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); | |
bc009e43 | 106 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 107 | |
52480137 | 108 | __visible bool kvm_rebooting; |
b7c4145b | 109 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 110 | |
54dee993 MT |
111 | static bool largepages_enabled = true; |
112 | ||
bf4bea8e | 113 | bool kvm_is_reserved_pfn(pfn_t pfn) |
cbff90a7 | 114 | { |
11feeb49 | 115 | if (pfn_valid(pfn)) |
bf4bea8e | 116 | return PageReserved(pfn_to_page(pfn)); |
cbff90a7 BAY |
117 | |
118 | return true; | |
119 | } | |
120 | ||
bccf2150 AK |
121 | /* |
122 | * Switches to specified vcpu, until a matching vcpu_put() | |
123 | */ | |
9fc77441 | 124 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 125 | { |
15ad7146 AK |
126 | int cpu; |
127 | ||
9fc77441 MT |
128 | if (mutex_lock_killable(&vcpu->mutex)) |
129 | return -EINTR; | |
15ad7146 AK |
130 | cpu = get_cpu(); |
131 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 132 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 133 | put_cpu(); |
9fc77441 | 134 | return 0; |
6aa8b732 AK |
135 | } |
136 | ||
313a3dc7 | 137 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 138 | { |
15ad7146 | 139 | preempt_disable(); |
313a3dc7 | 140 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
141 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
142 | preempt_enable(); | |
6aa8b732 AK |
143 | mutex_unlock(&vcpu->mutex); |
144 | } | |
145 | ||
d9e368d6 AK |
146 | static void ack_flush(void *_completed) |
147 | { | |
d9e368d6 AK |
148 | } |
149 | ||
445b8236 | 150 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 151 | { |
597a5f55 | 152 | int i, cpu, me; |
6ef7a1bc RR |
153 | cpumask_var_t cpus; |
154 | bool called = true; | |
d9e368d6 | 155 | struct kvm_vcpu *vcpu; |
d9e368d6 | 156 | |
79f55997 | 157 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 158 | |
3cba4130 | 159 | me = get_cpu(); |
988a2cae | 160 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 161 | kvm_make_request(req, vcpu); |
d9e368d6 | 162 | cpu = vcpu->cpu; |
6b7e2d09 XG |
163 | |
164 | /* Set ->requests bit before we read ->mode */ | |
165 | smp_mb(); | |
166 | ||
167 | if (cpus != NULL && cpu != -1 && cpu != me && | |
168 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 169 | cpumask_set_cpu(cpu, cpus); |
49846896 | 170 | } |
6ef7a1bc RR |
171 | if (unlikely(cpus == NULL)) |
172 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
173 | else if (!cpumask_empty(cpus)) | |
174 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
175 | else | |
176 | called = false; | |
3cba4130 | 177 | put_cpu(); |
6ef7a1bc | 178 | free_cpumask_var(cpus); |
49846896 | 179 | return called; |
d9e368d6 AK |
180 | } |
181 | ||
a6d51016 | 182 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 183 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 184 | { |
a086f6a1 XG |
185 | long dirty_count = kvm->tlbs_dirty; |
186 | ||
187 | smp_mb(); | |
445b8236 | 188 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
49846896 | 189 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 190 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 191 | } |
2ba9f0d8 | 192 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 193 | #endif |
2e53d63a | 194 | |
49846896 RR |
195 | void kvm_reload_remote_mmus(struct kvm *kvm) |
196 | { | |
445b8236 | 197 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 198 | } |
2e53d63a | 199 | |
d828199e MT |
200 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
201 | { | |
445b8236 | 202 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); |
d828199e MT |
203 | } |
204 | ||
3d81bc7e | 205 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
c7c9c56c | 206 | { |
445b8236 | 207 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); |
c7c9c56c YZ |
208 | } |
209 | ||
fb3f0f51 RR |
210 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
211 | { | |
212 | struct page *page; | |
213 | int r; | |
214 | ||
215 | mutex_init(&vcpu->mutex); | |
216 | vcpu->cpu = -1; | |
fb3f0f51 RR |
217 | vcpu->kvm = kvm; |
218 | vcpu->vcpu_id = id; | |
34bb10b7 | 219 | vcpu->pid = NULL; |
b6958ce4 | 220 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 221 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
222 | |
223 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
224 | if (!page) { | |
225 | r = -ENOMEM; | |
226 | goto fail; | |
227 | } | |
228 | vcpu->run = page_address(page); | |
229 | ||
4c088493 R |
230 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
231 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 232 | vcpu->preempted = false; |
4c088493 | 233 | |
e9b11c17 | 234 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 235 | if (r < 0) |
e9b11c17 | 236 | goto fail_free_run; |
fb3f0f51 RR |
237 | return 0; |
238 | ||
fb3f0f51 RR |
239 | fail_free_run: |
240 | free_page((unsigned long)vcpu->run); | |
241 | fail: | |
76fafa5e | 242 | return r; |
fb3f0f51 RR |
243 | } |
244 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
245 | ||
246 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
247 | { | |
34bb10b7 | 248 | put_pid(vcpu->pid); |
e9b11c17 | 249 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
250 | free_page((unsigned long)vcpu->run); |
251 | } | |
252 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
253 | ||
e930bffe AA |
254 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
255 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
256 | { | |
257 | return container_of(mn, struct kvm, mmu_notifier); | |
258 | } | |
259 | ||
260 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
261 | struct mm_struct *mm, | |
262 | unsigned long address) | |
263 | { | |
264 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 265 | int need_tlb_flush, idx; |
e930bffe AA |
266 | |
267 | /* | |
268 | * When ->invalidate_page runs, the linux pte has been zapped | |
269 | * already but the page is still allocated until | |
270 | * ->invalidate_page returns. So if we increase the sequence | |
271 | * here the kvm page fault will notice if the spte can't be | |
272 | * established because the page is going to be freed. If | |
273 | * instead the kvm page fault establishes the spte before | |
274 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
275 | * before returning. | |
276 | * | |
277 | * The sequence increase only need to be seen at spin_unlock | |
278 | * time, and not at spin_lock time. | |
279 | * | |
280 | * Increasing the sequence after the spin_unlock would be | |
281 | * unsafe because the kvm page fault could then establish the | |
282 | * pte after kvm_unmap_hva returned, without noticing the page | |
283 | * is going to be freed. | |
284 | */ | |
bc6678a3 | 285 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 286 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 287 | |
e930bffe | 288 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 289 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
290 | /* we've to flush the tlb before the pages can be freed */ |
291 | if (need_tlb_flush) | |
292 | kvm_flush_remote_tlbs(kvm); | |
293 | ||
565f3be2 | 294 | spin_unlock(&kvm->mmu_lock); |
fe71557a TC |
295 | |
296 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); | |
297 | ||
565f3be2 | 298 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
299 | } |
300 | ||
3da0dd43 IE |
301 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
302 | struct mm_struct *mm, | |
303 | unsigned long address, | |
304 | pte_t pte) | |
305 | { | |
306 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 307 | int idx; |
3da0dd43 | 308 | |
bc6678a3 | 309 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
310 | spin_lock(&kvm->mmu_lock); |
311 | kvm->mmu_notifier_seq++; | |
312 | kvm_set_spte_hva(kvm, address, pte); | |
313 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 314 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
315 | } |
316 | ||
e930bffe AA |
317 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
318 | struct mm_struct *mm, | |
319 | unsigned long start, | |
320 | unsigned long end) | |
321 | { | |
322 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 323 | int need_tlb_flush = 0, idx; |
e930bffe | 324 | |
bc6678a3 | 325 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
326 | spin_lock(&kvm->mmu_lock); |
327 | /* | |
328 | * The count increase must become visible at unlock time as no | |
329 | * spte can be established without taking the mmu_lock and | |
330 | * count is also read inside the mmu_lock critical section. | |
331 | */ | |
332 | kvm->mmu_notifier_count++; | |
b3ae2096 | 333 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 334 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
335 | /* we've to flush the tlb before the pages can be freed */ |
336 | if (need_tlb_flush) | |
337 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
338 | |
339 | spin_unlock(&kvm->mmu_lock); | |
340 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
341 | } |
342 | ||
343 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
344 | struct mm_struct *mm, | |
345 | unsigned long start, | |
346 | unsigned long end) | |
347 | { | |
348 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
349 | ||
350 | spin_lock(&kvm->mmu_lock); | |
351 | /* | |
352 | * This sequence increase will notify the kvm page fault that | |
353 | * the page that is going to be mapped in the spte could have | |
354 | * been freed. | |
355 | */ | |
356 | kvm->mmu_notifier_seq++; | |
a355aa54 | 357 | smp_wmb(); |
e930bffe AA |
358 | /* |
359 | * The above sequence increase must be visible before the | |
a355aa54 PM |
360 | * below count decrease, which is ensured by the smp_wmb above |
361 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
362 | */ |
363 | kvm->mmu_notifier_count--; | |
364 | spin_unlock(&kvm->mmu_lock); | |
365 | ||
366 | BUG_ON(kvm->mmu_notifier_count < 0); | |
367 | } | |
368 | ||
369 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
370 | struct mm_struct *mm, | |
57128468 ALC |
371 | unsigned long start, |
372 | unsigned long end) | |
e930bffe AA |
373 | { |
374 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 375 | int young, idx; |
e930bffe | 376 | |
bc6678a3 | 377 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 378 | spin_lock(&kvm->mmu_lock); |
e930bffe | 379 | |
57128468 | 380 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
381 | if (young) |
382 | kvm_flush_remote_tlbs(kvm); | |
383 | ||
565f3be2 TY |
384 | spin_unlock(&kvm->mmu_lock); |
385 | srcu_read_unlock(&kvm->srcu, idx); | |
386 | ||
e930bffe AA |
387 | return young; |
388 | } | |
389 | ||
8ee53820 AA |
390 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
391 | struct mm_struct *mm, | |
392 | unsigned long address) | |
393 | { | |
394 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
395 | int young, idx; | |
396 | ||
397 | idx = srcu_read_lock(&kvm->srcu); | |
398 | spin_lock(&kvm->mmu_lock); | |
399 | young = kvm_test_age_hva(kvm, address); | |
400 | spin_unlock(&kvm->mmu_lock); | |
401 | srcu_read_unlock(&kvm->srcu, idx); | |
402 | ||
403 | return young; | |
404 | } | |
405 | ||
85db06e5 MT |
406 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
407 | struct mm_struct *mm) | |
408 | { | |
409 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
410 | int idx; |
411 | ||
412 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 413 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 414 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
415 | } |
416 | ||
e930bffe AA |
417 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
418 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
419 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
420 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
421 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 422 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 423 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 424 | .release = kvm_mmu_notifier_release, |
e930bffe | 425 | }; |
4c07b0a4 AK |
426 | |
427 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
428 | { | |
429 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
430 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
431 | } | |
432 | ||
433 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
434 | ||
435 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
436 | { | |
437 | return 0; | |
438 | } | |
439 | ||
e930bffe AA |
440 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
441 | ||
a47d2b07 | 442 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
443 | { |
444 | int i; | |
a47d2b07 | 445 | struct kvm_memslots *slots; |
bf3e05bc | 446 | |
a47d2b07 PB |
447 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
448 | if (!slots) | |
449 | return NULL; | |
450 | ||
451 | /* | |
452 | * Init kvm generation close to the maximum to easily test the | |
453 | * code of handling generation number wrap-around. | |
454 | */ | |
455 | slots->generation = -150; | |
bf3e05bc | 456 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 457 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
458 | |
459 | return slots; | |
460 | } | |
461 | ||
462 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
463 | { | |
464 | if (!memslot->dirty_bitmap) | |
465 | return; | |
466 | ||
467 | kvfree(memslot->dirty_bitmap); | |
468 | memslot->dirty_bitmap = NULL; | |
469 | } | |
470 | ||
471 | /* | |
472 | * Free any memory in @free but not in @dont. | |
473 | */ | |
474 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
475 | struct kvm_memory_slot *dont) | |
476 | { | |
477 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
478 | kvm_destroy_dirty_bitmap(free); | |
479 | ||
480 | kvm_arch_free_memslot(kvm, free, dont); | |
481 | ||
482 | free->npages = 0; | |
483 | } | |
484 | ||
485 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
486 | { | |
487 | struct kvm_memory_slot *memslot; | |
488 | ||
489 | if (!slots) | |
490 | return; | |
491 | ||
492 | kvm_for_each_memslot(memslot, slots) | |
493 | kvm_free_memslot(kvm, memslot, NULL); | |
494 | ||
495 | kvfree(slots); | |
bf3e05bc XG |
496 | } |
497 | ||
e08b9637 | 498 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 499 | { |
d89f5eff JK |
500 | int r, i; |
501 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 502 | |
d89f5eff JK |
503 | if (!kvm) |
504 | return ERR_PTR(-ENOMEM); | |
505 | ||
e08b9637 | 506 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 507 | if (r) |
719d93cd | 508 | goto out_err_no_disable; |
10474ae8 AG |
509 | |
510 | r = hardware_enable_all(); | |
511 | if (r) | |
719d93cd | 512 | goto out_err_no_disable; |
10474ae8 | 513 | |
c77dcacb | 514 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 515 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 516 | #endif |
6aa8b732 | 517 | |
1e702d9a AW |
518 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
519 | ||
46a26bf5 | 520 | r = -ENOMEM; |
f481b069 PB |
521 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
522 | kvm->memslots[i] = kvm_alloc_memslots(); | |
523 | if (!kvm->memslots[i]) | |
524 | goto out_err_no_srcu; | |
525 | } | |
00f034a1 | 526 | |
