]>
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; |
19020f8a | 220 | vcpu->halt_poll_ns = 0; |
b6958ce4 | 221 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 222 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
223 | |
224 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
225 | if (!page) { | |
226 | r = -ENOMEM; | |
227 | goto fail; | |
228 | } | |
229 | vcpu->run = page_address(page); | |
230 | ||
4c088493 R |
231 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
232 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 233 | vcpu->preempted = false; |
4c088493 | 234 | |
e9b11c17 | 235 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 236 | if (r < 0) |
e9b11c17 | 237 | goto fail_free_run; |
fb3f0f51 RR |
238 | return 0; |
239 | ||
fb3f0f51 RR |
240 | fail_free_run: |
241 | free_page((unsigned long)vcpu->run); | |
242 | fail: | |
76fafa5e | 243 | return r; |
fb3f0f51 RR |
244 | } |
245 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
246 | ||
247 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
248 | { | |
34bb10b7 | 249 | put_pid(vcpu->pid); |
e9b11c17 | 250 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
251 | free_page((unsigned long)vcpu->run); |
252 | } | |
253 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
254 | ||
e930bffe AA |
255 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
256 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
257 | { | |
258 | return container_of(mn, struct kvm, mmu_notifier); | |
259 | } | |
260 | ||
261 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
262 | struct mm_struct *mm, | |
263 | unsigned long address) | |
264 | { | |
265 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 266 | int need_tlb_flush, idx; |
e930bffe AA |
267 | |
268 | /* | |
269 | * When ->invalidate_page runs, the linux pte has been zapped | |
270 | * already but the page is still allocated until | |
271 | * ->invalidate_page returns. So if we increase the sequence | |
272 | * here the kvm page fault will notice if the spte can't be | |
273 | * established because the page is going to be freed. If | |
274 | * instead the kvm page fault establishes the spte before | |
275 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
276 | * before returning. | |
277 | * | |
278 | * The sequence increase only need to be seen at spin_unlock | |
279 | * time, and not at spin_lock time. | |
280 | * | |
281 | * Increasing the sequence after the spin_unlock would be | |
282 | * unsafe because the kvm page fault could then establish the | |
283 | * pte after kvm_unmap_hva returned, without noticing the page | |
284 | * is going to be freed. | |
285 | */ | |
bc6678a3 | 286 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 287 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 288 | |
e930bffe | 289 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 290 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
291 | /* we've to flush the tlb before the pages can be freed */ |
292 | if (need_tlb_flush) | |
293 | kvm_flush_remote_tlbs(kvm); | |
294 | ||
565f3be2 | 295 | spin_unlock(&kvm->mmu_lock); |
fe71557a TC |
296 | |
297 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); | |
298 | ||
565f3be2 | 299 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
300 | } |
301 | ||
3da0dd43 IE |
302 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
303 | struct mm_struct *mm, | |
304 | unsigned long address, | |
305 | pte_t pte) | |
306 | { | |
307 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 308 | int idx; |
3da0dd43 | 309 | |
bc6678a3 | 310 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
311 | spin_lock(&kvm->mmu_lock); |
312 | kvm->mmu_notifier_seq++; | |
313 | kvm_set_spte_hva(kvm, address, pte); | |
314 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 315 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
316 | } |
317 | ||
e930bffe AA |
318 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
319 | struct mm_struct *mm, | |
320 | unsigned long start, | |
321 | unsigned long end) | |
322 | { | |
323 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 324 | int need_tlb_flush = 0, idx; |
e930bffe | 325 | |
bc6678a3 | 326 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
327 | spin_lock(&kvm->mmu_lock); |
328 | /* | |
329 | * The count increase must become visible at unlock time as no | |
330 | * spte can be established without taking the mmu_lock and | |
331 | * count is also read inside the mmu_lock critical section. | |
332 | */ | |
333 | kvm->mmu_notifier_count++; | |
b3ae2096 | 334 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 335 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
336 | /* we've to flush the tlb before the pages can be freed */ |
337 | if (need_tlb_flush) | |
338 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
339 | |
340 | spin_unlock(&kvm->mmu_lock); | |
341 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
342 | } |
343 | ||
344 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
345 | struct mm_struct *mm, | |
346 | unsigned long start, | |
347 | unsigned long end) | |
348 | { | |
349 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
350 | ||
351 | spin_lock(&kvm->mmu_lock); | |
352 | /* | |
353 | * This sequence increase will notify the kvm page fault that | |
354 | * the page that is going to be mapped in the spte could have | |
355 | * been freed. | |
356 | */ | |
357 | kvm->mmu_notifier_seq++; | |
a355aa54 | 358 | smp_wmb(); |
e930bffe AA |
359 | /* |
360 | * The above sequence increase must be visible before the | |
a355aa54 PM |
361 | * below count decrease, which is ensured by the smp_wmb above |
362 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
363 | */ |
364 | kvm->mmu_notifier_count--; | |
365 | spin_unlock(&kvm->mmu_lock); | |
366 | ||
367 | BUG_ON(kvm->mmu_notifier_count < 0); | |
368 | } | |
369 | ||
370 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
371 | struct mm_struct *mm, | |
57128468 ALC |
372 | unsigned long start, |
373 | unsigned long end) | |
e930bffe AA |
374 | { |
375 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 376 | int young, idx; |
e930bffe | 377 | |
bc6678a3 | 378 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 379 | spin_lock(&kvm->mmu_lock); |
e930bffe | 380 | |
57128468 | 381 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
382 | if (young) |
383 | kvm_flush_remote_tlbs(kvm); | |
384 | ||
565f3be2 TY |
385 | spin_unlock(&kvm->mmu_lock); |
386 | srcu_read_unlock(&kvm->srcu, idx); | |
387 | ||
e930bffe AA |
388 | return young; |
389 | } | |
390 | ||
8ee53820 AA |
391 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
392 | struct mm_struct *mm, | |
393 | unsigned long address) | |
394 | { | |
395 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
396 | int young, idx; | |
397 | ||
398 | idx = srcu_read_lock(&kvm->srcu); | |
399 | spin_lock(&kvm->mmu_lock); | |
400 | young = kvm_test_age_hva(kvm, address); | |
401 | spin_unlock(&kvm->mmu_lock); | |
402 | srcu_read_unlock(&kvm->srcu, idx); | |
403 | ||
404 | return young; | |
405 | } | |
406 | ||
85db06e5 MT |
407 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
408 | struct mm_struct *mm) | |
409 | { | |
410 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
411 | int idx; |
412 | ||
413 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 414 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 415 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
416 | } |
417 | ||
e930bffe AA |
418 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
419 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
420 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
421 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
422 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 423 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 424 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 425 | .release = kvm_mmu_notifier_release, |
e930bffe | 426 | }; |
4c07b0a4 AK |
427 | |
428 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
429 | { | |
430 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
431 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
432 | } | |
433 | ||
434 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
435 | ||
436 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
437 | { | |
438 | return 0; | |
439 | } | |
440 | ||
e930bffe AA |
441 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
442 | ||
a47d2b07 | 443 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
444 | { |
445 | int i; | |
a47d2b07 | 446 | struct kvm_memslots *slots; |
bf3e05bc | 447 | |
a47d2b07 PB |
448 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
449 | if (!slots) | |
450 | return NULL; | |
451 | ||
452 | /* | |
453 | * Init kvm generation close to the maximum to easily test the | |
454 | * code of handling generation number wrap-around. | |
455 | */ | |
456 | slots->generation = -150; | |
bf3e05bc | 457 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 458 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
459 | |
460 | return slots; | |
461 | } | |
462 | ||
463 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
464 | { | |
465 | if (!memslot->dirty_bitmap) | |
466 | return; | |
467 | ||
468 | kvfree(memslot->dirty_bitmap); | |
469 | memslot->dirty_bitmap = NULL; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Free any memory in @free but not in @dont. | |
474 | */ | |
475 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
476 | struct kvm_memory_slot *dont) | |
477 | { | |
478 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
479 | kvm_destroy_dirty_bitmap(free); | |
480 | ||
481 | kvm_arch_free_memslot(kvm, free, dont); | |
482 | ||
483 | free->npages = 0; | |
484 | } | |
485 | ||
486 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
487 | { | |
488 | struct kvm_memory_slot *memslot; | |
489 | ||
490 | if (!slots) | |
491 | return; | |
492 | ||
493 | kvm_for_each_memslot(memslot, slots) | |
494 | kvm_free_memslot(kvm, memslot, NULL); | |
495 | ||
496 | kvfree(slots); | |
bf3e05bc XG |
497 | } |
498 | ||
e08b9637 | 499 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 500 | { |
d89f5eff JK |
501 | int r, i; |
502 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 503 | |
d89f5eff JK |
504 | if (!kvm) |
505 | return ERR_PTR(-ENOMEM); | |
506 | ||
e08b9637 | 507 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 508 | if (r) |
719d93cd | 509 | goto out_err_no_disable; |
10474ae8 AG |
510 | |
511 | r = hardware_enable_all(); | |
512 | if (r) | |
719d93cd | 513 | goto out_err_no_disable; |
10474ae8 | 514 | |
c77dcacb | 515 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 516 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 517 | #endif |
