]>
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 | ||
af669ac6 | 19 | #include <kvm/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 | ||
0fa97788 | 69 | static unsigned int halt_poll_ns; |
f7819512 PB |
70 | module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR); |
71 | ||
fa40a821 MT |
72 | /* |
73 | * Ordering of locks: | |
74 | * | |
b7d409de | 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; |
f4fee932 | 83 | static int kvm_usage_count; |
10474ae8 | 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; |
e23a808b | 92 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 93 | |
bccf2150 AK |
94 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
95 | unsigned long arg); | |
de8e5d74 | 96 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
97 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
98 | unsigned long arg); | |
99 | #endif | |
10474ae8 AG |
100 | static int hardware_enable_all(void); |
101 | static void hardware_disable_all(void); | |
bccf2150 | 102 | |
e93f8a0f | 103 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e SH |
104 | |
105 | static void kvm_release_pfn_dirty(pfn_t pfn); | |
106 | static void mark_page_dirty_in_slot(struct kvm *kvm, | |
107 | struct kvm_memory_slot *memslot, gfn_t gfn); | |
e93f8a0f | 108 | |
52480137 | 109 | __visible bool kvm_rebooting; |
b7c4145b | 110 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 111 | |
54dee993 MT |
112 | static bool largepages_enabled = true; |
113 | ||
bf4bea8e | 114 | bool kvm_is_reserved_pfn(pfn_t pfn) |
cbff90a7 | 115 | { |
11feeb49 | 116 | if (pfn_valid(pfn)) |
bf4bea8e | 117 | return PageReserved(pfn_to_page(pfn)); |
cbff90a7 BAY |
118 | |
119 | return true; | |
120 | } | |
121 | ||
bccf2150 AK |
122 | /* |
123 | * Switches to specified vcpu, until a matching vcpu_put() | |
124 | */ | |
9fc77441 | 125 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 126 | { |
15ad7146 AK |
127 | int cpu; |
128 | ||
9fc77441 MT |
129 | if (mutex_lock_killable(&vcpu->mutex)) |
130 | return -EINTR; | |
15ad7146 AK |
131 | cpu = get_cpu(); |
132 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 133 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 134 | put_cpu(); |
9fc77441 | 135 | return 0; |
6aa8b732 AK |
136 | } |
137 | ||
313a3dc7 | 138 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 139 | { |
15ad7146 | 140 | preempt_disable(); |
313a3dc7 | 141 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
142 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
143 | preempt_enable(); | |
6aa8b732 AK |
144 | mutex_unlock(&vcpu->mutex); |
145 | } | |
146 | ||
d9e368d6 AK |
147 | static void ack_flush(void *_completed) |
148 | { | |
d9e368d6 AK |
149 | } |
150 | ||
445b8236 | 151 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 152 | { |
597a5f55 | 153 | int i, cpu, me; |
6ef7a1bc RR |
154 | cpumask_var_t cpus; |
155 | bool called = true; | |
d9e368d6 | 156 | struct kvm_vcpu *vcpu; |
d9e368d6 | 157 | |
79f55997 | 158 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 159 | |
3cba4130 | 160 | me = get_cpu(); |
988a2cae | 161 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 162 | kvm_make_request(req, vcpu); |
d9e368d6 | 163 | cpu = vcpu->cpu; |
6b7e2d09 XG |
164 | |
165 | /* Set ->requests bit before we read ->mode */ | |
166 | smp_mb(); | |
167 | ||
168 | if (cpus != NULL && cpu != -1 && cpu != me && | |
169 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 170 | cpumask_set_cpu(cpu, cpus); |
49846896 | 171 | } |
6ef7a1bc RR |
172 | if (unlikely(cpus == NULL)) |
173 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
174 | else if (!cpumask_empty(cpus)) | |
175 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
176 | else | |
177 | called = false; | |
3cba4130 | 178 | put_cpu(); |
6ef7a1bc | 179 | free_cpumask_var(cpus); |
49846896 | 180 | return called; |
d9e368d6 AK |
181 | } |
182 | ||
a6d51016 | 183 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 184 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 185 | { |
a086f6a1 XG |
186 | long dirty_count = kvm->tlbs_dirty; |
187 | ||
188 | smp_mb(); | |
445b8236 | 189 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
49846896 | 190 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 191 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 192 | } |
2ba9f0d8 | 193 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 194 | #endif |
2e53d63a | 195 | |
49846896 RR |
196 | void kvm_reload_remote_mmus(struct kvm *kvm) |
197 | { | |
445b8236 | 198 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 199 | } |
2e53d63a | 200 | |
d828199e MT |
201 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
202 | { | |
445b8236 | 203 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); |
d828199e MT |
204 | } |
205 | ||
3d81bc7e | 206 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
c7c9c56c | 207 | { |
445b8236 | 208 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); |
c7c9c56c YZ |
209 | } |
210 | ||
fb3f0f51 RR |
211 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
212 | { | |
213 | struct page *page; | |
214 | int r; | |
215 | ||
216 | mutex_init(&vcpu->mutex); | |
217 | vcpu->cpu = -1; | |
fb3f0f51 RR |
218 | vcpu->kvm = kvm; |
219 | vcpu->vcpu_id = id; | |
34bb10b7 | 220 | vcpu->pid = NULL; |
b6958ce4 | 221 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 222 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
223 | |
224 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
225 | if (!page) { | |
226 | r = -ENOMEM; | |
227 | goto fail; | |
228 | } | |
229 | vcpu->run = page_address(page); | |
230 | ||
4c088493 R |
231 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
232 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 233 | vcpu->preempted = false; |
4c088493 | 234 | |
e9b11c17 | 235 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 236 | if (r < 0) |
e9b11c17 | 237 | goto fail_free_run; |
fb3f0f51 RR |
238 | return 0; |
239 | ||
fb3f0f51 RR |
240 | fail_free_run: |
241 | free_page((unsigned long)vcpu->run); | |
242 | fail: | |
76fafa5e | 243 | return r; |
fb3f0f51 RR |
244 | } |
245 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
246 | ||
247 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
248 | { | |
34bb10b7 | 249 | put_pid(vcpu->pid); |
e9b11c17 | 250 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
251 | free_page((unsigned long)vcpu->run); |
252 | } | |
253 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
254 | ||
e930bffe AA |
255 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
256 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
257 | { | |
258 | return container_of(mn, struct kvm, mmu_notifier); | |
259 | } | |
260 | ||
261 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
262 | struct mm_struct *mm, | |
263 | unsigned long address) | |
264 | { | |
265 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 266 | int need_tlb_flush, idx; |
e930bffe AA |
267 | |
268 | /* | |
269 | * When ->invalidate_page runs, the linux pte has been zapped | |
270 | * already but the page is still allocated until | |
271 | * ->invalidate_page returns. So if we increase the sequence | |
272 | * here the kvm page fault will notice if the spte can't be | |
273 | * established because the page is going to be freed. If | |
274 | * instead the kvm page fault establishes the spte before | |
275 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
276 | * before returning. | |
277 | * | |
278 | * The sequence increase only need to be seen at spin_unlock | |
279 | * time, and not at spin_lock time. | |
280 | * | |
281 | * Increasing the sequence after the spin_unlock would be | |
282 | * unsafe because the kvm page fault could then establish the | |
283 | * pte after kvm_unmap_hva returned, without noticing the page | |
284 | * is going to be freed. | |
285 | */ | |
bc6678a3 | 286 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 287 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 288 | |
e930bffe | 289 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 290 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
291 | /* we've to flush the tlb before the pages can be freed */ |
292 | if (need_tlb_flush) | |
293 | kvm_flush_remote_tlbs(kvm); | |
294 | ||
565f3be2 | 295 | spin_unlock(&kvm->mmu_lock); |
fe71557a TC |
296 | |
297 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); | |
298 | ||
565f3be2 | 299 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
300 | } |
301 | ||
3da0dd43 IE |
302 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
303 | struct mm_struct *mm, | |
304 | unsigned long address, | |
305 | pte_t pte) | |
306 | { | |
307 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 308 | int idx; |
3da0dd43 | 309 | |
bc6678a3 | 310 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
311 | spin_lock(&kvm->mmu_lock); |
312 | kvm->mmu_notifier_seq++; | |
313 | kvm_set_spte_hva(kvm, address, pte); | |
314 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 315 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
316 | } |
317 | ||
e930bffe AA |
318 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
319 | struct mm_struct *mm, | |
320 | unsigned long start, | |
321 | unsigned long end) | |
322 | { | |
323 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 324 | int need_tlb_flush = 0, idx; |
e930bffe | 325 | |
bc6678a3 | 326 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
327 | spin_lock(&kvm->mmu_lock); |
328 | /* | |
329 | * The count increase must become visible at unlock time as no | |
330 | * spte can be established without taking the mmu_lock and | |
331 | * count is also read inside the mmu_lock critical section. | |
332 | */ | |
333 | kvm->mmu_notifier_count++; | |
b3ae2096 | 334 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 335 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
336 | /* we've to flush the tlb before the pages can be freed */ |
337 | if (need_tlb_flush) | |
338 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
339 | |
340 | spin_unlock(&kvm->mmu_lock); | |
341 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
342 | } |
343 | ||
344 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
345 | struct mm_struct *mm, | |
346 | unsigned long start, | |
347 | unsigned long end) | |
348 | { | |
349 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
350 | ||
351 | spin_lock(&kvm->mmu_lock); | |
352 | /* | |
353 | * This sequence increase will notify the kvm page fault that | |
354 | * the page that is going to be mapped in the spte could have | |
355 | * been freed. | |
356 | */ | |
357 | kvm->mmu_notifier_seq++; | |
a355aa54 | 358 | smp_wmb(); |
e930bffe AA |
359 | /* |
360 | * The above sequence increase must be visible before the | |
a355aa54 PM |
361 | * below count decrease, which is ensured by the smp_wmb above |
362 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
363 | */ |
364 | kvm->mmu_notifier_count--; | |
365 | spin_unlock(&kvm->mmu_lock); | |
366 | ||
367 | BUG_ON(kvm->mmu_notifier_count < 0); | |
368 | } | |
369 | ||
370 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
371 | struct mm_struct *mm, | |
57128468 ALC |
372 | unsigned long start, |
373 | unsigned long end) | |
e930bffe AA |
374 | { |
375 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 376 | int young, idx; |
e930bffe | 377 | |
bc6678a3 | 378 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 379 | spin_lock(&kvm->mmu_lock); |
e930bffe | 380 | |
57128468 | 381 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
382 | if (young) |
383 | kvm_flush_remote_tlbs(kvm); | |
384 | ||
565f3be2 TY |
385 | spin_unlock(&kvm->mmu_lock); |
386 | srcu_read_unlock(&kvm->srcu, idx); | |
387 | ||
e930bffe AA |
388 | return young; |
389 | } | |
390 | ||
8ee53820 AA |
391 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
392 | struct mm_struct *mm, | |
393 | unsigned long address) | |
394 | { | |
395 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
396 | int young, idx; | |
397 | ||
398 | idx = srcu_read_lock(&kvm->srcu); | |
399 | spin_lock(&kvm->mmu_lock); | |
400 | young = kvm_test_age_hva(kvm, address); | |
401 | spin_unlock(&kvm->mmu_lock); | |
402 | srcu_read_unlock(&kvm->srcu, idx); | |
403 | ||
404 | return young; | |
405 | } | |
406 | ||
85db06e5 MT |
407 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
408 | struct mm_struct *mm) | |
409 | { | |
410 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
411 | int idx; |
412 | ||
413 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 414 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 415 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
416 | } |
417 | ||
e930bffe AA |
418 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
419 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
420 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
421 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
422 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 423 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 424 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 425 | .release = kvm_mmu_notifier_release, |
e930bffe | 426 | }; |
4c07b0a4 AK |
427 | |
428 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
429 | { | |
430 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
431 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
432 | } | |
433 | ||
434 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
435 | ||
436 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
437 | { | |
438 | return 0; | |
439 | } | |
440 | ||
e930bffe AA |
441 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
442 | ||
a47d2b07 | 443 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
444 | { |
445 | int i; | |
a47d2b07 | 446 | struct kvm_memslots *slots; |
bf3e05bc | 447 | |
a47d2b07 PB |
448 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
449 | if (!slots) | |
450 | return NULL; | |
451 | ||
452 | /* | |
453 | * Init kvm generation close to the maximum to easily test the | |
454 | * code of handling generation number wrap-around. | |
455 | */ | |
456 | slots->generation = -150; | |
bf3e05bc | 457 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 458 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
459 | |
460 | return slots; | |
461 | } | |
462 | ||
463 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
464 | { | |
465 | if (!memslot->dirty_bitmap) | |
466 | return; | |
467 | ||
468 | kvfree(memslot->dirty_bitmap); | |
469 | memslot->dirty_bitmap = NULL; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Free any memory in @free but not in @dont. | |
474 | */ | |
475 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