bc6678a3 | 527 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
528 | goto out_err_no_srcu; |
529 | if (init_srcu_struct(&kvm->irq_srcu)) | |
530 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
531 | for (i = 0; i < KVM_NR_BUSES; i++) { |
532 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
533 | GFP_KERNEL); | |
57e7fbee | 534 | if (!kvm->buses[i]) |
e93f8a0f | 535 | goto out_err; |
e93f8a0f | 536 | } |
e930bffe | 537 | |
74b5c5bf | 538 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
539 | kvm->mm = current->mm; |
540 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 541 | kvm_eventfd_init(kvm); |
11ec2804 | 542 | mutex_init(&kvm->lock); |
60eead79 | 543 | mutex_init(&kvm->irq_lock); |
79fac95e | 544 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 545 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 546 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
547 | |
548 | r = kvm_init_mmu_notifier(kvm); | |
549 | if (r) | |
550 | goto out_err; | |
551 | ||
2f303b74 | 552 | spin_lock(&kvm_lock); |
5e58cfe4 | 553 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 554 | spin_unlock(&kvm_lock); |
d89f5eff | 555 | |
f17abe9a | 556 | return kvm; |
10474ae8 AG |
557 | |
558 | out_err: | |
719d93cd CB |
559 | cleanup_srcu_struct(&kvm->irq_srcu); |
560 | out_err_no_irq_srcu: | |
57e7fbee | 561 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 562 | out_err_no_srcu: |
10474ae8 | 563 | hardware_disable_all(); |
719d93cd | 564 | out_err_no_disable: |
e93f8a0f MT |
565 | for (i = 0; i < KVM_NR_BUSES; i++) |
566 | kfree(kvm->buses[i]); | |
f481b069 PB |
567 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
568 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
d89f5eff | 569 | kvm_arch_free_vm(kvm); |
10474ae8 | 570 | return ERR_PTR(r); |
f17abe9a AK |
571 | } |
572 | ||
92eca8fa TY |
573 | /* |
574 | * Avoid using vmalloc for a small buffer. | |
575 | * Should not be used when the size is statically known. | |
576 | */ | |
c1a7b32a | 577 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
578 | { |
579 | if (size > PAGE_SIZE) | |
580 | return vzalloc(size); | |
581 | else | |
582 | return kzalloc(size, GFP_KERNEL); | |
583 | } | |
584 | ||
07f0a7bd SW |
585 | static void kvm_destroy_devices(struct kvm *kvm) |
586 | { | |
587 | struct list_head *node, *tmp; | |
588 | ||
589 | list_for_each_safe(node, tmp, &kvm->devices) { | |
590 | struct kvm_device *dev = | |
591 | list_entry(node, struct kvm_device, vm_node); | |
592 | ||
593 | list_del(node); | |
594 | dev->ops->destroy(dev); | |
595 | } | |
596 | } | |
597 | ||
f17abe9a AK |
598 | static void kvm_destroy_vm(struct kvm *kvm) |
599 | { | |
e93f8a0f | 600 | int i; |
6d4e4c4f AK |
601 | struct mm_struct *mm = kvm->mm; |
602 | ||
ad8ba2cd | 603 | kvm_arch_sync_events(kvm); |
2f303b74 | 604 | spin_lock(&kvm_lock); |
133de902 | 605 | list_del(&kvm->vm_list); |
2f303b74 | 606 | spin_unlock(&kvm_lock); |
399ec807 | 607 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
608 | for (i = 0; i < KVM_NR_BUSES; i++) |
609 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 610 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
611 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
612 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 613 | #else |
2df72e9b | 614 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 615 | #endif |
d19a9cd2 | 616 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 617 | kvm_destroy_devices(kvm); |
f481b069 PB |
618 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
619 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
820b3fcd | 620 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
621 | cleanup_srcu_struct(&kvm->srcu); |
622 | kvm_arch_free_vm(kvm); | |
10474ae8 | 623 | hardware_disable_all(); |
6d4e4c4f | 624 | mmdrop(mm); |
f17abe9a AK |
625 | } |
626 | ||
d39f13b0 IE |
627 | void kvm_get_kvm(struct kvm *kvm) |
628 | { | |
629 | atomic_inc(&kvm->users_count); | |
630 | } | |
631 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
632 | ||
633 | void kvm_put_kvm(struct kvm *kvm) | |
634 | { | |
635 | if (atomic_dec_and_test(&kvm->users_count)) | |
636 | kvm_destroy_vm(kvm); | |
637 | } | |
638 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
639 | ||
640 | ||
f17abe9a AK |
641 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
642 | { | |
643 | struct kvm *kvm = filp->private_data; | |
644 | ||
721eecbf GH |
645 | kvm_irqfd_release(kvm); |
646 | ||
d39f13b0 | 647 | kvm_put_kvm(kvm); |
6aa8b732 AK |
648 | return 0; |
649 | } | |
650 | ||
515a0127 TY |
651 | /* |
652 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 653 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 654 | */ |
a36a57b1 TY |
655 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
656 | { | |
515a0127 | 657 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 658 | |
92eca8fa | 659 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
660 | if (!memslot->dirty_bitmap) |
661 | return -ENOMEM; | |
662 | ||
a36a57b1 TY |
663 | return 0; |
664 | } | |
665 | ||
bf3e05bc | 666 | /* |
0e60b079 IM |
667 | * Insert memslot and re-sort memslots based on their GFN, |
668 | * so binary search could be used to lookup GFN. | |
669 | * Sorting algorithm takes advantage of having initially | |
670 | * sorted array and known changed memslot position. | |
bf3e05bc | 671 | */ |
5cc15027 PB |
672 | static void update_memslots(struct kvm_memslots *slots, |
673 | struct kvm_memory_slot *new) | |
bf3e05bc | 674 | { |
8593176c PB |
675 | int id = new->id; |
676 | int i = slots->id_to_index[id]; | |
063584d4 | 677 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 678 | |
8593176c | 679 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 680 | if (!new->npages) { |
dbaff309 | 681 | WARN_ON(!mslots[i].npages); |
9c1a5d38 IM |
682 | if (mslots[i].npages) |
683 | slots->used_slots--; | |
684 | } else { | |
685 | if (!mslots[i].npages) | |
686 | slots->used_slots++; | |
687 | } | |
0e60b079 | 688 | |
7f379cff | 689 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
690 | new->base_gfn <= mslots[i + 1].base_gfn) { |
691 | if (!mslots[i + 1].npages) | |
692 | break; | |
7f379cff IM |
693 | mslots[i] = mslots[i + 1]; |
694 | slots->id_to_index[mslots[i].id] = i; | |
695 | i++; | |
696 | } | |
efbeec70 PB |
697 | |
698 | /* | |
699 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
700 | * so that it moves before all those with base_gfn == npages == 0. | |
701 | * | |
702 | * On the other hand, if new->npages is zero, the above loop has | |
703 | * already left i pointing to the beginning of the empty part of | |
704 | * mslots, and the ">=" would move the hole backwards in this | |
705 | * case---which is wrong. So skip the loop when deleting a slot. | |
706 | */ | |
707 | if (new->npages) { | |
708 | while (i > 0 && | |
709 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
710 | mslots[i] = mslots[i - 1]; | |
711 | slots->id_to_index[mslots[i].id] = i; | |
712 | i--; | |
713 | } | |
dbaff309 PB |
714 | } else |
715 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 716 | |
8593176c PB |
717 | mslots[i] = *new; |
718 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
719 | } |
720 | ||
09170a49 | 721 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 722 | { |
4d8b81ab XG |
723 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
724 | ||
0f8a4de3 | 725 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
726 | valid_flags |= KVM_MEM_READONLY; |
727 | #endif | |
728 | ||
729 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
730 | return -EINVAL; |
731 | ||
732 | return 0; | |
733 | } | |
734 | ||
7ec4fb44 | 735 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 736 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 737 | { |
f481b069 | 738 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
7ec4fb44 | 739 | |
ee3d1570 DM |
740 | /* |
741 | * Set the low bit in the generation, which disables SPTE caching | |
742 | * until the end of synchronize_srcu_expedited. | |
743 | */ | |
744 | WARN_ON(old_memslots->generation & 1); | |
745 | slots->generation = old_memslots->generation + 1; | |
746 | ||
f481b069 | 747 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 748 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 749 | |
ee3d1570 DM |
750 | /* |
751 | * Increment the new memslot generation a second time. This prevents | |
752 | * vm exits that race with memslot updates from caching a memslot | |
753 | * generation that will (potentially) be valid forever. | |
754 | */ | |
755 | slots->generation++; | |
756 | ||
15f46015 | 757 | kvm_arch_memslots_updated(kvm, slots); |
e59dbe09 TY |
758 | |
759 | return old_memslots; | |
7ec4fb44 GN |
760 | } |
761 | ||
6aa8b732 AK |
762 | /* |
763 | * Allocate some memory and give it an address in the guest physical address | |
764 | * space. | |
765 | * | |
766 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 767 | * |
02d5d55b | 768 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 769 | */ |
f78e0e2e | 770 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 771 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 772 | { |
8234b22e | 773 | int r; |
6aa8b732 | 774 | gfn_t base_gfn; |
28bcb112 | 775 | unsigned long npages; |
a843fac2 | 776 | struct kvm_memory_slot *slot; |
6aa8b732 | 777 | struct kvm_memory_slot old, new; |
b7f69c55 | 778 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 779 | int as_id, id; |
f64c0398 | 780 | enum kvm_mr_change change; |
6aa8b732 | 781 | |
a50d64d6 XG |
782 | r = check_memory_region_flags(mem); |
783 | if (r) | |
784 | goto out; | |
785 | ||
6aa8b732 | 786 | r = -EINVAL; |
f481b069 PB |
787 | as_id = mem->slot >> 16; |
788 | id = (u16)mem->slot; | |
789 | ||
6aa8b732 AK |
790 | /* General sanity checks */ |
791 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
792 | goto out; | |
793 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
794 | goto out; | |
fa3d315a | 795 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 796 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 797 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
798 | !access_ok(VERIFY_WRITE, |
799 | (void __user *)(unsigned long)mem->userspace_addr, | |
800 | mem->memory_size))) | |
78749809 | 801 | goto out; |
f481b069 | 802 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
803 | goto out; |
804 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
805 | goto out; | |
806 | ||
f481b069 | 807 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
808 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
809 | npages = mem->memory_size >> PAGE_SHIFT; | |
810 | ||
660c22c4 TY |
811 | if (npages > KVM_MEM_MAX_NR_PAGES) |
812 | goto out; | |
813 | ||
a843fac2 | 814 | new = old = *slot; |
6aa8b732 | 815 | |
f481b069 | 816 | new.id = id; |
6aa8b732 AK |
817 | new.base_gfn = base_gfn; |
818 | new.npages = npages; | |
819 | new.flags = mem->flags; | |
820 | ||
f64c0398 TY |
821 | if (npages) { |
822 | if (!old.npages) | |
823 | change = KVM_MR_CREATE; | |
824 | else { /* Modify an existing slot. */ | |
825 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
826 | (npages != old.npages) || |
827 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
828 | goto out; |
829 | ||
830 | if (base_gfn != old.base_gfn) | |
831 | change = KVM_MR_MOVE; | |
832 | else if (new.flags != old.flags) | |
833 | change = KVM_MR_FLAGS_ONLY; | |
834 | else { /* Nothing to change. */ | |
835 | r = 0; | |
836 | goto out; | |
837 | } | |
838 | } | |
09170a49 PB |
839 | } else { |
840 | if (!old.npages) | |
841 | goto out; | |
842 | ||
f64c0398 | 843 | change = KVM_MR_DELETE; |
09170a49 PB |
844 | new.base_gfn = 0; |
845 | new.flags = 0; | |
846 | } | |
6aa8b732 | 847 | |
f64c0398 | 848 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
849 | /* Check for overlaps */ |
850 | r = -EEXIST; | |
f481b069 | 851 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
a843fac2 | 852 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
f481b069 | 853 | (slot->id == id)) |
0a706bee TY |
854 | continue; |
855 | if (!((base_gfn + npages <= slot->base_gfn) || | |
856 | (base_gfn >= slot->base_gfn + slot->npages))) | |
857 | goto out; | |
858 | } | |
6aa8b732 | 859 | } |
6aa8b732 | 860 | |
6aa8b732 AK |
861 | /* Free page dirty bitmap if unneeded */ |
862 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 863 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
864 | |
865 | r = -ENOMEM; | |
f64c0398 | 866 | if (change == KVM_MR_CREATE) { |
189a2f7b | 867 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 868 | |
5587027c | 869 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 870 | goto out_free; |
6aa8b732 | 871 | } |
ec04b260 | 872 | |
6aa8b732 AK |
873 | /* Allocate page dirty bitmap if needed */ |
874 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 875 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 876 | goto out_free; |
6aa8b732 AK |
877 | } |
878 | ||
74496134 | 879 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
f2a81036 PB |
880 | if (!slots) |
881 | goto out_free; | |
f481b069 | 882 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 883 | |
f64c0398 | 884 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 885 | slot = id_to_memslot(slots, id); |
28a37544 XG |
886 | slot->flags |= KVM_MEMSLOT_INVALID; |
887 | ||
f481b069 | 888 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 889 | |
e40f193f AW |
890 | /* slot was deleted or moved, clear iommu mapping */ |
891 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
892 | /* From this point no new shadow pages pointing to a deleted, |
893 | * or moved, memslot will be created. | |
bc6678a3 MT |
894 | * |
895 | * validation of sp->gfn happens in: | |
b7d409de XL |
896 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
897 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 898 | */ |
2df72e9b | 899 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
900 | |
901 | /* | |
902 | * We can re-use the old_memslots from above, the only difference | |
903 | * from the currently installed memslots is the invalid flag. This | |
904 | * will get overwritten by update_memslots anyway. | |
905 | */ | |
b7f69c55 | 906 | slots = old_memslots; |
bc6678a3 | 907 | } |
34d4cb8f | 908 | |
7b6195a9 | 909 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 910 | if (r) |
b7f69c55 | 911 | goto out_slots; |
f7784b8e | 912 | |
a47d2b07 | 913 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 914 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 915 | new.dirty_bitmap = NULL; |
db3fe4eb | 916 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
917 | } |
918 | ||
5cc15027 | 919 | update_memslots(slots, &new); |
f481b069 | 920 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 921 | |
f36f3f28 | 922 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 923 | |
a47d2b07 | 924 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 925 | kvfree(old_memslots); |
bc6678a3 | 926 | |
261874b0 AW |
927 | /* |
928 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
929 | * un-mapped and re-mapped if their base changes. Since base change |