6aa8b732 | 518 | |
1e702d9a AW |
519 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
520 | ||
46a26bf5 | 521 | r = -ENOMEM; |
f481b069 PB |
522 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
523 | kvm->memslots[i] = kvm_alloc_memslots(); | |
524 | if (!kvm->memslots[i]) | |
525 | goto out_err_no_srcu; | |
526 | } | |
00f034a1 | 527 | |
bc6678a3 | 528 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
529 | goto out_err_no_srcu; |
530 | if (init_srcu_struct(&kvm->irq_srcu)) | |
531 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
532 | for (i = 0; i < KVM_NR_BUSES; i++) { |
533 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
534 | GFP_KERNEL); | |
57e7fbee | 535 | if (!kvm->buses[i]) |
e93f8a0f | 536 | goto out_err; |
e93f8a0f | 537 | } |
e930bffe | 538 | |
74b5c5bf | 539 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
540 | kvm->mm = current->mm; |
541 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 542 | kvm_eventfd_init(kvm); |
11ec2804 | 543 | mutex_init(&kvm->lock); |
60eead79 | 544 | mutex_init(&kvm->irq_lock); |
79fac95e | 545 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 546 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 547 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
548 | |
549 | r = kvm_init_mmu_notifier(kvm); | |
550 | if (r) | |
551 | goto out_err; | |
552 | ||
2f303b74 | 553 | spin_lock(&kvm_lock); |
5e58cfe4 | 554 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 555 | spin_unlock(&kvm_lock); |
d89f5eff | 556 | |
2ecd9d29 PZ |
557 | preempt_notifier_inc(); |
558 | ||
f17abe9a | 559 | return kvm; |
10474ae8 AG |
560 | |
561 | out_err: | |
719d93cd CB |
562 | cleanup_srcu_struct(&kvm->irq_srcu); |
563 | out_err_no_irq_srcu: | |
57e7fbee | 564 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 565 | out_err_no_srcu: |
10474ae8 | 566 | hardware_disable_all(); |
719d93cd | 567 | out_err_no_disable: |
e93f8a0f MT |
568 | for (i = 0; i < KVM_NR_BUSES; i++) |
569 | kfree(kvm->buses[i]); | |
f481b069 PB |
570 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
571 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
d89f5eff | 572 | kvm_arch_free_vm(kvm); |
10474ae8 | 573 | return ERR_PTR(r); |
f17abe9a AK |
574 | } |
575 | ||
92eca8fa TY |
576 | /* |
577 | * Avoid using vmalloc for a small buffer. | |
578 | * Should not be used when the size is statically known. | |
579 | */ | |
c1a7b32a | 580 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
581 | { |
582 | if (size > PAGE_SIZE) | |
583 | return vzalloc(size); | |
584 | else | |
585 | return kzalloc(size, GFP_KERNEL); | |
586 | } | |
587 | ||
07f0a7bd SW |
588 | static void kvm_destroy_devices(struct kvm *kvm) |
589 | { | |
590 | struct list_head *node, *tmp; | |
591 | ||
592 | list_for_each_safe(node, tmp, &kvm->devices) { | |
593 | struct kvm_device *dev = | |
594 | list_entry(node, struct kvm_device, vm_node); | |
595 | ||
596 | list_del(node); | |
597 | dev->ops->destroy(dev); | |
598 | } | |
599 | } | |
600 | ||
f17abe9a AK |
601 | static void kvm_destroy_vm(struct kvm *kvm) |
602 | { | |
e93f8a0f | 603 | int i; |
6d4e4c4f AK |
604 | struct mm_struct *mm = kvm->mm; |
605 | ||
ad8ba2cd | 606 | kvm_arch_sync_events(kvm); |
2f303b74 | 607 | spin_lock(&kvm_lock); |
133de902 | 608 | list_del(&kvm->vm_list); |
2f303b74 | 609 | spin_unlock(&kvm_lock); |
399ec807 | 610 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
611 | for (i = 0; i < KVM_NR_BUSES; i++) |
612 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 613 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
614 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
615 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 616 | #else |
2df72e9b | 617 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 618 | #endif |
d19a9cd2 | 619 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 620 | kvm_destroy_devices(kvm); |
f481b069 PB |
621 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
622 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
820b3fcd | 623 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
624 | cleanup_srcu_struct(&kvm->srcu); |
625 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 626 | preempt_notifier_dec(); |
10474ae8 | 627 | hardware_disable_all(); |
6d4e4c4f | 628 | mmdrop(mm); |
f17abe9a AK |
629 | } |
630 | ||
d39f13b0 IE |
631 | void kvm_get_kvm(struct kvm *kvm) |
632 | { | |
633 | atomic_inc(&kvm->users_count); | |
634 | } | |
635 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
636 | ||
637 | void kvm_put_kvm(struct kvm *kvm) | |
638 | { | |
639 | if (atomic_dec_and_test(&kvm->users_count)) | |
640 | kvm_destroy_vm(kvm); | |
641 | } | |
642 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
643 | ||
644 | ||
f17abe9a AK |
645 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
646 | { | |
647 | struct kvm *kvm = filp->private_data; | |
648 | ||
721eecbf GH |
649 | kvm_irqfd_release(kvm); |
650 | ||
d39f13b0 | 651 | kvm_put_kvm(kvm); |
6aa8b732 AK |
652 | return 0; |
653 | } | |
654 | ||
515a0127 TY |
655 | /* |
656 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 657 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 658 | */ |
a36a57b1 TY |
659 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
660 | { | |
515a0127 | 661 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 662 | |
92eca8fa | 663 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
664 | if (!memslot->dirty_bitmap) |
665 | return -ENOMEM; | |
666 | ||
a36a57b1 TY |
667 | return 0; |
668 | } | |
669 | ||
bf3e05bc | 670 | /* |
0e60b079 IM |
671 | * Insert memslot and re-sort memslots based on their GFN, |
672 | * so binary search could be used to lookup GFN. | |
673 | * Sorting algorithm takes advantage of having initially | |
674 | * sorted array and known changed memslot position. | |
bf3e05bc | 675 | */ |
5cc15027 PB |
676 | static void update_memslots(struct kvm_memslots *slots, |
677 | struct kvm_memory_slot *new) | |
bf3e05bc | 678 | { |
8593176c PB |
679 | int id = new->id; |
680 | int i = slots->id_to_index[id]; | |
063584d4 | 681 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 682 | |
8593176c | 683 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 684 | if (!new->npages) { |
dbaff309 | 685 | WARN_ON(!mslots[i].npages); |
9c1a5d38 IM |
686 | if (mslots[i].npages) |
687 | slots->used_slots--; | |
688 | } else { | |
689 | if (!mslots[i].npages) | |
690 | slots->used_slots++; | |
691 | } | |
0e60b079 | 692 | |
7f379cff | 693 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
694 | new->base_gfn <= mslots[i + 1].base_gfn) { |
695 | if (!mslots[i + 1].npages) | |
696 | break; | |
7f379cff IM |
697 | mslots[i] = mslots[i + 1]; |
698 | slots->id_to_index[mslots[i].id] = i; | |
699 | i++; | |
700 | } | |
efbeec70 PB |
701 | |
702 | /* | |
703 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
704 | * so that it moves before all those with base_gfn == npages == 0. | |
705 | * | |
706 | * On the other hand, if new->npages is zero, the above loop has | |
707 | * already left i pointing to the beginning of the empty part of | |
708 | * mslots, and the ">=" would move the hole backwards in this | |
709 | * case---which is wrong. So skip the loop when deleting a slot. | |
710 | */ | |
711 | if (new->npages) { | |
712 | while (i > 0 && | |
713 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
714 | mslots[i] = mslots[i - 1]; | |
715 | slots->id_to_index[mslots[i].id] = i; | |
716 | i--; | |
717 | } | |
dbaff309 PB |
718 | } else |
719 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 720 | |
8593176c PB |
721 | mslots[i] = *new; |
722 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
723 | } |
724 | ||
09170a49 | 725 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 726 | { |
4d8b81ab XG |
727 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
728 | ||
0f8a4de3 | 729 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
730 | valid_flags |= KVM_MEM_READONLY; |
731 | #endif | |
732 | ||
733 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
734 | return -EINVAL; |
735 | ||
736 | return 0; | |
737 | } | |
738 | ||
7ec4fb44 | 739 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 740 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 741 | { |
f481b069 | 742 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
7ec4fb44 | 743 | |
ee3d1570 DM |
744 | /* |
745 | * Set the low bit in the generation, which disables SPTE caching | |
746 | * until the end of synchronize_srcu_expedited. | |
747 | */ | |
748 | WARN_ON(old_memslots->generation & 1); | |
749 | slots->generation = old_memslots->generation + 1; | |
750 | ||
f481b069 | 751 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 752 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 753 | |
ee3d1570 DM |
754 | /* |
755 | * Increment the new memslot generation a second time. This prevents | |
756 | * vm exits that race with memslot updates from caching a memslot | |
757 | * generation that will (potentially) be valid forever. | |
758 | */ | |
759 | slots->generation++; | |
760 | ||
15f46015 | 761 | kvm_arch_memslots_updated(kvm, slots); |
e59dbe09 TY |
762 | |
763 | return old_memslots; | |
7ec4fb44 GN |
764 | } |
765 | ||
6aa8b732 AK |
766 | /* |
767 | * Allocate some memory and give it an address in the guest physical address | |
768 | * space. | |
769 | * | |
770 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 771 | * |
02d5d55b | 772 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 773 | */ |
f78e0e2e | 774 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 775 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 776 | { |
8234b22e | 777 | int r; |
6aa8b732 | 778 | gfn_t base_gfn; |
28bcb112 | 779 | unsigned long npages; |
a843fac2 | 780 | struct kvm_memory_slot *slot; |
6aa8b732 | 781 | struct kvm_memory_slot old, new; |
b7f69c55 | 782 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 783 | int as_id, id; |
f64c0398 | 784 | enum kvm_mr_change change; |
6aa8b732 | 785 | |
a50d64d6 XG |
786 | r = check_memory_region_flags(mem); |
787 | if (r) | |
788 | goto out; | |
789 | ||
6aa8b732 | 790 | r = -EINVAL; |
f481b069 PB |
791 | as_id = mem->slot >> 16; |
792 | id = (u16)mem->slot; | |
793 | ||
6aa8b732 AK |
794 | /* General sanity checks */ |
795 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
796 | goto out; | |
797 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
798 | goto out; | |
fa3d315a | 799 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 800 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 801 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
802 | !access_ok(VERIFY_WRITE, |
