0;
3414 }
3415 
3416 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3417 {
3418         return kvm_io_bus_cmp(p1, p2);
3419 }
3420 
3421 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3422                           gpa_t addr, int len)
3423 {
3424         bus->range[bus->dev_count++] = (struct kvm_io_range) {
3425                 .addr = addr,
3426                 .len = len,
3427                 .dev = dev,
3428         };
3429 
3430         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3431                 kvm_io_bus_sort_cmp, NULL);
3432 
3433         return 0;
3434 }
3435 
3436 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3437                              gpa_t addr, int len)
3438 {
3439         struct kvm_io_range *range, key;
3440         int off;
3441 
3442         key = (struct kvm_io_range) {
3443                 .addr = addr,
3444                 .len = len,
3445         };
3446 
3447         range = bsearch(&key, bus->range, bus->dev_count,
3448                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3449         if (range == NULL)
3450                 return -ENOENT;
3451 
3452         off = range - bus->range;
3453 
3454         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3455                 off--;
3456 
3457         return off;
3458 }
3459 
3460 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3461                               struct kvm_io_range *range, const void *val)
3462 {
3463         int idx;
3464 
3465         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3466         if (idx < 0)
3467                 return -EOPNOTSUPP;
3468 
3469         while (idx < bus->dev_count &&
3470                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3471                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3472                                         range->len, val))
3473                         return idx;
3474                 idx++;
3475         }
3476 
3477         return -EOPNOTSUPP;
3478 }
3479 
3480 /* kvm_io_bus_write - called under kvm->slots_lock */
3481 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3482                      int len, const void *val)
3483 {
3484         struct kvm_io_bus *bus;
3485         struct kvm_io_range range;
3486         int r;
3487 
3488         range = (struct kvm_io_range) {
3489                 .addr = addr,
3490                 .len = len,
3491         };
3492 
3493         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3494         if (!bus)
3495                 return -ENOMEM;
3496         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3497         return r < 0 ? r : 0;
3498 }
3499 
3500 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3501 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3502                             gpa_t addr, int len, const void *val, long cookie)
3503 {
3504         struct kvm_io_bus *bus;
3505         struct kvm_io_range range;
3506 
3507         range = (struct kvm_io_range) {
3508                 .addr = addr,
3509                 .len = len,
3510         };
3511 
3512         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3513         if (!bus)
3514                 return -ENOMEM;
3515 
3516         /* First try the device referenced by cookie. */
3517         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3518             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3519                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3520                                         val))
3521                         return cookie;
3522 
3523         /*
3524          * cookie contained garbage; fall back to search and return the
3525          * correct cookie value.
3526          */
3527         return __kvm_io_bus_write(vcpu, bus, &range, val);
3528 }
3529 
3530 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3531                              struct kvm_io_range *range, void *val)
3532 {
3533         int idx;
3534 
3535         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3536         if (idx < 0)
3537                 return -EOPNOTSUPP;
3538 
3539         while (idx < bus->dev_count &&
3540                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3541                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3542                                        range->len, val))
3543                         return idx;
3544                 idx++;
3545         }
3546 
3547         return -EOPNOTSUPP;
3548 }
3549 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3550 
3551 /* kvm_io_bus_read - called under kvm->slots_lock */
3552 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3553                     int len, void *val)
3554 {
3555         struct kvm_io_bus *bus;
3556         struct kvm_io_range range;
3557         int r;
3558 
3559         range = (struct kvm_io_range) {
3560                 .addr = addr,
