]>
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 | 54 | #include <asm/io.h> |
2ea75be3 | 55 | #include <asm/ioctl.h> |
e495606d | 56 | #include <asm/uaccess.h> |
3e021bf5 | 57 | #include <asm/pgtable.h> |
6aa8b732 | 58 | |
5f94c174 | 59 | #include "coalesced_mmio.h" |
af585b92 | 60 | #include "async_pf.h" |
3c3c29fd | 61 | #include "vfio.h" |
5f94c174 | 62 | |
229456fc MT |
63 | #define CREATE_TRACE_POINTS |
64 | #include <trace/events/kvm.h> | |
65 | ||
6aa8b732 AK |
66 | MODULE_AUTHOR("Qumranet"); |
67 | MODULE_LICENSE("GPL"); | |
68 | ||
f7819512 PB |
69 | unsigned int halt_poll_ns = 0; |
70 | module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR); | |
71 | ||
fa40a821 MT |
72 | /* |
73 | * Ordering of locks: | |
74 | * | |
fae3a353 | 75 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
76 | */ |
77 | ||
2f303b74 | 78 | DEFINE_SPINLOCK(kvm_lock); |
4a937f96 | 79 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 80 | LIST_HEAD(vm_list); |
133de902 | 81 | |
7f59f492 | 82 | static cpumask_var_t cpus_hardware_enabled; |
10474ae8 AG |
83 | static int kvm_usage_count = 0; |
84 | static atomic_t hardware_enable_failed; | |
1b6c0168 | 85 | |
c16f862d RR |
86 | struct kmem_cache *kvm_vcpu_cache; |
87 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 88 | |
15ad7146 AK |
89 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
90 | ||
76f7c879 | 91 | struct dentry *kvm_debugfs_dir; |
6aa8b732 | 92 | |
bccf2150 AK |
93 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
94 | unsigned long arg); | |
de8e5d74 | 95 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
96 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
97 | unsigned long arg); | |
98 | #endif | |
10474ae8 AG |
99 | static int hardware_enable_all(void); |
100 | static void hardware_disable_all(void); | |
bccf2150 | 101 | |
e93f8a0f | 102 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e SH |
103 | |
104 | static void kvm_release_pfn_dirty(pfn_t pfn); | |
105 | static void mark_page_dirty_in_slot(struct kvm *kvm, | |
106 | struct kvm_memory_slot *memslot, gfn_t gfn); | |
e93f8a0f | 107 | |
52480137 | 108 | __visible bool kvm_rebooting; |
b7c4145b | 109 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 110 | |
54dee993 MT |
111 | static bool largepages_enabled = true; |
112 | ||
bf4bea8e | 113 | bool kvm_is_reserved_pfn(pfn_t pfn) |
cbff90a7 | 114 | { |
11feeb49 | 115 | if (pfn_valid(pfn)) |
bf4bea8e | 116 | return PageReserved(pfn_to_page(pfn)); |
cbff90a7 BAY |
117 | |
118 | return true; | |
119 | } | |
120 | ||
bccf2150 AK |
121 | /* |
122 | * Switches to specified vcpu, until a matching vcpu_put() | |
123 | */ | |
9fc77441 | 124 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 125 | { |
15ad7146 AK |
126 | int cpu; |
127 | ||
9fc77441 MT |
128 | if (mutex_lock_killable(&vcpu->mutex)) |
129 | return -EINTR; | |
15ad7146 AK |
130 | cpu = get_cpu(); |
131 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 132 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 133 | put_cpu(); |
9fc77441 | 134 | return 0; |
6aa8b732 AK |
135 | } |
136 | ||
313a3dc7 | 137 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 138 | { |
15ad7146 | 139 | preempt_disable(); |
313a3dc7 | 140 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
141 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
142 | preempt_enable(); | |
6aa8b732 AK |
143 | mutex_unlock(&vcpu->mutex); |
144 | } | |
145 | ||
d9e368d6 AK |
146 | static void ack_flush(void *_completed) |
147 | { | |
d9e368d6 AK |
148 | } |
149 | ||
445b8236 | 150 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 151 | { |
597a5f55 | 152 | int i, cpu, me; |
6ef7a1bc RR |
153 | cpumask_var_t cpus; |
154 | bool called = true; | |
d9e368d6 | 155 | struct kvm_vcpu *vcpu; |
d9e368d6 | 156 | |
79f55997 | 157 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 158 | |
3cba4130 | 159 | me = get_cpu(); |
988a2cae | 160 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 161 | kvm_make_request(req, vcpu); |
d9e368d6 | 162 | cpu = vcpu->cpu; |
6b7e2d09 XG |
163 | |
164 | /* Set ->requests bit before we read ->mode */ | |
165 | smp_mb(); | |
166 | ||
167 | if (cpus != NULL && cpu != -1 && cpu != me && | |
168 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 169 | cpumask_set_cpu(cpu, cpus); |
49846896 | 170 | } |
6ef7a1bc RR |
171 | if (unlikely(cpus == NULL)) |
172 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
173 | else if (!cpumask_empty(cpus)) | |
174 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
175 | else | |
176 | called = false; | |
3cba4130 | 177 | put_cpu(); |
6ef7a1bc | 178 | free_cpumask_var(cpus); |
49846896 | 179 | return called; |
d9e368d6 AK |
180 | } |
181 | ||
a6d51016 | 182 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 183 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 184 | { |
a086f6a1 XG |
185 | long dirty_count = kvm->tlbs_dirty; |
186 | ||
187 | smp_mb(); | |
445b8236 | 188 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
49846896 | 189 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 190 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 191 | } |
2ba9f0d8 | 192 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 193 | #endif |
2e53d63a | 194 | |
49846896 RR |
195 | void kvm_reload_remote_mmus(struct kvm *kvm) |
196 | { | |
445b8236 | 197 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 198 | } |
2e53d63a | 199 | |
d828199e MT |
200 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
201 | { | |
445b8236 | 202 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); |
d828199e MT |
203 | } |
204 | ||
3d81bc7e | 205 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
c7c9c56c | 206 | { |
445b8236 | 207 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); |
c7c9c56c YZ |
208 | } |
209 | ||
fb3f0f51 RR |
210 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
211 | { | |
212 | struct page *page; | |
213 | int r; | |
214 | ||
215 | mutex_init(&vcpu->mutex); | |
216 | vcpu->cpu = -1; | |
fb3f0f51 RR |
217 | vcpu->kvm = kvm; |
218 | vcpu->vcpu_id = id; | |
34bb10b7 | 219 | vcpu->pid = NULL; |
b6958ce4 | 220 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 221 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
222 | |
223 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
224 | if (!page) { | |
225 | r = -ENOMEM; | |
226 | goto fail; | |
227 | } | |
228 | vcpu->run = page_address(page); | |
229 | ||
4c088493 R |
230 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
231 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 232 | vcpu->preempted = false; |
4c088493 | 233 | |
e9b11c17 | 234 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 235 | if (r < 0) |
e9b11c17 | 236 | goto fail_free_run; |
fb3f0f51 RR |
237 | return 0; |
238 | ||
fb3f0f51 RR |
239 | fail_free_run: |
240 | free_page((unsigned long)vcpu->run); | |
241 | fail: | |
76fafa5e | 242 | return r; |
fb3f0f51 RR |
243 | } |
244 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
245 | ||
246 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
247 | { | |
34bb10b7 | 248 | put_pid(vcpu->pid); |
e9b11c17 | 249 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
250 | free_page((unsigned long)vcpu->run); |
251 | } | |
252 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
253 | ||
e930bffe AA |
254 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
255 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
256 | { | |
257 | return container_of(mn, struct kvm, mmu_notifier); | |
258 | } | |
259 | ||
260 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
261 | struct mm_struct *mm, | |
262 | unsigned long address) | |
263 | { | |
264 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 265 | int need_tlb_flush, idx; |
e930bffe AA |
266 | |
267 | /* | |
268 | * When ->invalidate_page runs, the linux pte has been zapped | |
269 | * already but the page is still allocated until | |
270 | * ->invalidate_page returns. So if we increase the sequence | |
271 | * here the kvm page fault will notice if the spte can't be | |
272 | * established because the page is going to be freed. If | |
273 | * instead the kvm page fault establishes the spte before | |
274 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
275 | * before returning. | |
276 | * | |
277 | * The sequence increase only need to be seen at spin_unlock | |
278 | * time, and not at spin_lock time. | |
279 | * | |
280 | * Increasing the sequence after the spin_unlock would be | |
281 | * unsafe because the kvm page fault could then establish the | |
282 | * pte after kvm_unmap_hva returned, without noticing the page | |
283 | * is going to be freed. | |
284 | */ | |
bc6678a3 | 285 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 286 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 287 | |
e930bffe | 288 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 289 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
290 | /* we've to flush the tlb before the pages can be freed */ |
291 | if (need_tlb_flush) | |
292 | kvm_flush_remote_tlbs(kvm); | |
293 | ||
565f3be2 | 294 | spin_unlock(&kvm->mmu_lock); |
fe71557a TC |
295 | |
296 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); | |
297 | ||
565f3be2 | 298 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
299 | } |
300 | ||
3da0dd43 IE |
301 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
302 | struct mm_struct *mm, | |
303 | unsigned long address, | |
304 | pte_t pte) | |
305 | { | |
306 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 307 | int idx; |
3da0dd43 | 308 | |
bc6678a3 | 309 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
310 | spin_lock(&kvm->mmu_lock); |
311 | kvm->mmu_notifier_seq++; | |
312 | kvm_set_spte_hva(kvm, address, pte); | |
313 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 314 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
315 | } |
316 | ||
e930bffe AA |
317 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
318 | struct mm_struct *mm, | |
319 | unsigned long start, | |
320 | unsigned long end) | |
321 | { | |
322 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 323 | int need_tlb_flush = 0, idx; |
e930bffe | 324 | |
bc6678a3 | 325 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
326 | spin_lock(&kvm->mmu_lock); |
327 | /* | |
328 | * The count increase must become visible at unlock time as no | |
329 | * spte can be established without taking the mmu_lock and | |
330 | * count is also read inside the mmu_lock critical section. | |
331 | */ | |
332 | kvm->mmu_notifier_count++; | |
b3ae2096 | 333 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 334 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
335 | /* we've to flush the tlb before the pages can be freed */ |
336 | if (need_tlb_flush) | |
337 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
338 | |
339 | spin_unlock(&kvm->mmu_lock); | |
340 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
341 | } |
342 | ||
343 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
344 | struct mm_struct *mm, | |
345 | unsigned long start, | |
346 | unsigned long end) | |
347 | { | |
348 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
349 | ||
350 | spin_lock(&kvm->mmu_lock); | |
351 | /* | |
352 | * This sequence increase will notify the kvm page fault that | |
353 | * the page that is going to be mapped in the spte could have | |
354 | * been freed. | |
355 | */ | |
356 | kvm->mmu_notifier_seq++; | |
a355aa54 | 357 | smp_wmb(); |
e930bffe AA |
358 | /* |
359 | * The above sequence increase must be visible before the | |
a355aa54 PM |
360 | * below count decrease, which is ensured by the smp_wmb above |
361 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
362 | */ |
363 | kvm->mmu_notifier_count--; | |
364 | spin_unlock(&kvm->mmu_lock); | |
365 | ||
366 | BUG_ON(kvm->mmu_notifier_count < 0); | |
367 | } | |
368 | ||
369 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
370 | struct mm_struct *mm, | |
57128468 ALC |
371 | unsigned long start, |
372 | unsigned long end) | |
e930bffe AA |
373 | { |
374 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 375 | int young, idx; |
e930bffe | 376 | |
bc6678a3 | 377 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 378 | spin_lock(&kvm->mmu_lock); |
e930bffe | 379 | |
57128468 | 380 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
381 | if (young) |
382 | kvm_flush_remote_tlbs(kvm); | |
383 | ||
565f3be2 TY |
384 | spin_unlock(&kvm->mmu_lock); |
385 | srcu_read_unlock(&kvm->srcu, idx); | |
386 | ||
e930bffe AA |
387 | return young; |
388 | } | |
389 | ||
8ee53820 AA |
390 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
391 | struct mm_struct *mm, | |
392 | unsigned long address) | |
393 | { | |
394 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
395 | int young, idx; | |
396 | ||
397 | idx = srcu_read_lock(&kvm->srcu); | |
398 | spin_lock(&kvm->mmu_lock); | |
399 | young = kvm_test_age_hva(kvm, address); | |
400 | spin_unlock(&kvm->mmu_lock); | |
401 | srcu_read_unlock(&kvm->srcu, idx); | |
402 | ||
403 | return young; | |
404 | } | |
405 | ||
85db06e5 MT |
406 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
407 | struct mm_struct *mm) | |
408 | { | |
409 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
410 | int idx; |
411 | ||
412 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 413 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 414 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
415 | } |
416 | ||
e930bffe AA |
417 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
418 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
419 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
420 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
421 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 422 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 423 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 424 | .release = kvm_mmu_notifier_release, |
e930bffe | 425 | }; |
4c07b0a4 AK |
426 | |
427 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
428 | { | |
429 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
430 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
431 | } | |
432 | ||
433 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
434 | ||
435 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
436 | { | |
437 | return 0; | |
438 | } | |
439 | ||
e930bffe AA |
440 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
441 | ||
bf3e05bc XG |
442 | static void kvm_init_memslots_id(struct kvm *kvm) |
443 | { | |
444 | int i; | |
445 | struct kvm_memslots *slots = kvm->memslots; | |
446 | ||
447 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
f85e2cb5 | 448 | slots->id_to_index[i] = slots->memslots[i].id = i; |
bf3e05bc XG |
449 | } |
450 | ||
e08b9637 | 451 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 452 | { |