803 | (void __user *)(unsigned long)mem->userspace_addr, | |
804 | mem->memory_size))) | |
78749809 | 805 | goto out; |
f481b069 | 806 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
807 | goto out; |
808 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
809 | goto out; | |
810 | ||
f481b069 | 811 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
812 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
813 | npages = mem->memory_size >> PAGE_SHIFT; | |
814 | ||
660c22c4 TY |
815 | if (npages > KVM_MEM_MAX_NR_PAGES) |
816 | goto out; | |
817 | ||
a843fac2 | 818 | new = old = *slot; |
6aa8b732 | 819 | |
f481b069 | 820 | new.id = id; |
6aa8b732 AK |
821 | new.base_gfn = base_gfn; |
822 | new.npages = npages; | |
823 | new.flags = mem->flags; | |
824 | ||
f64c0398 TY |
825 | if (npages) { |
826 | if (!old.npages) | |
827 | change = KVM_MR_CREATE; | |
828 | else { /* Modify an existing slot. */ | |
829 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
830 | (npages != old.npages) || |
831 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
832 | goto out; |
833 | ||
834 | if (base_gfn != old.base_gfn) | |
835 | change = KVM_MR_MOVE; | |
836 | else if (new.flags != old.flags) | |
837 | change = KVM_MR_FLAGS_ONLY; | |
838 | else { /* Nothing to change. */ | |
839 | r = 0; | |
840 | goto out; | |
841 | } | |
842 | } | |
09170a49 PB |
843 | } else { |
844 | if (!old.npages) | |
845 | goto out; | |
846 | ||
f64c0398 | 847 | change = KVM_MR_DELETE; |
09170a49 PB |
848 | new.base_gfn = 0; |
849 | new.flags = 0; | |
850 | } | |
6aa8b732 | 851 | |
f64c0398 | 852 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
853 | /* Check for overlaps */ |
854 | r = -EEXIST; | |
f481b069 | 855 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
a843fac2 | 856 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
f481b069 | 857 | (slot->id == id)) |
0a706bee TY |
858 | continue; |
859 | if (!((base_gfn + npages <= slot->base_gfn) || | |
860 | (base_gfn >= slot->base_gfn + slot->npages))) | |
861 | goto out; | |
862 | } | |
6aa8b732 | 863 | } |
6aa8b732 | 864 | |
6aa8b732 AK |
865 | /* Free page dirty bitmap if unneeded */ |
866 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 867 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
868 | |
869 | r = -ENOMEM; | |
f64c0398 | 870 | if (change == KVM_MR_CREATE) { |
189a2f7b | 871 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 872 | |
5587027c | 873 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 874 | goto out_free; |
6aa8b732 | 875 | } |
ec04b260 | 876 | |
6aa8b732 AK |
877 | /* Allocate page dirty bitmap if needed */ |
878 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 879 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 880 | goto out_free; |
6aa8b732 AK |
881 | } |
882 | ||
74496134 | 883 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
f2a81036 PB |
884 | if (!slots) |
885 | goto out_free; | |
f481b069 | 886 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 887 | |
f64c0398 | 888 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 889 | slot = id_to_memslot(slots, id); |
28a37544 XG |
890 | slot->flags |= KVM_MEMSLOT_INVALID; |
891 | ||
f481b069 | 892 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 893 | |
e40f193f AW |
894 | /* slot was deleted or moved, clear iommu mapping */ |
895 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
896 | /* From this point no new shadow pages pointing to a deleted, |
897 | * or moved, memslot will be created. | |
bc6678a3 MT |
898 | * |
899 | * validation of sp->gfn happens in: | |
b7d409de XL |
900 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
901 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 902 | */ |
2df72e9b | 903 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
904 | |
905 | /* | |
906 | * We can re-use the old_memslots from above, the only difference | |
907 | * from the currently installed memslots is the invalid flag. This | |
908 | * will get overwritten by update_memslots anyway. | |
909 | */ | |
b7f69c55 | 910 | slots = old_memslots; |
bc6678a3 | 911 | } |
34d4cb8f | 912 | |
7b6195a9 | 913 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 914 | if (r) |
b7f69c55 | 915 | goto out_slots; |
f7784b8e | 916 | |
a47d2b07 | 917 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 918 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 919 | new.dirty_bitmap = NULL; |
db3fe4eb | 920 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
921 | } |
922 | ||
5cc15027 | 923 | update_memslots(slots, &new); |
f481b069 | 924 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 925 | |
f36f3f28 | 926 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 927 | |
a47d2b07 | 928 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 929 | kvfree(old_memslots); |
bc6678a3 | 930 | |
261874b0 AW |
931 | /* |
932 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
933 | * un-mapped and re-mapped if their base changes. Since base change |
934 | * unmapping is handled above with slot deletion, mapping alone is | |
935 | * needed here. Anything else the iommu might care about for existing | |
936 | * slots (size changes, userspace addr changes and read-only flag | |
937 | * changes) is disallowed above, so any other attribute changes getting | |
938 | * here can be skipped. | |
261874b0 | 939 | */ |
75d61fbc TY |
940 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
941 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 942 | return r; |
bc6678a3 MT |
943 | } |
944 | ||
6aa8b732 AK |
945 | return 0; |
946 | ||
e40f193f | 947 | out_slots: |
74496134 | 948 | kvfree(slots); |
f78e0e2e | 949 | out_free: |
a47d2b07 | 950 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
951 | out: |
952 | return r; | |
210c7c4d | 953 | } |
f78e0e2e SY |
954 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
955 | ||
956 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 957 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
958 | { |
959 | int r; | |
960 | ||
79fac95e | 961 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 962 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 963 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
964 | return r; |
965 | } | |
210c7c4d IE |
966 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
967 | ||
7940876e SH |
968 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
969 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 970 | { |
f481b069 | 971 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 972 | return -EINVAL; |
09170a49 | 973 | |
47ae31e2 | 974 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
975 | } |
976 | ||
5bb064dc ZX |
977 | int kvm_get_dirty_log(struct kvm *kvm, |
978 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 979 | { |
9f6b8029 | 980 | struct kvm_memslots *slots; |
6aa8b732 | 981 | struct kvm_memory_slot *memslot; |
f481b069 | 982 | int r, i, as_id, id; |
87bf6e7d | 983 | unsigned long n; |
6aa8b732 AK |
984 | unsigned long any = 0; |
985 | ||
6aa8b732 | 986 | r = -EINVAL; |
f481b069 PB |
987 | as_id = log->slot >> 16; |
988 | id = (u16)log->slot; | |
989 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
6aa8b732 AK |
990 | goto out; |
991 | ||
f481b069 PB |
992 | slots = __kvm_memslots(kvm, as_id); |
993 | memslot = id_to_memslot(slots, id); | |
6aa8b732 AK |
994 | r = -ENOENT; |
995 | if (!memslot->dirty_bitmap) | |
996 | goto out; | |
997 | ||
87bf6e7d | 998 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 999 | |
cd1a4a98 | 1000 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1001 | any = memslot->dirty_bitmap[i]; |
1002 | ||
1003 | r = -EFAULT; | |
1004 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
1005 | goto out; | |
1006 | ||
5bb064dc ZX |
1007 | if (any) |
1008 | *is_dirty = 1; | |
6aa8b732 AK |
1009 | |
1010 | r = 0; | |
6aa8b732 | 1011 | out: |
6aa8b732 AK |
1012 | return r; |
1013 | } | |
2ba9f0d8 | 1014 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1015 | |
ba0513b5 MS |
1016 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1017 | /** | |
1018 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1019 | * are dirty write protect them for next write. | |
1020 | * @kvm: pointer to kvm instance | |
1021 | * @log: slot id and address to which we copy the log | |
1022 | * @is_dirty: flag set if any page is dirty | |
1023 | * | |
1024 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1025 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1026 | * following order: | |
1027 | * | |
1028 | * 1. Take a snapshot of the bit and clear it if needed. | |
1029 | * 2. Write protect the corresponding page. | |
1030 | * 3. Copy the snapshot to the userspace. | |
1031 | * 4. Upon return caller flushes TLB's if needed. | |
1032 | * | |
1033 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1034 | * entry. This is not a problem because the page is reported dirty using | |
1035 | * the snapshot taken before and step 4 ensures that writes done after | |
1036 | * exiting to userspace will be logged for the next call. | |
1037 | * | |
1038 | */ | |
1039 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1040 | struct kvm_dirty_log *log, bool *is_dirty) | |
1041 | { | |
9f6b8029 | 1042 | struct kvm_memslots *slots; |
ba0513b5 | 1043 | struct kvm_memory_slot *memslot; |
f481b069 | 1044 | int r, i, as_id, id; |
ba0513b5 MS |
1045 | unsigned long n; |
1046 | unsigned long *dirty_bitmap; | |
1047 | unsigned long *dirty_bitmap_buffer; | |
1048 | ||
1049 | r = -EINVAL; | |
f481b069 PB |
1050 | as_id = log->slot >> 16; |
1051 | id = (u16)log->slot; | |
1052 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
ba0513b5 MS |
1053 | goto out; |
1054 | ||
f481b069 PB |
1055 | slots = __kvm_memslots(kvm, as_id); |
1056 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1057 | |
1058 | dirty_bitmap = memslot->dirty_bitmap; | |
1059 | r = -ENOENT; | |
1060 | if (!dirty_bitmap) | |
1061 | goto out; | |
1062 | ||
1063 | n = kvm_dirty_bitmap_bytes(memslot); | |
1064 | ||
1065 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1066 | memset(dirty_bitmap_buffer, 0, n); | |
1067 | ||
1068 | spin_lock(&kvm->mmu_lock); | |
1069 | *is_dirty = false; | |
1070 | for (i = 0; i < n / sizeof(long); i++) { | |
1071 | unsigned long mask; | |
1072 | gfn_t offset; | |
1073 | ||
1074 | if (!dirty_bitmap[i]) | |
1075 | continue; | |
1076 | ||
1077 | *is_dirty = true; | |
1078 | ||
1079 | mask = xchg(&dirty_bitmap[i], 0); | |
1080 | dirty_bitmap_buffer[i] = mask; | |
1081 | ||
58d2930f TY |
1082 | if (mask) { |
1083 | offset = i * BITS_PER_LONG; | |
1084 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1085 | offset, mask); | |
1086 | } | |
ba0513b5 MS |
1087 | } |
1088 | ||