930 | * unmapping is handled above with slot deletion, mapping alone is | |
931 | * needed here. Anything else the iommu might care about for existing | |
932 | * slots (size changes, userspace addr changes and read-only flag | |
933 | * changes) is disallowed above, so any other attribute changes getting | |
934 | * here can be skipped. | |
261874b0 | 935 | */ |
75d61fbc TY |
936 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
937 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 938 | return r; |
bc6678a3 MT |
939 | } |
940 | ||
6aa8b732 AK |
941 | return 0; |
942 | ||
e40f193f | 943 | out_slots: |
74496134 | 944 | kvfree(slots); |
f78e0e2e | 945 | out_free: |
a47d2b07 | 946 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
947 | out: |
948 | return r; | |
210c7c4d | 949 | } |
f78e0e2e SY |
950 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
951 | ||
952 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 953 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
954 | { |
955 | int r; | |
956 | ||
79fac95e | 957 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 958 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 959 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
960 | return r; |
961 | } | |
210c7c4d IE |
962 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
963 | ||
7940876e SH |
964 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
965 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 966 | { |
f481b069 | 967 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 968 | return -EINVAL; |
09170a49 | 969 | |
47ae31e2 | 970 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
971 | } |
972 | ||
5bb064dc ZX |
973 | int kvm_get_dirty_log(struct kvm *kvm, |
974 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 975 | { |
9f6b8029 | 976 | struct kvm_memslots *slots; |
6aa8b732 | 977 | struct kvm_memory_slot *memslot; |
f481b069 | 978 | int r, i, as_id, id; |
87bf6e7d | 979 | unsigned long n; |
6aa8b732 AK |
980 | unsigned long any = 0; |
981 | ||
6aa8b732 | 982 | r = -EINVAL; |
f481b069 PB |
983 | as_id = log->slot >> 16; |
984 | id = (u16)log->slot; | |
985 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
6aa8b732 AK |
986 | goto out; |
987 | ||
f481b069 PB |
988 | slots = __kvm_memslots(kvm, as_id); |
989 | memslot = id_to_memslot(slots, id); | |
6aa8b732 AK |
990 | r = -ENOENT; |
991 | if (!memslot->dirty_bitmap) | |
992 | goto out; | |
993 | ||
87bf6e7d | 994 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 995 | |
cd1a4a98 | 996 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
997 | any = memslot->dirty_bitmap[i]; |
998 | ||
999 | r = -EFAULT; | |
1000 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
1001 | goto out; | |
1002 | ||
5bb064dc ZX |
1003 | if (any) |
1004 | *is_dirty = 1; | |
6aa8b732 AK |
1005 | |
1006 | r = 0; | |
6aa8b732 | 1007 | out: |
6aa8b732 AK |
1008 | return r; |
1009 | } | |
2ba9f0d8 | 1010 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1011 | |
ba0513b5 MS |
1012 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1013 | /** | |
1014 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1015 | * are dirty write protect them for next write. | |
1016 | * @kvm: pointer to kvm instance | |
1017 | * @log: slot id and address to which we copy the log | |
1018 | * @is_dirty: flag set if any page is dirty | |
1019 | * | |
1020 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1021 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1022 | * following order: | |
1023 | * | |
1024 | * 1. Take a snapshot of the bit and clear it if needed. | |
1025 | * 2. Write protect the corresponding page. | |
1026 | * 3. Copy the snapshot to the userspace. | |
1027 | * 4. Upon return caller flushes TLB's if needed. | |
1028 | * | |
1029 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1030 | * entry. This is not a problem because the page is reported dirty using | |
1031 | * the snapshot taken before and step 4 ensures that writes done after | |
1032 | * exiting to userspace will be logged for the next call. | |
1033 | * | |
1034 | */ | |
1035 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1036 | struct kvm_dirty_log *log, bool *is_dirty) | |
1037 | { | |
9f6b8029 | 1038 | struct kvm_memslots *slots; |
ba0513b5 | 1039 | struct kvm_memory_slot *memslot; |
f481b069 | 1040 | int r, i, as_id, id; |
ba0513b5 MS |
1041 | unsigned long n; |
1042 | unsigned long *dirty_bitmap; | |
1043 | unsigned long *dirty_bitmap_buffer; | |
1044 | ||
1045 | r = -EINVAL; | |
f481b069 PB |
1046 | as_id = log->slot >> 16; |
1047 | id = (u16)log->slot; | |
1048 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
ba0513b5 MS |
1049 | goto out; |
1050 | ||
f481b069 PB |
1051 | slots = __kvm_memslots(kvm, as_id); |
1052 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1053 | |
1054 | dirty_bitmap = memslot->dirty_bitmap; | |
1055 | r = -ENOENT; | |
1056 | if (!dirty_bitmap) | |
1057 | goto out; | |
1058 | ||
1059 | n = kvm_dirty_bitmap_bytes(memslot); | |
1060 | ||
1061 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1062 | memset(dirty_bitmap_buffer, 0, n); | |
1063 | ||
1064 | spin_lock(&kvm->mmu_lock); | |
1065 | *is_dirty = false; | |
1066 | for (i = 0; i < n / sizeof(long); i++) { | |
1067 | unsigned long mask; | |
1068 | gfn_t offset; | |
1069 | ||
1070 | if (!dirty_bitmap[i]) | |
1071 | continue; | |
1072 | ||
1073 | *is_dirty = true; | |
1074 | ||
1075 | mask = xchg(&dirty_bitmap[i], 0); | |
1076 | dirty_bitmap_buffer[i] = mask; | |
1077 | ||
58d2930f TY |
1078 | if (mask) { |
1079 | offset = i * BITS_PER_LONG; | |
1080 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1081 | offset, mask); | |
1082 | } | |
ba0513b5 MS |
1083 | } |
1084 | ||
1085 | spin_unlock(&kvm->mmu_lock); | |
1086 | ||
1087 | r = -EFAULT; | |
1088 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1089 | goto out; | |
1090 | ||
1091 | r = 0; | |
1092 | out: | |
1093 | return r; | |
1094 | } | |
1095 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1096 | #endif | |
1097 | ||
db3fe4eb TY |
1098 | bool kvm_largepages_enabled(void) |
1099 | { | |
1100 | return largepages_enabled; | |
1101 | } | |
1102 | ||
54dee993 MT |
1103 | void kvm_disable_largepages(void) |
1104 | { | |
1105 | largepages_enabled = false; | |
1106 | } | |
1107 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1108 | ||
49c7754c GN |
1109 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1110 | { | |
1111 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1112 | } | |
a1f4d395 | 1113 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1114 | |
8e73485c PB |
1115 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1116 | { | |
1117 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1118 | } | |
1119 | ||
e0d62c7f IE |
1120 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
1121 | { | |
bf3e05bc | 1122 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1123 | |
bbacc0c1 | 1124 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
1125 | memslot->flags & KVM_MEMSLOT_INVALID) |
1126 | return 0; | |
e0d62c7f | 1127 | |
bf3e05bc | 1128 | return 1; |
e0d62c7f IE |
1129 | } |
1130 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1131 | ||
8f0b1ab6 JR |
1132 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1133 | { | |
1134 | struct vm_area_struct *vma; | |
1135 | unsigned long addr, size; | |
1136 | ||
1137 | size = PAGE_SIZE; | |
1138 | ||
1139 | addr = gfn_to_hva(kvm, gfn); | |
1140 | if (kvm_is_error_hva(addr)) | |
1141 | return PAGE_SIZE; | |
1142 | ||
1143 | down_read(¤t->mm->mmap_sem); | |
1144 | vma = find_vma(current->mm, addr); | |
1145 | if (!vma) | |
1146 | goto out; | |
1147 | ||
1148 | size = vma_kernel_pagesize(vma); | |
1149 | ||
1150 | out: | |
1151 | up_read(¤t->mm->mmap_sem); | |
1152 | ||
1153 | return size; | |
1154 | } | |
1155 | ||
4d8b81ab XG |
1156 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1157 | { | |
1158 | return slot->flags & KVM_MEM_READONLY; | |
1159 | } | |
1160 | ||
4d8b81ab XG |
1161 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1162 | gfn_t *nr_pages, bool write) | |
539cb660 | 1163 | { |
bc6678a3 | 1164 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1165 | return KVM_HVA_ERR_BAD; |
48987781 | 1166 | |
4d8b81ab XG |
1167 | if (memslot_is_readonly(slot) && write) |
1168 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1169 | |
1170 | if (nr_pages) | |
1171 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1172 | ||
4d8b81ab | 1173 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1174 | } |
48987781 | 1175 | |
4d8b81ab XG |
1176 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1177 | gfn_t *nr_pages) | |
1178 | { | |
1179 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1180 | } |
48987781 | 1181 | |
4d8b81ab | 1182 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1183 | gfn_t gfn) |
4d8b81ab XG |
1184 | { |
1185 | return gfn_to_hva_many(slot, gfn, NULL); | |
1186 | } | |
1187 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1188 | ||
48987781 XG |
1189 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1190 | { | |
49c7754c | 1191 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1192 | } |
0d150298 | 1193 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1194 | |
8e73485c PB |
1195 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1196 | { | |
1197 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1198 | } | |
1199 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1200 | ||
86ab8cff | 1201 | /* |
ba6a3541 PB |
1202 | * If writable is set to false, the hva returned by this function is only |
1203 | * allowed to be read. | |
86ab8cff | 1204 | */ |
64d83126 CD |
1205 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1206 | gfn_t gfn, bool *writable) | |
86ab8cff | 1207 | { |
a2ac07fe GN |
1208 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1209 | ||
1210 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1211 | *writable = !memslot_is_readonly(slot); |
1212 | ||
a2ac07fe | 1213 | return hva; |
86ab8cff XG |
1214 | } |
1215 | ||
64d83126 CD |
1216 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1217 | { | |
1218 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1219 | ||
1220 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1221 | } | |
1222 | ||
8e73485c PB |
1223 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1224 | { | |
1225 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1226 | ||
1227 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1228 | } | |
1229 | ||
39369f7a | 1230 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1231 | unsigned long start, int write, struct page **page) |
1232 | { | |
1233 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1234 | ||
1235 | if (write) | |
1236 | flags |= FOLL_WRITE; | |
1237 | ||
1238 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1239 | } | |
1240 | ||
fafc3dba HY |
1241 | static inline int check_user_page_hwpoison(unsigned long addr) |
1242 | { | |
1243 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1244 | ||
1245 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1246 | flags, NULL, NULL, NULL); | |
1247 | return rc == -EHWPOISON; | |
1248 | } | |
1249 | ||
2fc84311 XG |
1250 | /* |
1251 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1252 | * true indicates success, otherwise false is returned. | |
1253 | */ | |
1254 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1255 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1256 | { |
8d4e1288 | 1257 | struct page *page[1]; |
2fc84311 | 1258 | int npages; |
954bbbc2 | 1259 | |
2fc84311 XG |
1260 | if (!(async || atomic)) |
1261 | return false; | |
af585b92 | 1262 | |
12ce13fe XG |
1263 | /* |
1264 | * Fast pin a writable pfn only if it is a write fault request | |
1265 | * or the caller allows to map a writable pfn for a read fault | |
1266 | * request. | |
1267 | */ | |
1268 | if (!(write_fault || writable)) | |
1269 | return false; | |
612819c3 | 1270 | |
2fc84311 XG |
1271 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1272 | if (npages == 1) { | |
1273 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1274 | |
2fc84311 XG |
1275 | if (writable) |
1276 | *writable = true; | |
1277 | return true; | |
1278 | } | |
af585b92 | 1279 | |
2fc84311 XG |
1280 | return false; |
1281 | } | |
612819c3 | 1282 | |
2fc84311 XG |
1283 | /* |
1284 | * The slow path to get the pfn of the specified host virtual address, | |
1285 | * 1 indicates success, -errno is returned if error is detected. | |
1286 | */ | |
1287 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1288 | bool *writable, pfn_t *pfn) | |
1289 | { | |
1290 | struct page *page[1]; | |
1291 | int npages = 0; | |
612819c3 | 1292 | |
2fc84311 XG |
1293 | might_sleep(); |
1294 | ||
1295 | if (writable) | |
1296 | *writable = write_fault; | |
1297 | ||
1298 | if (async) { | |
1299 | down_read(¤t->mm->mmap_sem); | |
1300 | npages = get_user_page_nowait(current, current->mm, | |
1301 | addr, write_fault, page); | |
1302 | up_read(¤t->mm->mmap_sem); | |
0664e57f AA |
1303 | } else |
1304 | npages = __get_user_pages_unlocked(current, current->mm, addr, 1, | |
1305 | write_fault, 0, page, | |
1306 | FOLL_TOUCH|FOLL_HWPOISON); | |
2fc84311 XG |
1307 | if (npages != 1) |
1308 | return npages; | |
1309 | ||
1310 | /* map read fault as writable if possible */ | |
12ce13fe | 1311 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1312 | struct page *wpage[1]; |
1313 | ||
1314 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1315 | if (npages == 1) { | |
1316 | *writable = true; | |
1317 | put_page(page[0]); | |
1318 | page[0] = wpage[0]; | |
612819c3 | 1319 | } |
2fc84311 XG |
1320 | |
1321 | npages = 1; | |
887c08ac | 1322 | } |
2fc84311 XG |
1323 | *pfn = page_to_pfn(page[0]); |
1324 | return npages; | |
1325 | } | |
539cb660 | 1326 | |
4d8b81ab XG |
1327 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1328 | { | |
1329 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1330 | return false; | |
2e2e3738 | 1331 | |
4d8b81ab XG |
1332 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1333 | return false; | |
887c08ac | 1334 | |
4d8b81ab XG |
1335 | return true; |
1336 | } | |
bf998156 | 1337 | |
12ce13fe XG |
1338 | /* |
1339 | * Pin guest page in memory and return its pfn. | |
1340 | * @addr: host virtual address which maps memory to the guest | |
1341 | * @atomic: whether this function can sleep | |
1342 | * @async: whether this function need to wait IO complete if the | |
1343 | * host page is not in the memory | |
1344 | * @write_fault: whether we should get a writable host page | |
1345 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1346 | * | |
1347 | * The function will map a writable host page for these two cases: | |
1348 | * 1): @write_fault = true | |
1349 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1350 | * whether the mapping is writable. | |
1351 | */ | |
2fc84311 XG |
1352 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1353 | bool write_fault, bool *writable) | |
1354 | { | |
1355 | struct vm_area_struct *vma; | |
1356 | pfn_t pfn = 0; | |
1357 | int npages; | |
2e2e3738 | 1358 | |
2fc84311 XG |