d89f5eff JK |
453 | int r, i; |
454 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 455 | |
d89f5eff JK |
456 | if (!kvm) |
457 | return ERR_PTR(-ENOMEM); | |
458 | ||
e08b9637 | 459 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 460 | if (r) |
719d93cd | 461 | goto out_err_no_disable; |
10474ae8 AG |
462 | |
463 | r = hardware_enable_all(); | |
464 | if (r) | |
719d93cd | 465 | goto out_err_no_disable; |
10474ae8 | 466 | |
c77dcacb | 467 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 468 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 469 | #endif |
6aa8b732 | 470 | |
1e702d9a AW |
471 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
472 | ||
46a26bf5 | 473 | r = -ENOMEM; |
74496134 | 474 | kvm->memslots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
46a26bf5 | 475 | if (!kvm->memslots) |
719d93cd | 476 | goto out_err_no_srcu; |
00f034a1 PB |
477 | |
478 | /* | |
479 | * Init kvm generation close to the maximum to easily test the | |
480 | * code of handling generation number wrap-around. | |
481 | */ | |
482 | kvm->memslots->generation = -150; | |
483 | ||
bf3e05bc | 484 | kvm_init_memslots_id(kvm); |
bc6678a3 | 485 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
486 | goto out_err_no_srcu; |
487 | if (init_srcu_struct(&kvm->irq_srcu)) | |
488 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
489 | for (i = 0; i < KVM_NR_BUSES; i++) { |
490 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
491 | GFP_KERNEL); | |
57e7fbee | 492 | if (!kvm->buses[i]) |
e93f8a0f | 493 | goto out_err; |
e93f8a0f | 494 | } |
e930bffe | 495 | |
74b5c5bf | 496 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
497 | kvm->mm = current->mm; |
498 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 499 | kvm_eventfd_init(kvm); |
11ec2804 | 500 | mutex_init(&kvm->lock); |
60eead79 | 501 | mutex_init(&kvm->irq_lock); |
79fac95e | 502 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 503 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 504 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
505 | |
506 | r = kvm_init_mmu_notifier(kvm); | |
507 | if (r) | |
508 | goto out_err; | |
509 | ||
2f303b74 | 510 | spin_lock(&kvm_lock); |
5e58cfe4 | 511 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 512 | spin_unlock(&kvm_lock); |
d89f5eff | 513 | |
f17abe9a | 514 | return kvm; |
10474ae8 AG |
515 | |
516 | out_err: | |
719d93cd CB |
517 | cleanup_srcu_struct(&kvm->irq_srcu); |
518 | out_err_no_irq_srcu: | |
57e7fbee | 519 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 520 | out_err_no_srcu: |
10474ae8 | 521 | hardware_disable_all(); |
719d93cd | 522 | out_err_no_disable: |
e93f8a0f MT |
523 | for (i = 0; i < KVM_NR_BUSES; i++) |
524 | kfree(kvm->buses[i]); | |
74496134 | 525 | kvfree(kvm->memslots); |
d89f5eff | 526 | kvm_arch_free_vm(kvm); |
10474ae8 | 527 | return ERR_PTR(r); |
f17abe9a AK |
528 | } |
529 | ||
92eca8fa TY |
530 | /* |
531 | * Avoid using vmalloc for a small buffer. | |
532 | * Should not be used when the size is statically known. | |
533 | */ | |
c1a7b32a | 534 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
535 | { |
536 | if (size > PAGE_SIZE) | |
537 | return vzalloc(size); | |
538 | else | |
539 | return kzalloc(size, GFP_KERNEL); | |
540 | } | |
541 | ||
c1a7b32a | 542 | void kvm_kvfree(const void *addr) |
92eca8fa TY |
543 | { |
544 | if (is_vmalloc_addr(addr)) | |
545 | vfree(addr); | |
546 | else | |
547 | kfree(addr); | |
548 | } | |
549 | ||
a36a57b1 TY |
550 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
551 | { | |
552 | if (!memslot->dirty_bitmap) | |
553 | return; | |
554 | ||
92eca8fa | 555 | kvm_kvfree(memslot->dirty_bitmap); |
a36a57b1 TY |
556 | memslot->dirty_bitmap = NULL; |
557 | } | |
558 | ||
6aa8b732 AK |
559 | /* |
560 | * Free any memory in @free but not in @dont. | |
561 | */ | |
5587027c | 562 | static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free, |
6aa8b732 AK |
563 | struct kvm_memory_slot *dont) |
564 | { | |
6aa8b732 | 565 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
a36a57b1 | 566 | kvm_destroy_dirty_bitmap(free); |
6aa8b732 | 567 | |
5587027c | 568 | kvm_arch_free_memslot(kvm, free, dont); |
05da4558 | 569 | |
6aa8b732 | 570 | free->npages = 0; |
6aa8b732 AK |
571 | } |
572 | ||
7940876e | 573 | static void kvm_free_physmem(struct kvm *kvm) |
6aa8b732 | 574 | { |
46a26bf5 | 575 | struct kvm_memslots *slots = kvm->memslots; |
be6ba0f0 | 576 | struct kvm_memory_slot *memslot; |
46a26bf5 | 577 | |
be6ba0f0 | 578 | kvm_for_each_memslot(memslot, slots) |
5587027c | 579 | kvm_free_physmem_slot(kvm, memslot, NULL); |
6aa8b732 | 580 | |
74496134 | 581 | kvfree(kvm->memslots); |
6aa8b732 AK |
582 | } |
583 | ||
07f0a7bd SW |
584 | static void kvm_destroy_devices(struct kvm *kvm) |
585 | { | |
586 | struct list_head *node, *tmp; | |
587 | ||
588 | list_for_each_safe(node, tmp, &kvm->devices) { | |
589 | struct kvm_device *dev = | |
590 | list_entry(node, struct kvm_device, vm_node); | |
591 | ||
592 | list_del(node); | |
593 | dev->ops->destroy(dev); | |
594 | } | |
595 | } | |
596 | ||
f17abe9a AK |
597 | static void kvm_destroy_vm(struct kvm *kvm) |
598 | { | |
e93f8a0f | 599 | int i; |
6d4e4c4f AK |
600 | struct mm_struct *mm = kvm->mm; |
601 | ||
ad8ba2cd | 602 | kvm_arch_sync_events(kvm); |
2f303b74 | 603 | spin_lock(&kvm_lock); |
133de902 | 604 | list_del(&kvm->vm_list); |
2f303b74 | 605 | spin_unlock(&kvm_lock); |
399ec807 | 606 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
607 | for (i = 0; i < KVM_NR_BUSES; i++) |
608 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 609 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
610 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
611 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 612 | #else |
2df72e9b | 613 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 614 | #endif |
d19a9cd2 | 615 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 616 | kvm_destroy_devices(kvm); |
d89f5eff | 617 | kvm_free_physmem(kvm); |
820b3fcd | 618 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
619 | cleanup_srcu_struct(&kvm->srcu); |
620 | kvm_arch_free_vm(kvm); | |
10474ae8 | 621 | hardware_disable_all(); |
6d4e4c4f | 622 | mmdrop(mm); |
f17abe9a AK |
623 | } |
624 | ||
d39f13b0 IE |
625 | void kvm_get_kvm(struct kvm *kvm) |
626 | { | |
627 | atomic_inc(&kvm->users_count); | |
628 | } | |
629 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
630 | ||
631 | void kvm_put_kvm(struct kvm *kvm) | |
632 | { | |
633 | if (atomic_dec_and_test(&kvm->users_count)) | |
634 | kvm_destroy_vm(kvm); | |
635 | } | |
636 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
637 | ||
638 | ||
f17abe9a AK |
639 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
640 | { | |
641 | struct kvm *kvm = filp->private_data; | |
642 | ||
721eecbf GH |
643 | kvm_irqfd_release(kvm); |
644 | ||
d39f13b0 | 645 | kvm_put_kvm(kvm); |
6aa8b732 AK |
646 | return 0; |
647 | } | |
648 | ||
515a0127 TY |
649 | /* |
650 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 651 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 652 | */ |
a36a57b1 TY |
653 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
654 | { | |
515a0127 | 655 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 656 | |
92eca8fa | 657 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
658 | if (!memslot->dirty_bitmap) |
659 | return -ENOMEM; | |
660 | ||
a36a57b1 TY |
661 | return 0; |
662 | } | |
663 | ||
bf3e05bc | 664 | /* |
0e60b079 IM |
665 | * Insert memslot and re-sort memslots based on their GFN, |
666 | * so binary search could be used to lookup GFN. | |
667 | * Sorting algorithm takes advantage of having initially | |
668 | * sorted array and known changed memslot position. | |
bf3e05bc | 669 | */ |
5cc15027 PB |
670 | static void update_memslots(struct kvm_memslots *slots, |
671 | struct kvm_memory_slot *new) | |
bf3e05bc | 672 | { |
8593176c PB |
673 | int id = new->id; |
674 | int i = slots->id_to_index[id]; | |
063584d4 | 675 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 676 | |
8593176c | 677 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 678 | if (!new->npages) { |
dbaff309 | 679 | WARN_ON(!mslots[i].npages); |
0e60b079 | 680 | new->base_gfn = 0; |
b0165f1b | 681 | new->flags = 0; |
9c1a5d38 IM |
682 | if (mslots[i].npages) |
683 | slots->used_slots--; | |
684 | } else { | |
685 | if (!mslots[i].npages) | |
686 | slots->used_slots++; | |
687 | } | |
0e60b079 | 688 | |
7f379cff | 689 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
690 | new->base_gfn <= mslots[i + 1].base_gfn) { |
691 | if (!mslots[i + 1].npages) | |
692 | break; | |
7f379cff IM |
693 | mslots[i] = mslots[i + 1]; |
694 | slots->id_to_index[mslots[i].id] = i; | |
695 | i++; | |
696 | } | |
efbeec70 PB |
697 | |
698 | /* | |
699 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
700 | * so that it moves before all those with base_gfn == npages == 0. | |
701 | * | |
702 | * On the other hand, if new->npages is zero, the above loop has | |
703 | * already left i pointing to the beginning of the empty part of | |
704 | * mslots, and the ">=" would move the hole backwards in this | |
705 | * case---which is wrong. So skip the loop when deleting a slot. | |
706 | */ | |
707 | if (new->npages) { | |
708 | while (i > 0 && | |
709 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
710 | mslots[i] = mslots[i - 1]; | |
711 | slots->id_to_index[mslots[i].id] = i; | |
712 | i--; | |
713 | } | |
dbaff309 PB |
714 | } else |
715 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 716 | |
8593176c PB |
717 | mslots[i] = *new; |
718 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
719 | } |
720 | ||
a50d64d6 XG |
721 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
722 | { | |
4d8b81ab XG |
723 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
724 | ||
0f8a4de3 | 725 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
726 | valid_flags |= KVM_MEM_READONLY; |
727 | #endif | |
728 | ||
729 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
730 | return -EINVAL; |
731 | ||
732 | return 0; | |
733 | } | |
734 | ||
7ec4fb44 | 735 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
5cc15027 | 736 | struct kvm_memslots *slots) |
7ec4fb44 GN |
737 | { |
738 | struct kvm_memslots *old_memslots = kvm->memslots; | |
739 | ||
ee3d1570 DM |
740 | /* |
741 | * Set the low bit in the generation, which disables SPTE caching | |
742 | * until the end of synchronize_srcu_expedited. | |
743 | */ | |
744 | WARN_ON(old_memslots->generation & 1); | |
745 | slots->generation = old_memslots->generation + 1; | |
746 | ||
7ec4fb44 GN |
747 | rcu_assign_pointer(kvm->memslots, slots); |
748 | synchronize_srcu_expedited(&kvm->srcu); | |
e59dbe09 | 749 | |
ee3d1570 DM |
750 | /* |
751 | * Increment the new memslot generation a second time. This prevents | |
752 | * vm exits that race with memslot updates from caching a memslot | |
753 | * generation that will (potentially) be valid forever. | |
754 | */ | |
755 | slots->generation++; | |
756 | ||
e59dbe09 TY |
757 | kvm_arch_memslots_updated(kvm); |
758 | ||
759 | return old_memslots; | |
7ec4fb44 GN |
760 | } |
761 | ||
6aa8b732 AK |
762 | /* |
763 | * Allocate some memory and give it an address in the guest physical address | |
764 | * space. | |
765 | * | |
766 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 767 | * |
02d5d55b | 768 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 769 | */ |
f78e0e2e | 770 | int __kvm_set_memory_region(struct kvm *kvm, |
47ae31e2 | 771 | struct kvm_userspace_memory_region *mem) |
6aa8b732 | 772 | { |
8234b22e | 773 | int r; |
6aa8b732 | 774 | gfn_t base_gfn; |
28bcb112 | 775 | unsigned long npages; |
a843fac2 | 776 | struct kvm_memory_slot *slot; |
6aa8b732 | 777 | struct kvm_memory_slot old, new; |
b7f69c55 | 778 | struct kvm_memslots *slots = NULL, *old_memslots; |
f64c0398 | 779 | enum kvm_mr_change change; |
6aa8b732 | 780 | |
a50d64d6 XG |
781 | r = check_memory_region_flags(mem); |
782 | if (r) | |
783 | goto out; | |
784 | ||
6aa8b732 AK |
785 | r = -EINVAL; |
786 | /* General sanity checks */ | |
787 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
788 | goto out; | |
789 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
790 | goto out; | |
fa3d315a | 791 | /* We can read the guest memory with __xxx_user() later on. */ |
47ae31e2 | 792 | if ((mem->slot < KVM_USER_MEM_SLOTS) && |
fa3d315a | 793 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
794 | !access_ok(VERIFY_WRITE, |
795 | (void __user *)(unsigned long)mem->userspace_addr, | |
796 | mem->memory_size))) | |
78749809 | 797 | goto out; |
93a5cef0 | 798 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
799 | goto out; |
800 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
801 | goto out; | |
802 | ||
a843fac2 | 803 | slot = id_to_memslot(kvm->memslots, mem->slot); |
6aa8b732 AK |
804 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
805 | npages = mem->memory_size >> PAGE_SHIFT; | |
806 | ||
660c22c4 TY |
807 | if (npages > KVM_MEM_MAX_NR_PAGES) |
808 | goto out; | |
809 | ||
6aa8b732 AK |
810 | if (!npages) |
811 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
812 | ||
a843fac2 | 813 | new = old = *slot; |
6aa8b732 | 814 | |
e36d96f7 | 815 | new.id = mem->slot; |
6aa8b732 AK |
816 | new.base_gfn = base_gfn; |
817 | new.npages = npages; | |
818 | new.flags = mem->flags; | |
819 | ||
f64c0398 TY |
820 | if (npages) { |
821 | if (!old.npages) | |
822 | change = KVM_MR_CREATE; | |
823 | else { /* Modify an existing slot. */ | |
824 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
825 | (npages != old.npages) || |
826 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
827 | goto out; |
828 | ||
829 | if (base_gfn != old.base_gfn) | |
830 | change = KVM_MR_MOVE; | |
831 | else if (new.flags != old.flags) | |
832 | change = KVM_MR_FLAGS_ONLY; | |
833 | else { /* Nothing to change. */ | |
834 | r = 0; | |
835 | goto out; | |
836 | } | |
837 | } | |
838 | } else if (old.npages) { | |
839 | change = KVM_MR_DELETE; | |
840 | } else /* Modify a non-existent slot: disallowed. */ | |
0ea75e1d | 841 | goto out; |
6aa8b732 | 842 | |
f64c0398 | 843 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
844 | /* Check for overlaps */ |
845 | r = -EEXIST; | |
846 | kvm_for_each_memslot(slot, kvm->memslots) { | |
a843fac2 TY |
847 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
848 | (slot->id == mem->slot)) | |
0a706bee TY |
849 | continue; |
850 | if (!((base_gfn + npages <= slot->base_gfn) || | |
851 | (base_gfn >= slot->base_gfn + slot->npages))) | |
852 | goto out; | |
853 | } | |
6aa8b732 | 854 | } |
6aa8b732 | 855 | |
6aa8b732 AK |
856 | /* Free page dirty bitmap if unneeded */ |
857 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 858 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