1089 | spin_unlock(&kvm->mmu_lock); | |
1090 | ||
1091 | r = -EFAULT; | |
1092 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1093 | goto out; | |
1094 | ||
1095 | r = 0; | |
1096 | out: | |
1097 | return r; | |
1098 | } | |
1099 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1100 | #endif | |
1101 | ||
db3fe4eb TY |
1102 | bool kvm_largepages_enabled(void) |
1103 | { | |
1104 | return largepages_enabled; | |
1105 | } | |
1106 | ||
54dee993 MT |
1107 | void kvm_disable_largepages(void) |
1108 | { | |
1109 | largepages_enabled = false; | |
1110 | } | |
1111 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1112 | ||
49c7754c GN |
1113 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1114 | { | |
1115 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1116 | } | |
a1f4d395 | 1117 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1118 | |
8e73485c PB |
1119 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1120 | { | |
1121 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1122 | } | |
1123 | ||
e0d62c7f IE |
1124 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
1125 | { | |
bf3e05bc | 1126 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1127 | |
bbacc0c1 | 1128 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
1129 | memslot->flags & KVM_MEMSLOT_INVALID) |
1130 | return 0; | |
e0d62c7f | 1131 | |
bf3e05bc | 1132 | return 1; |
e0d62c7f IE |
1133 | } |
1134 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1135 | ||
8f0b1ab6 JR |
1136 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1137 | { | |
1138 | struct vm_area_struct *vma; | |
1139 | unsigned long addr, size; | |
1140 | ||
1141 | size = PAGE_SIZE; | |
1142 | ||
1143 | addr = gfn_to_hva(kvm, gfn); | |
1144 | if (kvm_is_error_hva(addr)) | |
1145 | return PAGE_SIZE; | |
1146 | ||
1147 | down_read(¤t->mm->mmap_sem); | |
1148 | vma = find_vma(current->mm, addr); | |
1149 | if (!vma) | |
1150 | goto out; | |
1151 | ||
1152 | size = vma_kernel_pagesize(vma); | |
1153 | ||
1154 | out: | |
1155 | up_read(¤t->mm->mmap_sem); | |
1156 | ||
1157 | return size; | |
1158 | } | |
1159 | ||
4d8b81ab XG |
1160 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1161 | { | |
1162 | return slot->flags & KVM_MEM_READONLY; | |
1163 | } | |
1164 | ||
4d8b81ab XG |
1165 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1166 | gfn_t *nr_pages, bool write) | |
539cb660 | 1167 | { |
bc6678a3 | 1168 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1169 | return KVM_HVA_ERR_BAD; |
48987781 | 1170 | |
4d8b81ab XG |
1171 | if (memslot_is_readonly(slot) && write) |
1172 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1173 | |
1174 | if (nr_pages) | |
1175 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1176 | ||
4d8b81ab | 1177 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1178 | } |
48987781 | 1179 | |
4d8b81ab XG |
1180 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1181 | gfn_t *nr_pages) | |
1182 | { | |
1183 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1184 | } |
48987781 | 1185 | |
4d8b81ab | 1186 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1187 | gfn_t gfn) |
4d8b81ab XG |
1188 | { |
1189 | return gfn_to_hva_many(slot, gfn, NULL); | |
1190 | } | |
1191 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1192 | ||
48987781 XG |
1193 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1194 | { | |
49c7754c | 1195 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1196 | } |
0d150298 | 1197 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1198 | |
8e73485c PB |
1199 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1200 | { | |
1201 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1202 | } | |
1203 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1204 | ||
86ab8cff | 1205 | /* |
ba6a3541 PB |
1206 | * If writable is set to false, the hva returned by this function is only |
1207 | * allowed to be read. | |
86ab8cff | 1208 | */ |
64d83126 CD |
1209 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1210 | gfn_t gfn, bool *writable) | |
86ab8cff | 1211 | { |
a2ac07fe GN |
1212 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1213 | ||
1214 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1215 | *writable = !memslot_is_readonly(slot); |
1216 | ||
a2ac07fe | 1217 | return hva; |
86ab8cff XG |
1218 | } |
1219 | ||
64d83126 CD |
1220 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1221 | { | |
1222 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1223 | ||
1224 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1225 | } | |
1226 | ||
8e73485c PB |
1227 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1228 | { | |
1229 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1230 | ||
1231 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1232 | } | |
1233 | ||
39369f7a | 1234 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1235 | unsigned long start, int write, struct page **page) |
1236 | { | |
1237 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1238 | ||
1239 | if (write) | |
1240 | flags |= FOLL_WRITE; | |
1241 | ||
1242 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1243 | } | |
1244 | ||
fafc3dba HY |
1245 | static inline int check_user_page_hwpoison(unsigned long addr) |
1246 | { | |
1247 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1248 | ||
1249 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1250 | flags, NULL, NULL, NULL); | |
1251 | return rc == -EHWPOISON; | |
1252 | } | |
1253 | ||
2fc84311 XG |
1254 | /* |
1255 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1256 | * true indicates success, otherwise false is returned. | |
1257 | */ | |
1258 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1259 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1260 | { |
8d4e1288 | 1261 | struct page *page[1]; |
2fc84311 | 1262 | int npages; |
954bbbc2 | 1263 | |
2fc84311 XG |
1264 | if (!(async || atomic)) |
1265 | return false; | |
af585b92 | 1266 | |
12ce13fe XG |
1267 | /* |
1268 | * Fast pin a writable pfn only if it is a write fault request | |
1269 | * or the caller allows to map a writable pfn for a read fault | |
1270 | * request. | |
1271 | */ | |
1272 | if (!(write_fault || writable)) | |
1273 | return false; | |
612819c3 | 1274 | |
2fc84311 XG |
1275 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1276 | if (npages == 1) { | |
1277 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1278 | |
2fc84311 XG |
1279 | if (writable) |
1280 | *writable = true; | |
1281 | return true; | |
1282 | } | |
af585b92 | 1283 | |
2fc84311 XG |
1284 | return false; |
1285 | } | |
612819c3 | 1286 | |
2fc84311 XG |
1287 | /* |
1288 | * The slow path to get the pfn of the specified host virtual address, | |
1289 | * 1 indicates success, -errno is returned if error is detected. | |
1290 | */ | |
1291 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1292 | bool *writable, pfn_t *pfn) | |
1293 | { | |
1294 | struct page *page[1]; | |
1295 | int npages = 0; | |
612819c3 | 1296 | |
2fc84311 XG |
1297 | might_sleep(); |
1298 | ||
1299 | if (writable) | |
1300 | *writable = write_fault; | |
1301 | ||
1302 | if (async) { | |
1303 | down_read(¤t->mm->mmap_sem); | |
1304 | npages = get_user_page_nowait(current, current->mm, | |
1305 | addr, write_fault, page); | |
1306 | up_read(¤t->mm->mmap_sem); | |
0664e57f AA |
1307 | } else |
1308 | npages = __get_user_pages_unlocked(current, current->mm, addr, 1, | |
1309 | write_fault, 0, page, | |
1310 | FOLL_TOUCH|FOLL_HWPOISON); | |
2fc84311 XG |
1311 | if (npages != 1) |
1312 | return npages; | |
1313 | ||
1314 | /* map read fault as writable if possible */ | |
12ce13fe | 1315 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1316 | struct page *wpage[1]; |
1317 | ||
1318 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1319 | if (npages == 1) { | |
1320 | *writable = true; | |
1321 | put_page(page[0]); | |
1322 | page[0] = wpage[0]; | |
612819c3 | 1323 | } |
2fc84311 XG |
1324 | |
1325 | npages = 1; | |
887c08ac | 1326 | } |
2fc84311 XG |
1327 | *pfn = page_to_pfn(page[0]); |
1328 | return npages; | |
1329 | } | |
539cb660 | 1330 | |
4d8b81ab XG |
1331 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1332 | { | |
1333 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1334 | return false; | |
2e2e3738 | 1335 | |
4d8b81ab XG |
1336 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1337 | return false; | |
887c08ac | 1338 | |
4d8b81ab XG |
1339 | return true; |
1340 | } | |
bf998156 | 1341 | |
12ce13fe XG |
1342 | /* |
1343 | * Pin guest page in memory and return its pfn. | |
1344 | * @addr: host virtual address which maps memory to the guest | |
1345 | * @atomic: whether this function can sleep | |
1346 | * @async: whether this function need to wait IO complete if the | |
1347 | * host page is not in the memory | |
1348 | * @write_fault: whether we should get a writable host page | |
1349 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1350 | * | |
1351 | * The function will map a writable host page for these two cases: | |
1352 | * 1): @write_fault = true | |
1353 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1354 | * whether the mapping is writable. | |
1355 | */ | |
2fc84311 XG |
1356 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1357 | bool write_fault, bool *writable) | |
1358 | { | |
1359 | struct vm_area_struct *vma; | |
1360 | pfn_t pfn = 0; | |
1361 | int npages; | |
2e2e3738 | 1362 | |
2fc84311 XG |
1363 | /* we can do it either atomically or asynchronously, not both */ |
1364 | BUG_ON(atomic && async); | |
8d4e1288 | 1365 | |
2fc84311 XG |
1366 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1367 | return pfn; | |
1368 | ||
1369 | if (atomic) | |
1370 | return KVM_PFN_ERR_FAULT; | |
1371 | ||
1372 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1373 | if (npages == 1) | |
1374 | return pfn; | |
8d4e1288 | 1375 | |
2fc84311 XG |
1376 | down_read(¤t->mm->mmap_sem); |
1377 | if (npages == -EHWPOISON || | |
1378 | (!async && check_user_page_hwpoison(addr))) { | |
1379 | pfn = KVM_PFN_ERR_HWPOISON; | |
1380 | goto exit; | |
1381 | } | |
1382 | ||
1383 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1384 | ||
1385 | if (vma == NULL) | |
1386 | pfn = KVM_PFN_ERR_FAULT; | |
1387 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1388 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1389 | vma->vm_pgoff; | |
bf4bea8e | 1390 | BUG_ON(!kvm_is_reserved_pfn(pfn)); |
2fc84311 | 1391 | } else { |
4d8b81ab | 1392 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1393 | *async = true; |
1394 | pfn = KVM_PFN_ERR_FAULT; | |
1395 | } | |
1396 | exit: | |
1397 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1398 | return pfn; |
35149e21 AL |
