Commit | Line | Data |
---|---|---|
6aa8b732 AK |
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. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
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 | ||
e2174021 | 19 | #include "iodev.h" |
6aa8b732 | 20 | |
edf88417 | 21 | #include <linux/kvm_host.h> |
6aa8b732 AK |
22 | #include <linux/kvm.h> |
23 | #include <linux/module.h> | |
24 | #include <linux/errno.h> | |
6aa8b732 | 25 | #include <linux/percpu.h> |
6aa8b732 AK |
26 | #include <linux/mm.h> |
27 | #include <linux/miscdevice.h> | |
28 | #include <linux/vmalloc.h> | |
6aa8b732 | 29 | #include <linux/reboot.h> |
6aa8b732 AK |
30 | #include <linux/debugfs.h> |
31 | #include <linux/highmem.h> | |
32 | #include <linux/file.h> | |
fb3600cc | 33 | #include <linux/syscore_ops.h> |
774c47f1 | 34 | #include <linux/cpu.h> |
e8edc6e0 | 35 | #include <linux/sched.h> |
d9e368d6 AK |
36 | #include <linux/cpumask.h> |
37 | #include <linux/smp.h> | |
d6d28168 | 38 | #include <linux/anon_inodes.h> |
04d2cc77 | 39 | #include <linux/profile.h> |
7aa81cc0 | 40 | #include <linux/kvm_para.h> |
6fc138d2 | 41 | #include <linux/pagemap.h> |
8d4e1288 | 42 | #include <linux/mman.h> |
35149e21 | 43 | #include <linux/swap.h> |
e56d532f | 44 | #include <linux/bitops.h> |
547de29e | 45 | #include <linux/spinlock.h> |
6ff5894c | 46 | #include <linux/compat.h> |
bc6678a3 | 47 | #include <linux/srcu.h> |
8f0b1ab6 | 48 | #include <linux/hugetlb.h> |
5a0e3ad6 | 49 | #include <linux/slab.h> |
743eeb0b SL |
50 | #include <linux/sort.h> |
51 | #include <linux/bsearch.h> | |
6aa8b732 | 52 | |
e495606d | 53 | #include <asm/processor.h> |
e495606d AK |
54 | #include <asm/io.h> |
55 | #include <asm/uaccess.h> | |
3e021bf5 | 56 | #include <asm/pgtable.h> |
6aa8b732 | 57 | |
5f94c174 | 58 | #include "coalesced_mmio.h" |
af585b92 | 59 | #include "async_pf.h" |
5f94c174 | 60 | |
229456fc MT |
61 | #define CREATE_TRACE_POINTS |
62 | #include <trace/events/kvm.h> | |
63 | ||
6aa8b732 AK |
64 | MODULE_AUTHOR("Qumranet"); |
65 | MODULE_LICENSE("GPL"); | |
66 | ||
fa40a821 MT |
67 | /* |
68 | * Ordering of locks: | |
69 | * | |
fae3a353 | 70 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
71 | */ |
72 | ||
e935b837 | 73 | DEFINE_RAW_SPINLOCK(kvm_lock); |
e9b11c17 | 74 | LIST_HEAD(vm_list); |
133de902 | 75 | |
7f59f492 | 76 | static cpumask_var_t cpus_hardware_enabled; |
10474ae8 AG |
77 | static int kvm_usage_count = 0; |
78 | static atomic_t hardware_enable_failed; | |
1b6c0168 | 79 | |
c16f862d RR |
80 | struct kmem_cache *kvm_vcpu_cache; |
81 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 82 | |
15ad7146 AK |
83 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
84 | ||
76f7c879 | 85 | struct dentry *kvm_debugfs_dir; |
6aa8b732 | 86 | |
bccf2150 AK |
87 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
88 | unsigned long arg); | |
1dda606c AG |
89 | #ifdef CONFIG_COMPAT |
90 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, | |
91 | unsigned long arg); | |
92 | #endif | |
10474ae8 AG |
93 | static int hardware_enable_all(void); |
94 | static void hardware_disable_all(void); | |
bccf2150 | 95 | |
e93f8a0f MT |
96 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
97 | ||
b7c4145b AK |
98 | bool kvm_rebooting; |
99 | EXPORT_SYMBOL_GPL(kvm_rebooting); | |
4ecac3fd | 100 | |
54dee993 MT |
101 | static bool largepages_enabled = true; |
102 | ||
a2766325 | 103 | bool kvm_is_mmio_pfn(pfn_t pfn) |
cbff90a7 | 104 | { |
fc5659c8 | 105 | if (pfn_valid(pfn)) { |
22e5c47e | 106 | int reserved; |
936a5fe6 | 107 | struct page *tail = pfn_to_page(pfn); |
22e5c47e AA |
108 | struct page *head = compound_trans_head(tail); |
109 | reserved = PageReserved(head); | |
936a5fe6 | 110 | if (head != tail) { |
936a5fe6 | 111 | /* |
22e5c47e AA |
112 | * "head" is not a dangling pointer |
113 | * (compound_trans_head takes care of that) | |
114 | * but the hugepage may have been splitted | |
115 | * from under us (and we may not hold a | |
116 | * reference count on the head page so it can | |
117 | * be reused before we run PageReferenced), so | |
118 | * we've to check PageTail before returning | |
119 | * what we just read. | |
936a5fe6 | 120 | */ |
22e5c47e AA |
121 | smp_rmb(); |
122 | if (PageTail(tail)) | |
123 | return reserved; | |
936a5fe6 AA |
124 | } |
125 | return PageReserved(tail); | |
fc5659c8 | 126 | } |
cbff90a7 BAY |
127 | |
128 | return true; | |
129 | } | |
130 | ||
bccf2150 AK |
131 | /* |
132 | * Switches to specified vcpu, until a matching vcpu_put() | |
133 | */ | |
313a3dc7 | 134 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 135 | { |
15ad7146 AK |
136 | int cpu; |
137 | ||
bccf2150 | 138 | mutex_lock(&vcpu->mutex); |
34bb10b7 RR |
139 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
140 | /* The thread running this VCPU changed. */ | |
141 | struct pid *oldpid = vcpu->pid; | |
142 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
143 | rcu_assign_pointer(vcpu->pid, newpid); | |
144 | synchronize_rcu(); | |
145 | put_pid(oldpid); | |
146 | } | |
15ad7146 AK |
147 | cpu = get_cpu(); |
148 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 149 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 150 | put_cpu(); |
6aa8b732 AK |
151 | } |
152 | ||
313a3dc7 | 153 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 154 | { |
15ad7146 | 155 | preempt_disable(); |
313a3dc7 | 156 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
157 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
158 | preempt_enable(); | |
6aa8b732 AK |
159 | mutex_unlock(&vcpu->mutex); |
160 | } | |
161 | ||
d9e368d6 AK |
162 | static void ack_flush(void *_completed) |
163 | { | |
d9e368d6 AK |
164 | } |
165 | ||
49846896 | 166 | static bool make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 167 | { |
597a5f55 | 168 | int i, cpu, me; |
6ef7a1bc RR |
169 | cpumask_var_t cpus; |
170 | bool called = true; | |
d9e368d6 | 171 | struct kvm_vcpu *vcpu; |
d9e368d6 | 172 | |
79f55997 | 173 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 174 | |
3cba4130 | 175 | me = get_cpu(); |
988a2cae | 176 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 177 | kvm_make_request(req, vcpu); |
d9e368d6 | 178 | cpu = vcpu->cpu; |
6b7e2d09 XG |
179 | |
180 | /* Set ->requests bit before we read ->mode */ | |
181 | smp_mb(); | |
182 | ||
183 | if (cpus != NULL && cpu != -1 && cpu != me && | |
184 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 185 | cpumask_set_cpu(cpu, cpus); |
49846896 | 186 | } |
6ef7a1bc RR |
187 | if (unlikely(cpus == NULL)) |
188 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
189 | else if (!cpumask_empty(cpus)) | |
190 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
191 | else | |
192 | called = false; | |
3cba4130 | 193 | put_cpu(); |
6ef7a1bc | 194 | free_cpumask_var(cpus); |
49846896 | 195 | return called; |
d9e368d6 AK |
196 | } |
197 | ||
49846896 | 198 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 199 | { |
bec87d6e | 200 | long dirty_count = kvm->tlbs_dirty; |
a4ee1ca4 XG |
201 | |
202 | smp_mb(); | |
49846896 RR |
203 | if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
204 | ++kvm->stat.remote_tlb_flush; | |
a4ee1ca4 | 205 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a MT |
206 | } |
207 | ||
49846896 RR |
208 | void kvm_reload_remote_mmus(struct kvm *kvm) |
209 | { | |
210 | make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); | |
211 | } | |
2e53d63a | 212 | |
fb3f0f51 RR |
213 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
214 | { | |
215 | struct page *page; | |
216 | int r; | |
217 | ||
218 | mutex_init(&vcpu->mutex); | |
219 | vcpu->cpu = -1; | |
fb3f0f51 RR |
220 | vcpu->kvm = kvm; |
221 | vcpu->vcpu_id = id; | |
34bb10b7 | 222 | vcpu->pid = NULL; |
b6958ce4 | 223 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 224 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
225 | |
226 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
227 | if (!page) { | |
228 | r = -ENOMEM; | |
229 | goto fail; | |
230 | } | |
231 | vcpu->run = page_address(page); | |
232 | ||
4c088493 R |
233 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
234 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
235 | ||
e9b11c17 | 236 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 237 | if (r < 0) |
e9b11c17 | 238 | goto fail_free_run; |
fb3f0f51 RR |
239 | return 0; |
240 | ||
fb3f0f51 RR |
241 | fail_free_run: |
242 | free_page((unsigned long)vcpu->run); | |
243 | fail: | |
76fafa5e | 244 | return r; |
fb3f0f51 RR |
245 | } |
246 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
247 | ||
248 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
249 | { | |
34bb10b7 | 250 | put_pid(vcpu->pid); |
e9b11c17 | 251 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
252 | free_page((unsigned long)vcpu->run); |
253 | } | |
254 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
255 | ||
e930bffe AA |
256 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
257 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
258 | { | |
259 | return container_of(mn, struct kvm, mmu_notifier); | |
260 | } | |
261 | ||
262 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
263 | struct mm_struct *mm, | |
264 | unsigned long address) | |
265 | { | |
266 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 267 | int need_tlb_flush, idx; |
e930bffe AA |
268 | |
269 | /* | |
270 | * When ->invalidate_page runs, the linux pte has been zapped | |
271 | * already but the page is still allocated until | |
272 | * ->invalidate_page returns. So if we increase the sequence | |
273 | * here the kvm page fault will notice if the spte can't be | |
274 | * established because the page is going to be freed. If | |
275 | * instead the kvm page fault establishes the spte before | |
276 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
277 | * before returning. | |
278 | * | |
279 | * The sequence increase only need to be seen at spin_unlock | |
280 | * time, and not at spin_lock time. | |
281 | * | |
282 | * Increasing the sequence after the spin_unlock would be | |
283 | * unsafe because the kvm page fault could then establish the | |
284 | * pte after kvm_unmap_hva returned, without noticing the page | |
285 | * is going to be freed. | |
286 | */ | |
bc6678a3 | 287 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 288 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 289 | |
e930bffe | 290 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 291 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
292 | /* we've to flush the tlb before the pages can be freed */ |
293 | if (need_tlb_flush) | |
294 | kvm_flush_remote_tlbs(kvm); | |
295 | ||
565f3be2 TY |
296 | spin_unlock(&kvm->mmu_lock); |
297 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
298 | } |
299 | ||
3da0dd43 IE |
300 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
301 | struct mm_struct *mm, | |
302 | unsigned long address, | |
303 | pte_t pte) | |
304 | { | |
305 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 306 | int idx; |
3da0dd43 | 307 | |
bc6678a3 | 308 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
309 | spin_lock(&kvm->mmu_lock); |
310 | kvm->mmu_notifier_seq++; | |
311 | kvm_set_spte_hva(kvm, address, pte); | |
312 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 313 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
314 | } |
315 | ||
e930bffe AA |
316 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
317 | struct mm_struct *mm, | |
318 | unsigned long start, | |
319 | unsigned long end) | |
320 | { | |
321 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 322 | int need_tlb_flush = 0, idx; |
e930bffe | 323 | |
bc6678a3 | 324 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
325 | spin_lock(&kvm->mmu_lock); |
326 | /* | |
327 | * The count increase must become visible at unlock time as no | |
328 | * spte can be established without taking the mmu_lock and | |
329 | * count is also read inside the mmu_lock critical section. | |
330 | */ | |
331 | kvm->mmu_notifier_count++; | |
b3ae2096 | 332 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 333 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
334 | /* we've to flush the tlb before the pages can be freed */ |
335 | if (need_tlb_flush) | |
336 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
337 | |
338 | spin_unlock(&kvm->mmu_lock); | |
339 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
340 | } |
341 | ||
342 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
343 | struct mm_struct *mm, | |
344 | unsigned long start, | |
345 | unsigned long end) | |
346 | { | |