859 | |
860 | r = -ENOMEM; | |
f64c0398 | 861 | if (change == KVM_MR_CREATE) { |
189a2f7b | 862 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 863 | |
5587027c | 864 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 865 | goto out_free; |
6aa8b732 | 866 | } |
ec04b260 | 867 | |
6aa8b732 AK |
868 | /* Allocate page dirty bitmap if needed */ |
869 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 870 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 871 | goto out_free; |
6aa8b732 AK |
872 | } |
873 | ||
74496134 | 874 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
f2a81036 PB |
875 | if (!slots) |
876 | goto out_free; | |
74496134 | 877 | memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots)); |
f2a81036 | 878 | |
f64c0398 | 879 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
28a37544 XG |
880 | slot = id_to_memslot(slots, mem->slot); |
881 | slot->flags |= KVM_MEMSLOT_INVALID; | |
882 | ||
5cc15027 | 883 | old_memslots = install_new_memslots(kvm, slots); |
bc6678a3 | 884 | |
e40f193f AW |
885 | /* slot was deleted or moved, clear iommu mapping */ |
886 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
887 | /* From this point no new shadow pages pointing to a deleted, |
888 | * or moved, memslot will be created. | |
bc6678a3 MT |
889 | * |
890 | * validation of sp->gfn happens in: | |
891 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
892 | * - kvm_is_visible_gfn (mmu_check_roots) | |
893 | */ | |
2df72e9b | 894 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
895 | |
896 | /* | |
897 | * We can re-use the old_memslots from above, the only difference | |
898 | * from the currently installed memslots is the invalid flag. This | |
899 | * will get overwritten by update_memslots anyway. | |
900 | */ | |
b7f69c55 | 901 | slots = old_memslots; |
bc6678a3 | 902 | } |
34d4cb8f | 903 | |
7b6195a9 | 904 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 905 | if (r) |
b7f69c55 | 906 | goto out_slots; |
f7784b8e | 907 | |
bc6678a3 | 908 | /* actual memory is freed via old in kvm_free_physmem_slot below */ |
f64c0398 | 909 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 910 | new.dirty_bitmap = NULL; |
db3fe4eb | 911 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
912 | } |
913 | ||
5cc15027 PB |
914 | update_memslots(slots, &new); |
915 | old_memslots = install_new_memslots(kvm, slots); | |
3ad82a7e | 916 | |
8482644a | 917 | kvm_arch_commit_memory_region(kvm, mem, &old, change); |
82ce2c96 | 918 | |
5587027c | 919 | kvm_free_physmem_slot(kvm, &old, &new); |
74496134 | 920 | kvfree(old_memslots); |
bc6678a3 | 921 | |
261874b0 AW |
922 | /* |
923 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
924 | * un-mapped and re-mapped if their base changes. Since base change |
925 | * unmapping is handled above with slot deletion, mapping alone is | |
926 | * needed here. Anything else the iommu might care about for existing | |
927 | * slots (size changes, userspace addr changes and read-only flag | |
928 | * changes) is disallowed above, so any other attribute changes getting | |
929 | * here can be skipped. | |
261874b0 | 930 | */ |
75d61fbc TY |
931 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
932 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 933 | return r; |
bc6678a3 MT |
934 | } |
935 | ||
6aa8b732 AK |
936 | return 0; |
937 | ||
e40f193f | 938 | out_slots: |
74496134 | 939 | kvfree(slots); |
f78e0e2e | 940 | out_free: |
5587027c | 941 | kvm_free_physmem_slot(kvm, &new, &old); |
6aa8b732 AK |
942 | out: |
943 | return r; | |
210c7c4d | 944 | } |
f78e0e2e SY |
945 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
946 | ||
947 | int kvm_set_memory_region(struct kvm *kvm, | |
47ae31e2 | 948 | struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
949 | { |
950 | int r; | |
951 | ||
79fac95e | 952 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 953 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 954 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
955 | return r; |
956 | } | |
210c7c4d IE |
957 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
958 | ||
7940876e SH |
959 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
960 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 961 | { |
bbacc0c1 | 962 | if (mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 963 | return -EINVAL; |
47ae31e2 | 964 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
965 | } |
966 | ||
5bb064dc ZX |
967 | int kvm_get_dirty_log(struct kvm *kvm, |
968 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 AK |
969 | { |
970 | struct kvm_memory_slot *memslot; | |
971 | int r, i; | |
87bf6e7d | 972 | unsigned long n; |
6aa8b732 AK |
973 | unsigned long any = 0; |
974 | ||
6aa8b732 | 975 | r = -EINVAL; |
bbacc0c1 | 976 | if (log->slot >= KVM_USER_MEM_SLOTS) |
6aa8b732 AK |
977 | goto out; |
978 | ||
28a37544 | 979 | memslot = id_to_memslot(kvm->memslots, log->slot); |
6aa8b732 AK |
980 | r = -ENOENT; |
981 | if (!memslot->dirty_bitmap) | |
982 | goto out; | |
983 | ||
87bf6e7d | 984 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 985 | |
cd1a4a98 | 986 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
987 | any = memslot->dirty_bitmap[i]; |
988 | ||
989 | r = -EFAULT; | |
990 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
991 | goto out; | |
992 | ||
5bb064dc ZX |
993 | if (any) |
994 | *is_dirty = 1; | |
6aa8b732 AK |
995 | |
996 | r = 0; | |
6aa8b732 | 997 | out: |
6aa8b732 AK |
998 | return r; |
999 | } | |
2ba9f0d8 | 1000 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1001 | |
ba0513b5 MS |
1002 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1003 | /** | |
1004 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1005 | * are dirty write protect them for next write. | |
1006 | * @kvm: pointer to kvm instance | |
1007 | * @log: slot id and address to which we copy the log | |
1008 | * @is_dirty: flag set if any page is dirty | |
1009 | * | |
1010 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1011 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1012 | * following order: | |
1013 | * | |
1014 | * 1. Take a snapshot of the bit and clear it if needed. | |
1015 | * 2. Write protect the corresponding page. | |
1016 | * 3. Copy the snapshot to the userspace. | |
1017 | * 4. Upon return caller flushes TLB's if needed. | |
1018 | * | |
1019 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1020 | * entry. This is not a problem because the page is reported dirty using | |
1021 | * the snapshot taken before and step 4 ensures that writes done after | |
1022 | * exiting to userspace will be logged for the next call. | |
1023 | * | |
1024 | */ | |
1025 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1026 | struct kvm_dirty_log *log, bool *is_dirty) | |
1027 | { | |
1028 | struct kvm_memory_slot *memslot; | |
1029 | int r, i; | |
1030 | unsigned long n; | |
1031 | unsigned long *dirty_bitmap; | |
1032 | unsigned long *dirty_bitmap_buffer; | |
1033 | ||
1034 | r = -EINVAL; | |
1035 | if (log->slot >= KVM_USER_MEM_SLOTS) | |
1036 | goto out; | |
1037 | ||
1038 | memslot = id_to_memslot(kvm->memslots, log->slot); | |
1039 | ||
1040 | dirty_bitmap = memslot->dirty_bitmap; | |
1041 | r = -ENOENT; | |
1042 | if (!dirty_bitmap) | |
1043 | goto out; | |
1044 | ||
1045 | n = kvm_dirty_bitmap_bytes(memslot); | |
1046 | ||
1047 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1048 | memset(dirty_bitmap_buffer, 0, n); | |
1049 | ||
1050 | spin_lock(&kvm->mmu_lock); | |
1051 | *is_dirty = false; | |
1052 | for (i = 0; i < n / sizeof(long); i++) { | |
1053 | unsigned long mask; | |
1054 | gfn_t offset; | |
1055 | ||
1056 | if (!dirty_bitmap[i]) | |
1057 | continue; | |
1058 | ||
1059 | *is_dirty = true; | |
1060 | ||
1061 | mask = xchg(&dirty_bitmap[i], 0); | |
1062 | dirty_bitmap_buffer[i] = mask; | |
1063 | ||
1064 | offset = i * BITS_PER_LONG; | |
3b0f1d01 | 1065 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, offset, |
ba0513b5 MS |
1066 | mask); |
1067 | } | |
1068 | ||
1069 | spin_unlock(&kvm->mmu_lock); | |
1070 | ||
1071 | r = -EFAULT; | |
1072 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1073 | goto out; | |
1074 | ||
1075 | r = 0; | |
1076 | out: | |
1077 | return r; | |
1078 | } | |
1079 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1080 | #endif | |
1081 | ||
db3fe4eb TY |
1082 | bool kvm_largepages_enabled(void) |
1083 | { | |
1084 | return largepages_enabled; | |
1085 | } | |
1086 | ||
54dee993 MT |
1087 | void kvm_disable_largepages(void) |
1088 | { | |
1089 | largepages_enabled = false; | |
1090 | } | |
1091 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1092 | ||
49c7754c GN |
1093 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1094 | { | |
1095 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1096 | } | |
a1f4d395 | 1097 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1098 | |
e0d62c7f IE |
1099 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
1100 | { | |
bf3e05bc | 1101 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1102 | |
bbacc0c1 | 1103 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
1104 | memslot->flags & KVM_MEMSLOT_INVALID) |
1105 | return 0; | |
e0d62c7f | 1106 | |
bf3e05bc | 1107 | return 1; |
e0d62c7f IE |
1108 | } |
1109 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1110 | ||
8f0b1ab6 JR |
1111 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1112 | { | |
1113 | struct vm_area_struct *vma; | |
1114 | unsigned long addr, size; | |
1115 | ||
1116 | size = PAGE_SIZE; | |
1117 | ||
1118 | addr = gfn_to_hva(kvm, gfn); | |
1119 | if (kvm_is_error_hva(addr)) | |
1120 | return PAGE_SIZE; | |
1121 | ||
1122 | down_read(¤t->mm->mmap_sem); | |
1123 | vma = find_vma(current->mm, addr); | |
1124 | if (!vma) | |
1125 | goto out; | |
1126 | ||
1127 | size = vma_kernel_pagesize(vma); | |
1128 | ||
1129 | out: | |
1130 | up_read(¤t->mm->mmap_sem); | |
1131 | ||
1132 | return size; | |
1133 | } | |
1134 | ||
4d8b81ab XG |
1135 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1136 | { | |
1137 | return slot->flags & KVM_MEM_READONLY; | |
1138 | } | |
1139 | ||
4d8b81ab XG |
1140 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1141 | gfn_t *nr_pages, bool write) | |
539cb660 | 1142 | { |
bc6678a3 | 1143 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1144 | return KVM_HVA_ERR_BAD; |
48987781 | 1145 | |
4d8b81ab XG |
1146 | if (memslot_is_readonly(slot) && write) |
1147 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1148 | |
1149 | if (nr_pages) | |
1150 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1151 | ||
4d8b81ab | 1152 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1153 | } |
48987781 | 1154 | |
4d8b81ab XG |
1155 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1156 | gfn_t *nr_pages) | |
1157 | { | |
1158 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1159 | } |
48987781 | 1160 | |
4d8b81ab | 1161 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1162 | gfn_t gfn) |
4d8b81ab XG |
1163 | { |
1164 | return gfn_to_hva_many(slot, gfn, NULL); | |
1165 | } | |
1166 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1167 | ||
48987781 XG |
1168 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1169 | { | |
49c7754c | 1170 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1171 | } |
0d150298 | 1172 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1173 | |
86ab8cff | 1174 | /* |
ba6a3541 PB |
1175 | * If writable is set to false, the hva returned by this function is only |
1176 | * allowed to be read. | |
86ab8cff | 1177 | */ |
64d83126 CD |
1178 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1179 | gfn_t gfn, bool *writable) | |
86ab8cff | 1180 | { |
a2ac07fe GN |
1181 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1182 | ||
1183 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1184 | *writable = !memslot_is_readonly(slot); |
1185 | ||
a2ac07fe | 1186 | return hva; |
86ab8cff XG |
1187 | } |
1188 | ||
64d83126 CD |
1189 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1190 | { | |
1191 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1192 | ||
1193 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1194 | } | |
1195 | ||
86ab8cff | 1196 | static int kvm_read_hva(void *data, void __user *hva, int len) |
8030089f | 1197 | { |
86ab8cff XG |
1198 | return __copy_from_user(data, hva, len); |
1199 | } | |
1200 | ||
1201 | static int kvm_read_hva_atomic(void *data, void __user *hva, int len) | |
1202 | { | |
1203 | return __copy_from_user_inatomic(data, hva, len); | |
8030089f GN |
1204 | } |
1205 | ||
39369f7a | 1206 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1207 | unsigned long start, int write, struct page **page) |
1208 | { | |
1209 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1210 | ||
1211 | if (write) | |
1212 | flags |= FOLL_WRITE; | |
1213 | ||
1214 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1215 | } | |
1216 | ||
fafc3dba HY |
1217 | static inline int check_user_page_hwpoison(unsigned long addr) |
1218 | { | |
1219 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1220 | ||
1221 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1222 | flags, NULL, NULL, NULL); | |
1223 | return rc == -EHWPOISON; | |
1224 | } | |
1225 | ||
2fc84311 XG |
1226 | /* |
1227 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1228 | * true indicates success, otherwise false is returned. | |
1229 | */ | |
1230 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1231 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1232 | { |
8d4e1288 | 1233 | struct page *page[1]; |
2fc84311 | 1234 | int npages; |
954bbbc2 | 1235 | |
2fc84311 XG |
1236 | if (!(async || atomic)) |
1237 | return false; | |
af585b92 | 1238 | |
12ce13fe XG |
1239 | /* |
1240 | * Fast pin a writable pfn only if it is a write fault request | |
1241 | * or the caller allows to map a writable pfn for a read fault | |
1242 | * request. | |
1243 | */ | |
1244 | if (!(write_fault || writable)) | |
1245 | return false; | |
612819c3 | 1246 | |
2fc84311 XG |
1247 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1248 | if (npages == 1) { | |
1249 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1250 | |
2fc84311 XG |
1251 | if (writable) |
1252 | *writable = true; | |
1253 | return true; | |
1254 | } | |
af585b92 | 1255 | |
2fc84311 XG |
1256 | return false; |
1257 | } | |
612819c3 | 1258 | |
2fc84311 XG |
1259 | /* |
1260 | * The slow path to get the pfn of the specified host virtual address, | |
1261 | * 1 indicates success, -errno is returned if error is detected. | |
1262 | */ | |
1263 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1264 | bool *writable, pfn_t *pfn) | |
1265 | { | |
1266 | struct page *page[1]; | |
1267 | int npages = 0; | |
612819c3 | 1268 | |
2fc84311 XG |
1269 | might_sleep(); |
1270 | ||
1271 | if (writable) | |