1399 | } |
1400 | ||
3520469d PB |
1401 | pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, |
1402 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1403 | { |
4d8b81ab XG |
1404 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1405 | ||
1406 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1407 | return KVM_PFN_ERR_RO_FAULT; | |
1408 | ||
1409 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1410 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1411 | |
1412 | /* Do not map writable pfn in the readonly memslot. */ | |
1413 | if (writable && memslot_is_readonly(slot)) { | |
1414 | *writable = false; | |
1415 | writable = NULL; | |
1416 | } | |
1417 | ||
1418 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1419 | writable); | |
887c08ac | 1420 | } |
3520469d | 1421 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1422 | |
612819c3 MT |
1423 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1424 | bool *writable) | |
1425 | { | |
e37afc6e PB |
1426 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1427 | write_fault, writable); | |
612819c3 MT |
1428 | } |
1429 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1430 | ||
d5661048 | 1431 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1432 | { |
4d8b81ab | 1433 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1434 | } |
e37afc6e | 1435 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1436 | |
037d92dc | 1437 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1438 | { |
4d8b81ab | 1439 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1440 | } |
037d92dc | 1441 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1442 | |
e37afc6e PB |
1443 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
1444 | { | |
1445 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1446 | } | |
1447 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1448 | ||
8e73485c PB |
1449 | pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
1450 | { | |
1451 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1452 | } | |
1453 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1454 | ||
e37afc6e PB |
1455 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1456 | { | |
1457 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1458 | } | |
1459 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1460 | ||
8e73485c PB |
1461 | pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
1462 | { | |
1463 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1464 | } | |
1465 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1466 | ||
d9ef13c2 PB |
1467 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1468 | struct page **pages, int nr_pages) | |
48987781 XG |
1469 | { |
1470 | unsigned long addr; | |
1471 | gfn_t entry; | |
1472 | ||
d9ef13c2 | 1473 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1474 | if (kvm_is_error_hva(addr)) |
1475 | return -1; | |
1476 | ||
1477 | if (entry < nr_pages) | |
1478 | return 0; | |
1479 | ||
1480 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1481 | } | |
1482 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1483 | ||
a2766325 XG |
1484 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1485 | { | |
81c52c56 | 1486 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1487 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1488 | |
bf4bea8e | 1489 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1490 | WARN_ON(1); |
6cede2e6 | 1491 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1492 | } |
a2766325 XG |
1493 | |
1494 | return pfn_to_page(pfn); | |
1495 | } | |
1496 | ||
35149e21 AL |
1497 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1498 | { | |
2e2e3738 AL |
1499 | pfn_t pfn; |
1500 | ||
1501 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1502 | |
a2766325 | 1503 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1504 | } |
1505 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1506 | ||
8e73485c PB |
1507 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1508 | { | |
1509 | pfn_t pfn; | |
1510 | ||
1511 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1512 | ||
1513 | return kvm_pfn_to_page(pfn); | |
1514 | } | |
1515 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1516 | ||
b4231d61 IE |
1517 | void kvm_release_page_clean(struct page *page) |
1518 | { | |
32cad84f XG |
1519 | WARN_ON(is_error_page(page)); |
1520 | ||
35149e21 | 1521 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1522 | } |
1523 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1524 | ||
35149e21 AL |
1525 | void kvm_release_pfn_clean(pfn_t pfn) |
1526 | { | |
bf4bea8e | 1527 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1528 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1529 | } |
1530 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1531 | ||
b4231d61 | 1532 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1533 | { |
a2766325 XG |
1534 | WARN_ON(is_error_page(page)); |
1535 | ||
35149e21 AL |
1536 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1537 | } | |
1538 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1539 | ||
7940876e | 1540 | static void kvm_release_pfn_dirty(pfn_t pfn) |
35149e21 AL |
1541 | { |
1542 | kvm_set_pfn_dirty(pfn); | |
1543 | kvm_release_pfn_clean(pfn); | |
1544 | } | |
35149e21 AL |
1545 | |
1546 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1547 | { | |
bf4bea8e | 1548 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 | 1549 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1550 | |
2e2e3738 AL |
1551 | if (!PageReserved(page)) |
1552 | SetPageDirty(page); | |
1553 | } | |
8a7ae055 | 1554 | } |
35149e21 AL |
1555 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1556 | ||
1557 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1558 | { | |
bf4bea8e | 1559 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1560 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1561 | } |
1562 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1563 | ||
1564 | void kvm_get_pfn(pfn_t pfn) | |
1565 | { | |
bf4bea8e | 1566 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1567 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1568 | } |
1569 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1570 | |
195aefde IE |
1571 | static int next_segment(unsigned long len, int offset) |
1572 | { | |
1573 | if (len > PAGE_SIZE - offset) | |
1574 | return PAGE_SIZE - offset; | |
1575 | else | |
1576 | return len; | |
1577 | } | |
1578 | ||
8e73485c PB |
1579 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1580 | void *data, int offset, int len) | |
195aefde | 1581 | { |
e0506bcb IE |
1582 | int r; |
1583 | unsigned long addr; | |
195aefde | 1584 | |
8e73485c | 1585 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1586 | if (kvm_is_error_hva(addr)) |
1587 | return -EFAULT; | |
3180a7fc | 1588 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1589 | if (r) |
195aefde | 1590 | return -EFAULT; |
195aefde IE |
1591 | return 0; |
1592 | } | |
8e73485c PB |
1593 | |
1594 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1595 | int len) | |
1596 | { | |
1597 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1598 | ||
1599 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1600 | } | |
195aefde IE |
1601 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1602 | ||
8e73485c PB |
1603 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1604 | int offset, int len) | |
1605 | { | |
1606 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1607 | ||
1608 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1609 | } | |
1610 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1611 | ||
195aefde IE |
1612 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1613 | { | |
1614 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1615 | int seg; | |
1616 | int offset = offset_in_page(gpa); | |
1617 | int ret; | |
1618 | ||
1619 | while ((seg = next_segment(len, offset)) != 0) { | |
1620 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1621 | if (ret < 0) | |
1622 | return ret; | |
1623 | offset = 0; | |
1624 | len -= seg; | |
1625 | data += seg; | |
1626 | ++gfn; | |
1627 | } | |
1628 | return 0; | |
1629 | } | |
1630 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1631 | ||
8e73485c | 1632 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 1633 | { |
7ec54588 | 1634 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 1635 | int seg; |
7ec54588 | 1636 | int offset = offset_in_page(gpa); |
8e73485c PB |
1637 | int ret; |
1638 | ||
1639 | while ((seg = next_segment(len, offset)) != 0) { | |
1640 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
1641 | if (ret < 0) | |
1642 | return ret; | |
1643 | offset = 0; | |
1644 | len -= seg; | |
1645 | data += seg; | |
1646 | ++gfn; | |
1647 | } | |
1648 | return 0; | |
1649 | } | |
1650 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 1651 | |
8e73485c PB |
1652 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1653 | void *data, int offset, unsigned long len) | |
1654 | { | |
1655 | int r; | |
1656 | unsigned long addr; | |
1657 | ||
1658 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
1659 | if (kvm_is_error_hva(addr)) |
1660 | return -EFAULT; | |
0aac03f0 | 1661 | pagefault_disable(); |
3180a7fc | 1662 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1663 | pagefault_enable(); |
7ec54588 MT |
1664 | if (r) |
1665 | return -EFAULT; | |
1666 | return 0; | |
1667 | } | |
7ec54588 | 1668 | |
8e73485c PB |
1669 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1670 | unsigned long len) | |
1671 | { | |
1672 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1673 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1674 | int offset = offset_in_page(gpa); | |
1675 | ||
1676 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1677 | } | |
1678 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
1679 | ||
1680 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
1681 | void *data, unsigned long len) | |
1682 | { | |
1683 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1684 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1685 | int offset = offset_in_page(gpa); | |
1686 | ||
1687 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1688 | } | |
1689 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
1690 | ||
1691 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
1692 | const void *data, int offset, int len) | |
195aefde | 1693 | { |
e0506bcb IE |
1694 | int r; |
1695 | unsigned long addr; | |
195aefde | 1696 | |
251eb841 | 1697 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
1698 | if (kvm_is_error_hva(addr)) |
1699 | return -EFAULT; | |
8b0cedff | 1700 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1701 | if (r) |