3561                 .len = len,
3562         };
3563 
3564         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3565         if (!bus)
3566                 return -ENOMEM;
3567         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3568         return r < 0 ? r : 0;
3569 }
3570 
3571 
3572 /* Caller must hold slots_lock. */
3573 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3574                             int len, struct kvm_io_device *dev)
3575 {
3576         struct kvm_io_bus *new_bus, *bus;
3577 
3578         bus = kvm_get_bus(kvm, bus_idx);
3579         if (!bus)
3580                 return -ENOMEM;
3581 
3582         /* exclude ioeventfd which is limited by maximum fd */
3583         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3584                 return -ENOSPC;
3585 
3586         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3587                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3588         if (!new_bus)
3589                 return -ENOMEM;
3590         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3591                sizeof(struct kvm_io_range)));
3592         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3593         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3594         synchronize_srcu_expedited(&kvm->srcu);
3595         kfree(bus);
3596 
3597         return 0;
3598 }
3599 
3600 /* Caller must hold slots_lock. */
3601 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3602                                struct kvm_io_device *dev)
3603 {
3604         int i;
3605         struct kvm_io_bus *new_bus, *bus;
3606 
3607         bus = kvm_get_bus(kvm, bus_idx);
3608         if (!bus)
3609                 return;
3610 
3611         for (i = 0; i < bus->dev_count; i++)
3612                 if (bus->range[i].dev == dev) {
3613                         break;
3614                 }
3615 
3616         if (i == bus->dev_count)
3617                 return;
3618 
3619         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3620                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3621         if (!new_bus)  {
3622                 pr_err("kvm: failed to shrink bus, removing it completely\n");
3623                 goto broken;
3624         }
3625 
3626         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3627         new_bus->dev_count--;
3628         memcpy(new_bus->range + i, bus->range + i + 1,
3629                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3630 
3631 broken:
3632         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3633         synchronize_srcu_expedited(&kvm->srcu);
3634         kfree(bus);
3635         return;
3636 }
3637 
3638 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3639                                          gpa_t addr)
3640 {
3641         struct kvm_io_bus *bus;
3642         int dev_idx, srcu_idx;
3643         struct kvm_io_device *iodev = NULL;
3644 
3645         srcu_idx = srcu_read_lock(&kvm->srcu);
3646 
3647         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3648         if (!bus)
3649                 goto out_unlock;
3650 
3651         dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3652         if (dev_idx < 0)
3653                 goto out_unlock;
3654 
3655         iodev = bus->range[dev_idx].dev;
3656 
3657 out_unlock:
3658         srcu_read_unlock(&kvm->srcu, srcu_idx);
3659 
3660         return iodev;
3661 }
3662 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3663 
3664 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3665                            int (*get)(void *, u64 *), int (*set)(void *, u64),
3666                            const char *fmt)
3667 {
3668         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3669                                           inode->i_private;
3670 
3671         /* The debugfs files are a reference to the kvm struct which
3672          * is still valid when kvm_destroy_vm is called.
3673          * To avoid the race between open and the removal of the debugfs
3674          * directory we test against the users count.
3675          */
3676         if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3677                 return -ENOENT;
3678 
3679         if (simple_attr_open(inode, file, get, set, fmt)) {
3680                 kvm_put_kvm(stat_data->kvm);
3681                 return -ENOMEM;
3682         }
3683 
3684         return 0;
3685 }
3686 
3687 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3688 {
3689         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3690                                           inode->i_private;
3691 
3692         simple_attr_release(inode, file);
3693         kvm_put_kvm(stat_data->kvm);
3694 
3695         return 0;
3696 }
3697 
3698 static int vm_stat_get_per_vm(void *data, u64 *val)
3699 {
3700         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3701 