476 | struct kvm_memory_slot *dont) | |
477 | { | |
478 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
479 | kvm_destroy_dirty_bitmap(free); | |
480 | ||
481 | kvm_arch_free_memslot(kvm, free, dont); | |
482 | ||
483 | free->npages = 0; | |
484 | } | |
485 | ||
486 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
487 | { | |
488 | struct kvm_memory_slot *memslot; | |
489 | ||
490 | if (!slots) | |
491 | return; | |
492 | ||
493 | kvm_for_each_memslot(memslot, slots) | |
494 | kvm_free_memslot(kvm, memslot, NULL); | |
495 | ||
496 | kvfree(slots); | |
bf3e05bc XG |
497 | } |
498 | ||
e08b9637 | 499 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 500 | { |
d89f5eff JK |
501 | int r, i; |
502 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 503 | |
d89f5eff JK |
504 | if (!kvm) |
505 | return ERR_PTR(-ENOMEM); | |
506 | ||
e08b9637 | 507 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 508 | if (r) |
719d93cd | 509 | goto out_err_no_disable; |
10474ae8 AG |
510 | |
511 | r = hardware_enable_all(); | |
512 | if (r) | |
719d93cd | 513 | goto out_err_no_disable; |
10474ae8 | 514 | |
c77dcacb | 515 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 516 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 517 | #endif |
6aa8b732 | 518 | |
1e702d9a AW |
519 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
520 | ||
46a26bf5 | 521 | r = -ENOMEM; |
a47d2b07 | 522 | kvm->memslots = kvm_alloc_memslots(); |
46a26bf5 | 523 | if (!kvm->memslots) |
719d93cd | 524 | goto out_err_no_srcu; |
00f034a1 | 525 | |
bc6678a3 | 526 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
527 | goto out_err_no_srcu; |
528 | if (init_srcu_struct(&kvm->irq_srcu)) | |
529 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
530 | for (i = 0; i < KVM_NR_BUSES; i++) { |
531 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
532 | GFP_KERNEL); | |
57e7fbee | 533 | if (!kvm->buses[i]) |
e93f8a0f | 534 | goto out_err; |
e93f8a0f | 535 | } |
e930bffe | 536 | |
74b5c5bf | 537 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
538 | kvm->mm = current->mm; |
539 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 540 | kvm_eventfd_init(kvm); |
11ec2804 | 541 | mutex_init(&kvm->lock); |
60eead79 | 542 | mutex_init(&kvm->irq_lock); |
79fac95e | 543 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 544 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 545 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
546 | |
547 | r = kvm_init_mmu_notifier(kvm); | |
548 | if (r) | |
549 | goto out_err; | |
550 | ||
2f303b74 | 551 | spin_lock(&kvm_lock); |
5e58cfe4 | 552 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 553 | spin_unlock(&kvm_lock); |
d89f5eff | 554 | |
f17abe9a | 555 | return kvm; |
10474ae8 AG |
556 | |
557 | out_err: | |
719d93cd CB |
558 | cleanup_srcu_struct(&kvm->irq_srcu); |
559 | out_err_no_irq_srcu: | |
57e7fbee | 560 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 561 | out_err_no_srcu: |
10474ae8 | 562 | hardware_disable_all(); |
719d93cd | 563 | out_err_no_disable: |
e93f8a0f MT |
564 | for (i = 0; i < KVM_NR_BUSES; i++) |
565 | kfree(kvm->buses[i]); | |
a47d2b07 | 566 | kvm_free_memslots(kvm, kvm->memslots); |
d89f5eff | 567 | kvm_arch_free_vm(kvm); |
10474ae8 | 568 | return ERR_PTR(r); |
f17abe9a AK |
569 | } |
570 | ||
92eca8fa TY |
571 | /* |
572 | * Avoid using vmalloc for a small buffer. | |
573 | * Should not be used when the size is statically known. | |
574 | */ | |
c1a7b32a | 575 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
576 | { |
577 | if (size > PAGE_SIZE) | |
578 | return vzalloc(size); | |
579 | else | |
580 | return kzalloc(size, GFP_KERNEL); | |
581 | } | |
582 | ||
07f0a7bd SW |
583 | static void kvm_destroy_devices(struct kvm *kvm) |
584 | { | |
585 | struct list_head *node, *tmp; | |
586 | ||
587 | list_for_each_safe(node, tmp, &kvm->devices) { | |
588 | struct kvm_device *dev = | |
589 | list_entry(node, struct kvm_device, vm_node); | |
590 | ||
591 | list_del(node); | |
592 | dev->ops->destroy(dev); | |
593 | } | |
594 | } | |
595 | ||
f17abe9a AK |
596 | static void kvm_destroy_vm(struct kvm *kvm) |
597 | { | |
e93f8a0f | 598 | int i; |
6d4e4c4f AK |
599 | struct mm_struct *mm = kvm->mm; |
600 | ||
ad8ba2cd | 601 | kvm_arch_sync_events(kvm); |
2f303b74 | 602 | spin_lock(&kvm_lock); |
133de902 | 603 | list_del(&kvm->vm_list); |
2f303b74 | 604 | spin_unlock(&kvm_lock); |
399ec807 | 605 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
606 | for (i = 0; i < KVM_NR_BUSES; i++) |
607 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 608 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
609 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
610 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 611 | #else |
2df72e9b | 612 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 613 | #endif |
d19a9cd2 | 614 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 615 | kvm_destroy_devices(kvm); |
a47d2b07 | 616 | kvm_free_memslots(kvm, kvm->memslots); |
820b3fcd | 617 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
618 | cleanup_srcu_struct(&kvm->srcu); |
619 | kvm_arch_free_vm(kvm); | |
10474ae8 | 620 | hardware_disable_all(); |
6d4e4c4f | 621 | mmdrop(mm); |
f17abe9a AK |
622 | } |
623 | ||
d39f13b0 IE |
624 | void kvm_get_kvm(struct kvm *kvm) |
625 | { | |
626 | atomic_inc(&kvm->users_count); | |
627 | } | |
628 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
629 | ||
630 | void kvm_put_kvm(struct kvm *kvm) | |
631 | { | |
632 | if (atomic_dec_and_test(&kvm->users_count)) | |
633 | kvm_destroy_vm(kvm); | |
634 | } | |
635 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
636 | ||
637 | ||
f17abe9a AK |
638 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
639 | { | |
640 | struct kvm *kvm = filp->private_data; | |
641 | ||
721eecbf GH |
642 | kvm_irqfd_release(kvm); |
643 | ||
d39f13b0 | 644 | kvm_put_kvm(kvm); |
6aa8b732 AK |
645 | return 0; |
646 | } | |
647 | ||
515a0127 TY |
648 | /* |
649 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 650 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 651 | */ |
a36a57b1 TY |
652 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
653 | { | |
515a0127 | 654 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 655 | |
92eca8fa | 656 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
657 | if (!memslot->dirty_bitmap) |
658 | return -ENOMEM; | |
659 | ||
a36a57b1 TY |
660 | return 0; |
661 | } | |
662 | ||
bf3e05bc | 663 | /* |
0e60b079 IM |
664 | * Insert memslot and re-sort memslots based on their GFN, |
665 | * so binary search could be used to lookup GFN. | |
666 | * Sorting algorithm takes advantage of having initially | |
667 | * sorted array and known changed memslot position. | |
bf3e05bc | 668 | */ |
5cc15027 PB |
669 | static void update_memslots(struct kvm_memslots *slots, |
670 | struct kvm_memory_slot *new) | |
bf3e05bc | 671 | { |
8593176c PB |
672 | int id = new->id; |
673 | int i = slots->id_to_index[id]; | |
063584d4 | 674 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 675 | |
8593176c | 676 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 677 | if (!new->npages) { |
dbaff309 | 678 | WARN_ON(!mslots[i].npages); |
0e60b079 | 679 | new->base_gfn = 0; |
b0165f1b | 680 | new->flags = 0; |
9c1a5d38 IM |
681 | if (mslots[i].npages) |
682 | slots->used_slots--; | |
683 | } else { | |
684 | if (!mslots[i].npages) | |
685 | slots->used_slots++; | |
686 | } | |
0e60b079 | 687 | |
7f379cff | 688 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
689 | new->base_gfn <= mslots[i + 1].base_gfn) { |
690 | if (!mslots[i + 1].npages) | |
691 | break; | |
7f379cff IM |
692 | mslots[i] = mslots[i + 1]; |
693 | slots->id_to_index[mslots[i].id] = i; | |
694 | i++; | |
695 | } | |
efbeec70 PB |
696 | |
697 | /* | |
698 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
699 | * so that it moves before all those with base_gfn == npages == 0. | |
700 | * | |
701 | * On the other hand, if new->npages is zero, the above loop has | |
702 | * already left i pointing to the beginning of the empty part of | |
703 | * mslots, and the ">=" would move the hole backwards in this | |
704 | * case---which is wrong. So skip the loop when deleting a slot. | |
705 | */ | |
706 | if (new->npages) { | |
707 | while (i > 0 && | |
708 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
709 | mslots[i] = mslots[i - 1]; | |
710 | slots->id_to_index[mslots[i].id] = i; | |
711 | i--; | |
712 | } | |
dbaff309 PB |
713 | } else |
714 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 715 | |
8593176c PB |
716 | mslots[i] = *new; |
717 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
718 | } |
719 | ||
a50d64d6 XG |
720 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
721 | { | |
4d8b81ab XG |
722 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
723 | ||
0f8a4de3 | 724 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
725 | valid_flags |= KVM_MEM_READONLY; |
726 | #endif | |
727 | ||
728 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
729 | return -EINVAL; |
730 | ||
731 | return 0; | |
732 | } | |
733 | ||
7ec4fb44 | 734 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
5cc15027 | 735 | struct kvm_memslots *slots) |
7ec4fb44 | 736 | { |
9f6b8029 | 737 | struct kvm_memslots *old_memslots = kvm_memslots(kvm); |
7ec4fb44 | 738 | |
ee3d1570 DM |
739 | /* |
740 | * Set the low bit in the generation, which disables SPTE caching | |
741 | * until the end of synchronize_srcu_expedited. | |
742 | */ | |
743 | WARN_ON(old_memslots->generation & 1); | |
744 | slots->generation = old_memslots->generation + 1; | |
745 | ||
7ec4fb44 GN |
746 | rcu_assign_pointer(kvm->memslots, slots); |
747 | synchronize_srcu_expedited(&kvm->srcu); | |
e59dbe09 | 748 | |
ee3d1570 DM |
749 | /* |
750 | * Increment the new memslot generation a second time. This prevents | |
751 | * vm exits that race with memslot updates from caching a memslot | |
752 | * generation that will (potentially) be valid forever. | |
753 | */ | |
754 | slots->generation++; | |
755 | ||
e59dbe09 TY |
756 | kvm_arch_memslots_updated(kvm); |
757 | ||
758 | return old_memslots; | |
7ec4fb44 GN |
759 | } |
760 | ||
6aa8b732 AK |
761 | /* |
762 | * Allocate some memory and give it an address in the guest physical address | |
763 | * space. | |
764 | * | |
765 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 766 | * |
02d5d55b | 767 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 768 | */ |
f78e0e2e | 769 | int __kvm_set_memory_region(struct kvm *kvm, |
47ae31e2 | 770 | struct kvm_userspace_memory_region *mem) |
6aa8b732 | 771 | { |
8234b22e | 772 | int r; |
6aa8b732 | 773 | gfn_t base_gfn; |
28bcb112 | 774 | unsigned long npages; |
a843fac2 | 775 | struct kvm_memory_slot *slot; |
6aa8b732 | 776 | struct kvm_memory_slot old, new; |
b7f69c55 | 777 | struct kvm_memslots *slots = NULL, *old_memslots; |
f64c0398 | 778 | enum kvm_mr_change change; |
6aa8b732 | 779 | |
a50d64d6 XG |
780 | r = check_memory_region_flags(mem); |
781 | if (r) | |
782 | goto out; | |
783 | ||
6aa8b732 AK |
784 | r = -EINVAL; |
785 | /* General sanity checks */ | |
786 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
787 | goto out; | |
788 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
789 | goto out; | |
fa3d315a | 790 | /* We can read the guest memory with __xxx_user() later on. */ |
47ae31e2 | 791 | if ((mem->slot < KVM_USER_MEM_SLOTS) && |
fa3d315a | 792 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
793 | !access_ok(VERIFY_WRITE, |
794 | (void __user *)(unsigned long)mem->userspace_addr, | |
795 | mem->memory_size))) | |
78749809 | 796 | goto out; |
93a5cef0 | 797 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
798 | goto out; |
799 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
800 | goto out; | |
801 | ||
9f6b8029 | 802 | slot = id_to_memslot(kvm_memslots(kvm), mem->slot); |
6aa8b732 AK |
803 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
804 | npages = mem->memory_size >> PAGE_SHIFT; | |
805 | ||
660c22c4 TY |
806 | if (npages > KVM_MEM_MAX_NR_PAGES) |
807 | goto out; | |
808 | ||
6aa8b732 AK |
809 | if (!npages) |
810 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
811 | ||
a843fac2 | 812 | new = old = *slot; |
6aa8b732 | 813 | |
e36d96f7 | 814 | new.id = mem->slot; |
6aa8b732 AK |
815 | new.base_gfn = base_gfn; |
816 | new.npages = npages; | |
817 | new.flags = mem->flags; | |
818 | ||
f64c0398 TY |
819 | if (npages) { |
820 | if (!old.npages) | |
821 | change = KVM_MR_CREATE; | |
822 | else { /* Modify an existing slot. */ | |
823 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
824 | (npages != old.npages) || |
825 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
826 | goto out; |
827 | ||
828 | if (base_gfn != old.base_gfn) | |
829 | change = KVM_MR_MOVE; | |
830 | else if (new.flags != old.flags) | |
831 | change = KVM_MR_FLAGS_ONLY; | |
832 | else { /* Nothing to change. */ | |
833 | r = 0; | |
834 | goto out; | |
835 | } | |
836 | } | |
837 | } else if (old.npages) { | |
838 | change = KVM_MR_DELETE; | |
839 | } else /* Modify a non-existent slot: disallowed. */ | |
0ea75e1d | 840 | goto out; |
6aa8b732 | 841 | |
f64c0398 | 842 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
843 | /* Check for overlaps */ |
844 | r = -EEXIST; | |
9f6b8029 | 845 | kvm_for_each_memslot(slot, kvm_memslots(kvm)) { |
a843fac2 TY |
846 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
847 | (slot->id == mem->slot)) | |
0a706bee TY |
848 | continue; |
849 | if (!((base_gfn + npages <= slot->base_gfn) || | |
850 | (base_gfn >= slot->base_gfn + slot->npages))) | |
851 | goto out; | |
852 | } | |
6aa8b732 | 853 | } |
6aa8b732 | 854 | |
6aa8b732 AK |
855 | /* Free page dirty bitmap if unneeded */ |
856 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 857 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
858 | |
859 | r = -ENOMEM; | |
f64c0398 | 860 | if (change == KVM_MR_CREATE) { |
189a2f7b | 861 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 862 | |
5587027c | 863 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 864 | goto out_free; |
6aa8b732 | 865 | } |
ec04b260 | 866 | |
6aa8b732 AK |
867 | /* Allocate page dirty bitmap if needed */ |
868 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 869 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 870 | goto out_free; |
6aa8b732 AK |
871 | } |
872 | ||
74496134 | 873 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
f2a81036 PB |
874 | if (!slots) |
875 | goto out_free; | |
9f6b8029 | 876 | memcpy(slots, kvm_memslots(kvm), sizeof(struct kvm_memslots)); |