347 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
348 | ||
349 | spin_lock(&kvm->mmu_lock); | |
350 | /* | |
351 | * This sequence increase will notify the kvm page fault that | |
352 | * the page that is going to be mapped in the spte could have | |
353 | * been freed. | |
354 | */ | |
355 | kvm->mmu_notifier_seq++; | |
a355aa54 | 356 | smp_wmb(); |
e930bffe AA |
357 | /* |
358 | * The above sequence increase must be visible before the | |
a355aa54 PM |
359 | * below count decrease, which is ensured by the smp_wmb above |
360 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
361 | */ |
362 | kvm->mmu_notifier_count--; | |
363 | spin_unlock(&kvm->mmu_lock); | |
364 | ||
365 | BUG_ON(kvm->mmu_notifier_count < 0); | |
366 | } | |
367 | ||
368 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
369 | struct mm_struct *mm, | |
370 | unsigned long address) | |
371 | { | |
372 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 373 | int young, idx; |
e930bffe | 374 | |
bc6678a3 | 375 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 376 | spin_lock(&kvm->mmu_lock); |
e930bffe | 377 | |
565f3be2 | 378 | young = kvm_age_hva(kvm, address); |
e930bffe AA |
379 | if (young) |
380 | kvm_flush_remote_tlbs(kvm); | |
381 | ||
565f3be2 TY |
382 | spin_unlock(&kvm->mmu_lock); |
383 | srcu_read_unlock(&kvm->srcu, idx); | |
384 | ||
e930bffe AA |
385 | return young; |
386 | } | |
387 | ||
8ee53820 AA |
388 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
389 | struct mm_struct *mm, | |
390 | unsigned long address) | |
391 | { | |
392 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
393 | int young, idx; | |
394 | ||
395 | idx = srcu_read_lock(&kvm->srcu); | |
396 | spin_lock(&kvm->mmu_lock); | |
397 | young = kvm_test_age_hva(kvm, address); | |
398 | spin_unlock(&kvm->mmu_lock); | |
399 | srcu_read_unlock(&kvm->srcu, idx); | |
400 | ||
401 | return young; | |
402 | } | |
403 | ||
85db06e5 MT |
404 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
405 | struct mm_struct *mm) | |
406 | { | |
407 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
408 | int idx; |
409 | ||
410 | idx = srcu_read_lock(&kvm->srcu); | |
85db06e5 | 411 | kvm_arch_flush_shadow(kvm); |
eda2beda | 412 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
413 | } |
414 | ||
e930bffe AA |
415 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
416 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
417 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
418 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
419 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 420 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 421 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 422 | .release = kvm_mmu_notifier_release, |
e930bffe | 423 | }; |
4c07b0a4 AK |
424 | |
425 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
426 | { | |
427 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
428 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
429 | } | |
430 | ||
431 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
432 | ||
433 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
434 | { | |
435 | return 0; | |
436 | } | |
437 | ||
e930bffe AA |
438 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
439 | ||
bf3e05bc XG |
440 | static void kvm_init_memslots_id(struct kvm *kvm) |
441 | { | |
442 | int i; | |
443 | struct kvm_memslots *slots = kvm->memslots; | |
444 | ||
445 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
f85e2cb5 | 446 | slots->id_to_index[i] = slots->memslots[i].id = i; |
bf3e05bc XG |
447 | } |
448 | ||
e08b9637 | 449 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 450 | { |
d89f5eff JK |
451 | int r, i; |
452 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 453 | |
d89f5eff JK |
454 | if (!kvm) |
455 | return ERR_PTR(-ENOMEM); | |
456 | ||
e08b9637 | 457 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff JK |
458 | if (r) |
459 | goto out_err_nodisable; | |
10474ae8 AG |
460 | |
461 | r = hardware_enable_all(); | |
462 | if (r) | |
463 | goto out_err_nodisable; | |
464 | ||
75858a84 AK |
465 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
466 | INIT_HLIST_HEAD(&kvm->mask_notifier_list); | |
136bdfee | 467 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 468 | #endif |
6aa8b732 | 469 | |
46a26bf5 MT |
470 | r = -ENOMEM; |
471 | kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); | |
472 | if (!kvm->memslots) | |
57e7fbee | 473 | goto out_err_nosrcu; |
bf3e05bc | 474 | kvm_init_memslots_id(kvm); |
bc6678a3 | 475 | if (init_srcu_struct(&kvm->srcu)) |
57e7fbee | 476 | goto out_err_nosrcu; |
e93f8a0f MT |
477 | for (i = 0; i < KVM_NR_BUSES; i++) { |
478 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
479 | GFP_KERNEL); | |
57e7fbee | 480 | if (!kvm->buses[i]) |
e93f8a0f | 481 | goto out_err; |
e93f8a0f | 482 | } |
e930bffe | 483 | |
74b5c5bf | 484 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
485 | kvm->mm = current->mm; |
486 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 487 | kvm_eventfd_init(kvm); |
11ec2804 | 488 | mutex_init(&kvm->lock); |
60eead79 | 489 | mutex_init(&kvm->irq_lock); |
79fac95e | 490 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 491 | atomic_set(&kvm->users_count, 1); |
74b5c5bf MW |
492 | |
493 | r = kvm_init_mmu_notifier(kvm); | |
494 | if (r) | |
495 | goto out_err; | |
496 | ||
e935b837 | 497 | raw_spin_lock(&kvm_lock); |
5e58cfe4 | 498 | list_add(&kvm->vm_list, &vm_list); |
e935b837 | 499 | raw_spin_unlock(&kvm_lock); |
d89f5eff | 500 | |
f17abe9a | 501 | return kvm; |
10474ae8 AG |
502 | |
503 | out_err: | |
57e7fbee JK |
504 | cleanup_srcu_struct(&kvm->srcu); |
505 | out_err_nosrcu: | |
10474ae8 AG |
506 | hardware_disable_all(); |
507 | out_err_nodisable: | |
e93f8a0f MT |
508 | for (i = 0; i < KVM_NR_BUSES; i++) |
509 | kfree(kvm->buses[i]); | |
46a26bf5 | 510 | kfree(kvm->memslots); |
d89f5eff | 511 | kvm_arch_free_vm(kvm); |
10474ae8 | 512 | return ERR_PTR(r); |
f17abe9a AK |
513 | } |
514 | ||
92eca8fa TY |
515 | /* |
516 | * Avoid using vmalloc for a small buffer. | |
517 | * Should not be used when the size is statically known. | |
518 | */ | |
c1a7b32a | 519 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
520 | { |
521 | if (size > PAGE_SIZE) | |
522 | return vzalloc(size); | |
523 | else | |
524 | return kzalloc(size, GFP_KERNEL); | |
525 | } | |
526 | ||
c1a7b32a | 527 | void kvm_kvfree(const void *addr) |
92eca8fa TY |
528 | { |
529 | if (is_vmalloc_addr(addr)) | |
530 | vfree(addr); | |
531 | else | |
532 | kfree(addr); | |
533 | } | |
534 | ||
a36a57b1 TY |
535 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
536 | { | |
537 | if (!memslot->dirty_bitmap) | |
538 | return; | |
539 | ||
92eca8fa | 540 | kvm_kvfree(memslot->dirty_bitmap); |
a36a57b1 TY |
541 | memslot->dirty_bitmap = NULL; |
542 | } | |
543 | ||
6aa8b732 AK |
544 | /* |
545 | * Free any memory in @free but not in @dont. | |
546 | */ | |
547 | static void kvm_free_physmem_slot(struct kvm_memory_slot *free, | |
548 | struct kvm_memory_slot *dont) | |
549 | { | |
6aa8b732 | 550 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
a36a57b1 | 551 | kvm_destroy_dirty_bitmap(free); |
6aa8b732 | 552 | |
db3fe4eb | 553 | kvm_arch_free_memslot(free, dont); |
05da4558 | 554 | |
6aa8b732 | 555 | free->npages = 0; |
6aa8b732 AK |
556 | } |
557 | ||
d19a9cd2 | 558 | void kvm_free_physmem(struct kvm *kvm) |
6aa8b732 | 559 | { |
46a26bf5 | 560 | struct kvm_memslots *slots = kvm->memslots; |
be6ba0f0 | 561 | struct kvm_memory_slot *memslot; |
46a26bf5 | 562 | |
be6ba0f0 XG |
563 | kvm_for_each_memslot(memslot, slots) |
564 | kvm_free_physmem_slot(memslot, NULL); | |
6aa8b732 | 565 | |
46a26bf5 | 566 | kfree(kvm->memslots); |
6aa8b732 AK |
567 | } |
568 | ||
f17abe9a AK |
569 | static void kvm_destroy_vm(struct kvm *kvm) |
570 | { | |
e93f8a0f | 571 | int i; |
6d4e4c4f AK |
572 | struct mm_struct *mm = kvm->mm; |
573 | ||
ad8ba2cd | 574 | kvm_arch_sync_events(kvm); |
e935b837 | 575 | raw_spin_lock(&kvm_lock); |
133de902 | 576 | list_del(&kvm->vm_list); |
e935b837 | 577 | raw_spin_unlock(&kvm_lock); |
399ec807 | 578 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
579 | for (i = 0; i < KVM_NR_BUSES; i++) |
580 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 581 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
582 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
583 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca GN |
584 | #else |
585 | kvm_arch_flush_shadow(kvm); | |
5f94c174 | 586 | #endif |
d19a9cd2 | 587 | kvm_arch_destroy_vm(kvm); |
d89f5eff JK |
588 | kvm_free_physmem(kvm); |
589 | cleanup_srcu_struct(&kvm->srcu); | |
590 | kvm_arch_free_vm(kvm); | |
10474ae8 | 591 | hardware_disable_all(); |
6d4e4c4f | 592 | mmdrop(mm); |
f17abe9a AK |
593 | } |
594 | ||
d39f13b0 IE |
595 | void kvm_get_kvm(struct kvm *kvm) |
596 | { | |
597 | atomic_inc(&kvm->users_count); | |
598 | } | |
599 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
600 | ||
601 | void kvm_put_kvm(struct kvm *kvm) | |
602 | { | |
603 | if (atomic_dec_and_test(&kvm->users_count)) | |
604 | kvm_destroy_vm(kvm); | |
605 | } | |
606 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
607 | ||
608 | ||
f17abe9a AK |
609 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
610 | { | |
611 | struct kvm *kvm = filp->private_data; | |
612 | ||
721eecbf GH |
613 | kvm_irqfd_release(kvm); |
614 | ||
d39f13b0 | 615 | kvm_put_kvm(kvm); |
6aa8b732 AK |
616 | return 0; |
617 | } | |
618 | ||
515a0127 TY |
619 | /* |
620 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 621 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 622 | */ |
a36a57b1 TY |
623 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
624 | { | |
189a2f7b | 625 | #ifndef CONFIG_S390 |
515a0127 | 626 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 627 | |
92eca8fa | 628 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
629 | if (!memslot->dirty_bitmap) |
630 | return -ENOMEM; | |
631 | ||
189a2f7b | 632 | #endif /* !CONFIG_S390 */ |
a36a57b1 TY |
633 | return 0; |
634 | } | |
635 | ||
bf3e05bc XG |
636 | static int cmp_memslot(const void *slot1, const void *slot2) |
637 | { | |
638 | struct kvm_memory_slot *s1, *s2; | |
639 | ||
640 | s1 = (struct kvm_memory_slot *)slot1; | |
641 | s2 = (struct kvm_memory_slot *)slot2; | |
642 | ||
643 | if (s1->npages < s2->npages) | |
644 | return 1; | |
645 | if (s1->npages > s2->npages) | |
646 | return -1; | |
647 | ||
648 | return 0; | |
649 | } | |
650 | ||
651 | /* | |
652 | * Sort the memslots base on its size, so the larger slots | |
653 | * will get better fit. | |
654 | */ | |
655 | static void sort_memslots(struct kvm_memslots *slots) | |
656 | { | |
f85e2cb5 XG |
657 | int i; |
658 | ||
bf3e05bc XG |
659 | sort(slots->memslots, KVM_MEM_SLOTS_NUM, |
660 | sizeof(struct kvm_memory_slot), cmp_memslot, NULL); | |
f85e2cb5 XG |
661 | |
662 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
663 | slots->id_to_index[slots->memslots[i].id] = i; | |
bf3e05bc XG |
664 | } |
665 | ||
be593d62 XG |
666 | void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new) |
667 | { | |
668 | if (new) { | |
669 | int id = new->id; | |
28a37544 | 670 | struct kvm_memory_slot *old = id_to_memslot(slots, id); |
bf3e05bc | 671 | unsigned long npages = old->npages; |
be593d62 | 672 | |
28a37544 | 673 | *old = *new; |
bf3e05bc XG |
674 | if (new->npages != npages) |
675 | sort_memslots(slots); | |
be593d62 XG |
676 | } |
677 | ||
678 | slots->generation++; | |
679 | } | |
680 | ||
a50d64d6 XG |
681 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
682 | { | |
683 | if (mem->flags & ~KVM_MEM_LOG_DIRTY_PAGES) | |
684 | return -EINVAL; | |
685 | ||
686 | return 0; | |
687 | } | |
688 | ||
6aa8b732 AK |
689 | /* |
690 | * Allocate some memory and give it an address in the guest physical address | |
691 | * space. | |
692 | * | |
693 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 694 | * |
10589a46 | 695 | * Must be called holding mmap_sem for write. |
6aa8b732 | 696 | */ |
f78e0e2e SY |
697 | int __kvm_set_memory_region(struct kvm *kvm, |
698 | struct kvm_userspace_memory_region *mem, | |
699 | int user_alloc) | |
6aa8b732 | 700 | { |
8234b22e | 701 | int r; |
6aa8b732 | 702 | gfn_t base_gfn; |
28bcb112 HC |
703 | unsigned long npages; |
704 | unsigned long i; | |
6aa8b732 AK |
705 | struct kvm_memory_slot *memslot; |
706 | struct kvm_memory_slot old, new; | |
bc6678a3 | 707 | struct kvm_memslots *slots, *old_memslots; |
6aa8b732 | 708 | |
a50d64d6 XG |
709 | r = check_memory_region_flags(mem); |
710 | if (r) | |