3702         *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3703 
3704         return 0;
3705 }
3706 
3707 static int vm_stat_clear_per_vm(void *data, u64 val)
3708 {
3709         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3710 
3711         if (val)
3712                 return -EINVAL;
3713 
3714         *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3715 
3716         return 0;
3717 }
3718 
3719 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3720 {
3721         __simple_attr_check_format("%llu\n", 0ull);
3722         return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3723                                 vm_stat_clear_per_vm, "%llu\n");
3724 }
3725 
3726 static const struct file_operations vm_stat_get_per_vm_fops = {
3727         .owner   = THIS_MODULE,
3728         .open    = vm_stat_get_per_vm_open,
3729         .release = kvm_debugfs_release,
3730         .read    = simple_attr_read,
3731         .write   = simple_attr_write,
3732         .llseek  = no_llseek,
3733 };
3734 
3735 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3736 {
3737         int i;
3738         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3739         struct kvm_vcpu *vcpu;
3740 
3741         *val = 0;
3742 
3743         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3744                 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3745 
3746         return 0;
3747 }
3748 
3749 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3750 {
3751         int i;
3752         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3753         struct kvm_vcpu *vcpu;
3754 
3755         if (val)
3756                 return -EINVAL;
3757 
3758         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3759                 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3760 
3761         return 0;
3762 }
3763 
3764 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3765 {
3766         __simple_attr_check_format("%llu\n", 0ull);
3767         return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3768                                  vcpu_stat_clear_per_vm, "%llu\n");
3769 }
3770 
3771 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3772         .owner   = THIS_MODULE,
3773         .open    = vcpu_stat_get_per_vm_open,
3774         .release = kvm_debugfs_release,
3775         .read    = simple_attr_read,
3776         .write   = simple_attr_write,
3777         .llseek  = no_llseek,
3778 };
3779 
3780 static const struct file_operations *stat_fops_per_vm[] = {
3781         [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3782         [KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
3783 };
3784 
3785 static int vm_stat_get(void *_offset, u64 *val)
3786 {
3787         unsigned offset = (long)_offset;
3788         struct kvm *kvm;
3789         struct kvm_stat_data stat_tmp = {.offset = offset};
3790         u64 tmp_val;
3791 
3792         *val = 0;
3793         spin_lock(&kvm_lock);
3794         list_for_each_entry(kvm, &vm_list, vm_list) {
3795                 stat_tmp.kvm = kvm;
3796                 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3797                 *val += tmp_val;
3798         }
3799         spin_unlock(&kvm_lock);
3800         return 0;
3801 }
3802 
3803 static int vm_stat_clear(void *_offset, u64 val)
3804 {
3805         unsigned offset = (long)_offset;
3806         struct kvm *kvm;
3807         struct kvm_stat_data stat_tmp = {.offset = offset};
3808 
3809         if (val)
3810                 return -EINVAL;
3811 
3812         spin_lock(&kvm_lock);
3813         list_for_each_entry(kvm, &vm_list, vm_list) {
3814                 stat_tmp.kvm = kvm;
3815                 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3816         }
3817         spin_unlock(&kvm_lock);
3818 
3819         return 0;
3820 }
3821 
3822 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3823 
3824 static int vcpu_stat_get(void *_offset, u64 *val)
3825 {
3826         unsigned offset = (long)_offset;
3827         struct kvm *kvm;
3828         struct kvm_stat_data stat_tmp = {.offset = offset};
3829         u64 tmp_val;
3830 
3831         *val = 0;
3832         spin_lock(&kvm_lock);
3833         list_for_each_entry(kvm, &vm_list, vm_list) {
3834                 stat_tmp.kvm = kvm;
3835                 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3836                 *val += tmp_val;
3837         }
3838         spin_unlock(&kvm_lock);
3839         return 0;
3840 }
3841 
3842 static int vcpu_stat_clear(void *_offset, u64 val)
3843 {
3844         unsigned offset = (long)_offset;
3845         struct kvm *kvm;
3846         struct kvm_stat_data stat_tmp = {.offset = offset};
3847 
3848         if (val)
3849                 return -EINVAL;
3850 
3851         spin_lock(&kvm_lock);
3852         list_for_each_entry(kvm, &vm_list, vm_list) {
3853                 stat_tmp.kvm = kvm;
3854                 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3855         }