1359 | /* we can do it either atomically or asynchronously, not both */ |
1360 | BUG_ON(atomic && async); | |
8d4e1288 | 1361 | |
2fc84311 XG |
1362 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1363 | return pfn; | |
1364 | ||
1365 | if (atomic) | |
1366 | return KVM_PFN_ERR_FAULT; | |
1367 | ||
1368 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1369 | if (npages == 1) | |
1370 | return pfn; | |
8d4e1288 | 1371 | |
2fc84311 XG |
1372 | down_read(¤t->mm->mmap_sem); |
1373 | if (npages == -EHWPOISON || | |
1374 | (!async && check_user_page_hwpoison(addr))) { | |
1375 | pfn = KVM_PFN_ERR_HWPOISON; | |
1376 | goto exit; | |
1377 | } | |
1378 | ||
1379 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1380 | ||
1381 | if (vma == NULL) | |
1382 | pfn = KVM_PFN_ERR_FAULT; | |
1383 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1384 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1385 | vma->vm_pgoff; | |
bf4bea8e | 1386 | BUG_ON(!kvm_is_reserved_pfn(pfn)); |
2fc84311 | 1387 | } else { |
4d8b81ab | 1388 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1389 | *async = true; |
1390 | pfn = KVM_PFN_ERR_FAULT; | |
1391 | } | |
1392 | exit: | |
1393 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1394 | return pfn; |
35149e21 AL |
1395 | } |
1396 | ||
3520469d PB |
1397 | pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, |
1398 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1399 | { |
4d8b81ab XG |
1400 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1401 | ||
1402 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1403 | return KVM_PFN_ERR_RO_FAULT; | |
1404 | ||
1405 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1406 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1407 | |
1408 | /* Do not map writable pfn in the readonly memslot. */ | |
1409 | if (writable && memslot_is_readonly(slot)) { | |
1410 | *writable = false; | |
1411 | writable = NULL; | |
1412 | } | |
1413 | ||
1414 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1415 | writable); | |
887c08ac | 1416 | } |
3520469d | 1417 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1418 | |
612819c3 MT |
1419 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1420 | bool *writable) | |
1421 | { | |
e37afc6e PB |
1422 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1423 | write_fault, writable); | |
612819c3 MT |
1424 | } |
1425 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1426 | ||
d5661048 | 1427 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1428 | { |
4d8b81ab | 1429 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1430 | } |
e37afc6e | 1431 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1432 | |
037d92dc | 1433 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1434 | { |
4d8b81ab | 1435 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1436 | } |
037d92dc | 1437 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1438 | |
e37afc6e PB |
1439 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
1440 | { | |
1441 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1442 | } | |
1443 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1444 | ||
8e73485c PB |
1445 | pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
1446 | { | |
1447 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1448 | } | |
1449 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1450 | ||
e37afc6e PB |
1451 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1452 | { | |
1453 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1454 | } | |
1455 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1456 | ||
8e73485c PB |
1457 | pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
1458 | { | |
1459 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1460 | } | |
1461 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1462 | ||
d9ef13c2 PB |
1463 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1464 | struct page **pages, int nr_pages) | |
48987781 XG |
1465 | { |
1466 | unsigned long addr; | |
1467 | gfn_t entry; | |
1468 | ||
d9ef13c2 | 1469 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1470 | if (kvm_is_error_hva(addr)) |
1471 | return -1; | |
1472 | ||
1473 | if (entry < nr_pages) | |
1474 | return 0; | |
1475 | ||
1476 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1477 | } | |
1478 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1479 | ||
a2766325 XG |
1480 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1481 | { | |
81c52c56 | 1482 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1483 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1484 | |
bf4bea8e | 1485 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1486 | WARN_ON(1); |
6cede2e6 | 1487 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1488 | } |
a2766325 XG |
1489 | |
1490 | return pfn_to_page(pfn); | |
1491 | } | |
1492 | ||
35149e21 AL |
1493 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1494 | { | |
2e2e3738 AL |
1495 | pfn_t pfn; |
1496 | ||
1497 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1498 | |
a2766325 | 1499 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1500 | } |
1501 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1502 | ||
8e73485c PB |
1503 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1504 | { | |
1505 | pfn_t pfn; | |
1506 | ||
1507 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1508 | ||
1509 | return kvm_pfn_to_page(pfn); | |
1510 | } | |
1511 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1512 | ||
b4231d61 IE |
1513 | void kvm_release_page_clean(struct page *page) |
1514 | { | |
32cad84f XG |
1515 | WARN_ON(is_error_page(page)); |
1516 | ||
35149e21 | 1517 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1518 | } |
1519 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1520 | ||
35149e21 AL |
1521 | void kvm_release_pfn_clean(pfn_t pfn) |
1522 | { | |
bf4bea8e | 1523 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1524 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1525 | } |
1526 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1527 | ||
b4231d61 | 1528 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1529 | { |
a2766325 XG |
1530 | WARN_ON(is_error_page(page)); |
1531 | ||
35149e21 AL |
1532 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1533 | } | |
1534 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1535 | ||
7940876e | 1536 | static void kvm_release_pfn_dirty(pfn_t pfn) |
35149e21 AL |
1537 | { |
1538 | kvm_set_pfn_dirty(pfn); | |
1539 | kvm_release_pfn_clean(pfn); | |
1540 | } | |
35149e21 AL |
1541 | |
1542 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1543 | { | |
bf4bea8e | 1544 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 | 1545 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1546 | |
2e2e3738 AL |
1547 | if (!PageReserved(page)) |
1548 | SetPageDirty(page); | |
1549 | } | |
8a7ae055 | 1550 | } |
35149e21 AL |
1551 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1552 | ||
1553 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1554 | { | |
bf4bea8e | 1555 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1556 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1557 | } |
1558 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1559 | ||
1560 | void kvm_get_pfn(pfn_t pfn) | |
1561 | { | |
bf4bea8e | 1562 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1563 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1564 | } |
1565 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1566 | |
195aefde IE |
1567 | static int next_segment(unsigned long len, int offset) |
1568 | { | |
1569 | if (len > PAGE_SIZE - offset) | |
1570 | return PAGE_SIZE - offset; | |
1571 | else | |
1572 | return len; | |
1573 | } | |
1574 | ||
8e73485c PB |
1575 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1576 | void *data, int offset, int len) | |
195aefde | 1577 | { |
e0506bcb IE |
1578 | int r; |
1579 | unsigned long addr; | |
195aefde | 1580 | |
8e73485c | 1581 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1582 | if (kvm_is_error_hva(addr)) |
1583 | return -EFAULT; | |
3180a7fc | 1584 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1585 | if (r) |
195aefde | 1586 | return -EFAULT; |
195aefde IE |
1587 | return 0; |
1588 | } | |
8e73485c PB |
1589 | |
1590 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1591 | int len) | |
1592 | { | |
1593 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1594 | ||
1595 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1596 | } | |
195aefde IE |
1597 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1598 | ||
8e73485c PB |
1599 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1600 | int offset, int len) | |
1601 | { | |
1602 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1603 | ||
1604 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1605 | } | |
1606 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1607 | ||
195aefde IE |
1608 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1609 | { | |
1610 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1611 | int seg; | |
1612 | int offset = offset_in_page(gpa); | |
1613 | int ret; | |
1614 | ||
1615 | while ((seg = next_segment(len, offset)) != 0) { | |
1616 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1617 | if (ret < 0) | |
1618 | return ret; | |
1619 | offset = 0; | |
1620 | len -= seg; | |
1621 | data += seg; | |
1622 | ++gfn; | |
1623 | } | |
1624 | return 0; | |
1625 | } | |
1626 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1627 | ||
8e73485c | 1628 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 1629 | { |
7ec54588 | 1630 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 1631 | int seg; |
7ec54588 | 1632 | int offset = offset_in_page(gpa); |
8e73485c PB |
1633 | int ret; |
1634 | ||
1635 | while ((seg = next_segment(len, offset)) != 0) { | |
1636 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
1637 | if (ret < 0) | |
1638 | return ret; | |
1639 | offset = 0; | |
1640 | len -= seg; | |
1641 | data += seg; | |
1642 | ++gfn; | |
1643 | } | |
1644 | return 0; | |
1645 | } | |
1646 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 1647 | |
8e73485c PB |
1648 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1649 | void *data, int offset, unsigned long len) | |
1650 | { | |
1651 | int r; | |
1652 | unsigned long addr; | |
1653 | ||
1654 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
1655 | if (kvm_is_error_hva(addr)) |
1656 | return -EFAULT; | |
0aac03f0 | 1657 | pagefault_disable(); |
3180a7fc | 1658 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1659 | pagefault_enable(); |
7ec54588 MT |
1660 | if (r) |
1661 | return -EFAULT; | |
1662 | return 0; | |
1663 | } | |
7ec54588 | 1664 | |
8e73485c PB |
1665 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1666 | unsigned long len) | |
1667 | { | |
1668 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1669 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1670 | int offset = offset_in_page(gpa); | |
1671 | ||
1672 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1673 | } | |
1674 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
1675 | ||
1676 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
1677 | void *data, unsigned long len) | |
1678 | { | |
1679 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1680 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1681 | int offset = offset_in_page(gpa); | |
1682 | ||
1683 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1684 | } | |
1685 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
1686 | ||
1687 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
1688 | const void *data, int offset, int len) | |
195aefde | 1689 | { |
e0506bcb IE |
1690 | int r; |
1691 | unsigned long addr; | |
195aefde | 1692 | |
251eb841 | 1693 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
1694 | if (kvm_is_error_hva(addr)) |
1695 | return -EFAULT; | |
8b0cedff | 1696 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1697 | if (r) |
195aefde | 1698 | return -EFAULT; |
bc009e43 | 1699 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
1700 | return 0; |
1701 | } | |
8e73485c PB |
1702 | |
1703 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
1704 | const void *data, int offset, int len) | |
1705 | { | |
1706 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1707 | ||
1708 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1709 | } | |
195aefde IE |
1710 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
1711 | ||
8e73485c PB |
1712 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
1713 | const void *data, int offset, int len) | |
1714 | { | |
1715 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1716 | ||
1717 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1718 | } | |
1719 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
1720 | ||
195aefde IE |
1721 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
1722 | unsigned long len) | |
1723 | { | |
1724 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1725 | int seg; | |
1726 | int offset = offset_in_page(gpa); | |
1727 | int ret; | |
1728 | ||
1729 | while ((seg = next_segment(len, offset)) != 0) { | |
1730 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1731 | if (ret < 0) | |
1732 | return ret; | |
1733 | offset = 0; | |
1734 | len -= seg; | |
1735 | data += seg; | |
1736 | ++gfn; | |
1737 | } | |
1738 | return 0; | |
1739 | } | |
ff651cb6 | 1740 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1741 | |
8e73485c PB |
1742 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
1743 | unsigned long len) | |
1744 | { | |
1745 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1746 | int seg; | |
1747 | int offset = offset_in_page(gpa); | |
1748 | int ret; | |
1749 | ||
1750 | while ((seg = next_segment(len, offset)) != 0) { | |
1751 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
1752 | if (ret < 0) | |
1753 | return ret; | |
1754 | offset = 0; | |
1755 | len -= seg; | |
1756 | data += seg; | |
1757 | ++gfn; | |
1758 | } | |
1759 | return 0; | |
1760 | } | |
1761 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
1762 | ||
49c7754c | 1763 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1764 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1765 | { |
1766 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1767 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1768 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1769 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1770 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1771 | gfn_t nr_pages_avail; | |
49c7754c GN |
1772 | |
1773 | ghc->gpa = gpa; | |
1774 | ghc->generation = slots->generation; | |
8f964525 AH |
1775 | ghc->len = len; |
1776 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
ca3f0874 RK |
1777 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
1778 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { | |
49c7754c | 1779 | ghc->hva += offset; |
8f964525 AH |
1780 | } else { |
1781 | /* | |
1782 | * If the requested region crosses two memslots, we still | |
1783 | * verify that the entire region is valid here. | |
1784 | */ | |
1785 | while (start_gfn <= end_gfn) { | |