711 | goto out; | |
712 | ||
6aa8b732 AK |
713 | r = -EINVAL; |
714 | /* General sanity checks */ | |
715 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
716 | goto out; | |
717 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
718 | goto out; | |
fa3d315a TY |
719 | /* We can read the guest memory with __xxx_user() later on. */ |
720 | if (user_alloc && | |
721 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || | |
9e3bb6b6 HC |
722 | !access_ok(VERIFY_WRITE, |
723 | (void __user *)(unsigned long)mem->userspace_addr, | |
724 | mem->memory_size))) | |
78749809 | 725 | goto out; |
93a5cef0 | 726 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
727 | goto out; |
728 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
729 | goto out; | |
730 | ||
28a37544 | 731 | memslot = id_to_memslot(kvm->memslots, mem->slot); |
6aa8b732 AK |
732 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
733 | npages = mem->memory_size >> PAGE_SHIFT; | |
734 | ||
660c22c4 TY |
735 | r = -EINVAL; |
736 | if (npages > KVM_MEM_MAX_NR_PAGES) | |
737 | goto out; | |
738 | ||
6aa8b732 AK |
739 | if (!npages) |
740 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
741 | ||
6aa8b732 AK |
742 | new = old = *memslot; |
743 | ||
e36d96f7 | 744 | new.id = mem->slot; |
6aa8b732 AK |
745 | new.base_gfn = base_gfn; |
746 | new.npages = npages; | |
747 | new.flags = mem->flags; | |
748 | ||
749 | /* Disallow changing a memory slot's size. */ | |
750 | r = -EINVAL; | |
751 | if (npages && old.npages && npages != old.npages) | |
f78e0e2e | 752 | goto out_free; |
6aa8b732 AK |
753 | |
754 | /* Check for overlaps */ | |
755 | r = -EEXIST; | |
756 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { | |
46a26bf5 | 757 | struct kvm_memory_slot *s = &kvm->memslots->memslots[i]; |
6aa8b732 | 758 | |
4cd481f6 | 759 | if (s == memslot || !s->npages) |
6aa8b732 AK |
760 | continue; |
761 | if (!((base_gfn + npages <= s->base_gfn) || | |
762 | (base_gfn >= s->base_gfn + s->npages))) | |
f78e0e2e | 763 | goto out_free; |
6aa8b732 | 764 | } |
6aa8b732 | 765 | |
6aa8b732 AK |
766 | /* Free page dirty bitmap if unneeded */ |
767 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 768 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
769 | |
770 | r = -ENOMEM; | |
771 | ||
772 | /* Allocate if a slot is being created */ | |
189a2f7b TY |
773 | if (npages && !old.npages) { |
774 | new.user_alloc = user_alloc; | |
775 | new.userspace_addr = mem->userspace_addr; | |
d89cc617 | 776 | |
db3fe4eb TY |
777 | if (kvm_arch_create_memslot(&new, npages)) |
778 | goto out_free; | |
6aa8b732 | 779 | } |
ec04b260 | 780 | |
6aa8b732 AK |
781 | /* Allocate page dirty bitmap if needed */ |
782 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 783 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 784 | goto out_free; |
bc6678a3 | 785 | /* destroy any largepage mappings for dirty tracking */ |
6aa8b732 AK |
786 | } |
787 | ||
bc6678a3 | 788 | if (!npages) { |
28a37544 XG |
789 | struct kvm_memory_slot *slot; |
790 | ||
bc6678a3 | 791 | r = -ENOMEM; |
6da64fdb TM |
792 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), |
793 | GFP_KERNEL); | |
bc6678a3 MT |
794 | if (!slots) |
795 | goto out_free; | |
28a37544 XG |
796 | slot = id_to_memslot(slots, mem->slot); |
797 | slot->flags |= KVM_MEMSLOT_INVALID; | |
798 | ||
be593d62 | 799 | update_memslots(slots, NULL); |
bc6678a3 MT |
800 | |
801 | old_memslots = kvm->memslots; | |
802 | rcu_assign_pointer(kvm->memslots, slots); | |
803 | synchronize_srcu_expedited(&kvm->srcu); | |
804 | /* From this point no new shadow pages pointing to a deleted | |
805 | * memslot will be created. | |
806 | * | |
807 | * validation of sp->gfn happens in: | |
808 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
809 | * - kvm_is_visible_gfn (mmu_check_roots) | |
810 | */ | |
34d4cb8f | 811 | kvm_arch_flush_shadow(kvm); |
bc6678a3 MT |
812 | kfree(old_memslots); |
813 | } | |
34d4cb8f | 814 | |
f7784b8e MT |
815 | r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc); |
816 | if (r) | |
817 | goto out_free; | |
818 | ||
32f6daad | 819 | /* map/unmap the pages in iommu page table */ |
bc6678a3 MT |
820 | if (npages) { |
821 | r = kvm_iommu_map_pages(kvm, &new); | |
822 | if (r) | |
823 | goto out_free; | |
32f6daad AW |
824 | } else |
825 | kvm_iommu_unmap_pages(kvm, &old); | |
604b38ac | 826 | |
bc6678a3 | 827 | r = -ENOMEM; |
6da64fdb TM |
828 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), |
829 | GFP_KERNEL); | |
bc6678a3 MT |
830 | if (!slots) |
831 | goto out_free; | |
bc6678a3 MT |
832 | |
833 | /* actual memory is freed via old in kvm_free_physmem_slot below */ | |
834 | if (!npages) { | |
bc6678a3 | 835 | new.dirty_bitmap = NULL; |
db3fe4eb | 836 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
837 | } |
838 | ||
be593d62 | 839 | update_memslots(slots, &new); |
bc6678a3 MT |
840 | old_memslots = kvm->memslots; |
841 | rcu_assign_pointer(kvm->memslots, slots); | |
842 | synchronize_srcu_expedited(&kvm->srcu); | |
3ad82a7e | 843 | |
f7784b8e | 844 | kvm_arch_commit_memory_region(kvm, mem, old, user_alloc); |
82ce2c96 | 845 | |
ce88decf XG |
846 | /* |
847 | * If the new memory slot is created, we need to clear all | |
848 | * mmio sptes. | |
849 | */ | |
850 | if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT) | |
851 | kvm_arch_flush_shadow(kvm); | |
852 | ||
bc6678a3 MT |
853 | kvm_free_physmem_slot(&old, &new); |
854 | kfree(old_memslots); | |
855 | ||
6aa8b732 AK |
856 | return 0; |
857 | ||
f78e0e2e | 858 | out_free: |
6aa8b732 AK |
859 | kvm_free_physmem_slot(&new, &old); |
860 | out: | |
861 | return r; | |
210c7c4d IE |
862 | |
863 | } | |
f78e0e2e SY |
864 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
865 | ||
866 | int kvm_set_memory_region(struct kvm *kvm, | |
867 | struct kvm_userspace_memory_region *mem, | |
868 | int user_alloc) | |
869 | { | |
870 | int r; | |
871 | ||
79fac95e | 872 | mutex_lock(&kvm->slots_lock); |
f78e0e2e | 873 | r = __kvm_set_memory_region(kvm, mem, user_alloc); |
79fac95e | 874 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
875 | return r; |
876 | } | |
210c7c4d IE |
877 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
878 | ||
1fe779f8 CO |
879 | int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
880 | struct | |
881 | kvm_userspace_memory_region *mem, | |
882 | int user_alloc) | |
210c7c4d | 883 | { |
e0d62c7f IE |
884 | if (mem->slot >= KVM_MEMORY_SLOTS) |
885 | return -EINVAL; | |
210c7c4d | 886 | return kvm_set_memory_region(kvm, mem, user_alloc); |
6aa8b732 AK |
887 | } |
888 | ||
5bb064dc ZX |
889 | int kvm_get_dirty_log(struct kvm *kvm, |
890 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 AK |
891 | { |
892 | struct kvm_memory_slot *memslot; | |
893 | int r, i; | |
87bf6e7d | 894 | unsigned long n; |
6aa8b732 AK |
895 | unsigned long any = 0; |
896 | ||
6aa8b732 AK |
897 | r = -EINVAL; |
898 | if (log->slot >= KVM_MEMORY_SLOTS) | |
899 | goto out; | |
900 | ||
28a37544 | 901 | memslot = id_to_memslot(kvm->memslots, log->slot); |
6aa8b732 AK |
902 | r = -ENOENT; |
903 | if (!memslot->dirty_bitmap) | |
904 | goto out; | |
905 | ||
87bf6e7d | 906 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 907 | |
cd1a4a98 | 908 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
909 | any = memslot->dirty_bitmap[i]; |
910 | ||
911 | r = -EFAULT; | |
912 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
913 | goto out; | |
914 | ||
5bb064dc ZX |
915 | if (any) |
916 | *is_dirty = 1; | |
6aa8b732 AK |
917 | |
918 | r = 0; | |
6aa8b732 | 919 | out: |
6aa8b732 AK |
920 | return r; |
921 | } | |
922 | ||
db3fe4eb TY |
923 | bool kvm_largepages_enabled(void) |
924 | { | |
925 | return largepages_enabled; | |
926 | } | |
927 | ||
54dee993 MT |
928 | void kvm_disable_largepages(void) |
929 | { | |
930 | largepages_enabled = false; | |
931 | } | |
932 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
933 | ||
f9d46eb0 IE |
934 | static inline unsigned long bad_hva(void) |
935 | { | |
936 | return PAGE_OFFSET; | |
937 | } | |
938 | ||
939 | int kvm_is_error_hva(unsigned long addr) | |
940 | { | |
941 | return addr == bad_hva(); | |
942 | } | |
943 | EXPORT_SYMBOL_GPL(kvm_is_error_hva); | |
944 | ||
49c7754c GN |
945 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
946 | { | |
947 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
948 | } | |
a1f4d395 | 949 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 950 | |
e0d62c7f IE |
951 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
952 | { | |
bf3e05bc | 953 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 954 | |
bf3e05bc XG |
955 | if (!memslot || memslot->id >= KVM_MEMORY_SLOTS || |
956 | memslot->flags & KVM_MEMSLOT_INVALID) | |
957 | return 0; | |
e0d62c7f | 958 | |
bf3e05bc | 959 | return 1; |
e0d62c7f IE |
960 | } |
961 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
962 | ||
8f0b1ab6 JR |
963 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
964 | { | |
965 | struct vm_area_struct *vma; | |
966 | unsigned long addr, size; | |
967 | ||
968 | size = PAGE_SIZE; | |
969 | ||
970 | addr = gfn_to_hva(kvm, gfn); | |
971 | if (kvm_is_error_hva(addr)) | |
972 | return PAGE_SIZE; | |
973 | ||
974 | down_read(¤t->mm->mmap_sem); | |
975 | vma = find_vma(current->mm, addr); | |
976 | if (!vma) | |
977 | goto out; | |
978 | ||
979 | size = vma_kernel_pagesize(vma); | |
980 | ||
981 | out: | |
982 | up_read(¤t->mm->mmap_sem); | |
983 | ||
984 | return size; | |
985 | } | |
986 | ||
49c7754c | 987 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
48987781 | 988 | gfn_t *nr_pages) |
539cb660 | 989 | { |
bc6678a3 | 990 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
539cb660 | 991 | return bad_hva(); |
48987781 XG |
992 | |
993 | if (nr_pages) | |
994 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
995 | ||
f5c98031 | 996 | return gfn_to_hva_memslot(slot, gfn); |
539cb660 | 997 | } |
48987781 XG |
998 | |
999 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) | |
1000 | { | |
49c7754c | 1001 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1002 | } |
0d150298 | 1003 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1004 | |
86ab8cff XG |
1005 | /* |
1006 | * The hva returned by this function is only allowed to be read. | |
1007 | * It should pair with kvm_read_hva() or kvm_read_hva_atomic(). | |
1008 | */ | |
1009 | static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn) | |
1010 | { | |
1011 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); | |
1012 | } | |
1013 | ||
1014 | static int kvm_read_hva(void *data, void __user *hva, int len) | |
1015 | { | |
1016 | return __copy_from_user(data, hva, len); | |
1017 | } | |
1018 | ||
1019 | static int kvm_read_hva_atomic(void *data, void __user *hva, int len) | |
1020 | { | |
1021 | return __copy_from_user_inatomic(data, hva, len); | |
1022 | } | |
1023 | ||
0857b9e9 GN |
1024 | int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
1025 | unsigned long start, int write, struct page **page) | |
1026 | { | |
1027 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1028 | ||
1029 | if (write) | |
1030 | flags |= FOLL_WRITE; | |
1031 | ||
1032 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1033 | } | |
1034 | ||
fafc3dba HY |
1035 | static inline int check_user_page_hwpoison(unsigned long addr) |
1036 | { | |
1037 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1038 | ||
1039 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1040 | flags, NULL, NULL, NULL); | |
1041 | return rc == -EHWPOISON; | |
1042 | } | |
1043 | ||
2fc84311 XG |
1044 | /* |
1045 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1046 | * true indicates success, otherwise false is returned. | |
1047 | */ | |
1048 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1049 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1050 | { |
8d4e1288 | 1051 | struct page *page[1]; |
2fc84311 | 1052 | int npages; |
954bbbc2 | 1053 | |
2fc84311 XG |
1054 | if (!(async || atomic)) |
1055 | return false; | |
af585b92 | 1056 | |
2fc84311 XG |
1057 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1058 | if (npages == 1) { | |
1059 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1060 | |
2fc84311 XG |
1061 | if (writable) |
1062 | *writable = true; | |
1063 | return true; | |
1064 | } | |
612819c3 | 1065 | |
2fc84311 XG |
1066 | return false; |
1067 | } | |
af585b92 | 1068 | |
2fc84311 XG |
1069 | /* |
1070 | * The slow path to get the pfn of the specified host virtual address, | |
1071 | * 1 indicates success, -errno is returned if error is detected. | |
1072 | */ | |