f2a81036 | 877 | |
f64c0398 | 878 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
28a37544 XG |
879 | slot = id_to_memslot(slots, mem->slot); |
880 | slot->flags |= KVM_MEMSLOT_INVALID; | |
881 | ||
5cc15027 | 882 | old_memslots = install_new_memslots(kvm, slots); |
bc6678a3 | 883 | |
e40f193f AW |
884 | /* slot was deleted or moved, clear iommu mapping */ |
885 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
886 | /* From this point no new shadow pages pointing to a deleted, |
887 | * or moved, memslot will be created. | |
bc6678a3 MT |
888 | * |
889 | * validation of sp->gfn happens in: | |
b7d409de XL |
890 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
891 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 892 | */ |
2df72e9b | 893 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
894 | |
895 | /* | |
896 | * We can re-use the old_memslots from above, the only difference | |
897 | * from the currently installed memslots is the invalid flag. This | |
898 | * will get overwritten by update_memslots anyway. | |
899 | */ | |
b7f69c55 | 900 | slots = old_memslots; |
bc6678a3 | 901 | } |
34d4cb8f | 902 | |
7b6195a9 | 903 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 904 | if (r) |
b7f69c55 | 905 | goto out_slots; |
f7784b8e | 906 | |
a47d2b07 | 907 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 908 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 909 | new.dirty_bitmap = NULL; |
db3fe4eb | 910 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
911 | } |
912 | ||
5cc15027 PB |
913 | update_memslots(slots, &new); |
914 | old_memslots = install_new_memslots(kvm, slots); | |
3ad82a7e | 915 | |
8482644a | 916 | kvm_arch_commit_memory_region(kvm, mem, &old, change); |
82ce2c96 | 917 | |
a47d2b07 | 918 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 919 | kvfree(old_memslots); |
bc6678a3 | 920 | |
261874b0 AW |
921 | /* |
922 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
923 | * un-mapped and re-mapped if their base changes. Since base change |
924 | * unmapping is handled above with slot deletion, mapping alone is | |
925 | * needed here. Anything else the iommu might care about for existing | |
926 | * slots (size changes, userspace addr changes and read-only flag | |
927 | * changes) is disallowed above, so any other attribute changes getting | |
928 | * here can be skipped. | |
261874b0 | 929 | */ |
75d61fbc TY |
930 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
931 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 932 | return r; |
bc6678a3 MT |
933 | } |
934 | ||
6aa8b732 AK |
935 | return 0; |
936 | ||
e40f193f | 937 | out_slots: |
74496134 | 938 | kvfree(slots); |
f78e0e2e | 939 | out_free: |
a47d2b07 | 940 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
941 | out: |
942 | return r; | |
210c7c4d | 943 | } |
f78e0e2e SY |
944 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
945 | ||
946 | int kvm_set_memory_region(struct kvm *kvm, | |
47ae31e2 | 947 | struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
948 | { |
949 | int r; | |
950 | ||
79fac95e | 951 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 952 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 953 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
954 | return r; |
955 | } | |
210c7c4d IE |
956 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
957 | ||
7940876e SH |
958 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
959 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 960 | { |
bbacc0c1 | 961 | if (mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 962 | return -EINVAL; |
47ae31e2 | 963 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
964 | } |
965 | ||
5bb064dc ZX |
966 | int kvm_get_dirty_log(struct kvm *kvm, |
967 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 968 | { |
9f6b8029 | 969 | struct kvm_memslots *slots; |
6aa8b732 AK |
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 | ||
9f6b8029 PB |
979 | slots = kvm_memslots(kvm); |
980 | memslot = id_to_memslot(slots, log->slot); | |
6aa8b732 AK |
981 | r = -ENOENT; |
982 | if (!memslot->dirty_bitmap) | |
983 | goto out; | |
984 | ||
87bf6e7d | 985 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 986 | |
cd1a4a98 | 987 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
988 | any = memslot->dirty_bitmap[i]; |
989 | ||
990 | r = -EFAULT; | |
991 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
992 | goto out; | |
993 | ||
5bb064dc ZX |
994 | if (any) |
995 | *is_dirty = 1; | |
6aa8b732 AK |
996 | |
997 | r = 0; | |
6aa8b732 | 998 | out: |
6aa8b732 AK |
999 | return r; |
1000 | } | |
2ba9f0d8 | 1001 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1002 | |
ba0513b5 MS |
1003 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1004 | /** | |
1005 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1006 | * are dirty write protect them for next write. | |
1007 | * @kvm: pointer to kvm instance | |
1008 | * @log: slot id and address to which we copy the log | |
1009 | * @is_dirty: flag set if any page is dirty | |
1010 | * | |
1011 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1012 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1013 | * following order: | |
1014 | * | |
1015 | * 1. Take a snapshot of the bit and clear it if needed. | |
1016 | * 2. Write protect the corresponding page. | |
1017 | * 3. Copy the snapshot to the userspace. | |
1018 | * 4. Upon return caller flushes TLB's if needed. | |
1019 | * | |
1020 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1021 | * entry. This is not a problem because the page is reported dirty using | |
1022 | * the snapshot taken before and step 4 ensures that writes done after | |
1023 | * exiting to userspace will be logged for the next call. | |
1024 | * | |
1025 | */ | |
1026 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1027 | struct kvm_dirty_log *log, bool *is_dirty) | |
1028 | { | |
9f6b8029 | 1029 | struct kvm_memslots *slots; |
ba0513b5 MS |
1030 | struct kvm_memory_slot *memslot; |
1031 | int r, i; | |
1032 | unsigned long n; | |
1033 | unsigned long *dirty_bitmap; | |
1034 | unsigned long *dirty_bitmap_buffer; | |
1035 | ||
1036 | r = -EINVAL; | |
1037 | if (log->slot >= KVM_USER_MEM_SLOTS) | |
1038 | goto out; | |
1039 | ||
9f6b8029 PB |
1040 | slots = kvm_memslots(kvm); |
1041 | memslot = id_to_memslot(slots, log->slot); | |
ba0513b5 MS |
1042 | |
1043 | dirty_bitmap = memslot->dirty_bitmap; | |
1044 | r = -ENOENT; | |
1045 | if (!dirty_bitmap) | |
1046 | goto out; | |
1047 | ||
1048 | n = kvm_dirty_bitmap_bytes(memslot); | |
1049 | ||
1050 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1051 | memset(dirty_bitmap_buffer, 0, n); | |
1052 | ||
1053 | spin_lock(&kvm->mmu_lock); | |
1054 | *is_dirty = false; | |
1055 | for (i = 0; i < n / sizeof(long); i++) { | |
1056 | unsigned long mask; | |
1057 | gfn_t offset; | |
1058 | ||
1059 | if (!dirty_bitmap[i]) | |
1060 | continue; | |
1061 | ||
1062 | *is_dirty = true; | |
1063 | ||
1064 | mask = xchg(&dirty_bitmap[i], 0); | |
1065 | dirty_bitmap_buffer[i] = mask; | |
1066 | ||
58d2930f TY |
1067 | if (mask) { |
1068 | offset = i * BITS_PER_LONG; | |
1069 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1070 | offset, mask); | |
1071 | } | |
ba0513b5 MS |
1072 | } |
1073 | ||
1074 | spin_unlock(&kvm->mmu_lock); | |
1075 | ||
1076 | r = -EFAULT; | |
1077 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1078 | goto out; | |
1079 | ||
1080 | r = 0; | |
1081 | out: | |
1082 | return r; | |
1083 | } | |
1084 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1085 | #endif | |
1086 | ||
db3fe4eb TY |
1087 | bool kvm_largepages_enabled(void) |
1088 | { | |
1089 | return largepages_enabled; | |
1090 | } | |
1091 | ||
54dee993 MT |
1092 | void kvm_disable_largepages(void) |
1093 | { | |
1094 | largepages_enabled = false; | |
1095 | } | |
1096 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1097 | ||
49c7754c GN |
1098 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1099 | { | |
1100 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1101 | } | |
a1f4d395 | 1102 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1103 | |
e0d62c7f IE |
1104 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
1105 | { | |
bf3e05bc | 1106 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1107 | |
bbacc0c1 | 1108 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
1109 | memslot->flags & KVM_MEMSLOT_INVALID) |
1110 | return 0; | |
e0d62c7f | 1111 | |
bf3e05bc | 1112 | return 1; |
e0d62c7f IE |
1113 | } |
1114 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1115 | ||
8f0b1ab6 JR |
1116 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1117 | { | |
1118 | struct vm_area_struct *vma; | |
1119 | unsigned long addr, size; | |
1120 | ||
1121 | size = PAGE_SIZE; | |
1122 | ||
1123 | addr = gfn_to_hva(kvm, gfn); | |
1124 | if (kvm_is_error_hva(addr)) | |
1125 | return PAGE_SIZE; | |
1126 | ||
1127 | down_read(¤t->mm->mmap_sem); | |
1128 | vma = find_vma(current->mm, addr); | |
1129 | if (!vma) | |
1130 | goto out; | |
1131 | ||
1132 | size = vma_kernel_pagesize(vma); | |
1133 | ||
1134 | out: | |
1135 | up_read(¤t->mm->mmap_sem); | |
1136 | ||
1137 | return size; | |
1138 | } | |
1139 | ||
4d8b81ab XG |
1140 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1141 | { | |
1142 | return slot->flags & KVM_MEM_READONLY; | |
1143 | } | |
1144 | ||
4d8b81ab XG |
1145 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1146 | gfn_t *nr_pages, bool write) | |
539cb660 | 1147 | { |
bc6678a3 | 1148 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1149 | return KVM_HVA_ERR_BAD; |
48987781 | 1150 | |
4d8b81ab XG |
1151 | if (memslot_is_readonly(slot) && write) |
1152 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1153 | |
1154 | if (nr_pages) | |
1155 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1156 | ||
4d8b81ab | 1157 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1158 | } |
48987781 | 1159 | |
4d8b81ab XG |
1160 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1161 | gfn_t *nr_pages) | |
1162 | { | |
1163 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1164 | } |
48987781 | 1165 | |
4d8b81ab | 1166 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1167 | gfn_t gfn) |
4d8b81ab XG |
1168 | { |
1169 | return gfn_to_hva_many(slot, gfn, NULL); | |
1170 | } | |
1171 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1172 | ||
48987781 XG |
1173 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1174 | { | |
49c7754c | 1175 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1176 | } |
0d150298 | 1177 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1178 | |
86ab8cff | 1179 | /* |
ba6a3541 PB |
1180 | * If writable is set to false, the hva returned by this function is only |
1181 | * allowed to be read. | |
86ab8cff | 1182 | */ |
64d83126 CD |
1183 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1184 | gfn_t gfn, bool *writable) | |
86ab8cff | 1185 | { |
a2ac07fe GN |
1186 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1187 | ||
1188 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1189 | *writable = !memslot_is_readonly(slot); |
1190 | ||
a2ac07fe | 1191 | return hva; |
86ab8cff XG |
1192 | } |
1193 | ||
64d83126 CD |
1194 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1195 | { | |
1196 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1197 | ||
1198 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1199 | } | |
1200 | ||
39369f7a | 1201 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1202 | unsigned long start, int write, struct page **page) |
1203 | { | |
1204 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1205 | ||
1206 | if (write) | |
1207 | flags |= FOLL_WRITE; | |
1208 | ||
1209 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1210 | } | |
1211 | ||
fafc3dba HY |
1212 | static inline int check_user_page_hwpoison(unsigned long addr) |
1213 | { | |
1214 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1215 | ||
1216 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1217 | flags, NULL, NULL, NULL); | |
1218 | return rc == -EHWPOISON; | |
1219 | } | |
1220 | ||
2fc84311 XG |
1221 | /* |
1222 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1223 | * true indicates success, otherwise false is returned. | |
1224 | */ | |
1225 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1226 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1227 | { |
8d4e1288 | 1228 | struct page *page[1]; |
2fc84311 | 1229 | int npages; |
954bbbc2 | 1230 | |
2fc84311 XG |
1231 | if (!(async || atomic)) |
1232 | return false; | |
af585b92 | 1233 | |
12ce13fe XG |
1234 | /* |
1235 | * Fast pin a writable pfn only if it is a write fault request | |
1236 | * or the caller allows to map a writable pfn for a read fault | |
1237 | * request. | |
1238 | */ | |
1239 | if (!(write_fault || writable)) | |
1240 | return false; | |
612819c3 | 1241 | |
2fc84311 XG |
1242 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1243 | if (npages == 1) { | |
1244 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1245 | |
2fc84311 XG |
1246 | if (writable) |
1247 | *writable = true; | |
1248 | return true; | |
1249 | } | |
af585b92 | 1250 | |
2fc84311 XG |
1251 | return false; |
1252 | } | |
612819c3 | 1253 | |
2fc84311 XG |
1254 | /* |
1255 | * The slow path to get the pfn of the specified host virtual address, | |
1256 | * 1 indicates success, -errno is returned if error is detected. | |
1257 | */ | |
1258 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1259 | bool *writable, pfn_t *pfn) | |
1260 | { | |
1261 | struct page *page[1]; | |
1262 | int npages = 0; | |
612819c3 | 1263 | |
2fc84311 XG |
1264 | might_sleep(); |
1265 | ||
1266 | if (writable) | |
1267 | *writable = write_fault; | |
1268 | ||
1269 | if (async) { | |
1270 | down_read(¤t->mm->mmap_sem); | |
1271 | npages = get_user_page_nowait(current, current->mm, | |
1272 | addr, write_fault, page); | |
1273 | up_read(¤t->mm->mmap_sem); | |
0664e57f AA |
1274 | } else |
1275 | npages = __get_user_pages_unlocked(current, current->mm, addr, 1, | |
1276 | write_fault, 0, page, | |
1277 | FOLL_TOUCH|FOLL_HWPOISON); | |
2fc84311 XG |
1278 | if (npages != 1) |
1279 | return npages; | |
1280 | ||
1281 | /* map read fault as writable if possible */ | |
12ce13fe | 1282 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1283 | struct page *wpage[1]; |
1284 | ||
1285 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1286 | if (npages == 1) { | |
1287 | *writable = true; | |
1288 | put_page(page[0]); | |
1289 | page[0] = wpage[0]; | |
612819c3 | 1290 | } |
2fc84311 XG |
1291 | |
1292 | npages = 1; | |