1786 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1787 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1788 | &nr_pages_avail); | |
1789 | if (kvm_is_error_hva(ghc->hva)) | |
1790 | return -EFAULT; | |
1791 | start_gfn += nr_pages_avail; | |
1792 | } | |
1793 | /* Use the slow path for cross page reads and writes. */ | |
1794 | ghc->memslot = NULL; | |
1795 | } | |
49c7754c GN |
1796 | return 0; |
1797 | } | |
1798 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1799 | ||
1800 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1801 | void *data, unsigned long len) | |
1802 | { | |
1803 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1804 | int r; | |
1805 | ||
8f964525 AH |
1806 | BUG_ON(len > ghc->len); |
1807 | ||
49c7754c | 1808 | if (slots->generation != ghc->generation) |
8f964525 AH |
1809 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1810 | ||
1811 | if (unlikely(!ghc->memslot)) | |
1812 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1813 | |
1814 | if (kvm_is_error_hva(ghc->hva)) | |
1815 | return -EFAULT; | |
1816 | ||
8b0cedff | 1817 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1818 | if (r) |
1819 | return -EFAULT; | |
bc009e43 | 1820 | mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT); |
49c7754c GN |
1821 | |
1822 | return 0; | |
1823 | } | |
1824 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1825 | ||
e03b644f GN |
1826 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1827 | void *data, unsigned long len) | |
1828 | { | |
1829 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1830 | int r; | |
1831 | ||
8f964525 AH |
1832 | BUG_ON(len > ghc->len); |
1833 | ||
e03b644f | 1834 | if (slots->generation != ghc->generation) |
8f964525 AH |
1835 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1836 | ||
1837 | if (unlikely(!ghc->memslot)) | |
1838 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1839 | |
1840 | if (kvm_is_error_hva(ghc->hva)) | |
1841 | return -EFAULT; | |
1842 | ||
1843 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1844 | if (r) | |
1845 | return -EFAULT; | |
1846 | ||
1847 | return 0; | |
1848 | } | |
1849 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1850 | ||
195aefde IE |
1851 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1852 | { | |
8a3caa6d HC |
1853 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
1854 | ||
1855 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
1856 | } |
1857 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1858 | ||
1859 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1860 | { | |
1861 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1862 | int seg; | |
1863 | int offset = offset_in_page(gpa); | |
1864 | int ret; | |
1865 | ||
bfda0e84 | 1866 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
1867 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
1868 | if (ret < 0) | |
1869 | return ret; | |
1870 | offset = 0; | |
1871 | len -= seg; | |
1872 | ++gfn; | |
1873 | } | |
1874 | return 0; | |
1875 | } | |
1876 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1877 | ||
bc009e43 | 1878 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 1879 | gfn_t gfn) |
6aa8b732 | 1880 | { |
7e9d619d RR |
1881 | if (memslot && memslot->dirty_bitmap) { |
1882 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1883 | |
b74ca3b3 | 1884 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1885 | } |
1886 | } | |
1887 | ||
49c7754c GN |
1888 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1889 | { | |
1890 | struct kvm_memory_slot *memslot; | |
1891 | ||
1892 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 1893 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 1894 | } |
2ba9f0d8 | 1895 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 1896 | |
8e73485c PB |
1897 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
1898 | { | |
1899 | struct kvm_memory_slot *memslot; | |
1900 | ||
1901 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1902 | mark_page_dirty_in_slot(memslot, gfn); | |
1903 | } | |
1904 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
1905 | ||
f7819512 PB |
1906 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
1907 | { | |
1908 | if (kvm_arch_vcpu_runnable(vcpu)) { | |
1909 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
1910 | return -EINTR; | |
1911 | } | |
1912 | if (kvm_cpu_has_pending_timer(vcpu)) | |
1913 | return -EINTR; | |
1914 | if (signal_pending(current)) | |
1915 | return -EINTR; | |
1916 | ||
1917 | return 0; | |
1918 | } | |
1919 | ||
b6958ce4 ED |
1920 | /* |
1921 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1922 | */ | |
8776e519 | 1923 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1924 | { |
f7819512 | 1925 | ktime_t start, cur; |
e5c239cf | 1926 | DEFINE_WAIT(wait); |
f7819512 PB |
1927 | bool waited = false; |
1928 | ||
1929 | start = cur = ktime_get(); | |
1930 | if (halt_poll_ns) { | |
1931 | ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns); | |
f95ef0cd | 1932 | |
f7819512 PB |
1933 | do { |
1934 | /* | |
1935 | * This sets KVM_REQ_UNHALT if an interrupt | |
1936 | * arrives. | |
1937 | */ | |
1938 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
1939 | ++vcpu->stat.halt_successful_poll; | |
1940 | goto out; | |
1941 | } | |
1942 | cur = ktime_get(); | |
1943 | } while (single_task_running() && ktime_before(cur, stop)); | |
1944 | } | |
e5c239cf MT |
1945 | |
1946 | for (;;) { | |
1947 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1948 | ||
f7819512 | 1949 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
1950 | break; |
1951 | ||
f7819512 | 1952 | waited = true; |
b6958ce4 | 1953 | schedule(); |
b6958ce4 | 1954 | } |
d3bef15f | 1955 | |
e5c239cf | 1956 | finish_wait(&vcpu->wq, &wait); |
f7819512 PB |
1957 | cur = ktime_get(); |
1958 | ||
1959 | out: | |
1960 | trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited); | |
b6958ce4 | 1961 | } |
2ba9f0d8 | 1962 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 1963 | |
8c84780d | 1964 | #ifndef CONFIG_S390 |
b6d33834 CD |
1965 | /* |
1966 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1967 | */ | |
1968 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1969 | { | |
1970 | int me; | |
1971 | int cpu = vcpu->cpu; | |
1972 | wait_queue_head_t *wqp; | |
1973 | ||
1974 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1975 | if (waitqueue_active(wqp)) { | |
1976 | wake_up_interruptible(wqp); | |
1977 | ++vcpu->stat.halt_wakeup; | |
1978 | } | |
1979 | ||
1980 | me = get_cpu(); | |
1981 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1982 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1983 | smp_send_reschedule(cpu); | |
1984 | put_cpu(); | |
1985 | } | |
a20ed54d | 1986 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1987 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1988 | |
fa93384f | 1989 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
1990 | { |
1991 | struct pid *pid; | |
1992 | struct task_struct *task = NULL; | |
fa93384f | 1993 | int ret = 0; |
41628d33 KW |
1994 | |
1995 | rcu_read_lock(); | |
1996 | pid = rcu_dereference(target->pid); | |
1997 | if (pid) | |
27fbe64b | 1998 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
1999 | rcu_read_unlock(); |
2000 | if (!task) | |
c45c528e | 2001 | return ret; |
c45c528e | 2002 | ret = yield_to(task, 1); |
41628d33 | 2003 | put_task_struct(task); |
c45c528e R |
2004 | |
2005 | return ret; | |
41628d33 KW |
2006 | } |
2007 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2008 | ||
06e48c51 R |
2009 | /* |
2010 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2011 | * Most eligible candidate to yield is decided by following heuristics: | |
2012 | * | |
2013 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2014 | * (preempted lock holder), indicated by @in_spin_loop. | |
2015 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2016 | * | |
2017 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2018 | * chance last time (mostly it has become eligible now since we have probably | |
2019 | * yielded to lockholder in last iteration. This is done by toggling | |
2020 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2021 | * | |
2022 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2023 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2024 | * burning. Giving priority for a potential lock-holder increases lock | |
2025 | * progress. | |
2026 | * | |
2027 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2028 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2029 | * and continue with next VCPU and so on. | |
2030 | */ | |
7940876e | 2031 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2032 | { |
4a55dd72 | 2033 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2034 | bool eligible; |
2035 | ||
2036 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2037 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2038 | |
2039 | if (vcpu->spin_loop.in_spin_loop) | |
2040 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2041 | ||
2042 | return eligible; | |
4a55dd72 SW |
2043 | #else |
2044 | return true; | |
06e48c51 | 2045 | #endif |
4a55dd72 | 2046 | } |
c45c528e | 2047 | |
217ece61 | 2048 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 2049 | { |
217ece61 RR |
2050 | struct kvm *kvm = me->kvm; |
2051 | struct kvm_vcpu *vcpu; | |
2052 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2053 | int yielded = 0; | |
c45c528e | 2054 | int try = 3; |
217ece61 RR |
2055 | int pass; |
2056 | int i; | |
d255f4f2 | 2057 | |
4c088493 | 2058 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2059 | /* |
2060 | * We boost the priority of a VCPU that is runnable but not | |
2061 | * currently running, because it got preempted by something | |
2062 | * else and called schedule in __vcpu_run. Hopefully that | |
2063 | * VCPU is holding the lock that we need and will release it. | |
2064 | * We approximate round-robin by starting at the last boosted VCPU. | |
2065 | */ | |
c45c528e | 2066 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2067 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2068 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2069 | i = last_boosted_vcpu; |
2070 | continue; | |
2071 | } else if (pass && i > last_boosted_vcpu) | |
2072 | break; | |
7bc7ae25 R |
2073 | if (!ACCESS_ONCE(vcpu->preempted)) |
2074 | continue; | |
217ece61 RR |
2075 | if (vcpu == me) |
2076 | continue; | |
98f4a146 | 2077 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 2078 | continue; |
06e48c51 R |
2079 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2080 | continue; | |
c45c528e R |
2081 | |
2082 | yielded = kvm_vcpu_yield_to(vcpu); | |
2083 | if (yielded > 0) { | |
217ece61 | 2084 | kvm->last_boosted_vcpu = i; |
217ece61 | 2085 | break; |
c45c528e R |
2086 | } else if (yielded < 0) { |
2087 | try--; | |
2088 | if (!try) | |
2089 | break; | |
217ece61 | 2090 | } |
217ece61 RR |
2091 | } |
2092 | } | |
4c088493 | 2093 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2094 | |
2095 | /* Ensure vcpu is not eligible during next spinloop */ | |
2096 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2097 | } |
2098 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2099 | ||
e4a533a4 | 2100 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
2101 | { |
2102 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
2103 | struct page *page; |
2104 | ||
e4a533a4 | 2105 | if (vmf->pgoff == 0) |
039576c0 | 2106 | page = virt_to_page(vcpu->run); |
09566765 | 2107 | #ifdef CONFIG_X86 |
e4a533a4 | 2108 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2109 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
2110 | #endif |
2111 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2112 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
2113 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2114 | #endif |
039576c0 | 2115 | else |
5b1c1493 | 2116 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2117 | get_page(page); |
e4a533a4 | 2118 | vmf->page = page; |
2119 | return 0; | |
9a2bb7f4 AK |
2120 | } |
2121 | ||
f0f37e2f | 2122 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2123 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2124 | }; |
2125 | ||
2126 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2127 | { | |
2128 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2129 | return 0; | |
2130 | } | |
2131 | ||
bccf2150 AK |
2132 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2133 | { | |
2134 | struct kvm_vcpu *vcpu = filp->private_data; | |
2135 | ||
66c0b394 | 2136 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2137 | return 0; |
2138 | } | |
2139 | ||
3d3aab1b | 2140 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2141 | .release = kvm_vcpu_release, |
2142 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2143 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2144 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2145 | #endif | |
9a2bb7f4 | 2146 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2147 | .llseek = noop_llseek, |
bccf2150 AK |
2148 | }; |
2149 | ||
2150 | /* | |
2151 | * Allocates an inode for the vcpu. | |
2152 | */ | |
2153 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2154 | { | |
24009b05 | 2155 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2156 | } |
2157 | ||
c5ea7660 AK |
2158 | /* |
2159 | * Creates some virtual cpus. Good luck creating more than one. | |
2160 | */ | |
73880c80 | 2161 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2162 | { |
2163 | int r; | |
988a2cae | 2164 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 2165 | |
338c7dba AH |
2166 | if (id >= KVM_MAX_VCPUS) |
2167 | return -EINVAL; | |
2168 | ||
73880c80 | 2169 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
2170 | if (IS_ERR(vcpu)) |
2171 | return PTR_ERR(vcpu); | |
c5ea7660 | 2172 | |
15ad7146 AK |
2173 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2174 | ||
26e5215f AK |
2175 | r = kvm_arch_vcpu_setup(vcpu); |
2176 | if (r) | |
d780592b | 2177 | goto vcpu_destroy; |
26e5215f | 2178 | |
11ec2804 | 2179 | mutex_lock(&kvm->lock); |
3e515705 AK |
2180 | if (!kvm_vcpu_compatible(vcpu)) { |
2181 | r = -EINVAL; | |
2182 | goto unlock_vcpu_destroy; | |
2183 | } | |
73880c80 GN |
2184 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
2185 | r = -EINVAL; | |
d780592b | 2186 | goto unlock_vcpu_destroy; |
fb3f0f51 | 2187 | } |
73880c80 | 2188 | |
988a2cae GN |
2189 | kvm_for_each_vcpu(r, v, kvm) |
2190 | if (v->vcpu_id == id) { | |
73880c80 | 2191 | r = -EEXIST; |
d780592b | 2192 | goto unlock_vcpu_destroy; |
73880c80 GN |
2193 | } |
2194 | ||
2195 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2196 | |
fb3f0f51 | 2197 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2198 | kvm_get_kvm(kvm); |
bccf2150 | 2199 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2200 | if (r < 0) { |
2201 | kvm_put_kvm(kvm); | |
d780592b | 2202 | goto unlock_vcpu_destroy; |
73880c80 GN |
2203 | } |
2204 | ||
2205 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
2206 | smp_wmb(); | |
2207 | atomic_inc(&kvm->online_vcpus); | |
2208 | ||