1073 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1074 | bool *writable, pfn_t *pfn) | |
1075 | { | |
1076 | struct page *page[1]; | |
1077 | int npages = 0; | |
612819c3 | 1078 | |
2fc84311 XG |
1079 | might_sleep(); |
1080 | ||
1081 | if (writable) | |
1082 | *writable = write_fault; | |
1083 | ||
1084 | if (async) { | |
1085 | down_read(¤t->mm->mmap_sem); | |
1086 | npages = get_user_page_nowait(current, current->mm, | |
1087 | addr, write_fault, page); | |
1088 | up_read(¤t->mm->mmap_sem); | |
1089 | } else | |
1090 | npages = get_user_pages_fast(addr, 1, write_fault, | |
1091 | page); | |
1092 | if (npages != 1) | |
1093 | return npages; | |
1094 | ||
1095 | /* map read fault as writable if possible */ | |
1096 | if (unlikely(!write_fault)) { | |
1097 | struct page *wpage[1]; | |
1098 | ||
1099 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1100 | if (npages == 1) { | |
1101 | *writable = true; | |
1102 | put_page(page[0]); | |
1103 | page[0] = wpage[0]; | |
612819c3 | 1104 | } |
2fc84311 XG |
1105 | |
1106 | npages = 1; | |
887c08ac | 1107 | } |
2fc84311 XG |
1108 | *pfn = page_to_pfn(page[0]); |
1109 | return npages; | |
1110 | } | |
539cb660 | 1111 | |
2fc84311 XG |
1112 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1113 | bool write_fault, bool *writable) | |
1114 | { | |
1115 | struct vm_area_struct *vma; | |
1116 | pfn_t pfn = 0; | |
1117 | int npages; | |
2e2e3738 | 1118 | |
2fc84311 XG |
1119 | /* we can do it either atomically or asynchronously, not both */ |
1120 | BUG_ON(atomic && async); | |
887c08ac | 1121 | |
2fc84311 | 1122 | BUG_ON(!write_fault && !writable); |
bf998156 | 1123 | |
2fc84311 XG |
1124 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1125 | return pfn; | |
1126 | ||
1127 | if (atomic) | |
1128 | return KVM_PFN_ERR_FAULT; | |
1129 | ||
1130 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1131 | if (npages == 1) | |
1132 | return pfn; | |
8d4e1288 | 1133 | |
2fc84311 XG |
1134 | down_read(¤t->mm->mmap_sem); |
1135 | if (npages == -EHWPOISON || | |
1136 | (!async && check_user_page_hwpoison(addr))) { | |
1137 | pfn = KVM_PFN_ERR_HWPOISON; | |
1138 | goto exit; | |
1139 | } | |
1140 | ||
1141 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1142 | ||
1143 | if (vma == NULL) | |
1144 | pfn = KVM_PFN_ERR_FAULT; | |
1145 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1146 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1147 | vma->vm_pgoff; | |
1148 | BUG_ON(!kvm_is_mmio_pfn(pfn)); | |
1149 | } else { | |
1150 | if (async && (vma->vm_flags & VM_WRITE)) | |
1151 | *async = true; | |
1152 | pfn = KVM_PFN_ERR_FAULT; | |
1153 | } | |
1154 | exit: | |
1155 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1156 | return pfn; |
35149e21 AL |
1157 | } |
1158 | ||
612819c3 MT |
1159 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async, |
1160 | bool write_fault, bool *writable) | |
506f0d6f MT |
1161 | { |
1162 | unsigned long addr; | |
1163 | ||
af585b92 GN |
1164 | if (async) |
1165 | *async = false; | |
1166 | ||
506f0d6f | 1167 | addr = gfn_to_hva(kvm, gfn); |
a2766325 | 1168 | if (kvm_is_error_hva(addr)) |
950e9509 | 1169 | return KVM_PFN_ERR_BAD; |
506f0d6f | 1170 | |
d5661048 | 1171 | return hva_to_pfn(addr, atomic, async, write_fault, writable); |
365fb3fd XG |
1172 | } |
1173 | ||
1174 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) | |
1175 | { | |
612819c3 | 1176 | return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL); |
365fb3fd XG |
1177 | } |
1178 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1179 | ||
612819c3 MT |
1180 | pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async, |
1181 | bool write_fault, bool *writable) | |
af585b92 | 1182 | { |
612819c3 | 1183 | return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable); |
af585b92 GN |
1184 | } |
1185 | EXPORT_SYMBOL_GPL(gfn_to_pfn_async); | |
1186 | ||
365fb3fd XG |
1187 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1188 | { | |
612819c3 | 1189 | return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL); |
506f0d6f | 1190 | } |
35149e21 AL |
1191 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
1192 | ||
612819c3 MT |
1193 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1194 | bool *writable) | |
1195 | { | |
1196 | return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable); | |
1197 | } | |
1198 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1199 | ||
d5661048 | 1200 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f MT |
1201 | { |
1202 | unsigned long addr = gfn_to_hva_memslot(slot, gfn); | |
d5661048 | 1203 | return hva_to_pfn(addr, false, NULL, true, NULL); |
506f0d6f MT |
1204 | } |
1205 | ||
037d92dc XG |
1206 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
1207 | { | |
1208 | unsigned long addr = gfn_to_hva_memslot(slot, gfn); | |
1209 | ||
1210 | return hva_to_pfn(addr, true, NULL, true, NULL); | |
1211 | } | |
1212 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); | |
1213 | ||
48987781 XG |
1214 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
1215 | int nr_pages) | |
1216 | { | |
1217 | unsigned long addr; | |
1218 | gfn_t entry; | |
1219 | ||
49c7754c | 1220 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
48987781 XG |
1221 | if (kvm_is_error_hva(addr)) |
1222 | return -1; | |
1223 | ||
1224 | if (entry < nr_pages) | |
1225 | return 0; | |
1226 | ||
1227 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1228 | } | |
1229 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1230 | ||
a2766325 XG |
1231 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1232 | { | |
cb9aaa30 XG |
1233 | if (is_error_pfn(pfn)) |
1234 | return KVM_ERR_PTR_BAD_PAGE; | |
a2766325 | 1235 | |
cb9aaa30 XG |
1236 | if (kvm_is_mmio_pfn(pfn)) { |
1237 | WARN_ON(1); | |
6cede2e6 | 1238 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1239 | } |
a2766325 XG |
1240 | |
1241 | return pfn_to_page(pfn); | |
1242 | } | |
1243 | ||
35149e21 AL |
1244 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1245 | { | |
2e2e3738 AL |
1246 | pfn_t pfn; |
1247 | ||
1248 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1249 | |
a2766325 | 1250 | return kvm_pfn_to_page(pfn); |
954bbbc2 | 1251 | } |
aab61cc0 | 1252 | |
954bbbc2 AK |
1253 | EXPORT_SYMBOL_GPL(gfn_to_page); |
1254 | ||
b4231d61 IE |
1255 | void kvm_release_page_clean(struct page *page) |
1256 | { | |
32cad84f XG |
1257 | WARN_ON(is_error_page(page)); |
1258 | ||
1259 | kvm_release_pfn_clean(page_to_pfn(page)); | |
b4231d61 IE |
1260 | } |
1261 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1262 | ||
35149e21 AL |
1263 | void kvm_release_pfn_clean(pfn_t pfn) |
1264 | { | |
cb9aaa30 XG |
1265 | WARN_ON(is_error_pfn(pfn)); |
1266 | ||
1267 | if (!kvm_is_mmio_pfn(pfn)) | |
2e2e3738 | 1268 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1269 | } |
1270 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1271 | ||
b4231d61 | 1272 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1273 | { |
a2766325 XG |
1274 | WARN_ON(is_error_page(page)); |
1275 | ||
35149e21 AL |
1276 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1277 | } | |
1278 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1279 | ||
1280 | void kvm_release_pfn_dirty(pfn_t pfn) | |
1281 | { | |
1282 | kvm_set_pfn_dirty(pfn); | |
1283 | kvm_release_pfn_clean(pfn); | |
1284 | } | |
1285 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); | |
1286 | ||
1287 | void kvm_set_page_dirty(struct page *page) | |
1288 | { | |
1289 | kvm_set_pfn_dirty(page_to_pfn(page)); | |
1290 | } | |
1291 | EXPORT_SYMBOL_GPL(kvm_set_page_dirty); | |
1292 | ||
1293 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1294 | { | |
c77fb9dc | 1295 | if (!kvm_is_mmio_pfn(pfn)) { |
2e2e3738 AL |
1296 | struct page *page = pfn_to_page(pfn); |
1297 | if (!PageReserved(page)) | |
1298 | SetPageDirty(page); | |
1299 | } | |
8a7ae055 | 1300 | } |
35149e21 AL |
1301 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1302 | ||
1303 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1304 | { | |
c77fb9dc | 1305 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1306 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1307 | } |
1308 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1309 | ||
1310 | void kvm_get_pfn(pfn_t pfn) | |
1311 | { | |
c77fb9dc | 1312 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1313 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1314 | } |
1315 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1316 | |
195aefde IE |
1317 | static int next_segment(unsigned long len, int offset) |
1318 | { | |
1319 | if (len > PAGE_SIZE - offset) | |
1320 | return PAGE_SIZE - offset; | |
1321 | else | |
1322 | return len; | |
1323 | } | |
1324 | ||
1325 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1326 | int len) | |
1327 | { | |
e0506bcb IE |
1328 | int r; |
1329 | unsigned long addr; | |
195aefde | 1330 | |
86ab8cff | 1331 | addr = gfn_to_hva_read(kvm, gfn); |
e0506bcb IE |
1332 | if (kvm_is_error_hva(addr)) |
1333 | return -EFAULT; | |
86ab8cff | 1334 | r = kvm_read_hva(data, (void __user *)addr + offset, len); |
e0506bcb | 1335 | if (r) |
195aefde | 1336 | return -EFAULT; |
195aefde IE |
1337 | return 0; |
1338 | } | |
1339 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); | |
1340 | ||
1341 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) | |
1342 | { | |
1343 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1344 | int seg; | |
1345 | int offset = offset_in_page(gpa); | |
1346 | int ret; | |
1347 | ||
1348 | while ((seg = next_segment(len, offset)) != 0) { | |
1349 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1350 | if (ret < 0) | |
1351 | return ret; | |
1352 | offset = 0; | |
1353 | len -= seg; | |
1354 | data += seg; | |
1355 | ++gfn; | |
1356 | } | |
1357 | return 0; | |
1358 | } | |
1359 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1360 | ||
7ec54588 MT |
1361 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1362 | unsigned long len) | |
1363 | { | |
1364 | int r; | |
1365 | unsigned long addr; | |
1366 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1367 | int offset = offset_in_page(gpa); | |
1368 | ||
86ab8cff | 1369 | addr = gfn_to_hva_read(kvm, gfn); |
7ec54588 MT |
1370 | if (kvm_is_error_hva(addr)) |
1371 | return -EFAULT; | |
0aac03f0 | 1372 | pagefault_disable(); |
86ab8cff | 1373 | r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1374 | pagefault_enable(); |
7ec54588 MT |
1375 | if (r) |
1376 | return -EFAULT; | |
1377 | return 0; | |
1378 | } | |
1379 | EXPORT_SYMBOL(kvm_read_guest_atomic); | |
1380 | ||
195aefde IE |
1381 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
1382 | int offset, int len) | |
1383 | { | |
e0506bcb IE |
1384 | int r; |
1385 | unsigned long addr; | |
195aefde | 1386 | |
e0506bcb IE |
1387 | addr = gfn_to_hva(kvm, gfn); |
1388 | if (kvm_is_error_hva(addr)) | |
1389 | return -EFAULT; | |
8b0cedff | 1390 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1391 | if (r) |
195aefde | 1392 | return -EFAULT; |
195aefde IE |
1393 | mark_page_dirty(kvm, gfn); |
1394 | return 0; | |
1395 | } | |
1396 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); | |
1397 | ||
1398 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, | |
1399 | unsigned long len) | |
1400 | { | |
1401 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1402 | int seg; | |
1403 | int offset = offset_in_page(gpa); | |
1404 | int ret; | |
1405 | ||
1406 | while ((seg = next_segment(len, offset)) != 0) { | |
1407 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1408 | if (ret < 0) | |
1409 | return ret; | |
1410 | offset = 0; | |
1411 | len -= seg; | |
1412 | data += seg; | |
1413 | ++gfn; | |
1414 | } | |
1415 | return 0; | |
1416 | } | |
1417 | ||
49c7754c GN |
1418 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1419 | gpa_t gpa) | |
1420 | { | |
1421 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1422 | int offset = offset_in_page(gpa); | |
1423 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1424 | ||
1425 | ghc->gpa = gpa; | |
1426 | ghc->generation = slots->generation; | |
9d4cba7f | 1427 | ghc->memslot = gfn_to_memslot(kvm, gfn); |
49c7754c GN |
1428 | ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL); |
1429 | if (!kvm_is_error_hva(ghc->hva)) | |
1430 | ghc->hva += offset; | |
1431 | else | |
1432 | return -EFAULT; | |
1433 | ||
1434 | return 0; | |
1435 | } | |
1436 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1437 | ||
1438 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1439 | void *data, unsigned long len) | |
1440 | { | |