3856         spin_unlock(&kvm_lock);
3857 
3858         return 0;
3859 }
3860 
3861 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3862                         "%llu\n");
3863 
3864 static const struct file_operations *stat_fops[] = {
3865         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3866         [KVM_STAT_VM]   = &vm_stat_fops,
3867 };
3868 
3869 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3870 {
3871         struct kobj_uevent_env *env;
3872         unsigned long long created, active;
3873 
3874         if (!kvm_dev.this_device || !kvm)
3875                 return;
3876 
3877         spin_lock(&kvm_lock);
3878         if (type == KVM_EVENT_CREATE_VM) {
3879                 kvm_createvm_count++;
3880                 kvm_active_vms++;
3881         } else if (type == KVM_EVENT_DESTROY_VM) {
3882                 kvm_active_vms--;
3883         }
3884         created = kvm_createvm_count;
3885         active = kvm_active_vms;
3886         spin_unlock(&kvm_lock);
3887 
3888         env = kzalloc(sizeof(*env), GFP_KERNEL);
3889         if (!env)
3890                 return;
3891 
3892         add_uevent_var(env, "CREATED=%llu", created);
3893         add_uevent_var(env, "COUNT=%llu", active);
3894 
3895         if (type == KVM_EVENT_CREATE_VM) {
3896                 add_uevent_var(env, "EVENT=create");
3897                 kvm->userspace_pid = task_pid_nr(current);
3898         } else if (type == KVM_EVENT_DESTROY_VM) {
3899                 add_uevent_var(env, "EVENT=destroy");
3900         }
3901         add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3902 
3903         if (kvm->debugfs_dentry) {
3904                 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3905 
3906                 if (p) {
3907                         tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3908                         if (!IS_ERR(tmp))
3909                                 add_uevent_var(env, "STATS_PATH=%s", tmp);
3910                         kfree(p);
3911                 }
3912         }
3913         /* no need for checks, since we are adding at most only 5 keys */
3914         env->envp[env->envp_idx++] = NULL;
3915         kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3916         kfree(env);
3917 }
3918 
3919 static int kvm_init_debug(void)
3920 {
3921         int r = -EEXIST;
3922         struct kvm_stats_debugfs_item *p;
3923 
3924         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3925         if (kvm_debugfs_dir == NULL)
3926                 goto out;
3927 
3928         kvm_debugfs_num_entries = 0;
3929         for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3930                 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3931                                          (void *)(long)p->offset,
3932                                          stat_fops[p->kind]))
3933                         goto out_dir;
3934         }
3935 
3936         return 0;
3937 
3938 out_dir:
3939         debugfs_remove_recursive(kvm_debugfs_dir);
3940 out:
3941         return r;
3942 }
3943 
3944 static int kvm_suspend(void)
3945 {
3946         if (kvm_usage_count)
3947                 hardware_disable_nolock(NULL);
3948         return 0;
3949 }
3950 
3951 static void kvm_resume(void)
3952 {
3953         if (kvm_usage_count) {
3954                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3955                 hardware_enable_nolock(NULL);
3956         }
3957 }
3958 
3959 static struct syscore_ops kvm_syscore_ops = {
3960         .suspend = kvm_suspend,
3961         .resume = kvm_resume,
3962 };
3963 
3964 static inline
3965 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3966 {
3967         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3968 }
3969 
3970 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3971 {
3972         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3973 
3974         if (vcpu->preempted)
3975                 vcpu->preempted = false;
3976 
3977         kvm_arch_sched_in(vcpu, cpu);
3978 
3979         kvm_arch_vcpu_load(vcpu, cpu);
3980 }
3981 
3982 static void kvm_sched_out(struct preempt_notifier *pn,
3983                           struct task_struct *next)
3984 {
3985         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3986 
3987         if (current->state == TASK_RUNNING)
3988                 vcpu->preempted = true;
3989         kvm_arch_vcpu_put(vcpu);
3990 }
3991 
3992 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3993                   struct module *module)
3994 {
3995         int r;
3996         int cpu;
3997 
3998         r = kvm_arch_init(opaque);
3999         if (r)
4000                 goto out_fail;
4001 
4002         /*
4003          * kvm_arch_init makes sure there's at most one caller
4004          * for architectures that support multiple implementations,
4005          * like intel and amd on x86.