887c08ac | 1293 | } |
2fc84311 XG |
1294 | *pfn = page_to_pfn(page[0]); |
1295 | return npages; | |
1296 | } | |
539cb660 | 1297 | |
4d8b81ab XG |
1298 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1299 | { | |
1300 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1301 | return false; | |
2e2e3738 | 1302 | |
4d8b81ab XG |
1303 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1304 | return false; | |
887c08ac | 1305 | |
4d8b81ab XG |
1306 | return true; |
1307 | } | |
bf998156 | 1308 | |
12ce13fe XG |
1309 | /* |
1310 | * Pin guest page in memory and return its pfn. | |
1311 | * @addr: host virtual address which maps memory to the guest | |
1312 | * @atomic: whether this function can sleep | |
1313 | * @async: whether this function need to wait IO complete if the | |
1314 | * host page is not in the memory | |
1315 | * @write_fault: whether we should get a writable host page | |
1316 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1317 | * | |
1318 | * The function will map a writable host page for these two cases: | |
1319 | * 1): @write_fault = true | |
1320 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1321 | * whether the mapping is writable. | |
1322 | */ | |
2fc84311 XG |
1323 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1324 | bool write_fault, bool *writable) | |
1325 | { | |
1326 | struct vm_area_struct *vma; | |
1327 | pfn_t pfn = 0; | |
1328 | int npages; | |
2e2e3738 | 1329 | |
2fc84311 XG |
1330 | /* we can do it either atomically or asynchronously, not both */ |
1331 | BUG_ON(atomic && async); | |
8d4e1288 | 1332 | |
2fc84311 XG |
1333 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1334 | return pfn; | |
1335 | ||
1336 | if (atomic) | |
1337 | return KVM_PFN_ERR_FAULT; | |
1338 | ||
1339 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1340 | if (npages == 1) | |
1341 | return pfn; | |
8d4e1288 | 1342 | |
2fc84311 XG |
1343 | down_read(¤t->mm->mmap_sem); |
1344 | if (npages == -EHWPOISON || | |
1345 | (!async && check_user_page_hwpoison(addr))) { | |
1346 | pfn = KVM_PFN_ERR_HWPOISON; | |
1347 | goto exit; | |
1348 | } | |
1349 | ||
1350 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1351 | ||
1352 | if (vma == NULL) | |
1353 | pfn = KVM_PFN_ERR_FAULT; | |
1354 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1355 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1356 | vma->vm_pgoff; | |
bf4bea8e | 1357 | BUG_ON(!kvm_is_reserved_pfn(pfn)); |
2fc84311 | 1358 | } else { |
4d8b81ab | 1359 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1360 | *async = true; |
1361 | pfn = KVM_PFN_ERR_FAULT; | |
1362 | } | |
1363 | exit: | |
1364 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1365 | return pfn; |
35149e21 AL |
1366 | } |
1367 | ||
3520469d PB |
1368 | pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, |
1369 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1370 | { |
4d8b81ab XG |
1371 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1372 | ||
1373 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1374 | return KVM_PFN_ERR_RO_FAULT; | |
1375 | ||
1376 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1377 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1378 | |
1379 | /* Do not map writable pfn in the readonly memslot. */ | |
1380 | if (writable && memslot_is_readonly(slot)) { | |
1381 | *writable = false; | |
1382 | writable = NULL; | |
1383 | } | |
1384 | ||
1385 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1386 | writable); | |
887c08ac | 1387 | } |
3520469d | 1388 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1389 | |
3520469d | 1390 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, |
612819c3 | 1391 | bool write_fault, bool *writable) |
506f0d6f | 1392 | { |
4d8b81ab | 1393 | struct kvm_memory_slot *slot; |
506f0d6f | 1394 | |
4d8b81ab | 1395 | slot = gfn_to_memslot(kvm, gfn); |
506f0d6f | 1396 | |
3520469d | 1397 | return __gfn_to_pfn_memslot(slot, gfn, atomic, NULL, write_fault, |
4d8b81ab | 1398 | writable); |
365fb3fd XG |
1399 | } |
1400 | ||
1401 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) | |
1402 | { | |
3520469d | 1403 | return __gfn_to_pfn(kvm, gfn, true, true, NULL); |
365fb3fd XG |
1404 | } |
1405 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1406 | ||
1407 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) | |
1408 | { | |
3520469d | 1409 | return __gfn_to_pfn(kvm, gfn, false, true, NULL); |
506f0d6f | 1410 | } |
35149e21 AL |
1411 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
1412 | ||
612819c3 MT |
1413 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1414 | bool *writable) | |
1415 | { | |
3520469d | 1416 | return __gfn_to_pfn(kvm, gfn, false, write_fault, writable); |
612819c3 MT |
1417 | } |
1418 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1419 | ||
d5661048 | 1420 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1421 | { |
4d8b81ab | 1422 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f MT |
1423 | } |
1424 | ||
037d92dc | 1425 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1426 | { |
4d8b81ab | 1427 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1428 | } |
037d92dc | 1429 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1430 | |
48987781 XG |
1431 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
1432 | int nr_pages) | |
1433 | { | |
1434 | unsigned long addr; | |
1435 | gfn_t entry; | |
1436 | ||
49c7754c | 1437 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
48987781 XG |
1438 | if (kvm_is_error_hva(addr)) |
1439 | return -1; | |
1440 | ||
1441 | if (entry < nr_pages) | |
1442 | return 0; | |
1443 | ||
1444 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1445 | } | |
1446 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1447 | ||
a2766325 XG |
1448 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1449 | { | |
81c52c56 | 1450 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1451 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1452 | |
bf4bea8e | 1453 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1454 | WARN_ON(1); |
6cede2e6 | 1455 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1456 | } |
a2766325 XG |
1457 | |
1458 | return pfn_to_page(pfn); | |
1459 | } | |
1460 | ||
35149e21 AL |
1461 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1462 | { | |
2e2e3738 AL |
1463 | pfn_t pfn; |
1464 | ||
1465 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1466 | |
a2766325 | 1467 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1468 | } |
1469 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1470 | ||
b4231d61 IE |
1471 | void kvm_release_page_clean(struct page *page) |
1472 | { | |
32cad84f XG |
1473 | WARN_ON(is_error_page(page)); |
1474 | ||
35149e21 | 1475 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1476 | } |
1477 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1478 | ||
35149e21 AL |
1479 | void kvm_release_pfn_clean(pfn_t pfn) |
1480 | { | |
bf4bea8e | 1481 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1482 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1483 | } |
1484 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1485 | ||
b4231d61 | 1486 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1487 | { |
a2766325 XG |
1488 | WARN_ON(is_error_page(page)); |
1489 | ||
35149e21 AL |
1490 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1491 | } | |
1492 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1493 | ||
7940876e | 1494 | static void kvm_release_pfn_dirty(pfn_t pfn) |
35149e21 AL |
1495 | { |
1496 | kvm_set_pfn_dirty(pfn); | |
1497 | kvm_release_pfn_clean(pfn); | |
1498 | } | |
35149e21 AL |
1499 | |
1500 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1501 | { | |
bf4bea8e | 1502 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 | 1503 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1504 | |
2e2e3738 AL |
1505 | if (!PageReserved(page)) |
1506 | SetPageDirty(page); | |
1507 | } | |
8a7ae055 | 1508 | } |
35149e21 AL |
1509 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1510 | ||
1511 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1512 | { | |
bf4bea8e | 1513 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1514 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1515 | } |
1516 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1517 | ||
1518 | void kvm_get_pfn(pfn_t pfn) | |
1519 | { | |
bf4bea8e | 1520 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1521 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1522 | } |
1523 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1524 | |
195aefde IE |
1525 | static int next_segment(unsigned long len, int offset) |
1526 | { | |
1527 | if (len > PAGE_SIZE - offset) | |
1528 | return PAGE_SIZE - offset; | |
1529 | else | |
1530 | return len; | |
1531 | } | |
1532 | ||
1533 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1534 | int len) | |
1535 | { | |
e0506bcb IE |
1536 | int r; |
1537 | unsigned long addr; | |
195aefde | 1538 | |
ba6a3541 | 1539 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
e0506bcb IE |
1540 | if (kvm_is_error_hva(addr)) |
1541 | return -EFAULT; | |
3180a7fc | 1542 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1543 | if (r) |
195aefde | 1544 | return -EFAULT; |
195aefde IE |
1545 | return 0; |
1546 | } | |
1547 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); | |
1548 | ||
1549 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) | |
1550 | { | |
1551 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1552 | int seg; | |
1553 | int offset = offset_in_page(gpa); | |
1554 | int ret; | |
1555 | ||
1556 | while ((seg = next_segment(len, offset)) != 0) { | |
1557 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1558 | if (ret < 0) | |
1559 | return ret; | |
1560 | offset = 0; | |
1561 | len -= seg; | |
1562 | data += seg; | |
1563 | ++gfn; | |
1564 | } | |
1565 | return 0; | |
1566 | } | |
1567 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1568 | ||
7ec54588 MT |
1569 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1570 | unsigned long len) | |
1571 | { | |
1572 | int r; | |
1573 | unsigned long addr; | |
1574 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1575 | int offset = offset_in_page(gpa); | |
1576 | ||
ba6a3541 | 1577 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
7ec54588 MT |
1578 | if (kvm_is_error_hva(addr)) |
1579 | return -EFAULT; | |
0aac03f0 | 1580 | pagefault_disable(); |
3180a7fc | 1581 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1582 | pagefault_enable(); |
7ec54588 MT |
1583 | if (r) |
1584 | return -EFAULT; | |
1585 | return 0; | |
1586 | } | |
1587 | EXPORT_SYMBOL(kvm_read_guest_atomic); | |
1588 | ||
195aefde IE |
1589 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
1590 | int offset, int len) | |
1591 | { | |
e0506bcb | 1592 | int r; |
251eb841 | 1593 | struct kvm_memory_slot *memslot; |
e0506bcb | 1594 | unsigned long addr; |
195aefde | 1595 | |
251eb841 RK |
1596 | memslot = gfn_to_memslot(kvm, gfn); |
1597 | addr = gfn_to_hva_memslot(memslot, gfn); | |
e0506bcb IE |
1598 | if (kvm_is_error_hva(addr)) |
1599 | return -EFAULT; | |
8b0cedff | 1600 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1601 | if (r) |
195aefde | 1602 | return -EFAULT; |
251eb841 | 1603 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
1604 | return 0; |
1605 | } | |
1606 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); | |
1607 | ||
1608 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, | |
1609 | unsigned long len) | |
1610 | { | |
1611 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1612 | int seg; | |
1613 | int offset = offset_in_page(gpa); | |
1614 | int ret; | |
1615 | ||
1616 | while ((seg = next_segment(len, offset)) != 0) { | |
1617 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1618 | if (ret < 0) | |
1619 | return ret; | |
1620 | offset = 0; | |
1621 | len -= seg; | |
1622 | data += seg; | |
1623 | ++gfn; | |
1624 | } | |
1625 | return 0; | |
1626 | } | |
ff651cb6 | 1627 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1628 | |
49c7754c | 1629 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1630 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1631 | { |
1632 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1633 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1634 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1635 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1636 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1637 | gfn_t nr_pages_avail; | |
49c7754c GN |
1638 | |
1639 | ghc->gpa = gpa; | |
1640 | ghc->generation = slots->generation; | |
8f964525 AH |
1641 | ghc->len = len; |
1642 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
ca3f0874 RK |
1643 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
1644 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { | |
49c7754c | 1645 | ghc->hva += offset; |
8f964525 AH |
1646 | } else { |
1647 | /* | |
1648 | * If the requested region crosses two memslots, we still | |
1649 | * verify that the entire region is valid here. | |
1650 | */ | |
1651 | while (start_gfn <= end_gfn) { | |
1652 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1653 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1654 | &nr_pages_avail); | |
1655 | if (kvm_is_error_hva(ghc->hva)) | |
1656 | return -EFAULT; | |
1657 | start_gfn += nr_pages_avail; | |
1658 | } | |
1659 | /* Use the slow path for cross page reads and writes. */ | |
1660 | ghc->memslot = NULL; | |
1661 | } | |
49c7754c GN |
1662 | return 0; |
1663 | } | |
1664 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1665 | ||
1666 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1667 | void *data, unsigned long len) | |
1668 | { | |
1669 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1670 | int r; | |
1671 | ||
8f964525 AH |
1672 | BUG_ON(len > ghc->len); |
1673 | ||
49c7754c | 1674 | if (slots->generation != ghc->generation) |
8f964525 AH |
1675 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1676 | ||
1677 | if (unlikely(!ghc->memslot)) | |
1678 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1679 | |
1680 | if (kvm_is_error_hva(ghc->hva)) | |
1681 | return -EFAULT; | |
1682 | ||
8b0cedff | 1683 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1684 | if (r) |
1685 | return -EFAULT; | |
1686 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); | |
1687 | ||
1688 | return 0; | |
1689 | } | |
1690 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1691 | ||
e03b644f GN |
1692 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1693 | void *data, unsigned long len) | |
1694 | { | |