1272 | *writable = write_fault; | |
1273 | ||
1274 | if (async) { | |
1275 | down_read(¤t->mm->mmap_sem); | |
1276 | npages = get_user_page_nowait(current, current->mm, | |
1277 | addr, write_fault, page); | |
1278 | up_read(¤t->mm->mmap_sem); | |
0664e57f AA |
1279 | } else |
1280 | npages = __get_user_pages_unlocked(current, current->mm, addr, 1, | |
1281 | write_fault, 0, page, | |
1282 | FOLL_TOUCH|FOLL_HWPOISON); | |
2fc84311 XG |
1283 | if (npages != 1) |
1284 | return npages; | |
1285 | ||
1286 | /* map read fault as writable if possible */ | |
12ce13fe | 1287 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1288 | struct page *wpage[1]; |
1289 | ||
1290 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1291 | if (npages == 1) { | |
1292 | *writable = true; | |
1293 | put_page(page[0]); | |
1294 | page[0] = wpage[0]; | |
612819c3 | 1295 | } |
2fc84311 XG |
1296 | |
1297 | npages = 1; | |
887c08ac | 1298 | } |
2fc84311 XG |
1299 | *pfn = page_to_pfn(page[0]); |
1300 | return npages; | |
1301 | } | |
539cb660 | 1302 | |
4d8b81ab XG |
1303 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1304 | { | |
1305 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1306 | return false; | |
2e2e3738 | 1307 | |
4d8b81ab XG |
1308 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1309 | return false; | |
887c08ac | 1310 | |
4d8b81ab XG |
1311 | return true; |
1312 | } | |
bf998156 | 1313 | |
12ce13fe XG |
1314 | /* |
1315 | * Pin guest page in memory and return its pfn. | |
1316 | * @addr: host virtual address which maps memory to the guest | |
1317 | * @atomic: whether this function can sleep | |
1318 | * @async: whether this function need to wait IO complete if the | |
1319 | * host page is not in the memory | |
1320 | * @write_fault: whether we should get a writable host page | |
1321 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1322 | * | |
1323 | * The function will map a writable host page for these two cases: | |
1324 | * 1): @write_fault = true | |
1325 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1326 | * whether the mapping is writable. | |
1327 | */ | |
2fc84311 XG |
1328 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1329 | bool write_fault, bool *writable) | |
1330 | { | |
1331 | struct vm_area_struct *vma; | |
1332 | pfn_t pfn = 0; | |
1333 | int npages; | |
2e2e3738 | 1334 | |
2fc84311 XG |
1335 | /* we can do it either atomically or asynchronously, not both */ |
1336 | BUG_ON(atomic && async); | |
8d4e1288 | 1337 | |
2fc84311 XG |
1338 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1339 | return pfn; | |
1340 | ||
1341 | if (atomic) | |
1342 | return KVM_PFN_ERR_FAULT; | |
1343 | ||
1344 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1345 | if (npages == 1) | |
1346 | return pfn; | |
8d4e1288 | 1347 | |
2fc84311 XG |
1348 | down_read(¤t->mm->mmap_sem); |
1349 | if (npages == -EHWPOISON || | |
1350 | (!async && check_user_page_hwpoison(addr))) { | |
1351 | pfn = KVM_PFN_ERR_HWPOISON; | |
1352 | goto exit; | |
1353 | } | |
1354 | ||
1355 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1356 | ||
1357 | if (vma == NULL) | |
1358 | pfn = KVM_PFN_ERR_FAULT; | |
1359 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1360 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1361 | vma->vm_pgoff; | |
bf4bea8e | 1362 | BUG_ON(!kvm_is_reserved_pfn(pfn)); |
2fc84311 | 1363 | } else { |
4d8b81ab | 1364 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1365 | *async = true; |
1366 | pfn = KVM_PFN_ERR_FAULT; | |
1367 | } | |
1368 | exit: | |
1369 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1370 | return pfn; |
35149e21 AL |
1371 | } |
1372 | ||
4d8b81ab XG |
1373 | static pfn_t |
1374 | __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, | |
1375 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1376 | { |
4d8b81ab XG |
1377 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1378 | ||
1379 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1380 | return KVM_PFN_ERR_RO_FAULT; | |
1381 | ||
1382 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1383 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1384 | |
1385 | /* Do not map writable pfn in the readonly memslot. */ | |
1386 | if (writable && memslot_is_readonly(slot)) { | |
1387 | *writable = false; | |
1388 | writable = NULL; | |
1389 | } | |
1390 | ||
1391 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1392 | writable); | |
887c08ac | 1393 | } |
887c08ac | 1394 | |
612819c3 MT |
1395 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async, |
1396 | bool write_fault, bool *writable) | |
506f0d6f | 1397 | { |
4d8b81ab | 1398 | struct kvm_memory_slot *slot; |
506f0d6f | 1399 | |
af585b92 GN |
1400 | if (async) |
1401 | *async = false; | |
1402 | ||
4d8b81ab | 1403 | slot = gfn_to_memslot(kvm, gfn); |
506f0d6f | 1404 | |
4d8b81ab XG |
1405 | return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault, |
1406 | writable); | |
365fb3fd XG |
1407 | } |
1408 | ||
1409 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) | |
1410 | { | |
612819c3 | 1411 | return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL); |
365fb3fd XG |
1412 | } |
1413 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1414 | ||
612819c3 MT |
1415 | pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async, |
1416 | bool write_fault, bool *writable) | |
af585b92 | 1417 | { |
612819c3 | 1418 | return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable); |
af585b92 GN |
1419 | } |
1420 | EXPORT_SYMBOL_GPL(gfn_to_pfn_async); | |
1421 | ||
365fb3fd XG |
1422 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1423 | { | |
612819c3 | 1424 | return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL); |
506f0d6f | 1425 | } |
35149e21 AL |
1426 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
1427 | ||
612819c3 MT |
1428 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1429 | bool *writable) | |
1430 | { | |
1431 | return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable); | |
1432 | } | |
1433 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1434 | ||
d5661048 | 1435 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1436 | { |
4d8b81ab | 1437 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f MT |
1438 | } |
1439 | ||
037d92dc | 1440 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1441 | { |
4d8b81ab | 1442 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1443 | } |
037d92dc | 1444 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1445 | |
48987781 XG |
1446 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
1447 | int nr_pages) | |
1448 | { | |
1449 | unsigned long addr; | |
1450 | gfn_t entry; | |
1451 | ||
49c7754c | 1452 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
48987781 XG |
1453 | if (kvm_is_error_hva(addr)) |
1454 | return -1; | |
1455 | ||
1456 | if (entry < nr_pages) | |
1457 | return 0; | |
1458 | ||
1459 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1460 | } | |
1461 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1462 | ||
a2766325 XG |
1463 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1464 | { | |
81c52c56 | 1465 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1466 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1467 | |
bf4bea8e | 1468 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1469 | WARN_ON(1); |
6cede2e6 | 1470 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1471 | } |
a2766325 XG |
1472 | |
1473 | return pfn_to_page(pfn); | |
1474 | } | |
1475 | ||
35149e21 AL |
1476 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1477 | { | |
2e2e3738 AL |
1478 | pfn_t pfn; |
1479 | ||
1480 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1481 | |
a2766325 | 1482 | return kvm_pfn_to_page(pfn); |
954bbbc2 | 1483 | } |
aab61cc0 | 1484 | |
954bbbc2 AK |
1485 | EXPORT_SYMBOL_GPL(gfn_to_page); |
1486 | ||
b4231d61 IE |
1487 | void kvm_release_page_clean(struct page *page) |
1488 | { | |
32cad84f XG |
1489 | WARN_ON(is_error_page(page)); |
1490 | ||
35149e21 | 1491 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1492 | } |
1493 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1494 | ||
35149e21 AL |
1495 | void kvm_release_pfn_clean(pfn_t pfn) |
1496 | { | |
bf4bea8e | 1497 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1498 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1499 | } |
1500 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1501 | ||
b4231d61 | 1502 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1503 | { |
a2766325 XG |
1504 | WARN_ON(is_error_page(page)); |
1505 | ||
35149e21 AL |
1506 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1507 | } | |
1508 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1509 | ||
7940876e | 1510 | static void kvm_release_pfn_dirty(pfn_t pfn) |
35149e21 AL |
1511 | { |
1512 | kvm_set_pfn_dirty(pfn); | |
1513 | kvm_release_pfn_clean(pfn); | |
1514 | } | |
35149e21 AL |
1515 | |
1516 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1517 | { | |
bf4bea8e | 1518 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 AL |
1519 | struct page *page = pfn_to_page(pfn); |
1520 | if (!PageReserved(page)) | |
1521 | SetPageDirty(page); | |
1522 | } | |
8a7ae055 | 1523 | } |
35149e21 AL |
1524 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1525 | ||
1526 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1527 | { | |
bf4bea8e | 1528 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1529 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1530 | } |
1531 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1532 | ||
1533 | void kvm_get_pfn(pfn_t pfn) | |
1534 | { | |
bf4bea8e | 1535 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1536 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1537 | } |
1538 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1539 | |
195aefde IE |
1540 | static int next_segment(unsigned long len, int offset) |
1541 | { | |
1542 | if (len > PAGE_SIZE - offset) | |
1543 | return PAGE_SIZE - offset; | |
1544 | else | |
1545 | return len; | |
1546 | } | |
1547 | ||
1548 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1549 | int len) | |
1550 | { | |
e0506bcb IE |
1551 | int r; |
1552 | unsigned long addr; | |
195aefde | 1553 | |
ba6a3541 | 1554 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
e0506bcb IE |
1555 | if (kvm_is_error_hva(addr)) |
1556 | return -EFAULT; | |
86ab8cff | 1557 | r = kvm_read_hva(data, (void __user *)addr + offset, len); |
e0506bcb | 1558 | if (r) |
195aefde | 1559 | return -EFAULT; |
195aefde IE |
1560 | return 0; |
1561 | } | |
1562 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); | |
1563 | ||
1564 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) | |
1565 | { | |
1566 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1567 | int seg; | |
1568 | int offset = offset_in_page(gpa); | |
1569 | int ret; | |
1570 | ||
1571 | while ((seg = next_segment(len, offset)) != 0) { | |
1572 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1573 | if (ret < 0) | |
1574 | return ret; | |
1575 | offset = 0; | |
1576 | len -= seg; | |
1577 | data += seg; | |
1578 | ++gfn; | |
1579 | } | |
1580 | return 0; | |
1581 | } | |
1582 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1583 | ||
7ec54588 MT |
1584 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1585 | unsigned long len) | |
1586 | { | |
1587 | int r; | |
1588 | unsigned long addr; | |
1589 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1590 | int offset = offset_in_page(gpa); | |
1591 | ||
ba6a3541 | 1592 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
7ec54588 MT |
1593 | if (kvm_is_error_hva(addr)) |
1594 | return -EFAULT; | |
0aac03f0 | 1595 | pagefault_disable(); |
86ab8cff | 1596 | r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1597 | pagefault_enable(); |
7ec54588 MT |
1598 | if (r) |
1599 | return -EFAULT; | |
1600 | return 0; | |
1601 | } | |
1602 | EXPORT_SYMBOL(kvm_read_guest_atomic); | |
1603 | ||
195aefde IE |
1604 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
1605 | int offset, int len) | |
1606 | { | |
e0506bcb IE |
1607 | int r; |
1608 | unsigned long addr; | |
195aefde | 1609 | |
e0506bcb IE |
1610 | addr = gfn_to_hva(kvm, gfn); |
1611 | if (kvm_is_error_hva(addr)) | |
1612 | return -EFAULT; | |
8b0cedff | 1613 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1614 | if (r) |
195aefde | 1615 | return -EFAULT; |
195aefde IE |
1616 | mark_page_dirty(kvm, gfn); |
1617 | return 0; | |
1618 | } | |
1619 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); | |
1620 | ||
1621 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, | |
1622 | unsigned long len) | |
1623 | { | |
1624 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1625 | int seg; | |
1626 | int offset = offset_in_page(gpa); | |
1627 | int ret; | |
1628 | ||
1629 | while ((seg = next_segment(len, offset)) != 0) { | |
1630 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1631 | if (ret < 0) | |
1632 | return ret; | |
1633 | offset = 0; | |
1634 | len -= seg; | |
1635 | data += seg; | |
1636 | ++gfn; | |
1637 | } | |
1638 | return 0; | |
1639 | } | |
ff651cb6 | 1640 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1641 | |
49c7754c | 1642 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1643 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1644 | { |
1645 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1646 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1647 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1648 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1649 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1650 | gfn_t nr_pages_avail; | |
49c7754c GN |
1651 | |
1652 | ghc->gpa = gpa; | |
1653 | ghc->generation = slots->generation; | |
8f964525 AH |
1654 | ghc->len = len; |
1655 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1656 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail); | |
1657 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) { | |
49c7754c | 1658 | ghc->hva += offset; |
8f964525 AH |
1659 | } else { |
1660 | /* | |
1661 | * If the requested region crosses two memslots, we still | |
1662 | * verify that the entire region is valid here. | |
1663 | */ | |
1664 | while (start_gfn <= end_gfn) { | |
1665 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1666 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1667 | &nr_pages_avail); | |
1668 | if (kvm_is_error_hva(ghc->hva)) | |
1669 | return -EFAULT; | |
1670 | start_gfn += nr_pages_avail; | |
1671 | } | |
1672 | /* Use the slow path for cross page reads and writes. */ | |
1673 | ghc->memslot = NULL; | |
1674 | } | |
49c7754c GN |
1675 | return 0; |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1678 | ||
1679 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1680 | void *data, unsigned long len) | |
1681 | { | |