195aefde | 1702 | return -EFAULT; |
bc009e43 | 1703 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
1704 | return 0; |
1705 | } | |
8e73485c PB |
1706 | |
1707 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
1708 | const void *data, int offset, int len) | |
1709 | { | |
1710 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1711 | ||
1712 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1713 | } | |
195aefde IE |
1714 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
1715 | ||
8e73485c PB |
1716 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
1717 | const void *data, int offset, int len) | |
1718 | { | |
1719 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1720 | ||
1721 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1722 | } | |
1723 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
1724 | ||
195aefde IE |
1725 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
1726 | unsigned long len) | |
1727 | { | |
1728 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1729 | int seg; | |
1730 | int offset = offset_in_page(gpa); | |
1731 | int ret; | |
1732 | ||
1733 | while ((seg = next_segment(len, offset)) != 0) { | |
1734 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1735 | if (ret < 0) | |
1736 | return ret; | |
1737 | offset = 0; | |
1738 | len -= seg; | |
1739 | data += seg; | |
1740 | ++gfn; | |
1741 | } | |
1742 | return 0; | |
1743 | } | |
ff651cb6 | 1744 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1745 | |
8e73485c PB |
1746 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
1747 | unsigned long len) | |
1748 | { | |
1749 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1750 | int seg; | |
1751 | int offset = offset_in_page(gpa); | |
1752 | int ret; | |
1753 | ||
1754 | while ((seg = next_segment(len, offset)) != 0) { | |
1755 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
1756 | if (ret < 0) | |
1757 | return ret; | |
1758 | offset = 0; | |
1759 | len -= seg; | |
1760 | data += seg; | |
1761 | ++gfn; | |
1762 | } | |
1763 | return 0; | |
1764 | } | |
1765 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
1766 | ||
49c7754c | 1767 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1768 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1769 | { |
1770 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1771 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1772 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1773 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1774 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1775 | gfn_t nr_pages_avail; | |
49c7754c GN |
1776 | |
1777 | ghc->gpa = gpa; | |
1778 | ghc->generation = slots->generation; | |
8f964525 AH |
1779 | ghc->len = len; |
1780 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
ca3f0874 RK |
1781 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
1782 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { | |
49c7754c | 1783 | ghc->hva += offset; |
8f964525 AH |
1784 | } else { |
1785 | /* | |
1786 | * If the requested region crosses two memslots, we still | |
1787 | * verify that the entire region is valid here. | |
1788 | */ | |
1789 | while (start_gfn <= end_gfn) { | |
1790 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1791 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1792 | &nr_pages_avail); | |
1793 | if (kvm_is_error_hva(ghc->hva)) | |
1794 | return -EFAULT; | |
1795 | start_gfn += nr_pages_avail; | |
1796 | } | |
1797 | /* Use the slow path for cross page reads and writes. */ | |
1798 | ghc->memslot = NULL; | |
1799 | } | |
49c7754c GN |
1800 | return 0; |
1801 | } | |
1802 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1803 | ||
1804 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1805 | void *data, unsigned long len) | |
1806 | { | |
1807 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1808 | int r; | |
1809 | ||
8f964525 AH |
1810 | BUG_ON(len > ghc->len); |
1811 | ||
49c7754c | 1812 | if (slots->generation != ghc->generation) |
8f964525 AH |
1813 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1814 | ||
1815 | if (unlikely(!ghc->memslot)) | |
1816 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1817 | |
1818 | if (kvm_is_error_hva(ghc->hva)) | |
1819 | return -EFAULT; | |
1820 | ||
8b0cedff | 1821 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1822 | if (r) |
1823 | return -EFAULT; | |
bc009e43 | 1824 | mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT); |
49c7754c GN |
1825 | |
1826 | return 0; | |
1827 | } | |
1828 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1829 | ||
e03b644f GN |
1830 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1831 | void *data, unsigned long len) | |
1832 | { | |
1833 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1834 | int r; | |
1835 | ||
8f964525 AH |
1836 | BUG_ON(len > ghc->len); |
1837 | ||
e03b644f | 1838 | if (slots->generation != ghc->generation) |
8f964525 AH |
1839 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1840 | ||
1841 | if (unlikely(!ghc->memslot)) | |
1842 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1843 | |
1844 | if (kvm_is_error_hva(ghc->hva)) | |
1845 | return -EFAULT; | |
1846 | ||
1847 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1848 | if (r) | |
1849 | return -EFAULT; | |
1850 | ||
1851 | return 0; | |
1852 | } | |
1853 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1854 | ||
195aefde IE |
1855 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1856 | { | |
8a3caa6d HC |
1857 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
1858 | ||
1859 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
1860 | } |
1861 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1862 | ||
1863 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1864 | { | |
1865 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1866 | int seg; | |
1867 | int offset = offset_in_page(gpa); | |
1868 | int ret; | |
1869 | ||
bfda0e84 | 1870 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
1871 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
1872 | if (ret < 0) | |
1873 | return ret; | |
1874 | offset = 0; | |
1875 | len -= seg; | |
1876 | ++gfn; | |
1877 | } | |
1878 | return 0; | |
1879 | } | |
1880 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1881 | ||
bc009e43 | 1882 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 1883 | gfn_t gfn) |
6aa8b732 | 1884 | { |
7e9d619d RR |
1885 | if (memslot && memslot->dirty_bitmap) { |
1886 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1887 | |
b74ca3b3 | 1888 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1889 | } |
1890 | } | |
1891 | ||
49c7754c GN |
1892 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1893 | { | |
1894 | struct kvm_memory_slot *memslot; | |
1895 | ||
1896 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 1897 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 1898 | } |
2ba9f0d8 | 1899 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 1900 | |
8e73485c PB |
1901 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
1902 | { | |
1903 | struct kvm_memory_slot *memslot; | |
1904 | ||
1905 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1906 | mark_page_dirty_in_slot(memslot, gfn); | |
1907 | } | |
1908 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
1909 | ||
f7819512 PB |
1910 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
1911 | { | |
1912 | if (kvm_arch_vcpu_runnable(vcpu)) { | |
1913 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
1914 | return -EINTR; | |
1915 | } | |
1916 | if (kvm_cpu_has_pending_timer(vcpu)) | |
1917 | return -EINTR; | |
1918 | if (signal_pending(current)) | |
1919 | return -EINTR; | |
1920 | ||
1921 | return 0; | |
1922 | } | |
1923 | ||
b6958ce4 ED |
1924 | /* |
1925 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1926 | */ | |
8776e519 | 1927 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1928 | { |
f7819512 | 1929 | ktime_t start, cur; |
e5c239cf | 1930 | DEFINE_WAIT(wait); |
f7819512 PB |
1931 | bool waited = false; |
1932 | ||
1933 | start = cur = ktime_get(); | |
19020f8a WL |
1934 | if (vcpu->halt_poll_ns) { |
1935 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); | |
f95ef0cd | 1936 | |
f7819512 PB |
1937 | do { |
1938 | /* | |
1939 | * This sets KVM_REQ_UNHALT if an interrupt | |
1940 | * arrives. | |
1941 | */ | |
1942 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
1943 | ++vcpu->stat.halt_successful_poll; | |
1944 | goto out; | |
1945 | } | |
1946 | cur = ktime_get(); | |
1947 | } while (single_task_running() && ktime_before(cur, stop)); | |
1948 | } | |
e5c239cf MT |
1949 | |
1950 | for (;;) { | |
1951 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1952 | ||
f7819512 | 1953 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
1954 | break; |
1955 | ||
f7819512 | 1956 | waited = true; |
b6958ce4 | 1957 | schedule(); |
b6958ce4 | 1958 | } |
d3bef15f | 1959 | |
e5c239cf | 1960 | finish_wait(&vcpu->wq, &wait); |
f7819512 PB |
1961 | cur = ktime_get(); |
1962 | ||
1963 | out: | |
1964 | trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited); | |
b6958ce4 | 1965 | } |
2ba9f0d8 | 1966 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 1967 | |
8c84780d | 1968 | #ifndef CONFIG_S390 |
b6d33834 CD |
1969 | /* |
1970 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1971 | */ | |
1972 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1973 | { | |
1974 | int me; | |
1975 | int cpu = vcpu->cpu; | |
1976 | wait_queue_head_t *wqp; | |
1977 | ||
1978 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1979 | if (waitqueue_active(wqp)) { | |
1980 | wake_up_interruptible(wqp); | |
1981 | ++vcpu->stat.halt_wakeup; | |
1982 | } | |
1983 | ||
1984 | me = get_cpu(); | |
1985 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1986 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1987 | smp_send_reschedule(cpu); | |
1988 | put_cpu(); | |
1989 | } | |
a20ed54d | 1990 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1991 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1992 | |
fa93384f | 1993 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
1994 | { |
1995 | struct pid *pid; | |
1996 | struct task_struct *task = NULL; | |
fa93384f | 1997 | int ret = 0; |
41628d33 KW |
1998 | |
1999 | rcu_read_lock(); | |
2000 | pid = rcu_dereference(target->pid); | |
2001 | if (pid) | |
27fbe64b | 2002 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2003 | rcu_read_unlock(); |
2004 | if (!task) | |
c45c528e | 2005 | return ret; |
c45c528e | 2006 | ret = yield_to(task, 1); |
41628d33 | 2007 | put_task_struct(task); |
c45c528e R |
2008 | |
2009 | return ret; | |
41628d33 KW |
2010 | } |