1695 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1696 | int r; | |
1697 | ||
8f964525 AH |
1698 | BUG_ON(len > ghc->len); |
1699 | ||
e03b644f | 1700 | if (slots->generation != ghc->generation) |
8f964525 AH |
1701 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1702 | ||
1703 | if (unlikely(!ghc->memslot)) | |
1704 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1705 | |
1706 | if (kvm_is_error_hva(ghc->hva)) | |
1707 | return -EFAULT; | |
1708 | ||
1709 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1710 | if (r) | |
1711 | return -EFAULT; | |
1712 | ||
1713 | return 0; | |
1714 | } | |
1715 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1716 | ||
195aefde IE |
1717 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1718 | { | |
8a3caa6d HC |
1719 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
1720 | ||
1721 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
1722 | } |
1723 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1724 | ||
1725 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1726 | { | |
1727 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1728 | int seg; | |
1729 | int offset = offset_in_page(gpa); | |
1730 | int ret; | |
1731 | ||
bfda0e84 | 1732 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
1733 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
1734 | if (ret < 0) | |
1735 | return ret; | |
1736 | offset = 0; | |
1737 | len -= seg; | |
1738 | ++gfn; | |
1739 | } | |
1740 | return 0; | |
1741 | } | |
1742 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1743 | ||
7940876e SH |
1744 | static void mark_page_dirty_in_slot(struct kvm *kvm, |
1745 | struct kvm_memory_slot *memslot, | |
1746 | gfn_t gfn) | |
6aa8b732 | 1747 | { |
7e9d619d RR |
1748 | if (memslot && memslot->dirty_bitmap) { |
1749 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1750 | |
b74ca3b3 | 1751 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1752 | } |
1753 | } | |
1754 | ||
49c7754c GN |
1755 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1756 | { | |
1757 | struct kvm_memory_slot *memslot; | |
1758 | ||
1759 | memslot = gfn_to_memslot(kvm, gfn); | |
1760 | mark_page_dirty_in_slot(kvm, memslot, gfn); | |
1761 | } | |
2ba9f0d8 | 1762 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 1763 | |
f7819512 PB |
1764 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
1765 | { | |
1766 | if (kvm_arch_vcpu_runnable(vcpu)) { | |
1767 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
1768 | return -EINTR; | |
1769 | } | |
1770 | if (kvm_cpu_has_pending_timer(vcpu)) | |
1771 | return -EINTR; | |
1772 | if (signal_pending(current)) | |
1773 | return -EINTR; | |
1774 | ||
1775 | return 0; | |
1776 | } | |
1777 | ||
b6958ce4 ED |
1778 | /* |
1779 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1780 | */ | |
8776e519 | 1781 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1782 | { |
f7819512 | 1783 | ktime_t start, cur; |
e5c239cf | 1784 | DEFINE_WAIT(wait); |
f7819512 PB |
1785 | bool waited = false; |
1786 | ||
1787 | start = cur = ktime_get(); | |
1788 | if (halt_poll_ns) { | |
1789 | ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns); | |
f95ef0cd | 1790 | |
f7819512 PB |
1791 | do { |
1792 | /* | |
1793 | * This sets KVM_REQ_UNHALT if an interrupt | |
1794 | * arrives. | |
1795 | */ | |
1796 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
1797 | ++vcpu->stat.halt_successful_poll; | |
1798 | goto out; | |
1799 | } | |
1800 | cur = ktime_get(); | |
1801 | } while (single_task_running() && ktime_before(cur, stop)); | |
1802 | } | |
e5c239cf MT |
1803 | |
1804 | for (;;) { | |
1805 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1806 | ||
f7819512 | 1807 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
1808 | break; |
1809 | ||
f7819512 | 1810 | waited = true; |
b6958ce4 | 1811 | schedule(); |
b6958ce4 | 1812 | } |
d3bef15f | 1813 | |
e5c239cf | 1814 | finish_wait(&vcpu->wq, &wait); |
f7819512 PB |
1815 | cur = ktime_get(); |
1816 | ||
1817 | out: | |
1818 | trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited); | |
b6958ce4 | 1819 | } |
2ba9f0d8 | 1820 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 1821 | |
8c84780d | 1822 | #ifndef CONFIG_S390 |
b6d33834 CD |
1823 | /* |
1824 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1825 | */ | |
1826 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1827 | { | |
1828 | int me; | |
1829 | int cpu = vcpu->cpu; | |
1830 | wait_queue_head_t *wqp; | |
1831 | ||
1832 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1833 | if (waitqueue_active(wqp)) { | |
1834 | wake_up_interruptible(wqp); | |
1835 | ++vcpu->stat.halt_wakeup; | |
1836 | } | |
1837 | ||
1838 | me = get_cpu(); | |
1839 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1840 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1841 | smp_send_reschedule(cpu); | |
1842 | put_cpu(); | |
1843 | } | |
a20ed54d | 1844 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1845 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1846 | |
fa93384f | 1847 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
1848 | { |
1849 | struct pid *pid; | |
1850 | struct task_struct *task = NULL; | |
fa93384f | 1851 | int ret = 0; |
41628d33 KW |
1852 | |
1853 | rcu_read_lock(); | |
1854 | pid = rcu_dereference(target->pid); | |
1855 | if (pid) | |
27fbe64b | 1856 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
1857 | rcu_read_unlock(); |
1858 | if (!task) | |
c45c528e | 1859 | return ret; |
c45c528e | 1860 | ret = yield_to(task, 1); |
41628d33 | 1861 | put_task_struct(task); |
c45c528e R |
1862 | |
1863 | return ret; | |
41628d33 KW |
1864 | } |
1865 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
1866 | ||
06e48c51 R |
1867 | /* |
1868 | * Helper that checks whether a VCPU is eligible for directed yield. | |
1869 | * Most eligible candidate to yield is decided by following heuristics: | |
1870 | * | |
1871 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
1872 | * (preempted lock holder), indicated by @in_spin_loop. | |
1873 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
1874 | * | |
1875 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
1876 | * chance last time (mostly it has become eligible now since we have probably | |
1877 | * yielded to lockholder in last iteration. This is done by toggling | |
1878 | * @dy_eligible each time a VCPU checked for eligibility.) | |
1879 | * | |
1880 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
1881 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
1882 | * burning. Giving priority for a potential lock-holder increases lock | |
1883 | * progress. | |
1884 | * | |
1885 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
1886 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
1887 | * and continue with next VCPU and so on. | |
1888 | */ | |
7940876e | 1889 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 1890 | { |
4a55dd72 | 1891 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
1892 | bool eligible; |
1893 | ||
1894 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 1895 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
1896 | |
1897 | if (vcpu->spin_loop.in_spin_loop) | |
1898 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
1899 | ||
1900 | return eligible; | |
4a55dd72 SW |
1901 | #else |
1902 | return true; | |
06e48c51 | 1903 | #endif |
4a55dd72 | 1904 | } |
c45c528e | 1905 | |
217ece61 | 1906 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 1907 | { |
217ece61 RR |
1908 | struct kvm *kvm = me->kvm; |
1909 | struct kvm_vcpu *vcpu; | |
1910 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
1911 | int yielded = 0; | |
c45c528e | 1912 | int try = 3; |
217ece61 RR |
1913 | int pass; |
1914 | int i; | |
d255f4f2 | 1915 | |
4c088493 | 1916 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
1917 | /* |
1918 | * We boost the priority of a VCPU that is runnable but not | |
1919 | * currently running, because it got preempted by something | |
1920 | * else and called schedule in __vcpu_run. Hopefully that | |
1921 | * VCPU is holding the lock that we need and will release it. | |
1922 | * We approximate round-robin by starting at the last boosted VCPU. | |
1923 | */ | |
c45c528e | 1924 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 1925 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 1926 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
1927 | i = last_boosted_vcpu; |
1928 | continue; | |
1929 | } else if (pass && i > last_boosted_vcpu) | |
1930 | break; | |
7bc7ae25 R |
1931 | if (!ACCESS_ONCE(vcpu->preempted)) |
1932 | continue; | |
217ece61 RR |
1933 | if (vcpu == me) |
1934 | continue; | |
98f4a146 | 1935 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 1936 | continue; |
06e48c51 R |
1937 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
1938 | continue; | |
c45c528e R |
1939 | |
1940 | yielded = kvm_vcpu_yield_to(vcpu); | |
1941 | if (yielded > 0) { | |
217ece61 | 1942 | kvm->last_boosted_vcpu = i; |
217ece61 | 1943 | break; |
c45c528e R |
1944 | } else if (yielded < 0) { |
1945 | try--; | |
1946 | if (!try) | |
1947 | break; | |
217ece61 | 1948 | } |
217ece61 RR |
1949 | } |
1950 | } | |
4c088493 | 1951 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
1952 | |
1953 | /* Ensure vcpu is not eligible during next spinloop */ | |
1954 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
1955 | } |
1956 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
1957 | ||
e4a533a4 | 1958 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
1959 | { |
1960 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
1961 | struct page *page; |
1962 | ||
e4a533a4 | 1963 | if (vmf->pgoff == 0) |
039576c0 | 1964 | page = virt_to_page(vcpu->run); |
09566765 | 1965 | #ifdef CONFIG_X86 |
e4a533a4 | 1966 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 1967 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
1968 | #endif |
1969 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
1970 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
1971 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 1972 | #endif |
039576c0 | 1973 | else |
5b1c1493 | 1974 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 1975 | get_page(page); |
e4a533a4 | 1976 | vmf->page = page; |
1977 | return 0; | |
9a2bb7f4 AK |
1978 | } |
1979 | ||
f0f37e2f | 1980 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 1981 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
1982 | }; |
1983 | ||
1984 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
1985 | { | |
1986 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
1987 | return 0; | |
1988 | } | |
1989 | ||
bccf2150 AK |
1990 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
1991 | { | |
1992 | struct kvm_vcpu *vcpu = filp->private_data; | |
1993 | ||
66c0b394 | 1994 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
1995 | return 0; |
1996 | } | |
1997 | ||
3d3aab1b | 1998 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
1999 | .release = kvm_vcpu_release, |
2000 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2001 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2002 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2003 | #endif | |
9a2bb7f4 | 2004 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2005 | .llseek = noop_llseek, |
bccf2150 AK |
2006 | }; |
2007 | ||
2008 | /* | |
2009 | * Allocates an inode for the vcpu. | |
2010 | */ | |
2011 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2012 | { | |
24009b05 | 2013 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2014 | } |
2015 | ||
c5ea7660 AK |
2016 | /* |
2017 | * Creates some virtual cpus. Good luck creating more than one. | |
2018 | */ | |
73880c80 | 2019 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2020 | { |
2021 | int r; | |
988a2cae | 2022 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 2023 | |
338c7dba AH |
2024 | if (id >= KVM_MAX_VCPUS) |
2025 | return -EINVAL; | |
2026 | ||
73880c80 | 2027 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
2028 | if (IS_ERR(vcpu)) |
2029 | return PTR_ERR(vcpu); | |
c5ea7660 | 2030 | |
15ad7146 AK |
2031 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2032 | ||
26e5215f AK |
2033 | r = kvm_arch_vcpu_setup(vcpu); |
2034 | if (r) | |
d780592b | 2035 | goto vcpu_destroy; |
26e5215f | 2036 | |
11ec2804 | 2037 | mutex_lock(&kvm->lock); |
3e515705 AK |
2038 | if (!kvm_vcpu_compatible(vcpu)) { |
2039 | r = -EINVAL; | |
2040 | goto unlock_vcpu_destroy; | |
2041 | } | |
73880c80 GN |
2042 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
2043 | r = -EINVAL; | |
d780592b | 2044 | goto unlock_vcpu_destroy; |
fb3f0f51 | 2045 | } |
73880c80 | 2046 | |
988a2cae GN |
2047 | kvm_for_each_vcpu(r, v, kvm) |
2048 | if (v->vcpu_id == id) { | |
73880c80 | 2049 | r = -EEXIST; |
d780592b | 2050 | goto unlock_vcpu_destroy; |
73880c80 GN |
2051 | } |
2052 | ||
2053 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2054 | |
fb3f0f51 | 2055 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2056 | kvm_get_kvm(kvm); |
bccf2150 | 2057 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2058 | if (r < 0) { |
2059 | kvm_put_kvm(kvm); | |
d780592b | 2060 | goto unlock_vcpu_destroy; |
73880c80 GN |
2061 | } |
2062 | ||
2063 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
2064 | smp_wmb(); | |
2065 | atomic_inc(&kvm->online_vcpus); | |
2066 | ||
73880c80 | 2067 | mutex_unlock(&kvm->lock); |
42897d86 | 2068 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2069 | return r; |
39c3b86e | 2070 | |
d780592b | 2071 | unlock_vcpu_destroy: |
7d8fece6 | 2072 | mutex_unlock(&kvm->lock); |
d780592b | 2073 | vcpu_destroy: |
d40ccc62 | 2074 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
2075 | return r; |
2076 | } | |
2077 | ||
1961d276 AK |
2078 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2079 | { | |
2080 | if (sigset) { | |
2081 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2082 | vcpu->sigset_active = 1; | |
2083 | vcpu->sigset = *sigset; | |
2084 | } else | |
2085 | vcpu->sigset_active = 0; | |
2086 | return 0; | |
2087 | } | |
2088 | ||
bccf2150 AK |
2089 | static long kvm_vcpu_ioctl(struct file *filp, |
2090 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2091 | { |
bccf2150 | 2092 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2093 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2094 | int r; |
fa3795a7 DH |
2095 | struct kvm_fpu *fpu = NULL; |
2096 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2097 | |
6d4e4c4f AK |
2098 | if (vcpu->kvm->mm != current->mm) |
2099 | return -EIO; | |
2122ff5e | 2100 | |
2ea75be3 DM |
2101 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2102 | return -EINVAL; | |