1682 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1683 | int r; | |
1684 | ||
8f964525 AH |
1685 | BUG_ON(len > ghc->len); |
1686 | ||
49c7754c | 1687 | if (slots->generation != ghc->generation) |
8f964525 AH |
1688 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1689 | ||
1690 | if (unlikely(!ghc->memslot)) | |
1691 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1692 | |
1693 | if (kvm_is_error_hva(ghc->hva)) | |
1694 | return -EFAULT; | |
1695 | ||
8b0cedff | 1696 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1697 | if (r) |
1698 | return -EFAULT; | |
1699 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); | |
1700 | ||
1701 | return 0; | |
1702 | } | |
1703 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1704 | ||
e03b644f GN |
1705 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1706 | void *data, unsigned long len) | |
1707 | { | |
1708 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1709 | int r; | |
1710 | ||
8f964525 AH |
1711 | BUG_ON(len > ghc->len); |
1712 | ||
e03b644f | 1713 | if (slots->generation != ghc->generation) |
8f964525 AH |
1714 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1715 | ||
1716 | if (unlikely(!ghc->memslot)) | |
1717 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1718 | |
1719 | if (kvm_is_error_hva(ghc->hva)) | |
1720 | return -EFAULT; | |
1721 | ||
1722 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1723 | if (r) | |
1724 | return -EFAULT; | |
1725 | ||
1726 | return 0; | |
1727 | } | |
1728 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1729 | ||
195aefde IE |
1730 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1731 | { | |
8a3caa6d HC |
1732 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
1733 | ||
1734 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
1735 | } |
1736 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1737 | ||
1738 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1739 | { | |
1740 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1741 | int seg; | |
1742 | int offset = offset_in_page(gpa); | |
1743 | int ret; | |
1744 | ||
1745 | while ((seg = next_segment(len, offset)) != 0) { | |
1746 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); | |
1747 | if (ret < 0) | |
1748 | return ret; | |
1749 | offset = 0; | |
1750 | len -= seg; | |
1751 | ++gfn; | |
1752 | } | |
1753 | return 0; | |
1754 | } | |
1755 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1756 | ||
7940876e SH |
1757 | static void mark_page_dirty_in_slot(struct kvm *kvm, |
1758 | struct kvm_memory_slot *memslot, | |
1759 | gfn_t gfn) | |
6aa8b732 | 1760 | { |
7e9d619d RR |
1761 | if (memslot && memslot->dirty_bitmap) { |
1762 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1763 | |
b74ca3b3 | 1764 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1765 | } |
1766 | } | |
1767 | ||
49c7754c GN |
1768 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1769 | { | |
1770 | struct kvm_memory_slot *memslot; | |
1771 | ||
1772 | memslot = gfn_to_memslot(kvm, gfn); | |
1773 | mark_page_dirty_in_slot(kvm, memslot, gfn); | |
1774 | } | |
2ba9f0d8 | 1775 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 1776 | |
f7819512 PB |
1777 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
1778 | { | |
1779 | if (kvm_arch_vcpu_runnable(vcpu)) { | |
1780 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
1781 | return -EINTR; | |
1782 | } | |
1783 | if (kvm_cpu_has_pending_timer(vcpu)) | |
1784 | return -EINTR; | |
1785 | if (signal_pending(current)) | |
1786 | return -EINTR; | |
1787 | ||
1788 | return 0; | |
1789 | } | |
1790 | ||
b6958ce4 ED |
1791 | /* |
1792 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1793 | */ | |
8776e519 | 1794 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1795 | { |
f7819512 | 1796 | ktime_t start, cur; |
e5c239cf | 1797 | DEFINE_WAIT(wait); |
f7819512 PB |
1798 | bool waited = false; |
1799 | ||
1800 | start = cur = ktime_get(); | |
1801 | if (halt_poll_ns) { | |
1802 | ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns); | |
1803 | do { | |
1804 | /* | |
1805 | * This sets KVM_REQ_UNHALT if an interrupt | |
1806 | * arrives. | |
1807 | */ | |
1808 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
1809 | ++vcpu->stat.halt_successful_poll; | |
1810 | goto out; | |
1811 | } | |
1812 | cur = ktime_get(); | |
1813 | } while (single_task_running() && ktime_before(cur, stop)); | |
1814 | } | |
e5c239cf MT |
1815 | |
1816 | for (;;) { | |
1817 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1818 | ||
f7819512 | 1819 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
1820 | break; |
1821 | ||
f7819512 | 1822 | waited = true; |
b6958ce4 | 1823 | schedule(); |
b6958ce4 | 1824 | } |
d3bef15f | 1825 | |
e5c239cf | 1826 | finish_wait(&vcpu->wq, &wait); |
f7819512 PB |
1827 | cur = ktime_get(); |
1828 | ||
1829 | out: | |
1830 | trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited); | |
b6958ce4 | 1831 | } |
2ba9f0d8 | 1832 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 1833 | |
8c84780d | 1834 | #ifndef CONFIG_S390 |
b6d33834 CD |
1835 | /* |
1836 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1837 | */ | |
1838 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1839 | { | |
1840 | int me; | |
1841 | int cpu = vcpu->cpu; | |
1842 | wait_queue_head_t *wqp; | |
1843 | ||
1844 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1845 | if (waitqueue_active(wqp)) { | |
1846 | wake_up_interruptible(wqp); | |
1847 | ++vcpu->stat.halt_wakeup; | |
1848 | } | |
1849 | ||
1850 | me = get_cpu(); | |
1851 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1852 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1853 | smp_send_reschedule(cpu); | |
1854 | put_cpu(); | |
1855 | } | |
a20ed54d | 1856 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1857 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1858 | |
fa93384f | 1859 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
1860 | { |
1861 | struct pid *pid; | |
1862 | struct task_struct *task = NULL; | |
fa93384f | 1863 | int ret = 0; |
41628d33 KW |
1864 | |
1865 | rcu_read_lock(); | |
1866 | pid = rcu_dereference(target->pid); | |
1867 | if (pid) | |
27fbe64b | 1868 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
1869 | rcu_read_unlock(); |
1870 | if (!task) | |
c45c528e | 1871 | return ret; |
c45c528e | 1872 | ret = yield_to(task, 1); |
41628d33 | 1873 | put_task_struct(task); |
c45c528e R |
1874 | |
1875 | return ret; | |
41628d33 KW |
1876 | } |
1877 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
1878 | ||
06e48c51 R |
1879 | /* |
1880 | * Helper that checks whether a VCPU is eligible for directed yield. | |
1881 | * Most eligible candidate to yield is decided by following heuristics: | |
1882 | * | |
1883 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
1884 | * (preempted lock holder), indicated by @in_spin_loop. | |
1885 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
1886 | * | |
1887 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
1888 | * chance last time (mostly it has become eligible now since we have probably | |
1889 | * yielded to lockholder in last iteration. This is done by toggling | |
1890 | * @dy_eligible each time a VCPU checked for eligibility.) | |
1891 | * | |
1892 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
1893 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
1894 | * burning. Giving priority for a potential lock-holder increases lock | |
1895 | * progress. | |
1896 | * | |
1897 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
1898 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
1899 | * and continue with next VCPU and so on. | |
1900 | */ | |
7940876e | 1901 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 1902 | { |
4a55dd72 | 1903 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
1904 | bool eligible; |
1905 | ||
1906 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 1907 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
1908 | |
1909 | if (vcpu->spin_loop.in_spin_loop) | |
1910 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
1911 | ||
1912 | return eligible; | |
4a55dd72 SW |
1913 | #else |
1914 | return true; | |
06e48c51 | 1915 | #endif |
4a55dd72 | 1916 | } |
c45c528e | 1917 | |
217ece61 | 1918 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 1919 | { |
217ece61 RR |
1920 | struct kvm *kvm = me->kvm; |
1921 | struct kvm_vcpu *vcpu; | |
1922 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
1923 | int yielded = 0; | |
c45c528e | 1924 | int try = 3; |
217ece61 RR |
1925 | int pass; |
1926 | int i; | |
d255f4f2 | 1927 | |
4c088493 | 1928 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
1929 | /* |
1930 | * We boost the priority of a VCPU that is runnable but not | |
1931 | * currently running, because it got preempted by something | |
1932 | * else and called schedule in __vcpu_run. Hopefully that | |
1933 | * VCPU is holding the lock that we need and will release it. | |
1934 | * We approximate round-robin by starting at the last boosted VCPU. | |
1935 | */ | |
c45c528e | 1936 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 1937 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 1938 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
1939 | i = last_boosted_vcpu; |
1940 | continue; | |
1941 | } else if (pass && i > last_boosted_vcpu) | |
1942 | break; | |
7bc7ae25 R |
1943 | if (!ACCESS_ONCE(vcpu->preempted)) |
1944 | continue; | |
217ece61 RR |
1945 | if (vcpu == me) |
1946 | continue; | |
98f4a146 | 1947 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 1948 | continue; |
06e48c51 R |
1949 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
1950 | continue; | |
c45c528e R |
1951 | |
1952 | yielded = kvm_vcpu_yield_to(vcpu); | |
1953 | if (yielded > 0) { | |
217ece61 | 1954 | kvm->last_boosted_vcpu = i; |
217ece61 | 1955 | break; |
c45c528e R |
1956 | } else if (yielded < 0) { |
1957 | try--; | |
1958 | if (!try) | |
1959 | break; | |
217ece61 | 1960 | } |
217ece61 RR |
1961 | } |
1962 | } | |
4c088493 | 1963 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
1964 | |
1965 | /* Ensure vcpu is not eligible during next spinloop */ | |
1966 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
1967 | } |
1968 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
1969 | ||
e4a533a4 | 1970 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
1971 | { |
1972 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
1973 | struct page *page; |
1974 | ||
e4a533a4 | 1975 | if (vmf->pgoff == 0) |
039576c0 | 1976 | page = virt_to_page(vcpu->run); |
09566765 | 1977 | #ifdef CONFIG_X86 |
e4a533a4 | 1978 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 1979 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
1980 | #endif |
1981 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
1982 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
1983 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 1984 | #endif |
039576c0 | 1985 | else |
5b1c1493 | 1986 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 1987 | get_page(page); |
e4a533a4 | 1988 | vmf->page = page; |
1989 | return 0; | |
9a2bb7f4 AK |
1990 | } |
1991 | ||
f0f37e2f | 1992 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 1993 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
1994 | }; |
1995 | ||
1996 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
1997 | { | |
1998 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
1999 | return 0; | |
2000 | } | |
2001 | ||
bccf2150 AK |
2002 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2003 | { | |
2004 | struct kvm_vcpu *vcpu = filp->private_data; | |
2005 | ||
66c0b394 | 2006 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2007 | return 0; |
2008 | } | |
2009 | ||
3d3aab1b | 2010 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2011 | .release = kvm_vcpu_release, |
2012 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2013 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2014 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2015 | #endif | |
9a2bb7f4 | 2016 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2017 | .llseek = noop_llseek, |
bccf2150 AK |
2018 | }; |
2019 | ||
2020 | /* | |
2021 | * Allocates an inode for the vcpu. | |
2022 | */ | |
2023 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2024 | { | |
24009b05 | 2025 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2026 | } |
2027 | ||
c5ea7660 AK |
2028 | /* |
2029 | * Creates some virtual cpus. Good luck creating more than one. | |
2030 | */ | |
73880c80 | 2031 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2032 | { |
2033 | int r; | |
988a2cae | 2034 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 2035 | |
338c7dba AH |
2036 | if (id >= KVM_MAX_VCPUS) |
2037 | return -EINVAL; | |
2038 | ||
73880c80 | 2039 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
2040 | if (IS_ERR(vcpu)) |
2041 | return PTR_ERR(vcpu); | |
c5ea7660 | 2042 | |
15ad7146 AK |
2043 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2044 | ||
26e5215f AK |
2045 | r = kvm_arch_vcpu_setup(vcpu); |
2046 | if (r) | |
d780592b | 2047 | goto vcpu_destroy; |
26e5215f | 2048 | |
11ec2804 | 2049 | mutex_lock(&kvm->lock); |
3e515705 AK |
2050 | if (!kvm_vcpu_compatible(vcpu)) { |
2051 | r = -EINVAL; | |
2052 | goto unlock_vcpu_destroy; | |
2053 | } | |
73880c80 GN |
2054 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
2055 | r = -EINVAL; | |
d780592b | 2056 | goto unlock_vcpu_destroy; |
fb3f0f51 | 2057 | } |
73880c80 | 2058 | |
988a2cae GN |
2059 | kvm_for_each_vcpu(r, v, kvm) |
2060 | if (v->vcpu_id == id) { | |
73880c80 | 2061 | r = -EEXIST; |
d780592b | 2062 | goto unlock_vcpu_destroy; |
73880c80 GN |
2063 | } |
2064 | ||
2065 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2066 | |
fb3f0f51 | 2067 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2068 | kvm_get_kvm(kvm); |
bccf2150 | 2069 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2070 | if (r < 0) { |
2071 | kvm_put_kvm(kvm); | |
d780592b | 2072 | goto unlock_vcpu_destroy; |
73880c80 GN |
2073 | } |
2074 | ||
2075 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
2076 | smp_wmb(); | |
2077 | atomic_inc(&kvm->online_vcpus); | |
2078 | ||
73880c80 | 2079 | mutex_unlock(&kvm->lock); |
42897d86 | 2080 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2081 | return r; |
39c3b86e | 2082 | |
d780592b | 2083 | unlock_vcpu_destroy: |
7d8fece6 | 2084 | mutex_unlock(&kvm->lock); |
d780592b | 2085 | vcpu_destroy: |
d40ccc62 | 2086 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
2087 | return r; |
2088 | } | |
2089 | ||
1961d276 AK |
2090 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2091 | { | |
2092 | if (sigset) { | |
2093 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2094 | vcpu->sigset_active = 1; | |
2095 | vcpu->sigset = *sigset; | |
2096 | } else | |
2097 | vcpu->sigset_active = 0; | |