1441 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1442 | int r; | |
1443 | ||
1444 | if (slots->generation != ghc->generation) | |
1445 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa); | |
1446 | ||
1447 | if (kvm_is_error_hva(ghc->hva)) | |
1448 | return -EFAULT; | |
1449 | ||
8b0cedff | 1450 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1451 | if (r) |
1452 | return -EFAULT; | |
1453 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); | |
1454 | ||
1455 | return 0; | |
1456 | } | |
1457 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1458 | ||
e03b644f GN |
1459 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1460 | void *data, unsigned long len) | |
1461 | { | |
1462 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1463 | int r; | |
1464 | ||
1465 | if (slots->generation != ghc->generation) | |
1466 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa); | |
1467 | ||
1468 | if (kvm_is_error_hva(ghc->hva)) | |
1469 | return -EFAULT; | |
1470 | ||
1471 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1472 | if (r) | |
1473 | return -EFAULT; | |
1474 | ||
1475 | return 0; | |
1476 | } | |
1477 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1478 | ||
195aefde IE |
1479 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1480 | { | |
3bcc8a8c HC |
1481 | return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page, |
1482 | offset, len); | |
195aefde IE |
1483 | } |
1484 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1485 | ||
1486 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1487 | { | |
1488 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1489 | int seg; | |
1490 | int offset = offset_in_page(gpa); | |
1491 | int ret; | |
1492 | ||
1493 | while ((seg = next_segment(len, offset)) != 0) { | |
1494 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); | |
1495 | if (ret < 0) | |
1496 | return ret; | |
1497 | offset = 0; | |
1498 | len -= seg; | |
1499 | ++gfn; | |
1500 | } | |
1501 | return 0; | |
1502 | } | |
1503 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1504 | ||
49c7754c GN |
1505 | void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot, |
1506 | gfn_t gfn) | |
6aa8b732 | 1507 | { |
7e9d619d RR |
1508 | if (memslot && memslot->dirty_bitmap) { |
1509 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1510 | |
93474b25 TY |
1511 | /* TODO: introduce set_bit_le() and use it */ |
1512 | test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
1513 | } |
1514 | } | |
1515 | ||
49c7754c GN |
1516 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1517 | { | |
1518 | struct kvm_memory_slot *memslot; | |
1519 | ||
1520 | memslot = gfn_to_memslot(kvm, gfn); | |
1521 | mark_page_dirty_in_slot(kvm, memslot, gfn); | |
1522 | } | |
1523 | ||
b6958ce4 ED |
1524 | /* |
1525 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1526 | */ | |
8776e519 | 1527 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1528 | { |
e5c239cf MT |
1529 | DEFINE_WAIT(wait); |
1530 | ||
1531 | for (;;) { | |
1532 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1533 | ||
a1b37100 | 1534 | if (kvm_arch_vcpu_runnable(vcpu)) { |
a8eeb04a | 1535 | kvm_make_request(KVM_REQ_UNHALT, vcpu); |
e5c239cf | 1536 | break; |
d7690175 | 1537 | } |
09cec754 GN |
1538 | if (kvm_cpu_has_pending_timer(vcpu)) |
1539 | break; | |
e5c239cf MT |
1540 | if (signal_pending(current)) |
1541 | break; | |
1542 | ||
b6958ce4 | 1543 | schedule(); |
b6958ce4 | 1544 | } |
d3bef15f | 1545 | |
e5c239cf | 1546 | finish_wait(&vcpu->wq, &wait); |
b6958ce4 ED |
1547 | } |
1548 | ||
8c84780d | 1549 | #ifndef CONFIG_S390 |
b6d33834 CD |
1550 | /* |
1551 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1552 | */ | |
1553 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1554 | { | |
1555 | int me; | |
1556 | int cpu = vcpu->cpu; | |
1557 | wait_queue_head_t *wqp; | |
1558 | ||
1559 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1560 | if (waitqueue_active(wqp)) { | |
1561 | wake_up_interruptible(wqp); | |
1562 | ++vcpu->stat.halt_wakeup; | |
1563 | } | |
1564 | ||
1565 | me = get_cpu(); | |
1566 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1567 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1568 | smp_send_reschedule(cpu); | |
1569 | put_cpu(); | |
1570 | } | |
8c84780d | 1571 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1572 | |
6aa8b732 AK |
1573 | void kvm_resched(struct kvm_vcpu *vcpu) |
1574 | { | |
3fca0365 YD |
1575 | if (!need_resched()) |
1576 | return; | |
6aa8b732 | 1577 | cond_resched(); |
6aa8b732 AK |
1578 | } |
1579 | EXPORT_SYMBOL_GPL(kvm_resched); | |
1580 | ||
41628d33 KW |
1581 | bool kvm_vcpu_yield_to(struct kvm_vcpu *target) |
1582 | { | |
1583 | struct pid *pid; | |
1584 | struct task_struct *task = NULL; | |
1585 | ||
1586 | rcu_read_lock(); | |
1587 | pid = rcu_dereference(target->pid); | |
1588 | if (pid) | |
1589 | task = get_pid_task(target->pid, PIDTYPE_PID); | |
1590 | rcu_read_unlock(); | |
1591 | if (!task) | |
1592 | return false; | |
1593 | if (task->flags & PF_VCPU) { | |
1594 | put_task_struct(task); | |
1595 | return false; | |
1596 | } | |
1597 | if (yield_to(task, 1)) { | |
1598 | put_task_struct(task); | |
1599 | return true; | |
1600 | } | |
1601 | put_task_struct(task); | |
1602 | return false; | |
1603 | } | |
1604 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
1605 | ||
06e48c51 R |
1606 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
1607 | /* | |
1608 | * Helper that checks whether a VCPU is eligible for directed yield. | |
1609 | * Most eligible candidate to yield is decided by following heuristics: | |
1610 | * | |
1611 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
1612 | * (preempted lock holder), indicated by @in_spin_loop. | |
1613 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
1614 | * | |
1615 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
1616 | * chance last time (mostly it has become eligible now since we have probably | |
1617 | * yielded to lockholder in last iteration. This is done by toggling | |
1618 | * @dy_eligible each time a VCPU checked for eligibility.) | |
1619 | * | |
1620 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
1621 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
1622 | * burning. Giving priority for a potential lock-holder increases lock | |
1623 | * progress. | |
1624 | * | |
1625 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
1626 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
1627 | * and continue with next VCPU and so on. | |
1628 | */ | |
1629 | bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) | |
1630 | { | |
1631 | bool eligible; | |
1632 | ||
1633 | eligible = !vcpu->spin_loop.in_spin_loop || | |
1634 | (vcpu->spin_loop.in_spin_loop && | |
1635 | vcpu->spin_loop.dy_eligible); | |
1636 | ||
1637 | if (vcpu->spin_loop.in_spin_loop) | |
1638 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
1639 | ||
1640 | return eligible; | |
1641 | } | |
1642 | #endif | |
217ece61 | 1643 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 1644 | { |
217ece61 RR |
1645 | struct kvm *kvm = me->kvm; |
1646 | struct kvm_vcpu *vcpu; | |
1647 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
1648 | int yielded = 0; | |
1649 | int pass; | |
1650 | int i; | |
d255f4f2 | 1651 | |
4c088493 | 1652 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
1653 | /* |
1654 | * We boost the priority of a VCPU that is runnable but not | |
1655 | * currently running, because it got preempted by something | |
1656 | * else and called schedule in __vcpu_run. Hopefully that | |
1657 | * VCPU is holding the lock that we need and will release it. | |
1658 | * We approximate round-robin by starting at the last boosted VCPU. | |
1659 | */ | |
1660 | for (pass = 0; pass < 2 && !yielded; pass++) { | |
1661 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
5cfc2aab | 1662 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
1663 | i = last_boosted_vcpu; |
1664 | continue; | |
1665 | } else if (pass && i > last_boosted_vcpu) | |
1666 | break; | |
1667 | if (vcpu == me) | |
1668 | continue; | |
1669 | if (waitqueue_active(&vcpu->wq)) | |
1670 | continue; | |
06e48c51 R |
1671 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
1672 | continue; | |
41628d33 | 1673 | if (kvm_vcpu_yield_to(vcpu)) { |
217ece61 RR |
1674 | kvm->last_boosted_vcpu = i; |
1675 | yielded = 1; | |
1676 | break; | |
1677 | } | |
217ece61 RR |
1678 | } |
1679 | } | |
4c088493 | 1680 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
1681 | |
1682 | /* Ensure vcpu is not eligible during next spinloop */ | |
1683 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
1684 | } |
1685 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
1686 | ||
e4a533a4 | 1687 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
1688 | { |
1689 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
1690 | struct page *page; |
1691 | ||
e4a533a4 | 1692 | if (vmf->pgoff == 0) |
039576c0 | 1693 | page = virt_to_page(vcpu->run); |
09566765 | 1694 | #ifdef CONFIG_X86 |
e4a533a4 | 1695 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 1696 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
1697 | #endif |
1698 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
1699 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
1700 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 1701 | #endif |
039576c0 | 1702 | else |
5b1c1493 | 1703 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 1704 | get_page(page); |
e4a533a4 | 1705 | vmf->page = page; |
1706 | return 0; | |
9a2bb7f4 AK |
1707 | } |
1708 | ||
f0f37e2f | 1709 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 1710 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
1711 | }; |
1712 | ||
1713 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
1714 | { | |
1715 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
1716 | return 0; | |
1717 | } | |
1718 | ||
bccf2150 AK |
1719 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
1720 | { | |
1721 | struct kvm_vcpu *vcpu = filp->private_data; | |
1722 | ||
66c0b394 | 1723 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
1724 | return 0; |
1725 | } | |
1726 | ||
3d3aab1b | 1727 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
1728 | .release = kvm_vcpu_release, |
1729 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
1dda606c AG |
1730 | #ifdef CONFIG_COMPAT |
1731 | .compat_ioctl = kvm_vcpu_compat_ioctl, | |
1732 | #endif | |
9a2bb7f4 | 1733 | .mmap = kvm_vcpu_mmap, |
6038f373 | 1734 | .llseek = noop_llseek, |
bccf2150 AK |
1735 | }; |
1736 | ||
1737 | /* | |
1738 | * Allocates an inode for the vcpu. | |
1739 | */ | |
1740 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
1741 | { | |
628ff7c1 | 1742 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR); |
bccf2150 AK |
1743 | } |
1744 | ||
c5ea7660 AK |
1745 | /* |
1746 | * Creates some virtual cpus. Good luck creating more than one. | |
1747 | */ | |
73880c80 | 1748 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
1749 | { |
1750 | int r; | |
988a2cae | 1751 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 1752 | |
73880c80 | 1753 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
1754 | if (IS_ERR(vcpu)) |
1755 | return PTR_ERR(vcpu); | |
c5ea7660 | 1756 | |
15ad7146 AK |
1757 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
1758 | ||
26e5215f AK |
1759 | r = kvm_arch_vcpu_setup(vcpu); |
1760 | if (r) | |
d780592b | 1761 | goto vcpu_destroy; |
26e5215f | 1762 | |
11ec2804 | 1763 | mutex_lock(&kvm->lock); |
3e515705 AK |
1764 | if (!kvm_vcpu_compatible(vcpu)) { |
1765 | r = -EINVAL; | |
1766 | goto unlock_vcpu_destroy; | |
1767 | } | |
73880c80 GN |
1768 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
1769 | r = -EINVAL; | |
d780592b | 1770 | goto unlock_vcpu_destroy; |
fb3f0f51 | 1771 | } |
73880c80 | 1772 | |
988a2cae GN |
1773 | kvm_for_each_vcpu(r, v, kvm) |
1774 | if (v->vcpu_id == id) { | |
73880c80 | 1775 | r = -EEXIST; |
d780592b | 1776 | goto unlock_vcpu_destroy; |
73880c80 GN |
1777 | } |
1778 | ||
1779 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 1780 | |
fb3f0f51 | 1781 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 1782 | kvm_get_kvm(kvm); |
bccf2150 | 1783 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
1784 | if (r < 0) { |
1785 | kvm_put_kvm(kvm); | |
d780592b | 1786 | goto unlock_vcpu_destroy; |
73880c80 GN |
1787 | } |
1788 | ||
1789 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
1790 | smp_wmb(); | |
1791 | atomic_inc(&kvm->online_vcpus); | |
1792 | ||
73880c80 | 1793 | mutex_unlock(&kvm->lock); |