2103 | ||
2f4d9b54 | 2104 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2105 | /* |
2106 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2107 | * so vcpu_load() would break it. | |
2108 | */ | |
47b43c52 | 2109 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
2122ff5e AK |
2110 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
2111 | #endif | |
2112 | ||
2113 | ||
9fc77441 MT |
2114 | r = vcpu_load(vcpu); |
2115 | if (r) | |
2116 | return r; | |
6aa8b732 | 2117 | switch (ioctl) { |
9a2bb7f4 | 2118 | case KVM_RUN: |
f0fe5108 AK |
2119 | r = -EINVAL; |
2120 | if (arg) | |
2121 | goto out; | |
7a72f7a1 CB |
2122 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
2123 | /* The thread running this VCPU changed. */ | |
2124 | struct pid *oldpid = vcpu->pid; | |
2125 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
f95ef0cd | 2126 | |
7a72f7a1 CB |
2127 | rcu_assign_pointer(vcpu->pid, newpid); |
2128 | if (oldpid) | |
2129 | synchronize_rcu(); | |
2130 | put_pid(oldpid); | |
2131 | } | |
b6c7a5dc | 2132 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2133 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2134 | break; |
6aa8b732 | 2135 | case KVM_GET_REGS: { |
3e4bb3ac | 2136 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2137 | |
3e4bb3ac XZ |
2138 | r = -ENOMEM; |
2139 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2140 | if (!kvm_regs) | |
6aa8b732 | 2141 | goto out; |
3e4bb3ac XZ |
2142 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2143 | if (r) | |
2144 | goto out_free1; | |
6aa8b732 | 2145 | r = -EFAULT; |
3e4bb3ac XZ |
2146 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2147 | goto out_free1; | |
6aa8b732 | 2148 | r = 0; |
3e4bb3ac XZ |
2149 | out_free1: |
2150 | kfree(kvm_regs); | |
6aa8b732 AK |
2151 | break; |
2152 | } | |
2153 | case KVM_SET_REGS: { | |
3e4bb3ac | 2154 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2155 | |
3e4bb3ac | 2156 | r = -ENOMEM; |
ff5c2c03 SL |
2157 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2158 | if (IS_ERR(kvm_regs)) { | |
2159 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2160 | goto out; |
ff5c2c03 | 2161 | } |
3e4bb3ac | 2162 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2163 | kfree(kvm_regs); |
6aa8b732 AK |
2164 | break; |
2165 | } | |
2166 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2167 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2168 | r = -ENOMEM; | |
2169 | if (!kvm_sregs) | |
2170 | goto out; | |
2171 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2172 | if (r) |
2173 | goto out; | |
2174 | r = -EFAULT; | |
fa3795a7 | 2175 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2176 | goto out; |
2177 | r = 0; | |
2178 | break; | |
2179 | } | |
2180 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2181 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2182 | if (IS_ERR(kvm_sregs)) { | |
2183 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2184 | kvm_sregs = NULL; |
6aa8b732 | 2185 | goto out; |
ff5c2c03 | 2186 | } |
fa3795a7 | 2187 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2188 | break; |
2189 | } | |
62d9f0db MT |
2190 | case KVM_GET_MP_STATE: { |
2191 | struct kvm_mp_state mp_state; | |
2192 | ||
2193 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2194 | if (r) | |
2195 | goto out; | |
2196 | r = -EFAULT; | |
893bdbf1 | 2197 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2198 | goto out; |
2199 | r = 0; | |
2200 | break; | |
2201 | } | |
2202 | case KVM_SET_MP_STATE: { | |
2203 | struct kvm_mp_state mp_state; | |
2204 | ||
2205 | r = -EFAULT; | |
893bdbf1 | 2206 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2207 | goto out; |
2208 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2209 | break; |
2210 | } | |
6aa8b732 AK |
2211 | case KVM_TRANSLATE: { |
2212 | struct kvm_translation tr; | |
2213 | ||
2214 | r = -EFAULT; | |
893bdbf1 | 2215 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2216 | goto out; |
8b006791 | 2217 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2218 | if (r) |
2219 | goto out; | |
2220 | r = -EFAULT; | |
893bdbf1 | 2221 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2222 | goto out; |
2223 | r = 0; | |
2224 | break; | |
2225 | } | |
d0bfb940 JK |
2226 | case KVM_SET_GUEST_DEBUG: { |
2227 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2228 | |
2229 | r = -EFAULT; | |
893bdbf1 | 2230 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2231 | goto out; |
d0bfb940 | 2232 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2233 | break; |
2234 | } | |
1961d276 AK |
2235 | case KVM_SET_SIGNAL_MASK: { |
2236 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2237 | struct kvm_signal_mask kvm_sigmask; | |
2238 | sigset_t sigset, *p; | |
2239 | ||
2240 | p = NULL; | |
2241 | if (argp) { | |
2242 | r = -EFAULT; | |
2243 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2244 | sizeof(kvm_sigmask))) |
1961d276 AK |
2245 | goto out; |
2246 | r = -EINVAL; | |
893bdbf1 | 2247 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2248 | goto out; |
2249 | r = -EFAULT; | |
2250 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2251 | sizeof(sigset))) |
1961d276 AK |
2252 | goto out; |
2253 | p = &sigset; | |
2254 | } | |
376d41ff | 2255 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2256 | break; |
2257 | } | |
b8836737 | 2258 | case KVM_GET_FPU: { |
fa3795a7 DH |
2259 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2260 | r = -ENOMEM; | |
2261 | if (!fpu) | |
2262 | goto out; | |
2263 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2264 | if (r) |
2265 | goto out; | |
2266 | r = -EFAULT; | |
fa3795a7 | 2267 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2268 | goto out; |
2269 | r = 0; | |
2270 | break; | |
2271 | } | |
2272 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2273 | fpu = memdup_user(argp, sizeof(*fpu)); |
2274 | if (IS_ERR(fpu)) { | |
2275 | r = PTR_ERR(fpu); | |
18595411 | 2276 | fpu = NULL; |
b8836737 | 2277 | goto out; |
ff5c2c03 | 2278 | } |
fa3795a7 | 2279 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2280 | break; |
2281 | } | |
bccf2150 | 2282 | default: |
313a3dc7 | 2283 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2284 | } |
2285 | out: | |
2122ff5e | 2286 | vcpu_put(vcpu); |
fa3795a7 DH |
2287 | kfree(fpu); |
2288 | kfree(kvm_sregs); | |
bccf2150 AK |
2289 | return r; |
2290 | } | |
2291 | ||
de8e5d74 | 2292 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2293 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2294 | unsigned int ioctl, unsigned long arg) | |
2295 | { | |
2296 | struct kvm_vcpu *vcpu = filp->private_data; | |
2297 | void __user *argp = compat_ptr(arg); | |
2298 | int r; | |
2299 | ||
2300 | if (vcpu->kvm->mm != current->mm) | |
2301 | return -EIO; | |
2302 | ||
2303 | switch (ioctl) { | |
2304 | case KVM_SET_SIGNAL_MASK: { | |
2305 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2306 | struct kvm_signal_mask kvm_sigmask; | |
2307 | compat_sigset_t csigset; | |
2308 | sigset_t sigset; | |
2309 | ||
2310 | if (argp) { | |
2311 | r = -EFAULT; | |
2312 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2313 | sizeof(kvm_sigmask))) |
1dda606c AG |
2314 | goto out; |
2315 | r = -EINVAL; | |
893bdbf1 | 2316 | if (kvm_sigmask.len != sizeof(csigset)) |
1dda606c AG |
2317 | goto out; |
2318 | r = -EFAULT; | |
2319 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
893bdbf1 | 2320 | sizeof(csigset))) |
1dda606c | 2321 | goto out; |
760a9a30 AC |
2322 | sigset_from_compat(&sigset, &csigset); |
2323 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2324 | } else | |
2325 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2326 | break; |
2327 | } | |
2328 | default: | |
2329 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2330 | } | |
2331 | ||
2332 | out: | |
2333 | return r; | |
2334 | } | |
2335 | #endif | |
2336 | ||
852b6d57 SW |
2337 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2338 | int (*accessor)(struct kvm_device *dev, | |
2339 | struct kvm_device_attr *attr), | |
2340 | unsigned long arg) | |
2341 | { | |
2342 | struct kvm_device_attr attr; | |
2343 | ||
2344 | if (!accessor) | |
2345 | return -EPERM; | |
2346 | ||
2347 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2348 | return -EFAULT; | |
2349 | ||
2350 | return accessor(dev, &attr); | |
2351 | } | |
2352 | ||
2353 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2354 | unsigned long arg) | |
2355 | { | |
2356 | struct kvm_device *dev = filp->private_data; | |
2357 | ||
2358 | switch (ioctl) { | |
2359 | case KVM_SET_DEVICE_ATTR: | |
2360 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2361 | case KVM_GET_DEVICE_ATTR: | |
2362 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2363 | case KVM_HAS_DEVICE_ATTR: | |
2364 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2365 | default: | |
2366 | if (dev->ops->ioctl) | |
2367 | return dev->ops->ioctl(dev, ioctl, arg); | |
2368 | ||
2369 | return -ENOTTY; | |
2370 | } | |
2371 | } | |
2372 | ||
852b6d57 SW |
2373 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2374 | { | |
2375 | struct kvm_device *dev = filp->private_data; | |
2376 | struct kvm *kvm = dev->kvm; | |
2377 | ||
852b6d57 SW |
2378 | kvm_put_kvm(kvm); |
2379 | return 0; | |
2380 | } | |
2381 | ||
2382 | static const struct file_operations kvm_device_fops = { | |
2383 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2384 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2385 | .compat_ioctl = kvm_device_ioctl, |
2386 | #endif | |
852b6d57 SW |
2387 | .release = kvm_device_release, |
2388 | }; | |
2389 | ||
2390 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2391 | { | |
2392 | if (filp->f_op != &kvm_device_fops) | |
2393 | return NULL; | |
2394 | ||
2395 | return filp->private_data; | |
2396 | } | |
2397 | ||
d60eacb0 | 2398 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2399 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2400 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2401 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2402 | #endif |
d60eacb0 | 2403 | |
5975a2e0 | 2404 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2405 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2406 | #endif |
d60eacb0 WD |
2407 | }; |
2408 | ||
2409 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2410 | { | |
2411 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2412 | return -ENOSPC; | |
2413 | ||
2414 | if (kvm_device_ops_table[type] != NULL) | |
2415 | return -EEXIST; | |
2416 | ||
2417 | kvm_device_ops_table[type] = ops; | |
2418 | return 0; | |
2419 | } | |
2420 | ||
571ee1b6 WL |
2421 | void kvm_unregister_device_ops(u32 type) |
2422 | { | |
2423 | if (kvm_device_ops_table[type] != NULL) | |
2424 | kvm_device_ops_table[type] = NULL; | |
2425 | } | |
2426 | ||
852b6d57 SW |
2427 | static int kvm_ioctl_create_device(struct kvm *kvm, |
2428 | struct kvm_create_device *cd) | |
2429 | { | |
2430 | struct kvm_device_ops *ops = NULL; | |
2431 | struct kvm_device *dev; | |
2432 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2433 | int ret; | |
2434 | ||
d60eacb0 WD |
2435 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
2436 | return -ENODEV; | |
2437 | ||
2438 | ops = kvm_device_ops_table[cd->type]; | |
2439 | if (ops == NULL) | |
852b6d57 | 2440 | return -ENODEV; |
852b6d57 SW |
2441 | |
2442 | if (test) | |
2443 | return 0; | |
2444 | ||
2445 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2446 | if (!dev) | |
2447 | return -ENOMEM; | |
2448 | ||
2449 | dev->ops = ops; | |
2450 | dev->kvm = kvm; | |
852b6d57 SW |
2451 | |
2452 | ret = ops->create(dev, cd->type); | |
2453 | if (ret < 0) { | |
2454 | kfree(dev); | |
2455 | return ret; | |
2456 | } | |
2457 | ||
24009b05 | 2458 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2459 | if (ret < 0) { |
2460 | ops->destroy(dev); | |
2461 | return ret; | |
2462 | } | |
2463 | ||
07f0a7bd | 2464 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2465 | kvm_get_kvm(kvm); |
2466 | cd->fd = ret; | |
2467 | return 0; | |
2468 | } | |
2469 | ||
92b591a4 AG |
2470 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2471 | { | |
2472 | switch (arg) { | |
2473 | case KVM_CAP_USER_MEMORY: | |
2474 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2475 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
2476 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE | |
2477 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2478 | #endif | |
2479 | case KVM_CAP_INTERNAL_ERROR_DATA: | |
2480 | #ifdef CONFIG_HAVE_KVM_MSI | |
2481 | case KVM_CAP_SIGNAL_MSI: | |
2482 | #endif | |
297e2105 | 2483 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 2484 | case KVM_CAP_IRQFD: |
92b591a4 AG |
2485 | case KVM_CAP_IRQFD_RESAMPLE: |
2486 | #endif | |
2487 | case KVM_CAP_CHECK_EXTENSION_VM: | |
2488 | return 1; | |
2489 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2490 | case KVM_CAP_IRQ_ROUTING: | |
2491 | return KVM_MAX_IRQ_ROUTES; | |
2492 | #endif | |
2493 | default: | |
2494 | break; | |
2495 | } | |
2496 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2497 | } | |
2498 | ||
bccf2150 AK |
2499 | static long kvm_vm_ioctl(struct file *filp, |
2500 | unsigned int ioctl, unsigned long arg) | |
2501 | { | |
2502 | struct kvm *kvm = filp->private_data; | |
2503 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2504 | int r; |
bccf2150 | 2505 | |
6d4e4c4f AK |
2506 | if (kvm->mm != current->mm) |
2507 | return -EIO; | |
bccf2150 AK |
2508 | switch (ioctl) { |
2509 | case KVM_CREATE_VCPU: | |
2510 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2511 | break; |
6fc138d2 IE |
2512 | case KVM_SET_USER_MEMORY_REGION: { |
2513 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2514 | ||
2515 | r = -EFAULT; | |
2516 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 2517 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
2518 | goto out; |
2519 | ||
47ae31e2 | 2520 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2521 | break; |
2522 | } | |
2523 | case KVM_GET_DIRTY_LOG: { | |
2524 | struct kvm_dirty_log log; | |
2525 | ||
2526 | r = -EFAULT; | |
893bdbf1 | 2527 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 2528 | goto out; |
2c6f5df9 | 2529 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2530 | break; |
2531 | } | |
5f94c174 LV |
2532 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2533 | case KVM_REGISTER_COALESCED_MMIO: { | |
2534 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2535 | |
5f94c174 | 2536 | r = -EFAULT; |
893bdbf1 | 2537 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2538 | goto out; |