2098 | return 0; | |
2099 | } | |
2100 | ||
bccf2150 AK |
2101 | static long kvm_vcpu_ioctl(struct file *filp, |
2102 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2103 | { |
bccf2150 | 2104 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2105 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2106 | int r; |
fa3795a7 DH |
2107 | struct kvm_fpu *fpu = NULL; |
2108 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2109 | |
6d4e4c4f AK |
2110 | if (vcpu->kvm->mm != current->mm) |
2111 | return -EIO; | |
2122ff5e | 2112 | |
2ea75be3 DM |
2113 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2114 | return -EINVAL; | |
2115 | ||
2f4d9b54 | 2116 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2117 | /* |
2118 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2119 | * so vcpu_load() would break it. | |
2120 | */ | |
2121 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT) | |
2122 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); | |
2123 | #endif | |
2124 | ||
2125 | ||
9fc77441 MT |
2126 | r = vcpu_load(vcpu); |
2127 | if (r) | |
2128 | return r; | |
6aa8b732 | 2129 | switch (ioctl) { |
9a2bb7f4 | 2130 | case KVM_RUN: |
f0fe5108 AK |
2131 | r = -EINVAL; |
2132 | if (arg) | |
2133 | goto out; | |
7a72f7a1 CB |
2134 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
2135 | /* The thread running this VCPU changed. */ | |
2136 | struct pid *oldpid = vcpu->pid; | |
2137 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
2138 | rcu_assign_pointer(vcpu->pid, newpid); | |
2139 | if (oldpid) | |
2140 | synchronize_rcu(); | |
2141 | put_pid(oldpid); | |
2142 | } | |
b6c7a5dc | 2143 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2144 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2145 | break; |
6aa8b732 | 2146 | case KVM_GET_REGS: { |
3e4bb3ac | 2147 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2148 | |
3e4bb3ac XZ |
2149 | r = -ENOMEM; |
2150 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2151 | if (!kvm_regs) | |
6aa8b732 | 2152 | goto out; |
3e4bb3ac XZ |
2153 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2154 | if (r) | |
2155 | goto out_free1; | |
6aa8b732 | 2156 | r = -EFAULT; |
3e4bb3ac XZ |
2157 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2158 | goto out_free1; | |
6aa8b732 | 2159 | r = 0; |
3e4bb3ac XZ |
2160 | out_free1: |
2161 | kfree(kvm_regs); | |
6aa8b732 AK |
2162 | break; |
2163 | } | |
2164 | case KVM_SET_REGS: { | |
3e4bb3ac | 2165 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2166 | |
3e4bb3ac | 2167 | r = -ENOMEM; |
ff5c2c03 SL |
2168 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2169 | if (IS_ERR(kvm_regs)) { | |
2170 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2171 | goto out; |
ff5c2c03 | 2172 | } |
3e4bb3ac | 2173 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2174 | kfree(kvm_regs); |
6aa8b732 AK |
2175 | break; |
2176 | } | |
2177 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2178 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2179 | r = -ENOMEM; | |
2180 | if (!kvm_sregs) | |
2181 | goto out; | |
2182 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2183 | if (r) |
2184 | goto out; | |
2185 | r = -EFAULT; | |
fa3795a7 | 2186 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2187 | goto out; |
2188 | r = 0; | |
2189 | break; | |
2190 | } | |
2191 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2192 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2193 | if (IS_ERR(kvm_sregs)) { | |
2194 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2195 | kvm_sregs = NULL; |
6aa8b732 | 2196 | goto out; |
ff5c2c03 | 2197 | } |
fa3795a7 | 2198 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2199 | break; |
2200 | } | |
62d9f0db MT |
2201 | case KVM_GET_MP_STATE: { |
2202 | struct kvm_mp_state mp_state; | |
2203 | ||
2204 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2205 | if (r) | |
2206 | goto out; | |
2207 | r = -EFAULT; | |
2208 | if (copy_to_user(argp, &mp_state, sizeof mp_state)) | |
2209 | goto out; | |
2210 | r = 0; | |
2211 | break; | |
2212 | } | |
2213 | case KVM_SET_MP_STATE: { | |
2214 | struct kvm_mp_state mp_state; | |
2215 | ||
2216 | r = -EFAULT; | |
2217 | if (copy_from_user(&mp_state, argp, sizeof mp_state)) | |
2218 | goto out; | |
2219 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2220 | break; |
2221 | } | |
6aa8b732 AK |
2222 | case KVM_TRANSLATE: { |
2223 | struct kvm_translation tr; | |
2224 | ||
2225 | r = -EFAULT; | |
2f366987 | 2226 | if (copy_from_user(&tr, argp, sizeof tr)) |
6aa8b732 | 2227 | goto out; |
8b006791 | 2228 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2229 | if (r) |
2230 | goto out; | |
2231 | r = -EFAULT; | |
2f366987 | 2232 | if (copy_to_user(argp, &tr, sizeof tr)) |
6aa8b732 AK |
2233 | goto out; |
2234 | r = 0; | |
2235 | break; | |
2236 | } | |
d0bfb940 JK |
2237 | case KVM_SET_GUEST_DEBUG: { |
2238 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2239 | |
2240 | r = -EFAULT; | |
2f366987 | 2241 | if (copy_from_user(&dbg, argp, sizeof dbg)) |
6aa8b732 | 2242 | goto out; |
d0bfb940 | 2243 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2244 | break; |
2245 | } | |
1961d276 AK |
2246 | case KVM_SET_SIGNAL_MASK: { |
2247 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2248 | struct kvm_signal_mask kvm_sigmask; | |
2249 | sigset_t sigset, *p; | |
2250 | ||
2251 | p = NULL; | |
2252 | if (argp) { | |
2253 | r = -EFAULT; | |
2254 | if (copy_from_user(&kvm_sigmask, argp, | |
2255 | sizeof kvm_sigmask)) | |
2256 | goto out; | |
2257 | r = -EINVAL; | |
2258 | if (kvm_sigmask.len != sizeof sigset) | |
2259 | goto out; | |
2260 | r = -EFAULT; | |
2261 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
2262 | sizeof sigset)) | |
2263 | goto out; | |
2264 | p = &sigset; | |
2265 | } | |
376d41ff | 2266 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2267 | break; |
2268 | } | |
b8836737 | 2269 | case KVM_GET_FPU: { |
fa3795a7 DH |
2270 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2271 | r = -ENOMEM; | |
2272 | if (!fpu) | |
2273 | goto out; | |
2274 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2275 | if (r) |
2276 | goto out; | |
2277 | r = -EFAULT; | |
fa3795a7 | 2278 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2279 | goto out; |
2280 | r = 0; | |
2281 | break; | |
2282 | } | |
2283 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2284 | fpu = memdup_user(argp, sizeof(*fpu)); |
2285 | if (IS_ERR(fpu)) { | |
2286 | r = PTR_ERR(fpu); | |
18595411 | 2287 | fpu = NULL; |
b8836737 | 2288 | goto out; |
ff5c2c03 | 2289 | } |
fa3795a7 | 2290 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2291 | break; |
2292 | } | |
bccf2150 | 2293 | default: |
313a3dc7 | 2294 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2295 | } |
2296 | out: | |
2122ff5e | 2297 | vcpu_put(vcpu); |
fa3795a7 DH |
2298 | kfree(fpu); |
2299 | kfree(kvm_sregs); | |
bccf2150 AK |
2300 | return r; |
2301 | } | |
2302 | ||
de8e5d74 | 2303 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2304 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2305 | unsigned int ioctl, unsigned long arg) | |
2306 | { | |
2307 | struct kvm_vcpu *vcpu = filp->private_data; | |
2308 | void __user *argp = compat_ptr(arg); | |
2309 | int r; | |
2310 | ||
2311 | if (vcpu->kvm->mm != current->mm) | |
2312 | return -EIO; | |
2313 | ||
2314 | switch (ioctl) { | |
2315 | case KVM_SET_SIGNAL_MASK: { | |
2316 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2317 | struct kvm_signal_mask kvm_sigmask; | |
2318 | compat_sigset_t csigset; | |
2319 | sigset_t sigset; | |
2320 | ||
2321 | if (argp) { | |
2322 | r = -EFAULT; | |
2323 | if (copy_from_user(&kvm_sigmask, argp, | |
2324 | sizeof kvm_sigmask)) | |
2325 | goto out; | |
2326 | r = -EINVAL; | |
2327 | if (kvm_sigmask.len != sizeof csigset) | |
2328 | goto out; | |
2329 | r = -EFAULT; | |
2330 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
2331 | sizeof csigset)) | |
2332 | goto out; | |
760a9a30 AC |
2333 | sigset_from_compat(&sigset, &csigset); |
2334 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2335 | } else | |
2336 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2337 | break; |
2338 | } | |
2339 | default: | |
2340 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2341 | } | |
2342 | ||
2343 | out: | |
2344 | return r; | |
2345 | } | |
2346 | #endif | |
2347 | ||
852b6d57 SW |
2348 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2349 | int (*accessor)(struct kvm_device *dev, | |
2350 | struct kvm_device_attr *attr), | |
2351 | unsigned long arg) | |
2352 | { | |
2353 | struct kvm_device_attr attr; | |
2354 | ||
2355 | if (!accessor) | |
2356 | return -EPERM; | |
2357 | ||
2358 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2359 | return -EFAULT; | |
2360 | ||
2361 | return accessor(dev, &attr); | |
2362 | } | |
2363 | ||
2364 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2365 | unsigned long arg) | |
2366 | { | |
2367 | struct kvm_device *dev = filp->private_data; | |
2368 | ||
2369 | switch (ioctl) { | |
2370 | case KVM_SET_DEVICE_ATTR: | |
2371 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2372 | case KVM_GET_DEVICE_ATTR: | |
2373 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2374 | case KVM_HAS_DEVICE_ATTR: | |
2375 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2376 | default: | |
2377 | if (dev->ops->ioctl) | |
2378 | return dev->ops->ioctl(dev, ioctl, arg); | |
2379 | ||
2380 | return -ENOTTY; | |
2381 | } | |
2382 | } | |
2383 | ||
852b6d57 SW |
2384 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2385 | { | |
2386 | struct kvm_device *dev = filp->private_data; | |
2387 | struct kvm *kvm = dev->kvm; | |
2388 | ||
852b6d57 SW |
2389 | kvm_put_kvm(kvm); |
2390 | return 0; | |
2391 | } | |
2392 | ||
2393 | static const struct file_operations kvm_device_fops = { | |
2394 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2395 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2396 | .compat_ioctl = kvm_device_ioctl, |
2397 | #endif | |
852b6d57 SW |
2398 | .release = kvm_device_release, |
2399 | }; | |
2400 | ||
2401 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2402 | { | |
2403 | if (filp->f_op != &kvm_device_fops) | |
2404 | return NULL; | |
2405 | ||
2406 | return filp->private_data; | |
2407 | } | |
2408 | ||
d60eacb0 | 2409 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2410 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2411 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2412 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2413 | #endif |
d60eacb0 | 2414 | |
5975a2e0 | 2415 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2416 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2417 | #endif |
d60eacb0 WD |
2418 | }; |
2419 | ||
2420 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2421 | { | |
2422 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2423 | return -ENOSPC; | |
2424 | ||
2425 | if (kvm_device_ops_table[type] != NULL) | |
2426 | return -EEXIST; | |
2427 | ||
2428 | kvm_device_ops_table[type] = ops; | |
2429 | return 0; | |
2430 | } | |
2431 | ||
571ee1b6 WL |
2432 | void kvm_unregister_device_ops(u32 type) |
2433 | { | |
2434 | if (kvm_device_ops_table[type] != NULL) | |
2435 | kvm_device_ops_table[type] = NULL; | |
2436 | } | |
2437 | ||
852b6d57 SW |
2438 | static int kvm_ioctl_create_device(struct kvm *kvm, |
2439 | struct kvm_create_device *cd) | |
2440 | { | |
2441 | struct kvm_device_ops *ops = NULL; | |
2442 | struct kvm_device *dev; | |
2443 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2444 | int ret; | |
2445 | ||
d60eacb0 WD |
2446 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
2447 | return -ENODEV; | |
2448 | ||
2449 | ops = kvm_device_ops_table[cd->type]; | |
2450 | if (ops == NULL) | |
852b6d57 | 2451 | return -ENODEV; |
852b6d57 SW |
2452 | |
2453 | if (test) | |
2454 | return 0; | |
2455 | ||
2456 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2457 | if (!dev) | |
2458 | return -ENOMEM; | |
2459 | ||
2460 | dev->ops = ops; | |
2461 | dev->kvm = kvm; | |
852b6d57 SW |
2462 | |
2463 | ret = ops->create(dev, cd->type); | |
2464 | if (ret < 0) { | |
2465 | kfree(dev); | |
2466 | return ret; | |
2467 | } | |
2468 | ||
24009b05 | 2469 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2470 | if (ret < 0) { |
2471 | ops->destroy(dev); | |
2472 | return ret; | |
2473 | } | |
2474 | ||
07f0a7bd | 2475 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2476 | kvm_get_kvm(kvm); |
2477 | cd->fd = ret; | |
2478 | return 0; | |
2479 | } | |
2480 | ||
92b591a4 AG |
2481 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2482 | { | |
2483 | switch (arg) { | |
2484 | case KVM_CAP_USER_MEMORY: | |
2485 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2486 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
2487 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE | |
2488 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2489 | #endif | |
2490 | case KVM_CAP_INTERNAL_ERROR_DATA: | |
2491 | #ifdef CONFIG_HAVE_KVM_MSI | |
2492 | case KVM_CAP_SIGNAL_MSI: | |
2493 | #endif | |
297e2105 | 2494 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 2495 | case KVM_CAP_IRQFD: |
92b591a4 AG |
2496 | case KVM_CAP_IRQFD_RESAMPLE: |
2497 | #endif | |
2498 | case KVM_CAP_CHECK_EXTENSION_VM: | |
2499 | return 1; | |
2500 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2501 | case KVM_CAP_IRQ_ROUTING: | |
2502 | return KVM_MAX_IRQ_ROUTES; | |
2503 | #endif | |
2504 | default: | |
2505 | break; | |
2506 | } | |
2507 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2508 | } | |
2509 | ||
bccf2150 AK |
2510 | static long kvm_vm_ioctl(struct file *filp, |
2511 | unsigned int ioctl, unsigned long arg) | |
2512 | { | |
2513 | struct kvm *kvm = filp->private_data; | |
2514 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2515 | int r; |
bccf2150 | 2516 | |
6d4e4c4f AK |
2517 | if (kvm->mm != current->mm) |
2518 | return -EIO; | |
bccf2150 AK |
2519 | switch (ioctl) { |
2520 | case KVM_CREATE_VCPU: | |
2521 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2522 | break; |
6fc138d2 IE |
2523 | case KVM_SET_USER_MEMORY_REGION: { |
2524 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2525 | ||
2526 | r = -EFAULT; | |
2527 | if (copy_from_user(&kvm_userspace_mem, argp, | |
2528 | sizeof kvm_userspace_mem)) | |
2529 | goto out; | |
2530 | ||
47ae31e2 | 2531 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2532 | break; |
2533 | } | |
2534 | case KVM_GET_DIRTY_LOG: { | |
2535 | struct kvm_dirty_log log; | |
2536 | ||
2537 | r = -EFAULT; | |
2f366987 | 2538 | if (copy_from_user(&log, argp, sizeof log)) |
6aa8b732 | 2539 | goto out; |
2c6f5df9 | 2540 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2541 | break; |
2542 | } | |
5f94c174 LV |