2011 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2012 | ||
06e48c51 R |
2013 | /* |
2014 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2015 | * Most eligible candidate to yield is decided by following heuristics: | |
2016 | * | |
2017 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2018 | * (preempted lock holder), indicated by @in_spin_loop. | |
2019 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2020 | * | |
2021 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2022 | * chance last time (mostly it has become eligible now since we have probably | |
2023 | * yielded to lockholder in last iteration. This is done by toggling | |
2024 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2025 | * | |
2026 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2027 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2028 | * burning. Giving priority for a potential lock-holder increases lock | |
2029 | * progress. | |
2030 | * | |
2031 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2032 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2033 | * and continue with next VCPU and so on. | |
2034 | */ | |
7940876e | 2035 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2036 | { |
4a55dd72 | 2037 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2038 | bool eligible; |
2039 | ||
2040 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2041 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2042 | |
2043 | if (vcpu->spin_loop.in_spin_loop) | |
2044 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2045 | ||
2046 | return eligible; | |
4a55dd72 SW |
2047 | #else |
2048 | return true; | |
06e48c51 | 2049 | #endif |
4a55dd72 | 2050 | } |
c45c528e | 2051 | |
217ece61 | 2052 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 2053 | { |
217ece61 RR |
2054 | struct kvm *kvm = me->kvm; |
2055 | struct kvm_vcpu *vcpu; | |
2056 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2057 | int yielded = 0; | |
c45c528e | 2058 | int try = 3; |
217ece61 RR |
2059 | int pass; |
2060 | int i; | |
d255f4f2 | 2061 | |
4c088493 | 2062 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2063 | /* |
2064 | * We boost the priority of a VCPU that is runnable but not | |
2065 | * currently running, because it got preempted by something | |
2066 | * else and called schedule in __vcpu_run. Hopefully that | |
2067 | * VCPU is holding the lock that we need and will release it. | |
2068 | * We approximate round-robin by starting at the last boosted VCPU. | |
2069 | */ | |
c45c528e | 2070 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2071 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2072 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2073 | i = last_boosted_vcpu; |
2074 | continue; | |
2075 | } else if (pass && i > last_boosted_vcpu) | |
2076 | break; | |
7bc7ae25 R |
2077 | if (!ACCESS_ONCE(vcpu->preempted)) |
2078 | continue; | |
217ece61 RR |
2079 | if (vcpu == me) |
2080 | continue; | |
98f4a146 | 2081 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 2082 | continue; |
06e48c51 R |
2083 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2084 | continue; | |
c45c528e R |
2085 | |
2086 | yielded = kvm_vcpu_yield_to(vcpu); | |
2087 | if (yielded > 0) { | |
217ece61 | 2088 | kvm->last_boosted_vcpu = i; |
217ece61 | 2089 | break; |
c45c528e R |
2090 | } else if (yielded < 0) { |
2091 | try--; | |
2092 | if (!try) | |
2093 | break; | |
217ece61 | 2094 | } |
217ece61 RR |
2095 | } |
2096 | } | |
4c088493 | 2097 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2098 | |
2099 | /* Ensure vcpu is not eligible during next spinloop */ | |
2100 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2101 | } |
2102 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2103 | ||
e4a533a4 | 2104 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
2105 | { |
2106 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
2107 | struct page *page; |
2108 | ||
e4a533a4 | 2109 | if (vmf->pgoff == 0) |
039576c0 | 2110 | page = virt_to_page(vcpu->run); |
09566765 | 2111 | #ifdef CONFIG_X86 |
e4a533a4 | 2112 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2113 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
2114 | #endif |
2115 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2116 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
2117 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2118 | #endif |
039576c0 | 2119 | else |
5b1c1493 | 2120 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2121 | get_page(page); |
e4a533a4 NP |
2122 | vmf->page = page; |
2123 | return 0; | |
9a2bb7f4 AK |
2124 | } |
2125 | ||
f0f37e2f | 2126 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2127 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2128 | }; |
2129 | ||
2130 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2131 | { | |
2132 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2133 | return 0; | |
2134 | } | |
2135 | ||
bccf2150 AK |
2136 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2137 | { | |
2138 | struct kvm_vcpu *vcpu = filp->private_data; | |
2139 | ||
66c0b394 | 2140 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2141 | return 0; |
2142 | } | |
2143 | ||
3d3aab1b | 2144 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2145 | .release = kvm_vcpu_release, |
2146 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2147 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2148 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2149 | #endif | |
9a2bb7f4 | 2150 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2151 | .llseek = noop_llseek, |
bccf2150 AK |
2152 | }; |
2153 | ||
2154 | /* | |
2155 | * Allocates an inode for the vcpu. | |
2156 | */ | |
2157 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2158 | { | |
24009b05 | 2159 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2160 | } |
2161 | ||
c5ea7660 AK |
2162 | /* |
2163 | * Creates some virtual cpus. Good luck creating more than one. | |
2164 | */ | |
73880c80 | 2165 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2166 | { |
2167 | int r; | |
988a2cae | 2168 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 2169 | |
338c7dba AH |
2170 | if (id >= KVM_MAX_VCPUS) |
2171 | return -EINVAL; | |
2172 | ||
73880c80 | 2173 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
2174 | if (IS_ERR(vcpu)) |
2175 | return PTR_ERR(vcpu); | |
c5ea7660 | 2176 | |
15ad7146 AK |
2177 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2178 | ||
26e5215f AK |
2179 | r = kvm_arch_vcpu_setup(vcpu); |
2180 | if (r) | |
d780592b | 2181 | goto vcpu_destroy; |
26e5215f | 2182 | |
11ec2804 | 2183 | mutex_lock(&kvm->lock); |
3e515705 AK |
2184 | if (!kvm_vcpu_compatible(vcpu)) { |
2185 | r = -EINVAL; | |
2186 | goto unlock_vcpu_destroy; | |
2187 | } | |
73880c80 GN |
2188 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
2189 | r = -EINVAL; | |
d780592b | 2190 | goto unlock_vcpu_destroy; |
fb3f0f51 | 2191 | } |
73880c80 | 2192 | |
988a2cae GN |
2193 | kvm_for_each_vcpu(r, v, kvm) |
2194 | if (v->vcpu_id == id) { | |
73880c80 | 2195 | r = -EEXIST; |
d780592b | 2196 | goto unlock_vcpu_destroy; |
73880c80 GN |
2197 | } |
2198 | ||
2199 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2200 | |
fb3f0f51 | 2201 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2202 | kvm_get_kvm(kvm); |
bccf2150 | 2203 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2204 | if (r < 0) { |
2205 | kvm_put_kvm(kvm); | |
d780592b | 2206 | goto unlock_vcpu_destroy; |
73880c80 GN |
2207 | } |
2208 | ||
2209 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
dd489240 PB |
2210 | |
2211 | /* | |
2212 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2213 | * before kvm->online_vcpu's incremented value. | |
2214 | */ | |
73880c80 GN |
2215 | smp_wmb(); |
2216 | atomic_inc(&kvm->online_vcpus); | |
2217 | ||
73880c80 | 2218 | mutex_unlock(&kvm->lock); |
42897d86 | 2219 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2220 | return r; |
39c3b86e | 2221 | |
d780592b | 2222 | unlock_vcpu_destroy: |
7d8fece6 | 2223 | mutex_unlock(&kvm->lock); |
d780592b | 2224 | vcpu_destroy: |
d40ccc62 | 2225 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
2226 | return r; |
2227 | } | |
2228 | ||
1961d276 AK |
2229 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2230 | { | |
2231 | if (sigset) { | |
2232 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2233 | vcpu->sigset_active = 1; | |
2234 | vcpu->sigset = *sigset; | |
2235 | } else | |
2236 | vcpu->sigset_active = 0; | |
2237 | return 0; | |
2238 | } | |
2239 | ||
bccf2150 AK |
2240 | static long kvm_vcpu_ioctl(struct file *filp, |
2241 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2242 | { |
bccf2150 | 2243 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2244 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2245 | int r; |
fa3795a7 DH |
2246 | struct kvm_fpu *fpu = NULL; |
2247 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2248 | |
6d4e4c4f AK |
2249 | if (vcpu->kvm->mm != current->mm) |
2250 | return -EIO; | |
2122ff5e | 2251 | |
2ea75be3 DM |
2252 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2253 | return -EINVAL; | |
2254 | ||
2f4d9b54 | 2255 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2256 | /* |
2257 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2258 | * so vcpu_load() would break it. | |
2259 | */ | |
47b43c52 | 2260 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
2122ff5e AK |
2261 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
2262 | #endif | |
2263 | ||
2264 | ||
9fc77441 MT |
2265 | r = vcpu_load(vcpu); |
2266 | if (r) | |
2267 | return r; | |
6aa8b732 | 2268 | switch (ioctl) { |
9a2bb7f4 | 2269 | case KVM_RUN: |
f0fe5108 AK |
2270 | r = -EINVAL; |
2271 | if (arg) | |
2272 | goto out; | |
7a72f7a1 CB |
2273 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
2274 | /* The thread running this VCPU changed. */ | |
2275 | struct pid *oldpid = vcpu->pid; | |
2276 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
f95ef0cd | 2277 | |
7a72f7a1 CB |
2278 | rcu_assign_pointer(vcpu->pid, newpid); |
2279 | if (oldpid) | |
2280 | synchronize_rcu(); | |
2281 | put_pid(oldpid); | |
2282 | } | |
b6c7a5dc | 2283 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2284 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2285 | break; |
6aa8b732 | 2286 | case KVM_GET_REGS: { |
3e4bb3ac | 2287 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2288 | |
3e4bb3ac XZ |
2289 | r = -ENOMEM; |
2290 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2291 | if (!kvm_regs) | |
6aa8b732 | 2292 | goto out; |
3e4bb3ac XZ |
2293 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2294 | if (r) | |