5f94c174 | 2539 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2540 | break; |
2541 | } | |
2542 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2543 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2544 | |
5f94c174 | 2545 | r = -EFAULT; |
893bdbf1 | 2546 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2547 | goto out; |
5f94c174 | 2548 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2549 | break; |
2550 | } | |
2551 | #endif | |
721eecbf GH |
2552 | case KVM_IRQFD: { |
2553 | struct kvm_irqfd data; | |
2554 | ||
2555 | r = -EFAULT; | |
893bdbf1 | 2556 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 2557 | goto out; |
d4db2935 | 2558 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2559 | break; |
2560 | } | |
d34e6b17 GH |
2561 | case KVM_IOEVENTFD: { |
2562 | struct kvm_ioeventfd data; | |
2563 | ||
2564 | r = -EFAULT; | |
893bdbf1 | 2565 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
2566 | goto out; |
2567 | r = kvm_ioeventfd(kvm, &data); | |
2568 | break; | |
2569 | } | |
73880c80 GN |
2570 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2571 | case KVM_SET_BOOT_CPU_ID: | |
2572 | r = 0; | |
894a9c55 | 2573 | mutex_lock(&kvm->lock); |
73880c80 GN |
2574 | if (atomic_read(&kvm->online_vcpus) != 0) |
2575 | r = -EBUSY; | |
2576 | else | |
2577 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2578 | mutex_unlock(&kvm->lock); |
73880c80 | 2579 | break; |
07975ad3 JK |
2580 | #endif |
2581 | #ifdef CONFIG_HAVE_KVM_MSI | |
2582 | case KVM_SIGNAL_MSI: { | |
2583 | struct kvm_msi msi; | |
2584 | ||
2585 | r = -EFAULT; | |
893bdbf1 | 2586 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
2587 | goto out; |
2588 | r = kvm_send_userspace_msi(kvm, &msi); | |
2589 | break; | |
2590 | } | |
23d43cf9 CD |
2591 | #endif |
2592 | #ifdef __KVM_HAVE_IRQ_LINE | |
2593 | case KVM_IRQ_LINE_STATUS: | |
2594 | case KVM_IRQ_LINE: { | |
2595 | struct kvm_irq_level irq_event; | |
2596 | ||
2597 | r = -EFAULT; | |
893bdbf1 | 2598 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
2599 | goto out; |
2600 | ||
aa2fbe6d YZ |
2601 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2602 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2603 | if (r) |
2604 | goto out; | |
2605 | ||
2606 | r = -EFAULT; | |
2607 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 2608 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
2609 | goto out; |
2610 | } | |
2611 | ||
2612 | r = 0; | |
2613 | break; | |
2614 | } | |
73880c80 | 2615 | #endif |
aa8d5944 AG |
2616 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2617 | case KVM_SET_GSI_ROUTING: { | |
2618 | struct kvm_irq_routing routing; | |
2619 | struct kvm_irq_routing __user *urouting; | |
2620 | struct kvm_irq_routing_entry *entries; | |
2621 | ||
2622 | r = -EFAULT; | |
2623 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2624 | goto out; | |
2625 | r = -EINVAL; | |
2626 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2627 | goto out; | |
2628 | if (routing.flags) | |
2629 | goto out; | |
2630 | r = -ENOMEM; | |
2631 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2632 | if (!entries) | |
2633 | goto out; | |
2634 | r = -EFAULT; | |
2635 | urouting = argp; | |
2636 | if (copy_from_user(entries, urouting->entries, | |
2637 | routing.nr * sizeof(*entries))) | |
2638 | goto out_free_irq_routing; | |
2639 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2640 | routing.flags); | |
a642a175 | 2641 | out_free_irq_routing: |
aa8d5944 AG |
2642 | vfree(entries); |
2643 | break; | |
2644 | } | |
2645 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2646 | case KVM_CREATE_DEVICE: { |
2647 | struct kvm_create_device cd; | |
2648 | ||
2649 | r = -EFAULT; | |
2650 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2651 | goto out; | |
2652 | ||
2653 | r = kvm_ioctl_create_device(kvm, &cd); | |
2654 | if (r) | |
2655 | goto out; | |
2656 | ||
2657 | r = -EFAULT; | |
2658 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2659 | goto out; | |
2660 | ||
2661 | r = 0; | |
2662 | break; | |
2663 | } | |
92b591a4 AG |
2664 | case KVM_CHECK_EXTENSION: |
2665 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
2666 | break; | |
f17abe9a | 2667 | default: |
1fe779f8 | 2668 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
2669 | } |
2670 | out: | |
2671 | return r; | |
2672 | } | |
2673 | ||
de8e5d74 | 2674 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2675 | struct compat_kvm_dirty_log { |
2676 | __u32 slot; | |
2677 | __u32 padding1; | |
2678 | union { | |
2679 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2680 | __u64 padding2; | |
2681 | }; | |
2682 | }; | |
2683 | ||
2684 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2685 | unsigned int ioctl, unsigned long arg) | |
2686 | { | |
2687 | struct kvm *kvm = filp->private_data; | |
2688 | int r; | |
2689 | ||
2690 | if (kvm->mm != current->mm) | |
2691 | return -EIO; | |
2692 | switch (ioctl) { | |
2693 | case KVM_GET_DIRTY_LOG: { | |
2694 | struct compat_kvm_dirty_log compat_log; | |
2695 | struct kvm_dirty_log log; | |
2696 | ||
2697 | r = -EFAULT; | |
2698 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2699 | sizeof(compat_log))) | |
2700 | goto out; | |
2701 | log.slot = compat_log.slot; | |
2702 | log.padding1 = compat_log.padding1; | |
2703 | log.padding2 = compat_log.padding2; | |
2704 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2705 | ||
2706 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2707 | break; |
2708 | } | |
2709 | default: | |
2710 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2711 | } | |
2712 | ||
2713 | out: | |
2714 | return r; | |
2715 | } | |
2716 | #endif | |
2717 | ||
3d3aab1b | 2718 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2719 | .release = kvm_vm_release, |
2720 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 2721 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2722 | .compat_ioctl = kvm_vm_compat_ioctl, |
2723 | #endif | |
6038f373 | 2724 | .llseek = noop_llseek, |
f17abe9a AK |
2725 | }; |
2726 | ||
e08b9637 | 2727 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2728 | { |
aac87636 | 2729 | int r; |
f17abe9a AK |
2730 | struct kvm *kvm; |
2731 | ||
e08b9637 | 2732 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2733 | if (IS_ERR(kvm)) |
2734 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2735 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2736 | r = kvm_coalesced_mmio_init(kvm); | |
2737 | if (r < 0) { | |
2738 | kvm_put_kvm(kvm); | |
2739 | return r; | |
2740 | } | |
2741 | #endif | |
24009b05 | 2742 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2743 | if (r < 0) |
66c0b394 | 2744 | kvm_put_kvm(kvm); |
f17abe9a | 2745 | |
aac87636 | 2746 | return r; |
f17abe9a AK |
2747 | } |
2748 | ||
2749 | static long kvm_dev_ioctl(struct file *filp, | |
2750 | unsigned int ioctl, unsigned long arg) | |
2751 | { | |
07c45a36 | 2752 | long r = -EINVAL; |
f17abe9a AK |
2753 | |
2754 | switch (ioctl) { | |
2755 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2756 | if (arg) |
2757 | goto out; | |
f17abe9a AK |
2758 | r = KVM_API_VERSION; |
2759 | break; | |
2760 | case KVM_CREATE_VM: | |
e08b9637 | 2761 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2762 | break; |
018d00d2 | 2763 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 2764 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 2765 | break; |
07c45a36 | 2766 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
2767 | if (arg) |
2768 | goto out; | |
adb1ff46 AK |
2769 | r = PAGE_SIZE; /* struct kvm_run */ |
2770 | #ifdef CONFIG_X86 | |
2771 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2772 | #endif |
2773 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2774 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2775 | #endif |
07c45a36 | 2776 | break; |
d4c9ff2d FEL |
2777 | case KVM_TRACE_ENABLE: |
2778 | case KVM_TRACE_PAUSE: | |
2779 | case KVM_TRACE_DISABLE: | |
2023a29c | 2780 | r = -EOPNOTSUPP; |
d4c9ff2d | 2781 | break; |
6aa8b732 | 2782 | default: |
043405e1 | 2783 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2784 | } |
2785 | out: | |
2786 | return r; | |
2787 | } | |
2788 | ||
6aa8b732 | 2789 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2790 | .unlocked_ioctl = kvm_dev_ioctl, |
2791 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2792 | .llseek = noop_llseek, |
6aa8b732 AK |
2793 | }; |
2794 | ||
2795 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2796 | KVM_MINOR, |
6aa8b732 AK |
2797 | "kvm", |
2798 | &kvm_chardev_ops, | |
2799 | }; | |
2800 | ||
75b7127c | 2801 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2802 | { |
2803 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2804 | int r; |
1b6c0168 | 2805 | |
7f59f492 | 2806 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2807 | return; |
10474ae8 | 2808 | |
7f59f492 | 2809 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 2810 | |
13a34e06 | 2811 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
2812 | |
2813 | if (r) { | |
2814 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2815 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 2816 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 2817 | } |
1b6c0168 AK |
2818 | } |
2819 | ||
4fa92fb2 | 2820 | static void hardware_enable(void) |
75b7127c | 2821 | { |
4a937f96 | 2822 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2823 | if (kvm_usage_count) |
2824 | hardware_enable_nolock(NULL); | |
4a937f96 | 2825 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2826 | } |
2827 | ||
2828 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2829 | { |
2830 | int cpu = raw_smp_processor_id(); | |
2831 | ||
7f59f492 | 2832 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2833 | return; |
7f59f492 | 2834 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 2835 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
2836 | } |
2837 | ||
4fa92fb2 | 2838 | static void hardware_disable(void) |
75b7127c | 2839 | { |
4a937f96 | 2840 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2841 | if (kvm_usage_count) |
2842 | hardware_disable_nolock(NULL); | |
4a937f96 | 2843 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2844 | } |
2845 | ||
10474ae8 AG |
2846 | static void hardware_disable_all_nolock(void) |
2847 | { | |
2848 | BUG_ON(!kvm_usage_count); | |
2849 | ||
2850 | kvm_usage_count--; | |
2851 | if (!kvm_usage_count) | |
75b7127c | 2852 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2853 | } |
2854 | ||
2855 | static void hardware_disable_all(void) | |
2856 | { | |
4a937f96 | 2857 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 2858 | hardware_disable_all_nolock(); |
4a937f96 | 2859 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2860 | } |
2861 | ||
2862 | static int hardware_enable_all(void) | |
2863 | { | |
2864 | int r = 0; | |
2865 | ||
4a937f96 | 2866 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
2867 | |
2868 | kvm_usage_count++; | |
2869 | if (kvm_usage_count == 1) { | |
2870 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 2871 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
2872 | |
2873 | if (atomic_read(&hardware_enable_failed)) { | |
2874 | hardware_disable_all_nolock(); | |
2875 | r = -EBUSY; | |
2876 | } | |
2877 | } | |
2878 | ||
4a937f96 | 2879 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2880 | |
2881 | return r; | |
2882 | } | |
2883 | ||
774c47f1 AK |
2884 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
2885 | void *v) | |
2886 | { | |
1a6f4d7f | 2887 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 2888 | switch (val) { |
cec9ad27 | 2889 | case CPU_DYING: |
4fa92fb2 | 2890 | hardware_disable(); |
6ec8a856 | 2891 | break; |
da908f2f | 2892 | case CPU_STARTING: |
4fa92fb2 | 2893 | hardware_enable(); |
774c47f1 AK |
2894 | break; |
2895 | } | |
2896 | return NOTIFY_OK; | |
2897 | } | |
2898 | ||
9a2b85c6 | 2899 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 2900 | void *v) |
9a2b85c6 | 2901 | { |
8e1c1815 SY |
2902 | /* |
2903 | * Some (well, at least mine) BIOSes hang on reboot if | |
2904 | * in vmx root mode. | |
2905 | * | |
2906 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
2907 | */ | |
1170adc6 | 2908 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 2909 | kvm_rebooting = true; |
75b7127c | 2910 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
2911 | return NOTIFY_OK; |
2912 | } | |
2913 | ||
2914 | static struct notifier_block kvm_reboot_notifier = { | |
2915 | .notifier_call = kvm_reboot, | |
2916 | .priority = 0, | |
2917 | }; | |
2918 | ||
e93f8a0f | 2919 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
2920 | { |
2921 | int i; | |
2922 | ||
2923 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 2924 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
2925 | |
2926 | kvm_iodevice_destructor(pos); | |
2927 | } | |
e93f8a0f | 2928 | kfree(bus); |
2eeb2e94 GH |
2929 | } |
2930 | ||
c21fbff1 | 2931 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 2932 | const struct kvm_io_range *r2) |
743eeb0b | 2933 | { |
743eeb0b SL |
2934 | if (r1->addr < r2->addr) |
2935 | return -1; | |
2936 | if (r1->addr + r1->len > r2->addr + r2->len) | |
2937 | return 1; | |
2938 | return 0; | |
2939 | } | |
2940 | ||
a343c9b7 PB |
2941 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
2942 | { | |
c21fbff1 | 2943 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
2944 | } |
2945 | ||
39369f7a | 2946 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
2947 | gpa_t addr, int len) |
2948 | { | |
743eeb0b SL |
2949 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
2950 | .addr = addr, | |
2951 | .len = len, | |
2952 | .dev = dev, | |
2953 | }; | |
2954 | ||
2955 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
2956 | kvm_io_bus_sort_cmp, NULL); | |
2957 | ||
2958 | return 0; | |
2959 | } | |
2960 | ||
39369f7a | 2961 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
2962 | gpa_t addr, int len) |
2963 | { | |
2964 | struct kvm_io_range *range, key; | |
2965 | int off; | |
2966 | ||
2967 | key = (struct kvm_io_range) { | |
2968 | .addr = addr, | |
2969 | .len = len, | |
2970 | }; | |
2971 | ||
2972 | range = bsearch(&key, bus->range, bus->dev_count, | |
2973 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
2974 | if (range == NULL) | |
2975 | return -ENOENT; | |
2976 | ||
2977 | off = range - bus->range; | |
2978 | ||
c21fbff1 | 2979 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
2980 | off--; |
2981 | ||
2982 | return off; | |
2983 | } | |
2984 | ||
e32edf4f | 2985 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
2986 | struct kvm_io_range *range, const void *val) |