fb3f0f51 | 1794 | return r; |
39c3b86e | 1795 | |
d780592b | 1796 | unlock_vcpu_destroy: |
7d8fece6 | 1797 | mutex_unlock(&kvm->lock); |
d780592b | 1798 | vcpu_destroy: |
d40ccc62 | 1799 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
1800 | return r; |
1801 | } | |
1802 | ||
1961d276 AK |
1803 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
1804 | { | |
1805 | if (sigset) { | |
1806 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1807 | vcpu->sigset_active = 1; | |
1808 | vcpu->sigset = *sigset; | |
1809 | } else | |
1810 | vcpu->sigset_active = 0; | |
1811 | return 0; | |
1812 | } | |
1813 | ||
bccf2150 AK |
1814 | static long kvm_vcpu_ioctl(struct file *filp, |
1815 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 1816 | { |
bccf2150 | 1817 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 1818 | void __user *argp = (void __user *)arg; |
313a3dc7 | 1819 | int r; |
fa3795a7 DH |
1820 | struct kvm_fpu *fpu = NULL; |
1821 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 1822 | |
6d4e4c4f AK |
1823 | if (vcpu->kvm->mm != current->mm) |
1824 | return -EIO; | |
2122ff5e AK |
1825 | |
1826 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) | |
1827 | /* | |
1828 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
1829 | * so vcpu_load() would break it. | |
1830 | */ | |
1831 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT) | |
1832 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); | |
1833 | #endif | |
1834 | ||
1835 | ||
1836 | vcpu_load(vcpu); | |
6aa8b732 | 1837 | switch (ioctl) { |
9a2bb7f4 | 1838 | case KVM_RUN: |
f0fe5108 AK |
1839 | r = -EINVAL; |
1840 | if (arg) | |
1841 | goto out; | |
b6c7a5dc | 1842 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 1843 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 1844 | break; |
6aa8b732 | 1845 | case KVM_GET_REGS: { |
3e4bb3ac | 1846 | struct kvm_regs *kvm_regs; |
6aa8b732 | 1847 | |
3e4bb3ac XZ |
1848 | r = -ENOMEM; |
1849 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
1850 | if (!kvm_regs) | |
6aa8b732 | 1851 | goto out; |
3e4bb3ac XZ |
1852 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
1853 | if (r) | |
1854 | goto out_free1; | |
6aa8b732 | 1855 | r = -EFAULT; |
3e4bb3ac XZ |
1856 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
1857 | goto out_free1; | |
6aa8b732 | 1858 | r = 0; |
3e4bb3ac XZ |
1859 | out_free1: |
1860 | kfree(kvm_regs); | |
6aa8b732 AK |
1861 | break; |
1862 | } | |
1863 | case KVM_SET_REGS: { | |
3e4bb3ac | 1864 | struct kvm_regs *kvm_regs; |
6aa8b732 | 1865 | |
3e4bb3ac | 1866 | r = -ENOMEM; |
ff5c2c03 SL |
1867 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
1868 | if (IS_ERR(kvm_regs)) { | |
1869 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 1870 | goto out; |
ff5c2c03 | 1871 | } |
3e4bb3ac | 1872 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
6aa8b732 | 1873 | if (r) |
3e4bb3ac | 1874 | goto out_free2; |
6aa8b732 | 1875 | r = 0; |
3e4bb3ac XZ |
1876 | out_free2: |
1877 | kfree(kvm_regs); | |
6aa8b732 AK |
1878 | break; |
1879 | } | |
1880 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
1881 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
1882 | r = -ENOMEM; | |
1883 | if (!kvm_sregs) | |
1884 | goto out; | |
1885 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
1886 | if (r) |
1887 | goto out; | |
1888 | r = -EFAULT; | |
fa3795a7 | 1889 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
1890 | goto out; |
1891 | r = 0; | |
1892 | break; | |
1893 | } | |
1894 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
1895 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
1896 | if (IS_ERR(kvm_sregs)) { | |
1897 | r = PTR_ERR(kvm_sregs); | |
6aa8b732 | 1898 | goto out; |
ff5c2c03 | 1899 | } |
fa3795a7 | 1900 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
1901 | if (r) |
1902 | goto out; | |
1903 | r = 0; | |
1904 | break; | |
1905 | } | |
62d9f0db MT |
1906 | case KVM_GET_MP_STATE: { |
1907 | struct kvm_mp_state mp_state; | |
1908 | ||
1909 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
1910 | if (r) | |
1911 | goto out; | |
1912 | r = -EFAULT; | |
1913 | if (copy_to_user(argp, &mp_state, sizeof mp_state)) | |
1914 | goto out; | |
1915 | r = 0; | |
1916 | break; | |
1917 | } | |
1918 | case KVM_SET_MP_STATE: { | |
1919 | struct kvm_mp_state mp_state; | |
1920 | ||
1921 | r = -EFAULT; | |
1922 | if (copy_from_user(&mp_state, argp, sizeof mp_state)) | |
1923 | goto out; | |
1924 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
1925 | if (r) | |
1926 | goto out; | |
1927 | r = 0; | |
1928 | break; | |
1929 | } | |
6aa8b732 AK |
1930 | case KVM_TRANSLATE: { |
1931 | struct kvm_translation tr; | |
1932 | ||
1933 | r = -EFAULT; | |
2f366987 | 1934 | if (copy_from_user(&tr, argp, sizeof tr)) |
6aa8b732 | 1935 | goto out; |
8b006791 | 1936 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
1937 | if (r) |
1938 | goto out; | |
1939 | r = -EFAULT; | |
2f366987 | 1940 | if (copy_to_user(argp, &tr, sizeof tr)) |
6aa8b732 AK |
1941 | goto out; |
1942 | r = 0; | |
1943 | break; | |
1944 | } | |
d0bfb940 JK |
1945 | case KVM_SET_GUEST_DEBUG: { |
1946 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
1947 | |
1948 | r = -EFAULT; | |
2f366987 | 1949 | if (copy_from_user(&dbg, argp, sizeof dbg)) |
6aa8b732 | 1950 | goto out; |
d0bfb940 | 1951 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
1952 | if (r) |
1953 | goto out; | |
1954 | r = 0; | |
1955 | break; | |
1956 | } | |
1961d276 AK |
1957 | case KVM_SET_SIGNAL_MASK: { |
1958 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
1959 | struct kvm_signal_mask kvm_sigmask; | |
1960 | sigset_t sigset, *p; | |
1961 | ||
1962 | p = NULL; | |
1963 | if (argp) { | |
1964 | r = -EFAULT; | |
1965 | if (copy_from_user(&kvm_sigmask, argp, | |
1966 | sizeof kvm_sigmask)) | |
1967 | goto out; | |
1968 | r = -EINVAL; | |
1969 | if (kvm_sigmask.len != sizeof sigset) | |
1970 | goto out; | |
1971 | r = -EFAULT; | |
1972 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
1973 | sizeof sigset)) | |
1974 | goto out; | |
1975 | p = &sigset; | |
1976 | } | |
376d41ff | 1977 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
1978 | break; |
1979 | } | |
b8836737 | 1980 | case KVM_GET_FPU: { |
fa3795a7 DH |
1981 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
1982 | r = -ENOMEM; | |
1983 | if (!fpu) | |
1984 | goto out; | |
1985 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
1986 | if (r) |
1987 | goto out; | |
1988 | r = -EFAULT; | |
fa3795a7 | 1989 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
1990 | goto out; |
1991 | r = 0; | |
1992 | break; | |
1993 | } | |
1994 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
1995 | fpu = memdup_user(argp, sizeof(*fpu)); |
1996 | if (IS_ERR(fpu)) { | |
1997 | r = PTR_ERR(fpu); | |
b8836737 | 1998 | goto out; |
ff5c2c03 | 1999 | } |
fa3795a7 | 2000 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2001 | if (r) |
2002 | goto out; | |
2003 | r = 0; | |
2004 | break; | |
2005 | } | |
bccf2150 | 2006 | default: |
313a3dc7 | 2007 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2008 | } |
2009 | out: | |
2122ff5e | 2010 | vcpu_put(vcpu); |
fa3795a7 DH |
2011 | kfree(fpu); |
2012 | kfree(kvm_sregs); | |
bccf2150 AK |
2013 | return r; |
2014 | } | |
2015 | ||
1dda606c AG |
2016 | #ifdef CONFIG_COMPAT |
2017 | static long kvm_vcpu_compat_ioctl(struct file *filp, | |
2018 | unsigned int ioctl, unsigned long arg) | |
2019 | { | |
2020 | struct kvm_vcpu *vcpu = filp->private_data; | |
2021 | void __user *argp = compat_ptr(arg); | |
2022 | int r; | |
2023 | ||
2024 | if (vcpu->kvm->mm != current->mm) | |
2025 | return -EIO; | |
2026 | ||
2027 | switch (ioctl) { | |
2028 | case KVM_SET_SIGNAL_MASK: { | |
2029 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2030 | struct kvm_signal_mask kvm_sigmask; | |
2031 | compat_sigset_t csigset; | |
2032 | sigset_t sigset; | |
2033 | ||
2034 | if (argp) { | |
2035 | r = -EFAULT; | |
2036 | if (copy_from_user(&kvm_sigmask, argp, | |
2037 | sizeof kvm_sigmask)) | |
2038 | goto out; | |
2039 | r = -EINVAL; | |
2040 | if (kvm_sigmask.len != sizeof csigset) | |
2041 | goto out; | |
2042 | r = -EFAULT; | |
2043 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
2044 | sizeof csigset)) | |
2045 | goto out; | |
2046 | } | |
2047 | sigset_from_compat(&sigset, &csigset); | |
2048 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2049 | break; | |
2050 | } | |
2051 | default: | |
2052 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2053 | } | |
2054 | ||
2055 | out: | |
2056 | return r; | |
2057 | } | |
2058 | #endif | |
2059 | ||
bccf2150 AK |
2060 | static long kvm_vm_ioctl(struct file *filp, |
2061 | unsigned int ioctl, unsigned long arg) | |
2062 | { | |
2063 | struct kvm *kvm = filp->private_data; | |
2064 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2065 | int r; |
bccf2150 | 2066 | |
6d4e4c4f AK |
2067 | if (kvm->mm != current->mm) |
2068 | return -EIO; | |
bccf2150 AK |
2069 | switch (ioctl) { |
2070 | case KVM_CREATE_VCPU: | |
2071 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
2072 | if (r < 0) | |
2073 | goto out; | |
2074 | break; | |
6fc138d2 IE |
2075 | case KVM_SET_USER_MEMORY_REGION: { |
2076 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2077 | ||
2078 | r = -EFAULT; | |
2079 | if (copy_from_user(&kvm_userspace_mem, argp, | |
2080 | sizeof kvm_userspace_mem)) | |
2081 | goto out; | |
2082 | ||
2083 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1); | |
6aa8b732 AK |
2084 | if (r) |
2085 | goto out; | |
2086 | break; | |
2087 | } | |
2088 | case KVM_GET_DIRTY_LOG: { | |
2089 | struct kvm_dirty_log log; | |
2090 | ||
2091 | r = -EFAULT; | |
2f366987 | 2092 | if (copy_from_user(&log, argp, sizeof log)) |
6aa8b732 | 2093 | goto out; |
2c6f5df9 | 2094 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2095 | if (r) |
2096 | goto out; | |
2097 | break; | |
2098 | } | |
5f94c174 LV |
2099 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2100 | case KVM_REGISTER_COALESCED_MMIO: { | |
2101 | struct kvm_coalesced_mmio_zone zone; | |
2102 | r = -EFAULT; | |
2103 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2104 | goto out; | |
5f94c174 LV |
2105 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
2106 | if (r) | |
2107 | goto out; | |
2108 | r = 0; | |
2109 | break; | |
2110 | } | |
2111 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2112 | struct kvm_coalesced_mmio_zone zone; | |
2113 | r = -EFAULT; | |
2114 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2115 | goto out; | |
5f94c174 LV |
2116 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
2117 | if (r) | |
2118 | goto out; | |
2119 | r = 0; | |
2120 | break; | |
2121 | } | |
2122 | #endif | |
721eecbf GH |
2123 | case KVM_IRQFD: { |
2124 | struct kvm_irqfd data; | |
2125 | ||
2126 | r = -EFAULT; | |
2127 | if (copy_from_user(&data, argp, sizeof data)) | |
2128 | goto out; | |
d4db2935 | 2129 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2130 | break; |
2131 | } | |
d34e6b17 GH |
2132 | case KVM_IOEVENTFD: { |
2133 | struct kvm_ioeventfd data; | |
2134 | ||
2135 | r = -EFAULT; | |
2136 | if (copy_from_user(&data, argp, sizeof data)) | |
2137 | goto out; | |
2138 | r = kvm_ioeventfd(kvm, &data); | |
2139 | break; | |
2140 | } | |
73880c80 GN |
2141 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2142 | case KVM_SET_BOOT_CPU_ID: | |
2143 | r = 0; | |
894a9c55 | 2144 | mutex_lock(&kvm->lock); |
73880c80 GN |
2145 | if (atomic_read(&kvm->online_vcpus) != 0) |
2146 | r = -EBUSY; | |
2147 | else | |
2148 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2149 | mutex_unlock(&kvm->lock); |
73880c80 | 2150 | break; |
07975ad3 JK |
2151 | #endif |
2152 | #ifdef CONFIG_HAVE_KVM_MSI | |
2153 | case KVM_SIGNAL_MSI: { | |
2154 | struct kvm_msi msi; | |
2155 | ||
2156 | r = -EFAULT; | |
2157 | if (copy_from_user(&msi, argp, sizeof msi)) | |
2158 | goto out; | |
2159 | r = kvm_send_userspace_msi(kvm, &msi); | |
2160 | break; | |
2161 | } | |
23d43cf9 CD |
2162 | #endif |
2163 | #ifdef __KVM_HAVE_IRQ_LINE | |
2164 | case KVM_IRQ_LINE_STATUS: | |
2165 | case KVM_IRQ_LINE: { | |
2166 | struct kvm_irq_level irq_event; | |
2167 | ||
2168 | r = -EFAULT; | |
2169 | if (copy_from_user(&irq_event, argp, sizeof irq_event)) | |
2170 | goto out; | |
2171 | ||
2172 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event); | |
2173 | if (r) | |
2174 | goto out; | |
2175 | ||
2176 | r = -EFAULT; | |
2177 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
2178 | if (copy_to_user(argp, &irq_event, sizeof irq_event)) | |
2179 | goto out; | |
2180 | } | |
2181 | ||