73880c80 | 2209 | mutex_unlock(&kvm->lock); |
42897d86 | 2210 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2211 | return r; |
39c3b86e | 2212 | |
d780592b | 2213 | unlock_vcpu_destroy: |
7d8fece6 | 2214 | mutex_unlock(&kvm->lock); |
d780592b | 2215 | vcpu_destroy: |
d40ccc62 | 2216 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
2217 | return r; |
2218 | } | |
2219 | ||
1961d276 AK |
2220 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2221 | { | |
2222 | if (sigset) { | |
2223 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2224 | vcpu->sigset_active = 1; | |
2225 | vcpu->sigset = *sigset; | |
2226 | } else | |
2227 | vcpu->sigset_active = 0; | |
2228 | return 0; | |
2229 | } | |
2230 | ||
bccf2150 AK |
2231 | static long kvm_vcpu_ioctl(struct file *filp, |
2232 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2233 | { |
bccf2150 | 2234 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2235 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2236 | int r; |
fa3795a7 DH |
2237 | struct kvm_fpu *fpu = NULL; |
2238 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2239 | |
6d4e4c4f AK |
2240 | if (vcpu->kvm->mm != current->mm) |
2241 | return -EIO; | |
2122ff5e | 2242 | |
2ea75be3 DM |
2243 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2244 | return -EINVAL; | |
2245 | ||
2f4d9b54 | 2246 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2247 | /* |
2248 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2249 | * so vcpu_load() would break it. | |
2250 | */ | |
47b43c52 | 2251 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
2122ff5e AK |
2252 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
2253 | #endif | |
2254 | ||
2255 | ||
9fc77441 MT |
2256 | r = vcpu_load(vcpu); |
2257 | if (r) | |
2258 | return r; | |
6aa8b732 | 2259 | switch (ioctl) { |
9a2bb7f4 | 2260 | case KVM_RUN: |
f0fe5108 AK |
2261 | r = -EINVAL; |
2262 | if (arg) | |
2263 | goto out; | |
7a72f7a1 CB |
2264 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
2265 | /* The thread running this VCPU changed. */ | |
2266 | struct pid *oldpid = vcpu->pid; | |
2267 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
f95ef0cd | 2268 | |
7a72f7a1 CB |
2269 | rcu_assign_pointer(vcpu->pid, newpid); |
2270 | if (oldpid) | |
2271 | synchronize_rcu(); | |
2272 | put_pid(oldpid); | |
2273 | } | |
b6c7a5dc | 2274 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2275 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2276 | break; |
6aa8b732 | 2277 | case KVM_GET_REGS: { |
3e4bb3ac | 2278 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2279 | |
3e4bb3ac XZ |
2280 | r = -ENOMEM; |
2281 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2282 | if (!kvm_regs) | |
6aa8b732 | 2283 | goto out; |
3e4bb3ac XZ |
2284 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2285 | if (r) | |
2286 | goto out_free1; | |
6aa8b732 | 2287 | r = -EFAULT; |
3e4bb3ac XZ |
2288 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2289 | goto out_free1; | |
6aa8b732 | 2290 | r = 0; |
3e4bb3ac XZ |
2291 | out_free1: |
2292 | kfree(kvm_regs); | |
6aa8b732 AK |
2293 | break; |
2294 | } | |
2295 | case KVM_SET_REGS: { | |
3e4bb3ac | 2296 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2297 | |
3e4bb3ac | 2298 | r = -ENOMEM; |
ff5c2c03 SL |
2299 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2300 | if (IS_ERR(kvm_regs)) { | |
2301 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2302 | goto out; |
ff5c2c03 | 2303 | } |
3e4bb3ac | 2304 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2305 | kfree(kvm_regs); |
6aa8b732 AK |
2306 | break; |
2307 | } | |
2308 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2309 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2310 | r = -ENOMEM; | |
2311 | if (!kvm_sregs) | |
2312 | goto out; | |
2313 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2314 | if (r) |
2315 | goto out; | |
2316 | r = -EFAULT; | |
fa3795a7 | 2317 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2318 | goto out; |
2319 | r = 0; | |
2320 | break; | |
2321 | } | |
2322 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2323 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2324 | if (IS_ERR(kvm_sregs)) { | |
2325 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2326 | kvm_sregs = NULL; |
6aa8b732 | 2327 | goto out; |
ff5c2c03 | 2328 | } |
fa3795a7 | 2329 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2330 | break; |
2331 | } | |
62d9f0db MT |
2332 | case KVM_GET_MP_STATE: { |
2333 | struct kvm_mp_state mp_state; | |
2334 | ||
2335 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2336 | if (r) | |
2337 | goto out; | |
2338 | r = -EFAULT; | |
893bdbf1 | 2339 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2340 | goto out; |
2341 | r = 0; | |
2342 | break; | |
2343 | } | |
2344 | case KVM_SET_MP_STATE: { | |
2345 | struct kvm_mp_state mp_state; | |
2346 | ||
2347 | r = -EFAULT; | |
893bdbf1 | 2348 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2349 | goto out; |
2350 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2351 | break; |
2352 | } | |
6aa8b732 AK |
2353 | case KVM_TRANSLATE: { |
2354 | struct kvm_translation tr; | |
2355 | ||
2356 | r = -EFAULT; | |
893bdbf1 | 2357 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2358 | goto out; |
8b006791 | 2359 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2360 | if (r) |
2361 | goto out; | |
2362 | r = -EFAULT; | |
893bdbf1 | 2363 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2364 | goto out; |
2365 | r = 0; | |
2366 | break; | |
2367 | } | |
d0bfb940 JK |
2368 | case KVM_SET_GUEST_DEBUG: { |
2369 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2370 | |
2371 | r = -EFAULT; | |
893bdbf1 | 2372 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2373 | goto out; |
d0bfb940 | 2374 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2375 | break; |
2376 | } | |
1961d276 AK |
2377 | case KVM_SET_SIGNAL_MASK: { |
2378 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2379 | struct kvm_signal_mask kvm_sigmask; | |
2380 | sigset_t sigset, *p; | |
2381 | ||
2382 | p = NULL; | |
2383 | if (argp) { | |
2384 | r = -EFAULT; | |
2385 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2386 | sizeof(kvm_sigmask))) |
1961d276 AK |
2387 | goto out; |
2388 | r = -EINVAL; | |
893bdbf1 | 2389 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2390 | goto out; |
2391 | r = -EFAULT; | |
2392 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2393 | sizeof(sigset))) |
1961d276 AK |
2394 | goto out; |
2395 | p = &sigset; | |
2396 | } | |
376d41ff | 2397 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2398 | break; |
2399 | } | |
b8836737 | 2400 | case KVM_GET_FPU: { |
fa3795a7 DH |
2401 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2402 | r = -ENOMEM; | |
2403 | if (!fpu) | |
2404 | goto out; | |
2405 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2406 | if (r) |
2407 | goto out; | |
2408 | r = -EFAULT; | |
fa3795a7 | 2409 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2410 | goto out; |
2411 | r = 0; | |
2412 | break; | |
2413 | } | |
2414 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2415 | fpu = memdup_user(argp, sizeof(*fpu)); |
2416 | if (IS_ERR(fpu)) { | |
2417 | r = PTR_ERR(fpu); | |
18595411 | 2418 | fpu = NULL; |
b8836737 | 2419 | goto out; |
ff5c2c03 | 2420 | } |
fa3795a7 | 2421 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2422 | break; |
2423 | } | |
bccf2150 | 2424 | default: |
313a3dc7 | 2425 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2426 | } |
2427 | out: | |
2122ff5e | 2428 | vcpu_put(vcpu); |
fa3795a7 DH |
2429 | kfree(fpu); |
2430 | kfree(kvm_sregs); | |
bccf2150 AK |
2431 | return r; |
2432 | } | |
2433 | ||
de8e5d74 | 2434 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2435 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2436 | unsigned int ioctl, unsigned long arg) | |
2437 | { | |
2438 | struct kvm_vcpu *vcpu = filp->private_data; | |
2439 | void __user *argp = compat_ptr(arg); | |
2440 | int r; | |
2441 | ||
2442 | if (vcpu->kvm->mm != current->mm) | |
2443 | return -EIO; | |
2444 | ||
2445 | switch (ioctl) { | |
2446 | case KVM_SET_SIGNAL_MASK: { | |
2447 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2448 | struct kvm_signal_mask kvm_sigmask; | |
2449 | compat_sigset_t csigset; | |
2450 | sigset_t sigset; | |
2451 | ||
2452 | if (argp) { | |
2453 | r = -EFAULT; | |
2454 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2455 | sizeof(kvm_sigmask))) |
1dda606c AG |
2456 | goto out; |
2457 | r = -EINVAL; | |
893bdbf1 | 2458 | if (kvm_sigmask.len != sizeof(csigset)) |
1dda606c AG |
2459 | goto out; |
2460 | r = -EFAULT; | |
2461 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
893bdbf1 | 2462 | sizeof(csigset))) |
1dda606c | 2463 | goto out; |
760a9a30 AC |
2464 | sigset_from_compat(&sigset, &csigset); |
2465 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2466 | } else | |
2467 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2468 | break; |
2469 | } | |
2470 | default: | |
2471 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2472 | } | |
2473 | ||
2474 | out: | |
2475 | return r; | |
2476 | } | |
2477 | #endif | |
2478 | ||
852b6d57 SW |
2479 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2480 | int (*accessor)(struct kvm_device *dev, | |
2481 | struct kvm_device_attr *attr), | |
2482 | unsigned long arg) | |
2483 | { | |
2484 | struct kvm_device_attr attr; | |
2485 | ||
2486 | if (!accessor) | |
2487 | return -EPERM; | |
2488 | ||
2489 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2490 | return -EFAULT; | |
2491 | ||
2492 | return accessor(dev, &attr); | |
2493 | } | |
2494 | ||
2495 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2496 | unsigned long arg) | |
2497 | { | |
2498 | struct kvm_device *dev = filp->private_data; | |
2499 | ||
2500 | switch (ioctl) { | |
2501 | case KVM_SET_DEVICE_ATTR: | |
2502 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2503 | case KVM_GET_DEVICE_ATTR: | |
2504 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2505 | case KVM_HAS_DEVICE_ATTR: | |
2506 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2507 | default: | |
2508 | if (dev->ops->ioctl) | |
2509 | return dev->ops->ioctl(dev, ioctl, arg); | |
2510 | ||
2511 | return -ENOTTY; | |
2512 | } | |
2513 | } | |
2514 | ||
852b6d57 SW |
2515 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2516 | { | |
2517 | struct kvm_device *dev = filp->private_data; | |
2518 | struct kvm *kvm = dev->kvm; | |
2519 | ||
852b6d57 SW |
2520 | kvm_put_kvm(kvm); |
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | static const struct file_operations kvm_device_fops = { | |
2525 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2526 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2527 | .compat_ioctl = kvm_device_ioctl, |
2528 | #endif | |
852b6d57 SW |
2529 | .release = kvm_device_release, |
2530 | }; | |
2531 | ||
2532 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2533 | { | |
2534 | if (filp->f_op != &kvm_device_fops) | |
2535 | return NULL; | |
2536 | ||
2537 | return filp->private_data; | |
2538 | } | |
2539 | ||
d60eacb0 | 2540 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2541 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2542 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2543 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2544 | #endif |
d60eacb0 | 2545 | |
5975a2e0 | 2546 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2547 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2548 | #endif |
d60eacb0 WD |
2549 | }; |
2550 | ||
2551 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2552 | { | |
2553 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2554 | return -ENOSPC; | |
2555 | ||
2556 | if (kvm_device_ops_table[type] != NULL) | |
2557 | return -EEXIST; | |
2558 | ||
2559 | kvm_device_ops_table[type] = ops; | |
2560 | return 0; | |
2561 | } | |
2562 | ||
571ee1b6 WL |
2563 | void kvm_unregister_device_ops(u32 type) |
2564 | { | |
2565 | if (kvm_device_ops_table[type] != NULL) | |
2566 | kvm_device_ops_table[type] = NULL; | |
2567 | } | |
2568 | ||
852b6d57 SW |
2569 | static int kvm_ioctl_create_device(struct kvm *kvm, |
2570 | struct kvm_create_device *cd) | |
2571 | { | |
2572 | struct kvm_device_ops *ops = NULL; | |
2573 | struct kvm_device *dev; | |
2574 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2575 | int ret; | |
2576 | ||
d60eacb0 WD |
2577 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
2578 | return -ENODEV; | |
2579 | ||
2580 | ops = kvm_device_ops_table[cd->type]; | |
2581 | if (ops == NULL) | |
852b6d57 | 2582 | return -ENODEV; |
852b6d57 SW |
2583 | |
2584 | if (test) | |
2585 | return 0; | |
2586 | ||
2587 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2588 | if (!dev) | |
2589 | return -ENOMEM; | |
2590 | ||
2591 | dev->ops = ops; | |
2592 | dev->kvm = kvm; | |
852b6d57 SW |
2593 | |
2594 | ret = ops->create(dev, cd->type); | |
2595 | if (ret < 0) { | |
2596 | kfree(dev); | |
2597 | return ret; | |
2598 | } | |
2599 | ||
24009b05 | 2600 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2601 | if (ret < 0) { |
2602 | ops->destroy(dev); | |
2603 | return ret; | |
2604 | } | |
2605 | ||
07f0a7bd | 2606 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2607 | kvm_get_kvm(kvm); |
2608 | cd->fd = ret; | |
2609 | return 0; | |
2610 | } | |
2611 | ||
92b591a4 AG |
2612 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2613 | { | |
2614 | switch (arg) { | |
2615 | case KVM_CAP_USER_MEMORY: | |
2616 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2617 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
2618 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE | |
2619 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2620 | #endif | |
2621 | case KVM_CAP_INTERNAL_ERROR_DATA: | |
2622 | #ifdef CONFIG_HAVE_KVM_MSI | |
2623 | case KVM_CAP_SIGNAL_MSI: | |
2624 | #endif | |
297e2105 | 2625 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 2626 | case KVM_CAP_IRQFD: |
92b591a4 AG |
2627 | case KVM_CAP_IRQFD_RESAMPLE: |
2628 | #endif | |
2629 | case KVM_CAP_CHECK_EXTENSION_VM: | |
2630 | return 1; | |
2631 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2632 | case KVM_CAP_IRQ_ROUTING: | |
2633 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
2634 | #endif |
2635 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
2636 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
2637 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 AG |
2638 | #endif |
2639 | default: | |
2640 | break; | |
2641 | } | |
2642 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2643 | } | |
2644 | ||
bccf2150 AK |
2645 | static long kvm_vm_ioctl(struct file *filp, |
2646 | unsigned int ioctl, unsigned long arg) | |
2647 | { | |
2648 | struct kvm *kvm = filp->private_data; | |
2649 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2650 | int r; |
bccf2150 | 2651 | |
6d4e4c4f AK |
2652 | if (kvm->mm != current->mm) |
2653 | return -EIO; | |
bccf2150 AK |
2654 | switch (ioctl) { |
2655 | case KVM_CREATE_VCPU: | |
2656 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2657 | break; |