2543 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2544 | case KVM_REGISTER_COALESCED_MMIO: { | |
2545 | struct kvm_coalesced_mmio_zone zone; | |
2546 | r = -EFAULT; | |
2547 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2548 | goto out; | |
5f94c174 | 2549 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2550 | break; |
2551 | } | |
2552 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2553 | struct kvm_coalesced_mmio_zone zone; | |
2554 | r = -EFAULT; | |
2555 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2556 | goto out; | |
5f94c174 | 2557 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2558 | break; |
2559 | } | |
2560 | #endif | |
721eecbf GH |
2561 | case KVM_IRQFD: { |
2562 | struct kvm_irqfd data; | |
2563 | ||
2564 | r = -EFAULT; | |
2565 | if (copy_from_user(&data, argp, sizeof data)) | |
2566 | goto out; | |
d4db2935 | 2567 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2568 | break; |
2569 | } | |
d34e6b17 GH |
2570 | case KVM_IOEVENTFD: { |
2571 | struct kvm_ioeventfd data; | |
2572 | ||
2573 | r = -EFAULT; | |
2574 | if (copy_from_user(&data, argp, sizeof data)) | |
2575 | goto out; | |
2576 | r = kvm_ioeventfd(kvm, &data); | |
2577 | break; | |
2578 | } | |
73880c80 GN |
2579 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2580 | case KVM_SET_BOOT_CPU_ID: | |
2581 | r = 0; | |
894a9c55 | 2582 | mutex_lock(&kvm->lock); |
73880c80 GN |
2583 | if (atomic_read(&kvm->online_vcpus) != 0) |
2584 | r = -EBUSY; | |
2585 | else | |
2586 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2587 | mutex_unlock(&kvm->lock); |
73880c80 | 2588 | break; |
07975ad3 JK |
2589 | #endif |
2590 | #ifdef CONFIG_HAVE_KVM_MSI | |
2591 | case KVM_SIGNAL_MSI: { | |
2592 | struct kvm_msi msi; | |
2593 | ||
2594 | r = -EFAULT; | |
2595 | if (copy_from_user(&msi, argp, sizeof msi)) | |
2596 | goto out; | |
2597 | r = kvm_send_userspace_msi(kvm, &msi); | |
2598 | break; | |
2599 | } | |
23d43cf9 CD |
2600 | #endif |
2601 | #ifdef __KVM_HAVE_IRQ_LINE | |
2602 | case KVM_IRQ_LINE_STATUS: | |
2603 | case KVM_IRQ_LINE: { | |
2604 | struct kvm_irq_level irq_event; | |
2605 | ||
2606 | r = -EFAULT; | |
2607 | if (copy_from_user(&irq_event, argp, sizeof irq_event)) | |
2608 | goto out; | |
2609 | ||
aa2fbe6d YZ |
2610 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2611 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2612 | if (r) |
2613 | goto out; | |
2614 | ||
2615 | r = -EFAULT; | |
2616 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
2617 | if (copy_to_user(argp, &irq_event, sizeof irq_event)) | |
2618 | goto out; | |
2619 | } | |
2620 | ||
2621 | r = 0; | |
2622 | break; | |
2623 | } | |
73880c80 | 2624 | #endif |
aa8d5944 AG |
2625 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2626 | case KVM_SET_GSI_ROUTING: { | |
2627 | struct kvm_irq_routing routing; | |
2628 | struct kvm_irq_routing __user *urouting; | |
2629 | struct kvm_irq_routing_entry *entries; | |
2630 | ||
2631 | r = -EFAULT; | |
2632 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2633 | goto out; | |
2634 | r = -EINVAL; | |
2635 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2636 | goto out; | |
2637 | if (routing.flags) | |
2638 | goto out; | |
2639 | r = -ENOMEM; | |
2640 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2641 | if (!entries) | |
2642 | goto out; | |
2643 | r = -EFAULT; | |
2644 | urouting = argp; | |
2645 | if (copy_from_user(entries, urouting->entries, | |
2646 | routing.nr * sizeof(*entries))) | |
2647 | goto out_free_irq_routing; | |
2648 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2649 | routing.flags); | |
2650 | out_free_irq_routing: | |
2651 | vfree(entries); | |
2652 | break; | |
2653 | } | |
2654 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2655 | case KVM_CREATE_DEVICE: { |
2656 | struct kvm_create_device cd; | |
2657 | ||
2658 | r = -EFAULT; | |
2659 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2660 | goto out; | |
2661 | ||
2662 | r = kvm_ioctl_create_device(kvm, &cd); | |
2663 | if (r) | |
2664 | goto out; | |
2665 | ||
2666 | r = -EFAULT; | |
2667 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2668 | goto out; | |
2669 | ||
2670 | r = 0; | |
2671 | break; | |
2672 | } | |
92b591a4 AG |
2673 | case KVM_CHECK_EXTENSION: |
2674 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
2675 | break; | |
f17abe9a | 2676 | default: |
1fe779f8 | 2677 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
2678 | } |
2679 | out: | |
2680 | return r; | |
2681 | } | |
2682 | ||
de8e5d74 | 2683 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2684 | struct compat_kvm_dirty_log { |
2685 | __u32 slot; | |
2686 | __u32 padding1; | |
2687 | union { | |
2688 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2689 | __u64 padding2; | |
2690 | }; | |
2691 | }; | |
2692 | ||
2693 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2694 | unsigned int ioctl, unsigned long arg) | |
2695 | { | |
2696 | struct kvm *kvm = filp->private_data; | |
2697 | int r; | |
2698 | ||
2699 | if (kvm->mm != current->mm) | |
2700 | return -EIO; | |
2701 | switch (ioctl) { | |
2702 | case KVM_GET_DIRTY_LOG: { | |
2703 | struct compat_kvm_dirty_log compat_log; | |
2704 | struct kvm_dirty_log log; | |
2705 | ||
2706 | r = -EFAULT; | |
2707 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2708 | sizeof(compat_log))) | |
2709 | goto out; | |
2710 | log.slot = compat_log.slot; | |
2711 | log.padding1 = compat_log.padding1; | |
2712 | log.padding2 = compat_log.padding2; | |
2713 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2714 | ||
2715 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2716 | break; |
2717 | } | |
2718 | default: | |
2719 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2720 | } | |
2721 | ||
2722 | out: | |
2723 | return r; | |
2724 | } | |
2725 | #endif | |
2726 | ||
3d3aab1b | 2727 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2728 | .release = kvm_vm_release, |
2729 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 2730 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2731 | .compat_ioctl = kvm_vm_compat_ioctl, |
2732 | #endif | |
6038f373 | 2733 | .llseek = noop_llseek, |
f17abe9a AK |
2734 | }; |
2735 | ||
e08b9637 | 2736 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2737 | { |
aac87636 | 2738 | int r; |
f17abe9a AK |
2739 | struct kvm *kvm; |
2740 | ||
e08b9637 | 2741 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2742 | if (IS_ERR(kvm)) |
2743 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2744 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2745 | r = kvm_coalesced_mmio_init(kvm); | |
2746 | if (r < 0) { | |
2747 | kvm_put_kvm(kvm); | |
2748 | return r; | |
2749 | } | |
2750 | #endif | |
24009b05 | 2751 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2752 | if (r < 0) |
66c0b394 | 2753 | kvm_put_kvm(kvm); |
f17abe9a | 2754 | |
aac87636 | 2755 | return r; |
f17abe9a AK |
2756 | } |
2757 | ||
2758 | static long kvm_dev_ioctl(struct file *filp, | |
2759 | unsigned int ioctl, unsigned long arg) | |
2760 | { | |
07c45a36 | 2761 | long r = -EINVAL; |
f17abe9a AK |
2762 | |
2763 | switch (ioctl) { | |
2764 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2765 | if (arg) |
2766 | goto out; | |
f17abe9a AK |
2767 | r = KVM_API_VERSION; |
2768 | break; | |
2769 | case KVM_CREATE_VM: | |
e08b9637 | 2770 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2771 | break; |
018d00d2 | 2772 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 2773 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 2774 | break; |
07c45a36 | 2775 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
2776 | if (arg) |
2777 | goto out; | |
adb1ff46 AK |
2778 | r = PAGE_SIZE; /* struct kvm_run */ |
2779 | #ifdef CONFIG_X86 | |
2780 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2781 | #endif |
2782 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2783 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2784 | #endif |
07c45a36 | 2785 | break; |
d4c9ff2d FEL |
2786 | case KVM_TRACE_ENABLE: |
2787 | case KVM_TRACE_PAUSE: | |
2788 | case KVM_TRACE_DISABLE: | |
2023a29c | 2789 | r = -EOPNOTSUPP; |
d4c9ff2d | 2790 | break; |
6aa8b732 | 2791 | default: |
043405e1 | 2792 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2793 | } |
2794 | out: | |
2795 | return r; | |
2796 | } | |
2797 | ||
6aa8b732 | 2798 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2799 | .unlocked_ioctl = kvm_dev_ioctl, |
2800 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2801 | .llseek = noop_llseek, |
6aa8b732 AK |
2802 | }; |
2803 | ||
2804 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2805 | KVM_MINOR, |
6aa8b732 AK |
2806 | "kvm", |
2807 | &kvm_chardev_ops, | |
2808 | }; | |
2809 | ||
75b7127c | 2810 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2811 | { |
2812 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2813 | int r; |
1b6c0168 | 2814 | |
7f59f492 | 2815 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2816 | return; |
10474ae8 | 2817 | |
7f59f492 | 2818 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 2819 | |
13a34e06 | 2820 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
2821 | |
2822 | if (r) { | |
2823 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2824 | atomic_inc(&hardware_enable_failed); | |
2825 | printk(KERN_INFO "kvm: enabling virtualization on " | |
2826 | "CPU%d failed\n", cpu); | |
2827 | } | |
1b6c0168 AK |
2828 | } |
2829 | ||
4fa92fb2 | 2830 | static void hardware_enable(void) |
75b7127c | 2831 | { |
4a937f96 | 2832 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2833 | if (kvm_usage_count) |
2834 | hardware_enable_nolock(NULL); | |
4a937f96 | 2835 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2836 | } |
2837 | ||
2838 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2839 | { |
2840 | int cpu = raw_smp_processor_id(); | |
2841 | ||
7f59f492 | 2842 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2843 | return; |
7f59f492 | 2844 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 2845 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
2846 | } |
2847 | ||
4fa92fb2 | 2848 | static void hardware_disable(void) |
75b7127c | 2849 | { |
4a937f96 | 2850 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2851 | if (kvm_usage_count) |
2852 | hardware_disable_nolock(NULL); | |
4a937f96 | 2853 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2854 | } |
2855 | ||
10474ae8 AG |
2856 | static void hardware_disable_all_nolock(void) |
2857 | { | |
2858 | BUG_ON(!kvm_usage_count); | |
2859 | ||
2860 | kvm_usage_count--; | |
2861 | if (!kvm_usage_count) | |
75b7127c | 2862 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2863 | } |
2864 | ||
2865 | static void hardware_disable_all(void) | |
2866 | { | |
4a937f96 | 2867 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 2868 | hardware_disable_all_nolock(); |
4a937f96 | 2869 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2870 | } |
2871 | ||
2872 | static int hardware_enable_all(void) | |
2873 | { | |
2874 | int r = 0; | |
2875 | ||
4a937f96 | 2876 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
2877 | |
2878 | kvm_usage_count++; | |
2879 | if (kvm_usage_count == 1) { | |
2880 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 2881 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
2882 | |
2883 | if (atomic_read(&hardware_enable_failed)) { | |
2884 | hardware_disable_all_nolock(); | |
2885 | r = -EBUSY; | |
2886 | } | |
2887 | } | |
2888 | ||
4a937f96 | 2889 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2890 | |
2891 | return r; | |
2892 | } | |
2893 | ||
774c47f1 AK |
2894 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
2895 | void *v) | |
2896 | { | |
2897 | int cpu = (long)v; | |
2898 | ||
1a6f4d7f | 2899 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 2900 | switch (val) { |
cec9ad27 | 2901 | case CPU_DYING: |
6ec8a856 AK |
2902 | printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n", |
2903 | cpu); | |
4fa92fb2 | 2904 | hardware_disable(); |
6ec8a856 | 2905 | break; |
da908f2f | 2906 | case CPU_STARTING: |
43934a38 JK |
2907 | printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n", |
2908 | cpu); | |
4fa92fb2 | 2909 | hardware_enable(); |
774c47f1 AK |
2910 | break; |
2911 | } | |
2912 | return NOTIFY_OK; | |
2913 | } | |
2914 | ||
9a2b85c6 | 2915 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 2916 | void *v) |
9a2b85c6 | 2917 | { |
8e1c1815 SY |
2918 | /* |
2919 | * Some (well, at least mine) BIOSes hang on reboot if | |
2920 | * in vmx root mode. | |
2921 | * | |
2922 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
2923 | */ | |
2924 | printk(KERN_INFO "kvm: exiting hardware virtualization\n"); | |
2925 | kvm_rebooting = true; | |
75b7127c | 2926 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
2927 | return NOTIFY_OK; |
2928 | } | |
2929 | ||
2930 | static struct notifier_block kvm_reboot_notifier = { | |
2931 | .notifier_call = kvm_reboot, | |
2932 | .priority = 0, | |
2933 | }; | |
2934 | ||
e93f8a0f | 2935 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
2936 | { |
2937 | int i; | |
2938 | ||
2939 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 2940 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
2941 | |
2942 | kvm_iodevice_destructor(pos); | |
2943 | } | |
e93f8a0f | 2944 | kfree(bus); |
2eeb2e94 GH |
2945 | } |
2946 | ||
c21fbff1 PB |
2947 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
2948 | const struct kvm_io_range *r2) | |
743eeb0b | 2949 | { |
743eeb0b SL |
2950 | if (r1->addr < r2->addr) |
2951 | return -1; | |
2952 | if (r1->addr + r1->len > r2->addr + r2->len) | |
2953 | return 1; | |
2954 | return 0; | |
2955 | } | |
2956 | ||
a343c9b7 PB |
2957 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
2958 | { | |
c21fbff1 | 2959 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
2960 | } |
2961 | ||
39369f7a | 2962 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
2963 | gpa_t addr, int len) |
2964 | { | |
743eeb0b SL |
2965 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
2966 | .addr = addr, | |
2967 | .len = len, | |
2968 | .dev = dev, | |
2969 | }; | |
2970 | ||
2971 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
2972 | kvm_io_bus_sort_cmp, NULL); | |
2973 | ||
2974 | return 0; | |
2975 | } | |
2976 | ||
39369f7a | 2977 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
2978 | gpa_t addr, int len) |
2979 | { | |
2980 | struct kvm_io_range *range, key; | |
2981 | int off; | |
2982 | ||
2983 | key = (struct kvm_io_range) { | |
2984 | .addr = addr, | |
2985 | .len = len, | |
2986 | }; | |
2987 | ||
2988 | range = bsearch(&key, bus->range, bus->dev_count, | |
2989 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
2990 | if (range == NULL) | |
2991 | return -ENOENT; | |
2992 | ||
2993 | off = range - bus->range; | |
2994 | ||
c21fbff1 | 2995 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