2182 | r = 0; | |
2183 | break; | |
2184 | } | |
73880c80 | 2185 | #endif |
f17abe9a | 2186 | default: |
1fe779f8 | 2187 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
bfd99ff5 AK |
2188 | if (r == -ENOTTY) |
2189 | r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg); | |
f17abe9a AK |
2190 | } |
2191 | out: | |
2192 | return r; | |
2193 | } | |
2194 | ||
6ff5894c AB |
2195 | #ifdef CONFIG_COMPAT |
2196 | struct compat_kvm_dirty_log { | |
2197 | __u32 slot; | |
2198 | __u32 padding1; | |
2199 | union { | |
2200 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2201 | __u64 padding2; | |
2202 | }; | |
2203 | }; | |
2204 | ||
2205 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2206 | unsigned int ioctl, unsigned long arg) | |
2207 | { | |
2208 | struct kvm *kvm = filp->private_data; | |
2209 | int r; | |
2210 | ||
2211 | if (kvm->mm != current->mm) | |
2212 | return -EIO; | |
2213 | switch (ioctl) { | |
2214 | case KVM_GET_DIRTY_LOG: { | |
2215 | struct compat_kvm_dirty_log compat_log; | |
2216 | struct kvm_dirty_log log; | |
2217 | ||
2218 | r = -EFAULT; | |
2219 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2220 | sizeof(compat_log))) | |
2221 | goto out; | |
2222 | log.slot = compat_log.slot; | |
2223 | log.padding1 = compat_log.padding1; | |
2224 | log.padding2 = compat_log.padding2; | |
2225 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2226 | ||
2227 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
2228 | if (r) | |
2229 | goto out; | |
2230 | break; | |
2231 | } | |
2232 | default: | |
2233 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2234 | } | |
2235 | ||
2236 | out: | |
2237 | return r; | |
2238 | } | |
2239 | #endif | |
2240 | ||
e4a533a4 | 2241 | static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
f17abe9a | 2242 | { |
777b3f49 MT |
2243 | struct page *page[1]; |
2244 | unsigned long addr; | |
2245 | int npages; | |
2246 | gfn_t gfn = vmf->pgoff; | |
f17abe9a | 2247 | struct kvm *kvm = vma->vm_file->private_data; |
f17abe9a | 2248 | |
777b3f49 MT |
2249 | addr = gfn_to_hva(kvm, gfn); |
2250 | if (kvm_is_error_hva(addr)) | |
e4a533a4 | 2251 | return VM_FAULT_SIGBUS; |
777b3f49 MT |
2252 | |
2253 | npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page, | |
2254 | NULL); | |
2255 | if (unlikely(npages != 1)) | |
e4a533a4 | 2256 | return VM_FAULT_SIGBUS; |
777b3f49 MT |
2257 | |
2258 | vmf->page = page[0]; | |
e4a533a4 | 2259 | return 0; |
f17abe9a AK |
2260 | } |
2261 | ||
f0f37e2f | 2262 | static const struct vm_operations_struct kvm_vm_vm_ops = { |
e4a533a4 | 2263 | .fault = kvm_vm_fault, |
f17abe9a AK |
2264 | }; |
2265 | ||
2266 | static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma) | |
2267 | { | |
2268 | vma->vm_ops = &kvm_vm_vm_ops; | |
2269 | return 0; | |
2270 | } | |
2271 | ||
3d3aab1b | 2272 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2273 | .release = kvm_vm_release, |
2274 | .unlocked_ioctl = kvm_vm_ioctl, | |
6ff5894c AB |
2275 | #ifdef CONFIG_COMPAT |
2276 | .compat_ioctl = kvm_vm_compat_ioctl, | |
2277 | #endif | |
f17abe9a | 2278 | .mmap = kvm_vm_mmap, |
6038f373 | 2279 | .llseek = noop_llseek, |
f17abe9a AK |
2280 | }; |
2281 | ||
e08b9637 | 2282 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2283 | { |
aac87636 | 2284 | int r; |
f17abe9a AK |
2285 | struct kvm *kvm; |
2286 | ||
e08b9637 | 2287 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2288 | if (IS_ERR(kvm)) |
2289 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2290 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2291 | r = kvm_coalesced_mmio_init(kvm); | |
2292 | if (r < 0) { | |
2293 | kvm_put_kvm(kvm); | |
2294 | return r; | |
2295 | } | |
2296 | #endif | |
aac87636 HC |
2297 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
2298 | if (r < 0) | |
66c0b394 | 2299 | kvm_put_kvm(kvm); |
f17abe9a | 2300 | |
aac87636 | 2301 | return r; |
f17abe9a AK |
2302 | } |
2303 | ||
1a811b61 AK |
2304 | static long kvm_dev_ioctl_check_extension_generic(long arg) |
2305 | { | |
2306 | switch (arg) { | |
ca9edaee | 2307 | case KVM_CAP_USER_MEMORY: |
1a811b61 | 2308 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
4cd481f6 | 2309 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: |
73880c80 GN |
2310 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2311 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2312 | #endif | |
a9c7399d | 2313 | case KVM_CAP_INTERNAL_ERROR_DATA: |
07975ad3 JK |
2314 | #ifdef CONFIG_HAVE_KVM_MSI |
2315 | case KVM_CAP_SIGNAL_MSI: | |
2316 | #endif | |
1a811b61 | 2317 | return 1; |
9900b4b4 | 2318 | #ifdef KVM_CAP_IRQ_ROUTING |
399ec807 | 2319 | case KVM_CAP_IRQ_ROUTING: |
36463146 | 2320 | return KVM_MAX_IRQ_ROUTES; |
399ec807 | 2321 | #endif |
1a811b61 AK |
2322 | default: |
2323 | break; | |
2324 | } | |
2325 | return kvm_dev_ioctl_check_extension(arg); | |
2326 | } | |
2327 | ||
f17abe9a AK |
2328 | static long kvm_dev_ioctl(struct file *filp, |
2329 | unsigned int ioctl, unsigned long arg) | |
2330 | { | |
07c45a36 | 2331 | long r = -EINVAL; |
f17abe9a AK |
2332 | |
2333 | switch (ioctl) { | |
2334 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2335 | r = -EINVAL; |
2336 | if (arg) | |
2337 | goto out; | |
f17abe9a AK |
2338 | r = KVM_API_VERSION; |
2339 | break; | |
2340 | case KVM_CREATE_VM: | |
e08b9637 | 2341 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2342 | break; |
018d00d2 | 2343 | case KVM_CHECK_EXTENSION: |
1a811b61 | 2344 | r = kvm_dev_ioctl_check_extension_generic(arg); |
5d308f45 | 2345 | break; |
07c45a36 AK |
2346 | case KVM_GET_VCPU_MMAP_SIZE: |
2347 | r = -EINVAL; | |
2348 | if (arg) | |
2349 | goto out; | |
adb1ff46 AK |
2350 | r = PAGE_SIZE; /* struct kvm_run */ |
2351 | #ifdef CONFIG_X86 | |
2352 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2353 | #endif |
2354 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2355 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2356 | #endif |
07c45a36 | 2357 | break; |
d4c9ff2d FEL |
2358 | case KVM_TRACE_ENABLE: |
2359 | case KVM_TRACE_PAUSE: | |
2360 | case KVM_TRACE_DISABLE: | |
2023a29c | 2361 | r = -EOPNOTSUPP; |
d4c9ff2d | 2362 | break; |
6aa8b732 | 2363 | default: |
043405e1 | 2364 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2365 | } |
2366 | out: | |
2367 | return r; | |
2368 | } | |
2369 | ||
6aa8b732 | 2370 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2371 | .unlocked_ioctl = kvm_dev_ioctl, |
2372 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2373 | .llseek = noop_llseek, |
6aa8b732 AK |
2374 | }; |
2375 | ||
2376 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2377 | KVM_MINOR, |
6aa8b732 AK |
2378 | "kvm", |
2379 | &kvm_chardev_ops, | |
2380 | }; | |
2381 | ||
75b7127c | 2382 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2383 | { |
2384 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2385 | int r; |
1b6c0168 | 2386 | |
7f59f492 | 2387 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2388 | return; |
10474ae8 | 2389 | |
7f59f492 | 2390 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 AG |
2391 | |
2392 | r = kvm_arch_hardware_enable(NULL); | |
2393 | ||
2394 | if (r) { | |
2395 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2396 | atomic_inc(&hardware_enable_failed); | |
2397 | printk(KERN_INFO "kvm: enabling virtualization on " | |
2398 | "CPU%d failed\n", cpu); | |
2399 | } | |
1b6c0168 AK |
2400 | } |
2401 | ||
75b7127c TY |
2402 | static void hardware_enable(void *junk) |
2403 | { | |
e935b837 | 2404 | raw_spin_lock(&kvm_lock); |
75b7127c | 2405 | hardware_enable_nolock(junk); |
e935b837 | 2406 | raw_spin_unlock(&kvm_lock); |
75b7127c TY |
2407 | } |
2408 | ||
2409 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2410 | { |
2411 | int cpu = raw_smp_processor_id(); | |
2412 | ||
7f59f492 | 2413 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2414 | return; |
7f59f492 | 2415 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
e9b11c17 | 2416 | kvm_arch_hardware_disable(NULL); |
1b6c0168 AK |
2417 | } |
2418 | ||
75b7127c TY |
2419 | static void hardware_disable(void *junk) |
2420 | { | |
e935b837 | 2421 | raw_spin_lock(&kvm_lock); |
75b7127c | 2422 | hardware_disable_nolock(junk); |
e935b837 | 2423 | raw_spin_unlock(&kvm_lock); |
75b7127c TY |
2424 | } |
2425 | ||
10474ae8 AG |
2426 | static void hardware_disable_all_nolock(void) |
2427 | { | |
2428 | BUG_ON(!kvm_usage_count); | |
2429 | ||
2430 | kvm_usage_count--; | |
2431 | if (!kvm_usage_count) | |
75b7127c | 2432 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2433 | } |
2434 | ||
2435 | static void hardware_disable_all(void) | |
2436 | { | |
e935b837 | 2437 | raw_spin_lock(&kvm_lock); |
10474ae8 | 2438 | hardware_disable_all_nolock(); |
e935b837 | 2439 | raw_spin_unlock(&kvm_lock); |
10474ae8 AG |
2440 | } |
2441 | ||
2442 | static int hardware_enable_all(void) | |
2443 | { | |
2444 | int r = 0; | |
2445 | ||
e935b837 | 2446 | raw_spin_lock(&kvm_lock); |
10474ae8 AG |
2447 | |
2448 | kvm_usage_count++; | |
2449 | if (kvm_usage_count == 1) { | |
2450 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 2451 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
2452 | |
2453 | if (atomic_read(&hardware_enable_failed)) { | |
2454 | hardware_disable_all_nolock(); | |
2455 | r = -EBUSY; | |
2456 | } | |
2457 | } | |
2458 | ||
e935b837 | 2459 | raw_spin_unlock(&kvm_lock); |
10474ae8 AG |
2460 | |
2461 | return r; | |
2462 | } | |
2463 | ||
774c47f1 AK |
2464 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
2465 | void *v) | |
2466 | { | |
2467 | int cpu = (long)v; | |
2468 | ||
10474ae8 AG |
2469 | if (!kvm_usage_count) |
2470 | return NOTIFY_OK; | |
2471 | ||
1a6f4d7f | 2472 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 2473 | switch (val) { |
cec9ad27 | 2474 | case CPU_DYING: |
6ec8a856 AK |
2475 | printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n", |
2476 | cpu); | |
2477 | hardware_disable(NULL); | |
2478 | break; | |
da908f2f | 2479 | case CPU_STARTING: |
43934a38 JK |
2480 | printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n", |
2481 | cpu); | |
da908f2f | 2482 | hardware_enable(NULL); |
774c47f1 AK |
2483 | break; |
2484 | } | |
2485 | return NOTIFY_OK; | |
2486 | } | |
2487 | ||
4ecac3fd | 2488 | |
b7c4145b | 2489 | asmlinkage void kvm_spurious_fault(void) |
4ecac3fd | 2490 | { |
4ecac3fd AK |
2491 | /* Fault while not rebooting. We want the trace. */ |
2492 | BUG(); | |
2493 | } | |
b7c4145b | 2494 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); |
4ecac3fd | 2495 | |
9a2b85c6 | 2496 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 2497 | void *v) |
9a2b85c6 | 2498 | { |
8e1c1815 SY |
2499 | /* |
2500 | * Some (well, at least mine) BIOSes hang on reboot if | |
2501 | * in vmx root mode. | |
2502 | * | |
2503 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
2504 | */ | |
2505 | printk(KERN_INFO "kvm: exiting hardware virtualization\n"); | |
2506 | kvm_rebooting = true; | |
75b7127c | 2507 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
2508 | return NOTIFY_OK; |
2509 | } | |
2510 | ||
2511 | static struct notifier_block kvm_reboot_notifier = { | |
2512 | .notifier_call = kvm_reboot, | |
2513 | .priority = 0, | |
2514 | }; | |
2515 | ||
e93f8a0f | 2516 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
2517 | { |
2518 | int i; | |
2519 | ||
2520 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 2521 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
2522 | |
2523 | kvm_iodevice_destructor(pos); | |
2524 | } | |
e93f8a0f | 2525 | kfree(bus); |
2eeb2e94 GH |
2526 | } |
2527 | ||
743eeb0b SL |
2528 | int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
2529 | { | |
2530 | const struct kvm_io_range *r1 = p1; | |
2531 | const struct kvm_io_range *r2 = p2; | |
2532 | ||
2533 | if (r1->addr < r2->addr) | |
2534 | return -1; | |
2535 | if (r1->addr + r1->len > r2->addr + r2->len) | |
2536 | return 1; | |
2537 | return 0; | |
2538 | } | |
2539 | ||
2540 | int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, | |
2541 | gpa_t addr, int len) | |
2542 | { | |
743eeb0b SL |
2543 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
2544 | .addr = addr, | |
2545 | .len = len, | |
2546 | .dev = dev, | |
2547 | }; | |
2548 | ||
2549 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
2550 | kvm_io_bus_sort_cmp, NULL); | |
2551 | ||
2552 | return 0; | |
2553 | } | |
2554 | ||
2555 | int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, | |
2556 | gpa_t addr, int len) | |
2557 | { | |
2558 | struct kvm_io_range *range, key; | |
2559 | int off; | |
2560 | ||
2561 | key = (struct kvm_io_range) { | |