6fc138d2 IE |
2658 | case KVM_SET_USER_MEMORY_REGION: { |
2659 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2660 | ||
2661 | r = -EFAULT; | |
2662 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 2663 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
2664 | goto out; |
2665 | ||
47ae31e2 | 2666 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2667 | break; |
2668 | } | |
2669 | case KVM_GET_DIRTY_LOG: { | |
2670 | struct kvm_dirty_log log; | |
2671 | ||
2672 | r = -EFAULT; | |
893bdbf1 | 2673 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 2674 | goto out; |
2c6f5df9 | 2675 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2676 | break; |
2677 | } | |
5f94c174 LV |
2678 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2679 | case KVM_REGISTER_COALESCED_MMIO: { | |
2680 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2681 | |
5f94c174 | 2682 | r = -EFAULT; |
893bdbf1 | 2683 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2684 | goto out; |
5f94c174 | 2685 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2686 | break; |
2687 | } | |
2688 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2689 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2690 | |
5f94c174 | 2691 | r = -EFAULT; |
893bdbf1 | 2692 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2693 | goto out; |
5f94c174 | 2694 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2695 | break; |
2696 | } | |
2697 | #endif | |
721eecbf GH |
2698 | case KVM_IRQFD: { |
2699 | struct kvm_irqfd data; | |
2700 | ||
2701 | r = -EFAULT; | |
893bdbf1 | 2702 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 2703 | goto out; |
d4db2935 | 2704 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2705 | break; |
2706 | } | |
d34e6b17 GH |
2707 | case KVM_IOEVENTFD: { |
2708 | struct kvm_ioeventfd data; | |
2709 | ||
2710 | r = -EFAULT; | |
893bdbf1 | 2711 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
2712 | goto out; |
2713 | r = kvm_ioeventfd(kvm, &data); | |
2714 | break; | |
2715 | } | |
73880c80 GN |
2716 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2717 | case KVM_SET_BOOT_CPU_ID: | |
2718 | r = 0; | |
894a9c55 | 2719 | mutex_lock(&kvm->lock); |
73880c80 GN |
2720 | if (atomic_read(&kvm->online_vcpus) != 0) |
2721 | r = -EBUSY; | |
2722 | else | |
2723 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2724 | mutex_unlock(&kvm->lock); |
73880c80 | 2725 | break; |
07975ad3 JK |
2726 | #endif |
2727 | #ifdef CONFIG_HAVE_KVM_MSI | |
2728 | case KVM_SIGNAL_MSI: { | |
2729 | struct kvm_msi msi; | |
2730 | ||
2731 | r = -EFAULT; | |
893bdbf1 | 2732 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
2733 | goto out; |
2734 | r = kvm_send_userspace_msi(kvm, &msi); | |
2735 | break; | |
2736 | } | |
23d43cf9 CD |
2737 | #endif |
2738 | #ifdef __KVM_HAVE_IRQ_LINE | |
2739 | case KVM_IRQ_LINE_STATUS: | |
2740 | case KVM_IRQ_LINE: { | |
2741 | struct kvm_irq_level irq_event; | |
2742 | ||
2743 | r = -EFAULT; | |
893bdbf1 | 2744 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
2745 | goto out; |
2746 | ||
aa2fbe6d YZ |
2747 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2748 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2749 | if (r) |
2750 | goto out; | |
2751 | ||
2752 | r = -EFAULT; | |
2753 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 2754 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
2755 | goto out; |
2756 | } | |
2757 | ||
2758 | r = 0; | |
2759 | break; | |
2760 | } | |
73880c80 | 2761 | #endif |
aa8d5944 AG |
2762 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2763 | case KVM_SET_GSI_ROUTING: { | |
2764 | struct kvm_irq_routing routing; | |
2765 | struct kvm_irq_routing __user *urouting; | |
2766 | struct kvm_irq_routing_entry *entries; | |
2767 | ||
2768 | r = -EFAULT; | |
2769 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2770 | goto out; | |
2771 | r = -EINVAL; | |
2772 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2773 | goto out; | |
2774 | if (routing.flags) | |
2775 | goto out; | |
2776 | r = -ENOMEM; | |
2777 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2778 | if (!entries) | |
2779 | goto out; | |
2780 | r = -EFAULT; | |
2781 | urouting = argp; | |
2782 | if (copy_from_user(entries, urouting->entries, | |
2783 | routing.nr * sizeof(*entries))) | |
2784 | goto out_free_irq_routing; | |
2785 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2786 | routing.flags); | |
a642a175 | 2787 | out_free_irq_routing: |
aa8d5944 AG |
2788 | vfree(entries); |
2789 | break; | |
2790 | } | |
2791 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2792 | case KVM_CREATE_DEVICE: { |
2793 | struct kvm_create_device cd; | |
2794 | ||
2795 | r = -EFAULT; | |
2796 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2797 | goto out; | |
2798 | ||
2799 | r = kvm_ioctl_create_device(kvm, &cd); | |
2800 | if (r) | |
2801 | goto out; | |
2802 | ||
2803 | r = -EFAULT; | |
2804 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2805 | goto out; | |
2806 | ||
2807 | r = 0; | |
2808 | break; | |
2809 | } | |
92b591a4 AG |
2810 | case KVM_CHECK_EXTENSION: |
2811 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
2812 | break; | |
f17abe9a | 2813 | default: |
1fe779f8 | 2814 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
2815 | } |
2816 | out: | |
2817 | return r; | |
2818 | } | |
2819 | ||
de8e5d74 | 2820 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2821 | struct compat_kvm_dirty_log { |
2822 | __u32 slot; | |
2823 | __u32 padding1; | |
2824 | union { | |
2825 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2826 | __u64 padding2; | |
2827 | }; | |
2828 | }; | |
2829 | ||
2830 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2831 | unsigned int ioctl, unsigned long arg) | |
2832 | { | |
2833 | struct kvm *kvm = filp->private_data; | |
2834 | int r; | |
2835 | ||
2836 | if (kvm->mm != current->mm) | |
2837 | return -EIO; | |
2838 | switch (ioctl) { | |
2839 | case KVM_GET_DIRTY_LOG: { | |
2840 | struct compat_kvm_dirty_log compat_log; | |
2841 | struct kvm_dirty_log log; | |
2842 | ||
2843 | r = -EFAULT; | |
2844 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2845 | sizeof(compat_log))) | |
2846 | goto out; | |
2847 | log.slot = compat_log.slot; | |
2848 | log.padding1 = compat_log.padding1; | |
2849 | log.padding2 = compat_log.padding2; | |
2850 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2851 | ||
2852 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2853 | break; |
2854 | } | |
2855 | default: | |
2856 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2857 | } | |
2858 | ||
2859 | out: | |
2860 | return r; | |
2861 | } | |
2862 | #endif | |
2863 | ||
3d3aab1b | 2864 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2865 | .release = kvm_vm_release, |
2866 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 2867 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2868 | .compat_ioctl = kvm_vm_compat_ioctl, |
2869 | #endif | |
6038f373 | 2870 | .llseek = noop_llseek, |
f17abe9a AK |
2871 | }; |
2872 | ||
e08b9637 | 2873 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2874 | { |
aac87636 | 2875 | int r; |
f17abe9a AK |
2876 | struct kvm *kvm; |
2877 | ||
e08b9637 | 2878 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2879 | if (IS_ERR(kvm)) |
2880 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2881 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2882 | r = kvm_coalesced_mmio_init(kvm); | |
2883 | if (r < 0) { | |
2884 | kvm_put_kvm(kvm); | |
2885 | return r; | |
2886 | } | |
2887 | #endif | |
24009b05 | 2888 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2889 | if (r < 0) |
66c0b394 | 2890 | kvm_put_kvm(kvm); |
f17abe9a | 2891 | |
aac87636 | 2892 | return r; |
f17abe9a AK |
2893 | } |
2894 | ||
2895 | static long kvm_dev_ioctl(struct file *filp, | |
2896 | unsigned int ioctl, unsigned long arg) | |
2897 | { | |
07c45a36 | 2898 | long r = -EINVAL; |
f17abe9a AK |
2899 | |
2900 | switch (ioctl) { | |
2901 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2902 | if (arg) |
2903 | goto out; | |
f17abe9a AK |
2904 | r = KVM_API_VERSION; |
2905 | break; | |
2906 | case KVM_CREATE_VM: | |
e08b9637 | 2907 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2908 | break; |
018d00d2 | 2909 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 2910 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 2911 | break; |
07c45a36 | 2912 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
2913 | if (arg) |
2914 | goto out; | |
adb1ff46 AK |
2915 | r = PAGE_SIZE; /* struct kvm_run */ |
2916 | #ifdef CONFIG_X86 | |
2917 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2918 | #endif |
2919 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2920 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2921 | #endif |
07c45a36 | 2922 | break; |
d4c9ff2d FEL |
2923 | case KVM_TRACE_ENABLE: |
2924 | case KVM_TRACE_PAUSE: | |
2925 | case KVM_TRACE_DISABLE: | |
2023a29c | 2926 | r = -EOPNOTSUPP; |
d4c9ff2d | 2927 | break; |
6aa8b732 | 2928 | default: |
043405e1 | 2929 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2930 | } |
2931 | out: | |
2932 | return r; | |
2933 | } | |
2934 | ||
6aa8b732 | 2935 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2936 | .unlocked_ioctl = kvm_dev_ioctl, |
2937 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2938 | .llseek = noop_llseek, |
6aa8b732 AK |
2939 | }; |
2940 | ||
2941 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2942 | KVM_MINOR, |
6aa8b732 AK |
2943 | "kvm", |
2944 | &kvm_chardev_ops, | |
2945 | }; | |
2946 | ||
75b7127c | 2947 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2948 | { |
2949 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2950 | int r; |
1b6c0168 | 2951 | |
7f59f492 | 2952 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2953 | return; |
10474ae8 | 2954 | |
7f59f492 | 2955 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 2956 | |
13a34e06 | 2957 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
2958 | |
2959 | if (r) { | |
2960 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2961 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 2962 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 2963 | } |
1b6c0168 AK |
2964 | } |
2965 | ||
4fa92fb2 | 2966 | static void hardware_enable(void) |
75b7127c | 2967 | { |
4a937f96 | 2968 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2969 | if (kvm_usage_count) |
2970 | hardware_enable_nolock(NULL); | |
4a937f96 | 2971 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2972 | } |
2973 | ||
2974 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2975 | { |
2976 | int cpu = raw_smp_processor_id(); | |
2977 | ||
7f59f492 | 2978 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2979 | return; |
7f59f492 | 2980 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 2981 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
2982 | } |
2983 | ||
4fa92fb2 | 2984 | static void hardware_disable(void) |
75b7127c | 2985 | { |
4a937f96 | 2986 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2987 | if (kvm_usage_count) |
2988 | hardware_disable_nolock(NULL); | |
4a937f96 | 2989 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2990 | } |
2991 | ||
10474ae8 AG |
2992 | static void hardware_disable_all_nolock(void) |
2993 | { | |
2994 | BUG_ON(!kvm_usage_count); | |
2995 | ||
2996 | kvm_usage_count--; | |
2997 | if (!kvm_usage_count) | |
75b7127c | 2998 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2999 | } |
3000 | ||
3001 | static void hardware_disable_all(void) | |
3002 | { | |
4a937f96 | 3003 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3004 | hardware_disable_all_nolock(); |
4a937f96 | 3005 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3006 | } |
3007 | ||
3008 | static int hardware_enable_all(void) | |
3009 | { | |
3010 | int r = 0; | |
3011 | ||
4a937f96 | 3012 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3013 | |
3014 | kvm_usage_count++; | |
3015 | if (kvm_usage_count == 1) { | |
3016 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3017 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3018 | |
3019 | if (atomic_read(&hardware_enable_failed)) { | |
3020 | hardware_disable_all_nolock(); | |
3021 | r = -EBUSY; | |
3022 | } | |
3023 | } | |
3024 | ||
4a937f96 | 3025 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3026 | |
3027 | return r; | |
3028 | } | |
3029 | ||
774c47f1 AK |
3030 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
3031 | void *v) | |
3032 | { | |
1a6f4d7f | 3033 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 3034 | switch (val) { |
cec9ad27 | 3035 | case CPU_DYING: |
4fa92fb2 | 3036 | hardware_disable(); |
6ec8a856 | 3037 | break; |
da908f2f | 3038 | case CPU_STARTING: |
4fa92fb2 | 3039 | hardware_enable(); |
774c47f1 AK |
3040 | break; |
3041 | } | |
3042 | return NOTIFY_OK; | |
3043 | } | |
3044 | ||
9a2b85c6 | 3045 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3046 | void *v) |
9a2b85c6 | 3047 | { |
8e1c1815 SY |
3048 | /* |
3049 | * Some (well, at least mine) BIOSes hang on reboot if | |
3050 | * in vmx root mode. | |
3051 | * | |
3052 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3053 | */ | |
1170adc6 | 3054 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3055 | kvm_rebooting = true; |
75b7127c | 3056 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3057 | return NOTIFY_OK; |
3058 | } | |
3059 | ||
3060 | static struct notifier_block kvm_reboot_notifier = { | |
3061 | .notifier_call = kvm_reboot, | |
3062 | .priority = 0, | |
3063 | }; | |
3064 | ||
e93f8a0f | 3065 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3066 | { |
3067 | int i; | |
3068 | ||
3069 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3070 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3071 | |
3072 | kvm_iodevice_destructor(pos); | |
3073 | } | |
e93f8a0f | 3074 | kfree(bus); |
2eeb2e94 GH |
3075 | } |
3076 | ||
c21fbff1 | 3077 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3078 | const struct kvm_io_range *r2) |
743eeb0b | 3079 | { |
743eeb0b SL |
3080 | if (r1->addr < r2->addr) |
3081 | return -1; | |
3082 | if (r1->addr + r1->len > r2->addr + r2->len) | |
3083 | return 1; | |
3084 | return 0; | |
3085 | } | |
3086 | ||
a343c9b7 PB |
3087 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3088 | { | |
c21fbff1 | 3089 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3090 | } |
3091 | ||
39369f7a | 3092 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
3093 | gpa_t addr, int len) |
3094 | { | |
743eeb0b SL |
3095 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
3096 | .addr = addr, | |
3097 | .len = len, | |
3098 | .dev = dev, | |
3099 | }; | |
3100 | ||
3101 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
3102 | kvm_io_bus_sort_cmp, NULL); | |
3103 | ||
3104 | return 0; | |
3105 | } | |
3106 | ||
39369f7a | 3107 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3108 | gpa_t addr, int len) |
3109 | { | |
3110 | struct kvm_io_range *range, key; | |
3111 | int off; | |
3112 | ||
3113 | key = (struct kvm_io_range) { | |
3114 | .addr = addr, | |
3115 | .len = len, | |