4006          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4007          * conflicts in case kvm is already setup for another implementation.
4008          */
4009         r = kvm_irqfd_init();
4010         if (r)
4011                 goto out_irqfd;
4012 
4013         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4014                 r = -ENOMEM;
4015                 goto out_free_0;
4016         }
4017 
4018         r = kvm_arch_hardware_setup();
4019         if (r < 0)
4020                 goto out_free_0a;
4021 
4022         for_each_online_cpu(cpu) {
4023                 smp_call_function_single(cpu,
4024                                 kvm_arch_check_processor_compat,
4025                                 &r, 1);
4026                 if (r < 0)
4027                         goto out_free_1;
4028         }
4029 
4030         r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4031                                       kvm_starting_cpu, kvm_dying_cpu);
4032         if (r)
4033                 goto out_free_2;
4034         register_reboot_notifier(&kvm_reboot_notifier);
4035 
4036         /* A kmem cache lets us meet the alignment requirements of fx_save. */
4037         if (!vcpu_align)
4038                 vcpu_align = __alignof__(struct kvm_vcpu);
4039         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4040                                            SLAB_ACCOUNT, NULL);
4041         if (!kvm_vcpu_cache) {
4042                 r = -ENOMEM;
4043                 goto out_free_3;
4044         }
4045 
4046         r = kvm_async_pf_init();
4047         if (r)
4048                 goto out_free;
4049 
4050         kvm_chardev_ops.owner = module;
4051         kvm_vm_fops.owner = module;
4052         kvm_vcpu_fops.owner = module;
4053 
4054         r = misc_register(&kvm_dev);
4055         if (r) {
4056                 pr_err("kvm: misc device register failed\n");
4057                 goto out_unreg;
4058         }
4059 
4060         register_syscore_ops(&kvm_syscore_ops);
4061 
4062         kvm_preempt_ops.sched_in = kvm_sched_in;
4063         kvm_preempt_ops.sched_out = kvm_sched_out;
4064 
4065         r = kvm_init_debug();
4066         if (r) {
4067                 pr_err("kvm: create debugfs files failed\n");
4068                 goto out_undebugfs;
4069         }
4070 
4071         r = kvm_vfio_ops_init();
4072         WARN_ON(r);
4073 
4074         return 0;
4075 
4076 out_undebugfs:
4077         unregister_syscore_ops(&kvm_syscore_ops);
4078         misc_deregister(&kvm_dev);
4079 out_unreg:
4080         kvm_async_pf_deinit();
4081 out_free:
4082         kmem_cache_destroy(kvm_vcpu_cache);
4083 out_free_3:
4084         unregister_reboot_notifier(&kvm_reboot_notifier);
4085         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4086 out_free_2:
4087 out_free_1:
4088         kvm_arch_hardware_unsetup();
4089 out_free_0a:
4090         free_cpumask_var(cpus_hardware_enabled);
4091 out_free_0:
4092         kvm_irqfd_exit();
4093 out_irqfd:
4094         kvm_arch_exit();
4095 out_fail:
4096         return r;
4097 }
4098 EXPORT_SYMBOL_GPL(kvm_init);
4099 
4100 void kvm_exit(void)
4101 {
4102         debugfs_remove_recursive(kvm_debugfs_dir);
4103         misc_deregister(&kvm_dev);
4104         kmem_cache_destroy(kvm_vcpu_cache);
4105         kvm_async_pf_deinit();
4106         unregister_syscore_ops(&kvm_syscore_ops);
4107         unregister_reboot_notifier(&kvm_reboot_notifier);
4108         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4109         on_each_cpu(hardware_disable_nolock, NULL, 1);
4110         kvm_arch_hardware_unsetup();
4111         kvm_arch_exit();
4112         kvm_irqfd_exit();
4113         free_cpumask_var(cpus_hardware_enabled);
4114         kvm_vfio_ops_exit();
4115 }
4116 EXPORT_SYMBOL_GPL(kvm_exit);
4117 

~ [ 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