2987 | { | |
2988 | int idx; | |
2989 | ||
2990 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
2991 | if (idx < 0) | |
2992 | return -EOPNOTSUPP; | |
2993 | ||
2994 | while (idx < bus->dev_count && | |
c21fbff1 | 2995 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 2996 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
2997 | range->len, val)) |
2998 | return idx; | |
2999 | idx++; | |
3000 | } | |
3001 | ||
3002 | return -EOPNOTSUPP; | |
3003 | } | |
3004 | ||
bda9020e | 3005 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3006 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3007 | int len, const void *val) |
2eeb2e94 | 3008 | { |
90d83dc3 | 3009 | struct kvm_io_bus *bus; |
743eeb0b | 3010 | struct kvm_io_range range; |
126a5af5 | 3011 | int r; |
743eeb0b SL |
3012 | |
3013 | range = (struct kvm_io_range) { | |
3014 | .addr = addr, | |
3015 | .len = len, | |
3016 | }; | |
90d83dc3 | 3017 | |
e32edf4f NN |
3018 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3019 | r = __kvm_io_bus_write(vcpu, bus, &range, val); | |
126a5af5 CH |
3020 | return r < 0 ? r : 0; |
3021 | } | |
3022 | ||
3023 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3024 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3025 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3026 | { |
3027 | struct kvm_io_bus *bus; | |
3028 | struct kvm_io_range range; | |
3029 | ||
3030 | range = (struct kvm_io_range) { | |
3031 | .addr = addr, | |
3032 | .len = len, | |
3033 | }; | |
3034 | ||
e32edf4f | 3035 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
126a5af5 CH |
3036 | |
3037 | /* First try the device referenced by cookie. */ | |
3038 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3039 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3040 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3041 | val)) |
3042 | return cookie; | |
3043 | ||
3044 | /* | |
3045 | * cookie contained garbage; fall back to search and return the | |
3046 | * correct cookie value. | |
3047 | */ | |
e32edf4f | 3048 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3049 | } |
3050 | ||
e32edf4f NN |
3051 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3052 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3053 | { |
3054 | int idx; | |
3055 | ||
3056 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3057 | if (idx < 0) |
3058 | return -EOPNOTSUPP; | |
3059 | ||
3060 | while (idx < bus->dev_count && | |
c21fbff1 | 3061 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3062 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3063 | range->len, val)) |
3064 | return idx; | |
743eeb0b SL |
3065 | idx++; |
3066 | } | |
3067 | ||
bda9020e MT |
3068 | return -EOPNOTSUPP; |
3069 | } | |
68c3b4d1 | 3070 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3071 | |
bda9020e | 3072 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3073 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3074 | int len, void *val) |
bda9020e | 3075 | { |
90d83dc3 | 3076 | struct kvm_io_bus *bus; |
743eeb0b | 3077 | struct kvm_io_range range; |
126a5af5 | 3078 | int r; |
743eeb0b SL |
3079 | |
3080 | range = (struct kvm_io_range) { | |
3081 | .addr = addr, | |
3082 | .len = len, | |
3083 | }; | |
e93f8a0f | 3084 | |
e32edf4f NN |
3085 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3086 | r = __kvm_io_bus_read(vcpu, bus, &range, val); | |
126a5af5 CH |
3087 | return r < 0 ? r : 0; |
3088 | } | |
743eeb0b | 3089 | |
2eeb2e94 | 3090 | |
79fac95e | 3091 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3092 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3093 | int len, struct kvm_io_device *dev) | |
6c474694 | 3094 | { |
e93f8a0f | 3095 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3096 | |
e93f8a0f | 3097 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3098 | /* exclude ioeventfd which is limited by maximum fd */ |
3099 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3100 | return -ENOSPC; |
2eeb2e94 | 3101 | |
a1300716 AK |
3102 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
3103 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
3104 | if (!new_bus) |
3105 | return -ENOMEM; | |
a1300716 AK |
3106 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3107 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3108 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3109 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3110 | synchronize_srcu_expedited(&kvm->srcu); | |
3111 | kfree(bus); | |
090b7aff GH |
3112 | |
3113 | return 0; | |
3114 | } | |
3115 | ||
79fac95e | 3116 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3117 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3118 | struct kvm_io_device *dev) | |
090b7aff | 3119 | { |
e93f8a0f MT |
3120 | int i, r; |
3121 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3122 | |
cdfca7b3 | 3123 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 3124 | r = -ENOENT; |
a1300716 AK |
3125 | for (i = 0; i < bus->dev_count; i++) |
3126 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3127 | r = 0; |
090b7aff GH |
3128 | break; |
3129 | } | |
e93f8a0f | 3130 | |
a1300716 | 3131 | if (r) |
e93f8a0f | 3132 | return r; |
a1300716 AK |
3133 | |
3134 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
3135 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
3136 | if (!new_bus) | |
3137 | return -ENOMEM; | |
3138 | ||
3139 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3140 | new_bus->dev_count--; | |
3141 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3142 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3143 | |
3144 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3145 | synchronize_srcu_expedited(&kvm->srcu); | |
3146 | kfree(bus); | |
3147 | return r; | |
2eeb2e94 GH |
3148 | } |
3149 | ||
774c47f1 AK |
3150 | static struct notifier_block kvm_cpu_notifier = { |
3151 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3152 | }; |
3153 | ||
8b88b099 | 3154 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3155 | { |
3156 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3157 | struct kvm *kvm; |
3158 | ||
8b88b099 | 3159 | *val = 0; |
2f303b74 | 3160 | spin_lock(&kvm_lock); |
ba1389b7 | 3161 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3162 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3163 | spin_unlock(&kvm_lock); |
8b88b099 | 3164 | return 0; |
ba1389b7 AK |
3165 | } |
3166 | ||
3167 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3168 | ||
8b88b099 | 3169 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3170 | { |
3171 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3172 | struct kvm *kvm; |
3173 | struct kvm_vcpu *vcpu; | |
3174 | int i; | |
3175 | ||
8b88b099 | 3176 | *val = 0; |
2f303b74 | 3177 | spin_lock(&kvm_lock); |
1165f5fe | 3178 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3179 | kvm_for_each_vcpu(i, vcpu, kvm) |
3180 | *val += *(u32 *)((void *)vcpu + offset); | |
3181 | ||
2f303b74 | 3182 | spin_unlock(&kvm_lock); |
8b88b099 | 3183 | return 0; |
1165f5fe AK |
3184 | } |
3185 | ||
ba1389b7 AK |
3186 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3187 | ||
828c0950 | 3188 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3189 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3190 | [KVM_STAT_VM] = &vm_stat_fops, | |
3191 | }; | |
1165f5fe | 3192 | |
4f69b680 | 3193 | static int kvm_init_debug(void) |
6aa8b732 | 3194 | { |
0c8eb04a | 3195 | int r = -EEXIST; |
6aa8b732 AK |
3196 | struct kvm_stats_debugfs_item *p; |
3197 | ||
76f7c879 | 3198 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3199 | if (kvm_debugfs_dir == NULL) |
3200 | goto out; | |
3201 | ||
3202 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3203 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3204 | (void *)(long)p->offset, |
ba1389b7 | 3205 | stat_fops[p->kind]); |
4f69b680 H |
3206 | if (p->dentry == NULL) |
3207 | goto out_dir; | |
3208 | } | |
3209 | ||
3210 | return 0; | |
3211 | ||
3212 | out_dir: | |
3213 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3214 | out: | |
3215 | return r; | |
6aa8b732 AK |
3216 | } |
3217 | ||
3218 | static void kvm_exit_debug(void) | |
3219 | { | |
3220 | struct kvm_stats_debugfs_item *p; | |
3221 | ||
3222 | for (p = debugfs_entries; p->name; ++p) | |
3223 | debugfs_remove(p->dentry); | |
76f7c879 | 3224 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3225 | } |
3226 | ||
fb3600cc | 3227 | static int kvm_suspend(void) |
59ae6c6b | 3228 | { |
10474ae8 | 3229 | if (kvm_usage_count) |
75b7127c | 3230 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3231 | return 0; |
3232 | } | |
3233 | ||
fb3600cc | 3234 | static void kvm_resume(void) |
59ae6c6b | 3235 | { |
ca84d1a2 | 3236 | if (kvm_usage_count) { |
4a937f96 | 3237 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3238 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3239 | } |
59ae6c6b AK |
3240 | } |
3241 | ||
fb3600cc | 3242 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3243 | .suspend = kvm_suspend, |
3244 | .resume = kvm_resume, | |
3245 | }; | |
3246 | ||
15ad7146 AK |
3247 | static inline |
3248 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3249 | { | |
3250 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3251 | } | |
3252 | ||
3253 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3254 | { | |
3255 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 3256 | |
3a08a8f9 R |
3257 | if (vcpu->preempted) |
3258 | vcpu->preempted = false; | |
15ad7146 | 3259 | |
e790d9ef RK |
3260 | kvm_arch_sched_in(vcpu, cpu); |
3261 | ||
e9b11c17 | 3262 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3263 | } |
3264 | ||
3265 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3266 | struct task_struct *next) | |
3267 | { | |
3268 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3269 | ||
3a08a8f9 R |
3270 | if (current->state == TASK_RUNNING) |
3271 | vcpu->preempted = true; | |
e9b11c17 | 3272 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3273 | } |
3274 | ||
0ee75bea | 3275 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3276 | struct module *module) |
6aa8b732 AK |
3277 | { |
3278 | int r; | |
002c7f7c | 3279 | int cpu; |
6aa8b732 | 3280 | |
f8c16bba ZX |
3281 | r = kvm_arch_init(opaque); |
3282 | if (r) | |
d2308784 | 3283 | goto out_fail; |
cb498ea2 | 3284 | |
7dac16c3 AH |
3285 | /* |
3286 | * kvm_arch_init makes sure there's at most one caller | |
3287 | * for architectures that support multiple implementations, | |
3288 | * like intel and amd on x86. | |
3289 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3290 | * conflicts in case kvm is already setup for another implementation. | |
3291 | */ | |
3292 | r = kvm_irqfd_init(); | |
3293 | if (r) | |
3294 | goto out_irqfd; | |
3295 | ||
8437a617 | 3296 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3297 | r = -ENOMEM; |
3298 | goto out_free_0; | |
3299 | } | |
3300 | ||
e9b11c17 | 3301 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3302 | if (r < 0) |
7f59f492 | 3303 | goto out_free_0a; |
6aa8b732 | 3304 | |
002c7f7c YS |
3305 | for_each_online_cpu(cpu) { |
3306 | smp_call_function_single(cpu, | |
e9b11c17 | 3307 | kvm_arch_check_processor_compat, |
8691e5a8 | 3308 | &r, 1); |
002c7f7c | 3309 | if (r < 0) |
d2308784 | 3310 | goto out_free_1; |
002c7f7c YS |
3311 | } |
3312 | ||
774c47f1 AK |
3313 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3314 | if (r) | |
d2308784 | 3315 | goto out_free_2; |
6aa8b732 AK |
3316 | register_reboot_notifier(&kvm_reboot_notifier); |
3317 | ||
c16f862d | 3318 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3319 | if (!vcpu_align) |
3320 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3321 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3322 | 0, NULL); |
c16f862d RR |
3323 | if (!kvm_vcpu_cache) { |
3324 | r = -ENOMEM; | |
fb3600cc | 3325 | goto out_free_3; |
c16f862d RR |
3326 | } |
3327 | ||
af585b92 GN |
3328 | r = kvm_async_pf_init(); |
3329 | if (r) | |
3330 | goto out_free; | |
3331 | ||
6aa8b732 | 3332 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3333 | kvm_vm_fops.owner = module; |
3334 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3335 | |
3336 | r = misc_register(&kvm_dev); | |
3337 | if (r) { | |
1170adc6 | 3338 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 3339 | goto out_unreg; |
6aa8b732 AK |
3340 | } |
3341 | ||
fb3600cc RW |
3342 | register_syscore_ops(&kvm_syscore_ops); |
3343 | ||
15ad7146 AK |
3344 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3345 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3346 | ||
4f69b680 H |
3347 | r = kvm_init_debug(); |
3348 | if (r) { | |
1170adc6 | 3349 | pr_err("kvm: create debugfs files failed\n"); |
4f69b680 H |
3350 | goto out_undebugfs; |
3351 | } | |
0ea4ed8e | 3352 | |
3c3c29fd PB |
3353 | r = kvm_vfio_ops_init(); |
3354 | WARN_ON(r); | |
3355 | ||
c7addb90 | 3356 | return 0; |
6aa8b732 | 3357 | |
4f69b680 H |
3358 | out_undebugfs: |
3359 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3360 | misc_deregister(&kvm_dev); |
af585b92 GN |
3361 | out_unreg: |
3362 | kvm_async_pf_deinit(); | |
6aa8b732 | 3363 | out_free: |
c16f862d | 3364 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3365 | out_free_3: |
6aa8b732 | 3366 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3367 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3368 | out_free_2: |
d2308784 | 3369 | out_free_1: |
e9b11c17 | 3370 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3371 | out_free_0a: |
3372 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3373 | out_free_0: |
a0f155e9 CH |
3374 | kvm_irqfd_exit(); |
3375 | out_irqfd: | |
7dac16c3 AH |
3376 | kvm_arch_exit(); |
3377 | out_fail: | |
6aa8b732 AK |
3378 | return r; |
3379 | } | |
cb498ea2 | 3380 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3381 | |
cb498ea2 | 3382 | void kvm_exit(void) |
6aa8b732 | 3383 | { |
0ea4ed8e | 3384 | kvm_exit_debug(); |
6aa8b732 | 3385 | misc_deregister(&kvm_dev); |
c16f862d | 3386 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3387 | kvm_async_pf_deinit(); |
fb3600cc | 3388 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3389 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3390 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3391 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3392 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3393 | kvm_arch_exit(); |
a0f155e9 | 3394 | kvm_irqfd_exit(); |
7f59f492 | 3395 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 3396 | kvm_vfio_ops_exit(); |
6aa8b732 | 3397 | } |
cb498ea2 | 3398 | EXPORT_SYMBOL_GPL(kvm_exit); |