3116 | }; | |
3117 | ||
3118 | range = bsearch(&key, bus->range, bus->dev_count, | |
3119 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3120 | if (range == NULL) | |
3121 | return -ENOENT; | |
3122 | ||
3123 | off = range - bus->range; | |
3124 | ||
c21fbff1 | 3125 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3126 | off--; |
3127 | ||
3128 | return off; | |
3129 | } | |
3130 | ||
e32edf4f | 3131 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3132 | struct kvm_io_range *range, const void *val) |
3133 | { | |
3134 | int idx; | |
3135 | ||
3136 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3137 | if (idx < 0) | |
3138 | return -EOPNOTSUPP; | |
3139 | ||
3140 | while (idx < bus->dev_count && | |
c21fbff1 | 3141 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3142 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3143 | range->len, val)) |
3144 | return idx; | |
3145 | idx++; | |
3146 | } | |
3147 | ||
3148 | return -EOPNOTSUPP; | |
3149 | } | |
3150 | ||
bda9020e | 3151 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3152 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3153 | int len, const void *val) |
2eeb2e94 | 3154 | { |
90d83dc3 | 3155 | struct kvm_io_bus *bus; |
743eeb0b | 3156 | struct kvm_io_range range; |
126a5af5 | 3157 | int r; |
743eeb0b SL |
3158 | |
3159 | range = (struct kvm_io_range) { | |
3160 | .addr = addr, | |
3161 | .len = len, | |
3162 | }; | |
90d83dc3 | 3163 | |
e32edf4f NN |
3164 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3165 | r = __kvm_io_bus_write(vcpu, bus, &range, val); | |
126a5af5 CH |
3166 | return r < 0 ? r : 0; |
3167 | } | |
3168 | ||
3169 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3170 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3171 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3172 | { |
3173 | struct kvm_io_bus *bus; | |
3174 | struct kvm_io_range range; | |
3175 | ||
3176 | range = (struct kvm_io_range) { | |
3177 | .addr = addr, | |
3178 | .len = len, | |
3179 | }; | |
3180 | ||
e32edf4f | 3181 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
126a5af5 CH |
3182 | |
3183 | /* First try the device referenced by cookie. */ | |
3184 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3185 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3186 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3187 | val)) |
3188 | return cookie; | |
3189 | ||
3190 | /* | |
3191 | * cookie contained garbage; fall back to search and return the | |
3192 | * correct cookie value. | |
3193 | */ | |
e32edf4f | 3194 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3195 | } |
3196 | ||
e32edf4f NN |
3197 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3198 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3199 | { |
3200 | int idx; | |
3201 | ||
3202 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3203 | if (idx < 0) |
3204 | return -EOPNOTSUPP; | |
3205 | ||
3206 | while (idx < bus->dev_count && | |
c21fbff1 | 3207 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3208 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3209 | range->len, val)) |
3210 | return idx; | |
743eeb0b SL |
3211 | idx++; |
3212 | } | |
3213 | ||
bda9020e MT |
3214 | return -EOPNOTSUPP; |
3215 | } | |
68c3b4d1 | 3216 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3217 | |
bda9020e | 3218 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3219 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3220 | int len, void *val) |
bda9020e | 3221 | { |
90d83dc3 | 3222 | struct kvm_io_bus *bus; |
743eeb0b | 3223 | struct kvm_io_range range; |
126a5af5 | 3224 | int r; |
743eeb0b SL |
3225 | |
3226 | range = (struct kvm_io_range) { | |
3227 | .addr = addr, | |
3228 | .len = len, | |
3229 | }; | |
e93f8a0f | 3230 | |
e32edf4f NN |
3231 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3232 | r = __kvm_io_bus_read(vcpu, bus, &range, val); | |
126a5af5 CH |
3233 | return r < 0 ? r : 0; |
3234 | } | |
743eeb0b | 3235 | |
2eeb2e94 | 3236 | |
79fac95e | 3237 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3238 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3239 | int len, struct kvm_io_device *dev) | |
6c474694 | 3240 | { |
e93f8a0f | 3241 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3242 | |
e93f8a0f | 3243 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3244 | /* exclude ioeventfd which is limited by maximum fd */ |
3245 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3246 | return -ENOSPC; |
2eeb2e94 | 3247 | |
a1300716 AK |
3248 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
3249 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
3250 | if (!new_bus) |
3251 | return -ENOMEM; | |
a1300716 AK |
3252 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3253 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3254 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3255 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3256 | synchronize_srcu_expedited(&kvm->srcu); | |
3257 | kfree(bus); | |
090b7aff GH |
3258 | |
3259 | return 0; | |
3260 | } | |
3261 | ||
79fac95e | 3262 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3263 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3264 | struct kvm_io_device *dev) | |
090b7aff | 3265 | { |
e93f8a0f MT |
3266 | int i, r; |
3267 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3268 | |
cdfca7b3 | 3269 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 3270 | r = -ENOENT; |
a1300716 AK |
3271 | for (i = 0; i < bus->dev_count; i++) |
3272 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3273 | r = 0; |
090b7aff GH |
3274 | break; |
3275 | } | |
e93f8a0f | 3276 | |
a1300716 | 3277 | if (r) |
e93f8a0f | 3278 | return r; |
a1300716 AK |
3279 | |
3280 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
3281 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
3282 | if (!new_bus) | |
3283 | return -ENOMEM; | |
3284 | ||
3285 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3286 | new_bus->dev_count--; | |
3287 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3288 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3289 | |
3290 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3291 | synchronize_srcu_expedited(&kvm->srcu); | |
3292 | kfree(bus); | |
3293 | return r; | |
2eeb2e94 GH |
3294 | } |
3295 | ||
774c47f1 AK |
3296 | static struct notifier_block kvm_cpu_notifier = { |
3297 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3298 | }; |
3299 | ||
8b88b099 | 3300 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3301 | { |
3302 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3303 | struct kvm *kvm; |
3304 | ||
8b88b099 | 3305 | *val = 0; |
2f303b74 | 3306 | spin_lock(&kvm_lock); |
ba1389b7 | 3307 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3308 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3309 | spin_unlock(&kvm_lock); |
8b88b099 | 3310 | return 0; |
ba1389b7 AK |
3311 | } |
3312 | ||
3313 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3314 | ||
8b88b099 | 3315 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3316 | { |
3317 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3318 | struct kvm *kvm; |
3319 | struct kvm_vcpu *vcpu; | |
3320 | int i; | |
3321 | ||
8b88b099 | 3322 | *val = 0; |
2f303b74 | 3323 | spin_lock(&kvm_lock); |
1165f5fe | 3324 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3325 | kvm_for_each_vcpu(i, vcpu, kvm) |
3326 | *val += *(u32 *)((void *)vcpu + offset); | |
3327 | ||
2f303b74 | 3328 | spin_unlock(&kvm_lock); |
8b88b099 | 3329 | return 0; |
1165f5fe AK |
3330 | } |
3331 | ||
ba1389b7 AK |
3332 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3333 | ||
828c0950 | 3334 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3335 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3336 | [KVM_STAT_VM] = &vm_stat_fops, | |
3337 | }; | |
1165f5fe | 3338 | |
4f69b680 | 3339 | static int kvm_init_debug(void) |
6aa8b732 | 3340 | { |
0c8eb04a | 3341 | int r = -EEXIST; |
6aa8b732 AK |
3342 | struct kvm_stats_debugfs_item *p; |
3343 | ||
76f7c879 | 3344 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3345 | if (kvm_debugfs_dir == NULL) |
3346 | goto out; | |
3347 | ||
3348 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3349 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3350 | (void *)(long)p->offset, |
ba1389b7 | 3351 | stat_fops[p->kind]); |
4f69b680 H |
3352 | if (p->dentry == NULL) |
3353 | goto out_dir; | |
3354 | } | |
3355 | ||
3356 | return 0; | |
3357 | ||
3358 | out_dir: | |
3359 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3360 | out: | |
3361 | return r; | |
6aa8b732 AK |
3362 | } |
3363 | ||
3364 | static void kvm_exit_debug(void) | |
3365 | { | |
3366 | struct kvm_stats_debugfs_item *p; | |
3367 | ||
3368 | for (p = debugfs_entries; p->name; ++p) | |
3369 | debugfs_remove(p->dentry); | |
76f7c879 | 3370 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3371 | } |
3372 | ||
fb3600cc | 3373 | static int kvm_suspend(void) |
59ae6c6b | 3374 | { |
10474ae8 | 3375 | if (kvm_usage_count) |
75b7127c | 3376 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3377 | return 0; |
3378 | } | |
3379 | ||
fb3600cc | 3380 | static void kvm_resume(void) |
59ae6c6b | 3381 | { |
ca84d1a2 | 3382 | if (kvm_usage_count) { |
4a937f96 | 3383 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3384 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3385 | } |
59ae6c6b AK |
3386 | } |
3387 | ||
fb3600cc | 3388 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3389 | .suspend = kvm_suspend, |
3390 | .resume = kvm_resume, | |
3391 | }; | |
3392 | ||
15ad7146 AK |
3393 | static inline |
3394 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3395 | { | |
3396 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3397 | } | |
3398 | ||
3399 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3400 | { | |
3401 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 3402 | |
3a08a8f9 R |
3403 | if (vcpu->preempted) |
3404 | vcpu->preempted = false; | |
15ad7146 | 3405 | |
e790d9ef RK |
3406 | kvm_arch_sched_in(vcpu, cpu); |
3407 | ||
e9b11c17 | 3408 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3409 | } |
3410 | ||
3411 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3412 | struct task_struct *next) | |
3413 | { | |
3414 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3415 | ||
3a08a8f9 R |
3416 | if (current->state == TASK_RUNNING) |
3417 | vcpu->preempted = true; | |
e9b11c17 | 3418 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3419 | } |
3420 | ||
0ee75bea | 3421 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3422 | struct module *module) |
6aa8b732 AK |
3423 | { |
3424 | int r; | |
002c7f7c | 3425 | int cpu; |
6aa8b732 | 3426 | |
f8c16bba ZX |
3427 | r = kvm_arch_init(opaque); |
3428 | if (r) | |
d2308784 | 3429 | goto out_fail; |
cb498ea2 | 3430 | |
7dac16c3 AH |
3431 | /* |
3432 | * kvm_arch_init makes sure there's at most one caller | |
3433 | * for architectures that support multiple implementations, | |
3434 | * like intel and amd on x86. | |
3435 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3436 | * conflicts in case kvm is already setup for another implementation. | |
3437 | */ | |
3438 | r = kvm_irqfd_init(); | |
3439 | if (r) | |
3440 | goto out_irqfd; | |
3441 | ||
8437a617 | 3442 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3443 | r = -ENOMEM; |
3444 | goto out_free_0; | |
3445 | } | |
3446 | ||
e9b11c17 | 3447 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3448 | if (r < 0) |
7f59f492 | 3449 | goto out_free_0a; |
6aa8b732 | 3450 | |
002c7f7c YS |
3451 | for_each_online_cpu(cpu) { |
3452 | smp_call_function_single(cpu, | |
e9b11c17 | 3453 | kvm_arch_check_processor_compat, |
8691e5a8 | 3454 | &r, 1); |
002c7f7c | 3455 | if (r < 0) |
d2308784 | 3456 | goto out_free_1; |
002c7f7c YS |
3457 | } |
3458 | ||
774c47f1 AK |
3459 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3460 | if (r) | |
d2308784 | 3461 | goto out_free_2; |
6aa8b732 AK |
3462 | register_reboot_notifier(&kvm_reboot_notifier); |
3463 | ||
c16f862d | 3464 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3465 | if (!vcpu_align) |
3466 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3467 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3468 | 0, NULL); |
c16f862d RR |
3469 | if (!kvm_vcpu_cache) { |
3470 | r = -ENOMEM; | |
fb3600cc | 3471 | goto out_free_3; |
c16f862d RR |
3472 | } |
3473 | ||
af585b92 GN |
3474 | r = kvm_async_pf_init(); |
3475 | if (r) | |
3476 | goto out_free; | |
3477 | ||
6aa8b732 | 3478 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3479 | kvm_vm_fops.owner = module; |
3480 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3481 | |
3482 | r = misc_register(&kvm_dev); | |
3483 | if (r) { | |
1170adc6 | 3484 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 3485 | goto out_unreg; |
6aa8b732 AK |
3486 | } |
3487 | ||
fb3600cc RW |
3488 | register_syscore_ops(&kvm_syscore_ops); |
3489 | ||
15ad7146 AK |
3490 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3491 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3492 | ||
4f69b680 H |
3493 | r = kvm_init_debug(); |
3494 | if (r) { | |
1170adc6 | 3495 | pr_err("kvm: create debugfs files failed\n"); |
4f69b680 H |
3496 | goto out_undebugfs; |
3497 | } | |
0ea4ed8e | 3498 | |
3c3c29fd PB |
3499 | r = kvm_vfio_ops_init(); |
3500 | WARN_ON(r); | |
3501 | ||
c7addb90 | 3502 | return 0; |
6aa8b732 | 3503 | |
4f69b680 H |
3504 | out_undebugfs: |
3505 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3506 | misc_deregister(&kvm_dev); |
af585b92 GN |
3507 | out_unreg: |
3508 | kvm_async_pf_deinit(); | |
6aa8b732 | 3509 | out_free: |
c16f862d | 3510 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3511 | out_free_3: |
6aa8b732 | 3512 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3513 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3514 | out_free_2: |
d2308784 | 3515 | out_free_1: |
e9b11c17 | 3516 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3517 | out_free_0a: |
3518 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3519 | out_free_0: |
a0f155e9 CH |
3520 | kvm_irqfd_exit(); |
3521 | out_irqfd: | |
7dac16c3 AH |
3522 | kvm_arch_exit(); |
3523 | out_fail: | |
6aa8b732 AK |
3524 | return r; |
3525 | } | |
cb498ea2 | 3526 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3527 | |
cb498ea2 | 3528 | void kvm_exit(void) |
6aa8b732 | 3529 | { |
0ea4ed8e | 3530 | kvm_exit_debug(); |
6aa8b732 | 3531 | misc_deregister(&kvm_dev); |
c16f862d | 3532 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3533 | kvm_async_pf_deinit(); |
fb3600cc | 3534 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3535 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3536 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3537 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3538 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3539 | kvm_arch_exit(); |
a0f155e9 | 3540 | kvm_irqfd_exit(); |
7f59f492 | 3541 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 3542 | kvm_vfio_ops_exit(); |
6aa8b732 | 3543 | } |
cb498ea2 | 3544 | EXPORT_SYMBOL_GPL(kvm_exit); |