2996 | off--; |
2997 | ||
2998 | return off; | |
2999 | } | |
3000 | ||
126a5af5 CH |
3001 | static int __kvm_io_bus_write(struct kvm_io_bus *bus, |
3002 | struct kvm_io_range *range, const void *val) | |
3003 | { | |
3004 | int idx; | |
3005 | ||
3006 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3007 | if (idx < 0) | |
3008 | return -EOPNOTSUPP; | |
3009 | ||
3010 | while (idx < bus->dev_count && | |
c21fbff1 | 3011 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
3012 | if (!kvm_iodevice_write(bus->range[idx].dev, range->addr, |
3013 | range->len, val)) | |
3014 | return idx; | |
3015 | idx++; | |
3016 | } | |
3017 | ||
3018 | return -EOPNOTSUPP; | |
3019 | } | |
3020 | ||
bda9020e | 3021 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e93f8a0f | 3022 | int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3023 | int len, const void *val) |
2eeb2e94 | 3024 | { |
90d83dc3 | 3025 | struct kvm_io_bus *bus; |
743eeb0b | 3026 | struct kvm_io_range range; |
126a5af5 | 3027 | int r; |
743eeb0b SL |
3028 | |
3029 | range = (struct kvm_io_range) { | |
3030 | .addr = addr, | |
3031 | .len = len, | |
3032 | }; | |
90d83dc3 LJ |
3033 | |
3034 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
126a5af5 CH |
3035 | r = __kvm_io_bus_write(bus, &range, val); |
3036 | return r < 0 ? r : 0; | |
3037 | } | |
3038 | ||
3039 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
3040 | int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, | |
3041 | int len, const void *val, long cookie) | |
3042 | { | |
3043 | struct kvm_io_bus *bus; | |
3044 | struct kvm_io_range range; | |
3045 | ||
3046 | range = (struct kvm_io_range) { | |
3047 | .addr = addr, | |
3048 | .len = len, | |
3049 | }; | |
3050 | ||
3051 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
3052 | ||
3053 | /* First try the device referenced by cookie. */ | |
3054 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3055 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
126a5af5 CH |
3056 | if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len, |
3057 | val)) | |
3058 | return cookie; | |
3059 | ||
3060 | /* | |
3061 | * cookie contained garbage; fall back to search and return the | |
3062 | * correct cookie value. | |
3063 | */ | |
3064 | return __kvm_io_bus_write(bus, &range, val); | |
3065 | } | |
3066 | ||
3067 | static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range, | |
3068 | void *val) | |
3069 | { | |
3070 | int idx; | |
3071 | ||
3072 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3073 | if (idx < 0) |
3074 | return -EOPNOTSUPP; | |
3075 | ||
3076 | while (idx < bus->dev_count && | |
c21fbff1 | 3077 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
3078 | if (!kvm_iodevice_read(bus->range[idx].dev, range->addr, |
3079 | range->len, val)) | |
3080 | return idx; | |
743eeb0b SL |
3081 | idx++; |
3082 | } | |
3083 | ||
bda9020e MT |
3084 | return -EOPNOTSUPP; |
3085 | } | |
68c3b4d1 | 3086 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3087 | |
bda9020e | 3088 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e93f8a0f MT |
3089 | int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3090 | int len, void *val) | |
bda9020e | 3091 | { |
90d83dc3 | 3092 | struct kvm_io_bus *bus; |
743eeb0b | 3093 | struct kvm_io_range range; |
126a5af5 | 3094 | int r; |
743eeb0b SL |
3095 | |
3096 | range = (struct kvm_io_range) { | |
3097 | .addr = addr, | |
3098 | .len = len, | |
3099 | }; | |
e93f8a0f | 3100 | |
90d83dc3 | 3101 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); |
126a5af5 CH |
3102 | r = __kvm_io_bus_read(bus, &range, val); |
3103 | return r < 0 ? r : 0; | |
3104 | } | |
743eeb0b | 3105 | |
2eeb2e94 | 3106 | |
79fac95e | 3107 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3108 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3109 | int len, struct kvm_io_device *dev) | |
6c474694 | 3110 | { |
e93f8a0f | 3111 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3112 | |
e93f8a0f | 3113 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3114 | /* exclude ioeventfd which is limited by maximum fd */ |
3115 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3116 | return -ENOSPC; |
2eeb2e94 | 3117 | |
a1300716 AK |
3118 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
3119 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
3120 | if (!new_bus) |
3121 | return -ENOMEM; | |
a1300716 AK |
3122 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3123 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3124 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3125 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3126 | synchronize_srcu_expedited(&kvm->srcu); | |
3127 | kfree(bus); | |
090b7aff GH |
3128 | |
3129 | return 0; | |
3130 | } | |
3131 | ||
79fac95e | 3132 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3133 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3134 | struct kvm_io_device *dev) | |
090b7aff | 3135 | { |
e93f8a0f MT |
3136 | int i, r; |
3137 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3138 | |
cdfca7b3 | 3139 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 3140 | r = -ENOENT; |
a1300716 AK |
3141 | for (i = 0; i < bus->dev_count; i++) |
3142 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3143 | r = 0; |
090b7aff GH |
3144 | break; |
3145 | } | |
e93f8a0f | 3146 | |
a1300716 | 3147 | if (r) |
e93f8a0f | 3148 | return r; |
a1300716 AK |
3149 | |
3150 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
3151 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
3152 | if (!new_bus) | |
3153 | return -ENOMEM; | |
3154 | ||
3155 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3156 | new_bus->dev_count--; | |
3157 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3158 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3159 | |
3160 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3161 | synchronize_srcu_expedited(&kvm->srcu); | |
3162 | kfree(bus); | |
3163 | return r; | |
2eeb2e94 GH |
3164 | } |
3165 | ||
774c47f1 AK |
3166 | static struct notifier_block kvm_cpu_notifier = { |
3167 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3168 | }; |
3169 | ||
8b88b099 | 3170 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3171 | { |
3172 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3173 | struct kvm *kvm; |
3174 | ||
8b88b099 | 3175 | *val = 0; |
2f303b74 | 3176 | spin_lock(&kvm_lock); |
ba1389b7 | 3177 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3178 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3179 | spin_unlock(&kvm_lock); |
8b88b099 | 3180 | return 0; |
ba1389b7 AK |
3181 | } |
3182 | ||
3183 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3184 | ||
8b88b099 | 3185 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3186 | { |
3187 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3188 | struct kvm *kvm; |
3189 | struct kvm_vcpu *vcpu; | |
3190 | int i; | |
3191 | ||
8b88b099 | 3192 | *val = 0; |
2f303b74 | 3193 | spin_lock(&kvm_lock); |
1165f5fe | 3194 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3195 | kvm_for_each_vcpu(i, vcpu, kvm) |
3196 | *val += *(u32 *)((void *)vcpu + offset); | |
3197 | ||
2f303b74 | 3198 | spin_unlock(&kvm_lock); |
8b88b099 | 3199 | return 0; |
1165f5fe AK |
3200 | } |
3201 | ||
ba1389b7 AK |
3202 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3203 | ||
828c0950 | 3204 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3205 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3206 | [KVM_STAT_VM] = &vm_stat_fops, | |
3207 | }; | |
1165f5fe | 3208 | |
4f69b680 | 3209 | static int kvm_init_debug(void) |
6aa8b732 | 3210 | { |
0c8eb04a | 3211 | int r = -EEXIST; |
6aa8b732 AK |
3212 | struct kvm_stats_debugfs_item *p; |
3213 | ||
76f7c879 | 3214 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3215 | if (kvm_debugfs_dir == NULL) |
3216 | goto out; | |
3217 | ||
3218 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3219 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3220 | (void *)(long)p->offset, |
ba1389b7 | 3221 | stat_fops[p->kind]); |
4f69b680 H |
3222 | if (p->dentry == NULL) |
3223 | goto out_dir; | |
3224 | } | |
3225 | ||
3226 | return 0; | |
3227 | ||
3228 | out_dir: | |
3229 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3230 | out: | |
3231 | return r; | |
6aa8b732 AK |
3232 | } |
3233 | ||
3234 | static void kvm_exit_debug(void) | |
3235 | { | |
3236 | struct kvm_stats_debugfs_item *p; | |
3237 | ||
3238 | for (p = debugfs_entries; p->name; ++p) | |
3239 | debugfs_remove(p->dentry); | |
76f7c879 | 3240 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3241 | } |
3242 | ||
fb3600cc | 3243 | static int kvm_suspend(void) |
59ae6c6b | 3244 | { |
10474ae8 | 3245 | if (kvm_usage_count) |
75b7127c | 3246 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3247 | return 0; |
3248 | } | |
3249 | ||
fb3600cc | 3250 | static void kvm_resume(void) |
59ae6c6b | 3251 | { |
ca84d1a2 | 3252 | if (kvm_usage_count) { |
4a937f96 | 3253 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3254 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3255 | } |
59ae6c6b AK |
3256 | } |
3257 | ||
fb3600cc | 3258 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3259 | .suspend = kvm_suspend, |
3260 | .resume = kvm_resume, | |
3261 | }; | |
3262 | ||
15ad7146 AK |
3263 | static inline |
3264 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3265 | { | |
3266 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3267 | } | |
3268 | ||
3269 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3270 | { | |
3271 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3a08a8f9 R |
3272 | if (vcpu->preempted) |
3273 | vcpu->preempted = false; | |
15ad7146 | 3274 | |
e790d9ef RK |
3275 | kvm_arch_sched_in(vcpu, cpu); |
3276 | ||
e9b11c17 | 3277 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3278 | } |
3279 | ||
3280 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3281 | struct task_struct *next) | |
3282 | { | |
3283 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3284 | ||
3a08a8f9 R |
3285 | if (current->state == TASK_RUNNING) |
3286 | vcpu->preempted = true; | |
e9b11c17 | 3287 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3288 | } |
3289 | ||
0ee75bea | 3290 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3291 | struct module *module) |
6aa8b732 AK |
3292 | { |
3293 | int r; | |
002c7f7c | 3294 | int cpu; |
6aa8b732 | 3295 | |
f8c16bba ZX |
3296 | r = kvm_arch_init(opaque); |
3297 | if (r) | |
d2308784 | 3298 | goto out_fail; |
cb498ea2 | 3299 | |
7dac16c3 AH |
3300 | /* |
3301 | * kvm_arch_init makes sure there's at most one caller | |
3302 | * for architectures that support multiple implementations, | |
3303 | * like intel and amd on x86. | |
3304 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3305 | * conflicts in case kvm is already setup for another implementation. | |
3306 | */ | |
3307 | r = kvm_irqfd_init(); | |
3308 | if (r) | |
3309 | goto out_irqfd; | |
3310 | ||
8437a617 | 3311 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3312 | r = -ENOMEM; |
3313 | goto out_free_0; | |
3314 | } | |
3315 | ||
e9b11c17 | 3316 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3317 | if (r < 0) |
7f59f492 | 3318 | goto out_free_0a; |
6aa8b732 | 3319 | |
002c7f7c YS |
3320 | for_each_online_cpu(cpu) { |
3321 | smp_call_function_single(cpu, | |
e9b11c17 | 3322 | kvm_arch_check_processor_compat, |
8691e5a8 | 3323 | &r, 1); |
002c7f7c | 3324 | if (r < 0) |
d2308784 | 3325 | goto out_free_1; |
002c7f7c YS |
3326 | } |
3327 | ||
774c47f1 AK |
3328 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3329 | if (r) | |
d2308784 | 3330 | goto out_free_2; |
6aa8b732 AK |
3331 | register_reboot_notifier(&kvm_reboot_notifier); |
3332 | ||
c16f862d | 3333 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3334 | if (!vcpu_align) |
3335 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3336 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3337 | 0, NULL); |
c16f862d RR |
3338 | if (!kvm_vcpu_cache) { |
3339 | r = -ENOMEM; | |
fb3600cc | 3340 | goto out_free_3; |
c16f862d RR |
3341 | } |
3342 | ||
af585b92 GN |
3343 | r = kvm_async_pf_init(); |
3344 | if (r) | |
3345 | goto out_free; | |
3346 | ||
6aa8b732 | 3347 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3348 | kvm_vm_fops.owner = module; |
3349 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3350 | |
3351 | r = misc_register(&kvm_dev); | |
3352 | if (r) { | |
d77c26fc | 3353 | printk(KERN_ERR "kvm: misc device register failed\n"); |
af585b92 | 3354 | goto out_unreg; |
6aa8b732 AK |
3355 | } |
3356 | ||
fb3600cc RW |
3357 | register_syscore_ops(&kvm_syscore_ops); |
3358 | ||
15ad7146 AK |
3359 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3360 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3361 | ||
4f69b680 H |
3362 | r = kvm_init_debug(); |
3363 | if (r) { | |
3364 | printk(KERN_ERR "kvm: create debugfs files failed\n"); | |
3365 | goto out_undebugfs; | |
3366 | } | |
0ea4ed8e | 3367 | |
3c3c29fd PB |
3368 | r = kvm_vfio_ops_init(); |
3369 | WARN_ON(r); | |
3370 | ||
c7addb90 | 3371 | return 0; |
6aa8b732 | 3372 | |
4f69b680 H |
3373 | out_undebugfs: |
3374 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3375 | misc_deregister(&kvm_dev); |
af585b92 GN |
3376 | out_unreg: |
3377 | kvm_async_pf_deinit(); | |
6aa8b732 | 3378 | out_free: |
c16f862d | 3379 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3380 | out_free_3: |
6aa8b732 | 3381 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3382 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3383 | out_free_2: |
d2308784 | 3384 | out_free_1: |
e9b11c17 | 3385 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3386 | out_free_0a: |
3387 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3388 | out_free_0: |
a0f155e9 CH |
3389 | kvm_irqfd_exit(); |
3390 | out_irqfd: | |
7dac16c3 AH |
3391 | kvm_arch_exit(); |
3392 | out_fail: | |
6aa8b732 AK |
3393 | return r; |
3394 | } | |
cb498ea2 | 3395 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3396 | |
cb498ea2 | 3397 | void kvm_exit(void) |
6aa8b732 | 3398 | { |
0ea4ed8e | 3399 | kvm_exit_debug(); |
6aa8b732 | 3400 | misc_deregister(&kvm_dev); |
c16f862d | 3401 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3402 | kvm_async_pf_deinit(); |
fb3600cc | 3403 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3404 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3405 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3406 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3407 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3408 | kvm_arch_exit(); |
a0f155e9 | 3409 | kvm_irqfd_exit(); |
7f59f492 | 3410 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 3411 | kvm_vfio_ops_exit(); |
6aa8b732 | 3412 | } |
cb498ea2 | 3413 | EXPORT_SYMBOL_GPL(kvm_exit); |