2562 | .addr = addr, | |
2563 | .len = len, | |
2564 | }; | |
2565 | ||
2566 | range = bsearch(&key, bus->range, bus->dev_count, | |
2567 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
2568 | if (range == NULL) | |
2569 | return -ENOENT; | |
2570 | ||
2571 | off = range - bus->range; | |
2572 | ||
2573 | while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0) | |
2574 | off--; | |
2575 | ||
2576 | return off; | |
2577 | } | |
2578 | ||
bda9020e | 2579 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e93f8a0f | 2580 | int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 2581 | int len, const void *val) |
2eeb2e94 | 2582 | { |
743eeb0b | 2583 | int idx; |
90d83dc3 | 2584 | struct kvm_io_bus *bus; |
743eeb0b SL |
2585 | struct kvm_io_range range; |
2586 | ||
2587 | range = (struct kvm_io_range) { | |
2588 | .addr = addr, | |
2589 | .len = len, | |
2590 | }; | |
90d83dc3 LJ |
2591 | |
2592 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
743eeb0b SL |
2593 | idx = kvm_io_bus_get_first_dev(bus, addr, len); |
2594 | if (idx < 0) | |
2595 | return -EOPNOTSUPP; | |
2596 | ||
2597 | while (idx < bus->dev_count && | |
2598 | kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) { | |
2599 | if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val)) | |
bda9020e | 2600 | return 0; |
743eeb0b SL |
2601 | idx++; |
2602 | } | |
2603 | ||
bda9020e MT |
2604 | return -EOPNOTSUPP; |
2605 | } | |
2eeb2e94 | 2606 | |
bda9020e | 2607 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e93f8a0f MT |
2608 | int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2609 | int len, void *val) | |
bda9020e | 2610 | { |
743eeb0b | 2611 | int idx; |
90d83dc3 | 2612 | struct kvm_io_bus *bus; |
743eeb0b SL |
2613 | struct kvm_io_range range; |
2614 | ||
2615 | range = (struct kvm_io_range) { | |
2616 | .addr = addr, | |
2617 | .len = len, | |
2618 | }; | |
e93f8a0f | 2619 | |
90d83dc3 | 2620 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); |
743eeb0b SL |
2621 | idx = kvm_io_bus_get_first_dev(bus, addr, len); |
2622 | if (idx < 0) | |
2623 | return -EOPNOTSUPP; | |
2624 | ||
2625 | while (idx < bus->dev_count && | |
2626 | kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) { | |
2627 | if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val)) | |
bda9020e | 2628 | return 0; |
743eeb0b SL |
2629 | idx++; |
2630 | } | |
2631 | ||
bda9020e | 2632 | return -EOPNOTSUPP; |
2eeb2e94 GH |
2633 | } |
2634 | ||
79fac95e | 2635 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
2636 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2637 | int len, struct kvm_io_device *dev) | |
6c474694 | 2638 | { |
e93f8a0f | 2639 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 2640 | |
e93f8a0f | 2641 | bus = kvm->buses[bus_idx]; |
a1300716 | 2642 | if (bus->dev_count > NR_IOBUS_DEVS - 1) |
090b7aff | 2643 | return -ENOSPC; |
2eeb2e94 | 2644 | |
a1300716 AK |
2645 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
2646 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
2647 | if (!new_bus) |
2648 | return -ENOMEM; | |
a1300716 AK |
2649 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
2650 | sizeof(struct kvm_io_range))); | |
743eeb0b | 2651 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
2652 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
2653 | synchronize_srcu_expedited(&kvm->srcu); | |
2654 | kfree(bus); | |
090b7aff GH |
2655 | |
2656 | return 0; | |
2657 | } | |
2658 | ||
79fac95e | 2659 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
2660 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
2661 | struct kvm_io_device *dev) | |
090b7aff | 2662 | { |
e93f8a0f MT |
2663 | int i, r; |
2664 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 2665 | |
cdfca7b3 | 2666 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 2667 | r = -ENOENT; |
a1300716 AK |
2668 | for (i = 0; i < bus->dev_count; i++) |
2669 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 2670 | r = 0; |
090b7aff GH |
2671 | break; |
2672 | } | |
e93f8a0f | 2673 | |
a1300716 | 2674 | if (r) |
e93f8a0f | 2675 | return r; |
a1300716 AK |
2676 | |
2677 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
2678 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
2679 | if (!new_bus) | |
2680 | return -ENOMEM; | |
2681 | ||
2682 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
2683 | new_bus->dev_count--; | |
2684 | memcpy(new_bus->range + i, bus->range + i + 1, | |
2685 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
2686 | |
2687 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
2688 | synchronize_srcu_expedited(&kvm->srcu); | |
2689 | kfree(bus); | |
2690 | return r; | |
2eeb2e94 GH |
2691 | } |
2692 | ||
774c47f1 AK |
2693 | static struct notifier_block kvm_cpu_notifier = { |
2694 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
2695 | }; |
2696 | ||
8b88b099 | 2697 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
2698 | { |
2699 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
2700 | struct kvm *kvm; |
2701 | ||
8b88b099 | 2702 | *val = 0; |
e935b837 | 2703 | raw_spin_lock(&kvm_lock); |
ba1389b7 | 2704 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 2705 | *val += *(u32 *)((void *)kvm + offset); |
e935b837 | 2706 | raw_spin_unlock(&kvm_lock); |
8b88b099 | 2707 | return 0; |
ba1389b7 AK |
2708 | } |
2709 | ||
2710 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
2711 | ||
8b88b099 | 2712 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
2713 | { |
2714 | unsigned offset = (long)_offset; | |
1165f5fe AK |
2715 | struct kvm *kvm; |
2716 | struct kvm_vcpu *vcpu; | |
2717 | int i; | |
2718 | ||
8b88b099 | 2719 | *val = 0; |
e935b837 | 2720 | raw_spin_lock(&kvm_lock); |
1165f5fe | 2721 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
2722 | kvm_for_each_vcpu(i, vcpu, kvm) |
2723 | *val += *(u32 *)((void *)vcpu + offset); | |
2724 | ||
e935b837 | 2725 | raw_spin_unlock(&kvm_lock); |
8b88b099 | 2726 | return 0; |
1165f5fe AK |
2727 | } |
2728 | ||
ba1389b7 AK |
2729 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
2730 | ||
828c0950 | 2731 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
2732 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
2733 | [KVM_STAT_VM] = &vm_stat_fops, | |
2734 | }; | |
1165f5fe | 2735 | |
4f69b680 | 2736 | static int kvm_init_debug(void) |
6aa8b732 | 2737 | { |
4f69b680 | 2738 | int r = -EFAULT; |
6aa8b732 AK |
2739 | struct kvm_stats_debugfs_item *p; |
2740 | ||
76f7c879 | 2741 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
2742 | if (kvm_debugfs_dir == NULL) |
2743 | goto out; | |
2744 | ||
2745 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 2746 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 2747 | (void *)(long)p->offset, |
ba1389b7 | 2748 | stat_fops[p->kind]); |
4f69b680 H |
2749 | if (p->dentry == NULL) |
2750 | goto out_dir; | |
2751 | } | |
2752 | ||
2753 | return 0; | |
2754 | ||
2755 | out_dir: | |
2756 | debugfs_remove_recursive(kvm_debugfs_dir); | |
2757 | out: | |
2758 | return r; | |
6aa8b732 AK |
2759 | } |
2760 | ||
2761 | static void kvm_exit_debug(void) | |
2762 | { | |
2763 | struct kvm_stats_debugfs_item *p; | |
2764 | ||
2765 | for (p = debugfs_entries; p->name; ++p) | |
2766 | debugfs_remove(p->dentry); | |
76f7c879 | 2767 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
2768 | } |
2769 | ||
fb3600cc | 2770 | static int kvm_suspend(void) |
59ae6c6b | 2771 | { |
10474ae8 | 2772 | if (kvm_usage_count) |
75b7127c | 2773 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
2774 | return 0; |
2775 | } | |
2776 | ||
fb3600cc | 2777 | static void kvm_resume(void) |
59ae6c6b | 2778 | { |
ca84d1a2 | 2779 | if (kvm_usage_count) { |
e935b837 | 2780 | WARN_ON(raw_spin_is_locked(&kvm_lock)); |
75b7127c | 2781 | hardware_enable_nolock(NULL); |
ca84d1a2 | 2782 | } |
59ae6c6b AK |
2783 | } |
2784 | ||
fb3600cc | 2785 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
2786 | .suspend = kvm_suspend, |
2787 | .resume = kvm_resume, | |
2788 | }; | |
2789 | ||
15ad7146 AK |
2790 | static inline |
2791 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
2792 | { | |
2793 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
2794 | } | |
2795 | ||
2796 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
2797 | { | |
2798 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
2799 | ||
e9b11c17 | 2800 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
2801 | } |
2802 | ||
2803 | static void kvm_sched_out(struct preempt_notifier *pn, | |
2804 | struct task_struct *next) | |
2805 | { | |
2806 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
2807 | ||
e9b11c17 | 2808 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
2809 | } |
2810 | ||
0ee75bea | 2811 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 2812 | struct module *module) |
6aa8b732 AK |
2813 | { |
2814 | int r; | |
002c7f7c | 2815 | int cpu; |
6aa8b732 | 2816 | |
f8c16bba ZX |
2817 | r = kvm_arch_init(opaque); |
2818 | if (r) | |
d2308784 | 2819 | goto out_fail; |
cb498ea2 | 2820 | |
8437a617 | 2821 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
2822 | r = -ENOMEM; |
2823 | goto out_free_0; | |
2824 | } | |
2825 | ||
e9b11c17 | 2826 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 2827 | if (r < 0) |
7f59f492 | 2828 | goto out_free_0a; |
6aa8b732 | 2829 | |
002c7f7c YS |
2830 | for_each_online_cpu(cpu) { |
2831 | smp_call_function_single(cpu, | |
e9b11c17 | 2832 | kvm_arch_check_processor_compat, |
8691e5a8 | 2833 | &r, 1); |
002c7f7c | 2834 | if (r < 0) |
d2308784 | 2835 | goto out_free_1; |
002c7f7c YS |
2836 | } |
2837 | ||
774c47f1 AK |
2838 | r = register_cpu_notifier(&kvm_cpu_notifier); |
2839 | if (r) | |
d2308784 | 2840 | goto out_free_2; |
6aa8b732 AK |
2841 | register_reboot_notifier(&kvm_reboot_notifier); |
2842 | ||
c16f862d | 2843 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
2844 | if (!vcpu_align) |
2845 | vcpu_align = __alignof__(struct kvm_vcpu); | |
2846 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 2847 | 0, NULL); |
c16f862d RR |
2848 | if (!kvm_vcpu_cache) { |
2849 | r = -ENOMEM; | |
fb3600cc | 2850 | goto out_free_3; |
c16f862d RR |
2851 | } |
2852 | ||
af585b92 GN |
2853 | r = kvm_async_pf_init(); |
2854 | if (r) | |
2855 | goto out_free; | |
2856 | ||
6aa8b732 | 2857 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
2858 | kvm_vm_fops.owner = module; |
2859 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
2860 | |
2861 | r = misc_register(&kvm_dev); | |
2862 | if (r) { | |
d77c26fc | 2863 | printk(KERN_ERR "kvm: misc device register failed\n"); |
af585b92 | 2864 | goto out_unreg; |
6aa8b732 AK |
2865 | } |
2866 | ||
fb3600cc RW |
2867 | register_syscore_ops(&kvm_syscore_ops); |
2868 | ||
15ad7146 AK |
2869 | kvm_preempt_ops.sched_in = kvm_sched_in; |
2870 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
2871 | ||
4f69b680 H |
2872 | r = kvm_init_debug(); |
2873 | if (r) { | |
2874 | printk(KERN_ERR "kvm: create debugfs files failed\n"); | |
2875 | goto out_undebugfs; | |
2876 | } | |
0ea4ed8e | 2877 | |
c7addb90 | 2878 | return 0; |
6aa8b732 | 2879 | |
4f69b680 H |
2880 | out_undebugfs: |
2881 | unregister_syscore_ops(&kvm_syscore_ops); | |
af585b92 GN |
2882 | out_unreg: |
2883 | kvm_async_pf_deinit(); | |
6aa8b732 | 2884 | out_free: |
c16f862d | 2885 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 2886 | out_free_3: |
6aa8b732 | 2887 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 2888 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 2889 | out_free_2: |
d2308784 | 2890 | out_free_1: |
e9b11c17 | 2891 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
2892 | out_free_0a: |
2893 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 2894 | out_free_0: |
f8c16bba | 2895 | kvm_arch_exit(); |
d2308784 | 2896 | out_fail: |
6aa8b732 AK |
2897 | return r; |
2898 | } | |
cb498ea2 | 2899 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 2900 | |
cb498ea2 | 2901 | void kvm_exit(void) |
6aa8b732 | 2902 | { |
0ea4ed8e | 2903 | kvm_exit_debug(); |
6aa8b732 | 2904 | misc_deregister(&kvm_dev); |
c16f862d | 2905 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 2906 | kvm_async_pf_deinit(); |
fb3600cc | 2907 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 2908 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 2909 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 2910 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 2911 | kvm_arch_hardware_unsetup(); |
f8c16bba | 2912 | kvm_arch_exit(); |
7f59f492 | 2913 | free_cpumask_var(cpus_hardware_enabled); |
6aa8b732 | 2914 | } |
cb498ea2 | 2915 | EXPORT_SYMBOL_GPL(kvm_exit); |