2295 | goto out_free1; | |
6aa8b732 | 2296 | r = -EFAULT; |
3e4bb3ac XZ |
2297 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2298 | goto out_free1; | |
6aa8b732 | 2299 | r = 0; |
3e4bb3ac XZ |
2300 | out_free1: |
2301 | kfree(kvm_regs); | |
6aa8b732 AK |
2302 | break; |
2303 | } | |
2304 | case KVM_SET_REGS: { | |
3e4bb3ac | 2305 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2306 | |
3e4bb3ac | 2307 | r = -ENOMEM; |
ff5c2c03 SL |
2308 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2309 | if (IS_ERR(kvm_regs)) { | |
2310 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2311 | goto out; |
ff5c2c03 | 2312 | } |
3e4bb3ac | 2313 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2314 | kfree(kvm_regs); |
6aa8b732 AK |
2315 | break; |
2316 | } | |
2317 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2318 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2319 | r = -ENOMEM; | |
2320 | if (!kvm_sregs) | |
2321 | goto out; | |
2322 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2323 | if (r) |
2324 | goto out; | |
2325 | r = -EFAULT; | |
fa3795a7 | 2326 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2327 | goto out; |
2328 | r = 0; | |
2329 | break; | |
2330 | } | |
2331 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2332 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2333 | if (IS_ERR(kvm_sregs)) { | |
2334 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2335 | kvm_sregs = NULL; |
6aa8b732 | 2336 | goto out; |
ff5c2c03 | 2337 | } |
fa3795a7 | 2338 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2339 | break; |
2340 | } | |
62d9f0db MT |
2341 | case KVM_GET_MP_STATE: { |
2342 | struct kvm_mp_state mp_state; | |
2343 | ||
2344 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2345 | if (r) | |
2346 | goto out; | |
2347 | r = -EFAULT; | |
893bdbf1 | 2348 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2349 | goto out; |
2350 | r = 0; | |
2351 | break; | |
2352 | } | |
2353 | case KVM_SET_MP_STATE: { | |
2354 | struct kvm_mp_state mp_state; | |
2355 | ||
2356 | r = -EFAULT; | |
893bdbf1 | 2357 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2358 | goto out; |
2359 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2360 | break; |
2361 | } | |
6aa8b732 AK |
2362 | case KVM_TRANSLATE: { |
2363 | struct kvm_translation tr; | |
2364 | ||
2365 | r = -EFAULT; | |
893bdbf1 | 2366 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2367 | goto out; |
8b006791 | 2368 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2369 | if (r) |
2370 | goto out; | |
2371 | r = -EFAULT; | |
893bdbf1 | 2372 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2373 | goto out; |
2374 | r = 0; | |
2375 | break; | |
2376 | } | |
d0bfb940 JK |
2377 | case KVM_SET_GUEST_DEBUG: { |
2378 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2379 | |
2380 | r = -EFAULT; | |
893bdbf1 | 2381 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2382 | goto out; |
d0bfb940 | 2383 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2384 | break; |
2385 | } | |
1961d276 AK |
2386 | case KVM_SET_SIGNAL_MASK: { |
2387 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2388 | struct kvm_signal_mask kvm_sigmask; | |
2389 | sigset_t sigset, *p; | |
2390 | ||
2391 | p = NULL; | |
2392 | if (argp) { | |
2393 | r = -EFAULT; | |
2394 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2395 | sizeof(kvm_sigmask))) |
1961d276 AK |
2396 | goto out; |
2397 | r = -EINVAL; | |
893bdbf1 | 2398 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2399 | goto out; |
2400 | r = -EFAULT; | |
2401 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2402 | sizeof(sigset))) |
1961d276 AK |
2403 | goto out; |
2404 | p = &sigset; | |
2405 | } | |
376d41ff | 2406 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2407 | break; |
2408 | } | |
b8836737 | 2409 | case KVM_GET_FPU: { |
fa3795a7 DH |
2410 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2411 | r = -ENOMEM; | |
2412 | if (!fpu) | |
2413 | goto out; | |
2414 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2415 | if (r) |
2416 | goto out; | |
2417 | r = -EFAULT; | |
fa3795a7 | 2418 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2419 | goto out; |
2420 | r = 0; | |
2421 | break; | |
2422 | } | |
2423 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2424 | fpu = memdup_user(argp, sizeof(*fpu)); |
2425 | if (IS_ERR(fpu)) { | |
2426 | r = PTR_ERR(fpu); | |
18595411 | 2427 | fpu = NULL; |
b8836737 | 2428 | goto out; |
ff5c2c03 | 2429 | } |
fa3795a7 | 2430 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2431 | break; |
2432 | } | |
bccf2150 | 2433 | default: |
313a3dc7 | 2434 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2435 | } |
2436 | out: | |
2122ff5e | 2437 | vcpu_put(vcpu); |
fa3795a7 DH |
2438 | kfree(fpu); |
2439 | kfree(kvm_sregs); | |
bccf2150 AK |
2440 | return r; |
2441 | } | |
2442 | ||
de8e5d74 | 2443 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2444 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2445 | unsigned int ioctl, unsigned long arg) | |
2446 | { | |
2447 | struct kvm_vcpu *vcpu = filp->private_data; | |
2448 | void __user *argp = compat_ptr(arg); | |
2449 | int r; | |
2450 | ||
2451 | if (vcpu->kvm->mm != current->mm) | |
2452 | return -EIO; | |
2453 | ||
2454 | switch (ioctl) { | |
2455 | case KVM_SET_SIGNAL_MASK: { | |
2456 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2457 | struct kvm_signal_mask kvm_sigmask; | |
2458 | compat_sigset_t csigset; | |
2459 | sigset_t sigset; | |
2460 | ||
2461 | if (argp) { | |
2462 | r = -EFAULT; | |
2463 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2464 | sizeof(kvm_sigmask))) |
1dda606c AG |
2465 | goto out; |
2466 | r = -EINVAL; | |
893bdbf1 | 2467 | if (kvm_sigmask.len != sizeof(csigset)) |
1dda606c AG |
2468 | goto out; |
2469 | r = -EFAULT; | |
2470 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
893bdbf1 | 2471 | sizeof(csigset))) |
1dda606c | 2472 | goto out; |
760a9a30 AC |
2473 | sigset_from_compat(&sigset, &csigset); |
2474 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2475 | } else | |
2476 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2477 | break; |
2478 | } | |
2479 | default: | |
2480 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2481 | } | |
2482 | ||
2483 | out: | |
2484 | return r; | |
2485 | } | |
2486 | #endif | |
2487 | ||
852b6d57 SW |
2488 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2489 | int (*accessor)(struct kvm_device *dev, | |
2490 | struct kvm_device_attr *attr), | |
2491 | unsigned long arg) | |
2492 | { | |
2493 | struct kvm_device_attr attr; | |
2494 | ||
2495 | if (!accessor) | |
2496 | return -EPERM; | |
2497 | ||
2498 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2499 | return -EFAULT; | |
2500 | ||
2501 | return accessor(dev, &attr); | |
2502 | } | |
2503 | ||
2504 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2505 | unsigned long arg) | |
2506 | { | |
2507 | struct kvm_device *dev = filp->private_data; | |
2508 | ||
2509 | switch (ioctl) { | |
2510 | case KVM_SET_DEVICE_ATTR: | |
2511 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2512 | case KVM_GET_DEVICE_ATTR: | |
2513 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2514 | case KVM_HAS_DEVICE_ATTR: | |
2515 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2516 | default: | |
2517 | if (dev->ops->ioctl) | |
2518 | return dev->ops->ioctl(dev, ioctl, arg); | |
2519 | ||
2520 | return -ENOTTY; | |
2521 | } | |
2522 | } | |
2523 | ||
852b6d57 SW |
2524 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2525 | { | |
2526 | struct kvm_device *dev = filp->private_data; | |
2527 | struct kvm *kvm = dev->kvm; | |
2528 | ||
852b6d57 SW |
2529 | kvm_put_kvm(kvm); |
2530 | return 0; | |
2531 | } | |
2532 | ||
2533 | static const struct file_operations kvm_device_fops = { | |
2534 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2535 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2536 | .compat_ioctl = kvm_device_ioctl, |
2537 | #endif | |
852b6d57 SW |
2538 | .release = kvm_device_release, |
2539 | }; | |
2540 | ||
2541 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2542 | { | |
2543 | if (filp->f_op != &kvm_device_fops) | |
2544 | return NULL; | |
2545 | ||
2546 | return filp->private_data; | |
2547 | } | |
2548 | ||
d60eacb0 | 2549 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2550 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2551 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2552 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2553 | #endif |
d60eacb0 | 2554 | |
5975a2e0 | 2555 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2556 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2557 | #endif |
d60eacb0 WD |
2558 | }; |
2559 | ||
2560 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2561 | { | |
2562 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2563 | return -ENOSPC; | |
2564 | ||
2565 | if (kvm_device_ops_table[type] != NULL) | |
2566 | return -EEXIST; | |
2567 | ||
2568 | kvm_device_ops_table[type] = ops; | |
2569 | return 0; | |
2570 | } | |
2571 | ||
571ee1b6 WL |
2572 | void kvm_unregister_device_ops(u32 type) |
2573 | { | |
2574 | if (kvm_device_ops_table[type] != NULL) | |
2575 | kvm_device_ops_table[type] = NULL; | |
2576 | } | |
2577 | ||
852b6d57 SW |
2578 | static int kvm_ioctl_create_device(struct kvm *kvm, |
2579 | struct kvm_create_device *cd) | |
2580 | { | |
2581 | struct kvm_device_ops *ops = NULL; | |
2582 | struct kvm_device *dev; | |
2583 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2584 | int ret; | |
2585 | ||
d60eacb0 WD |
2586 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
2587 | return -ENODEV; | |
2588 | ||
2589 | ops = kvm_device_ops_table[cd->type]; | |
2590 | if (ops == NULL) | |
852b6d57 | 2591 | return -ENODEV; |
852b6d57 SW |
2592 | |
2593 | if (test) | |
2594 | return 0; | |
2595 | ||
2596 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2597 | if (!dev) | |
2598 | return -ENOMEM; | |
2599 | ||
2600 | dev->ops = ops; | |
2601 | dev->kvm = kvm; | |
852b6d57 SW |
2602 | |
2603 | ret = ops->create(dev, cd->type); | |
2604 | if (ret < 0) { | |
2605 | kfree(dev); | |
2606 | return ret; | |
2607 | } | |
2608 | ||
24009b05 | 2609 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2610 | if (ret < 0) { |
2611 | ops->destroy(dev); | |
2612 | return ret; | |
2613 | } | |
2614 | ||
07f0a7bd | 2615 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2616 | kvm_get_kvm(kvm); |
2617 | cd->fd = ret; | |
2618 | return 0; | |
2619 | } | |
2620 | ||
92b591a4 AG |
2621 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2622 | { | |
2623 | switch (arg) { | |
2624 | case KVM_CAP_USER_MEMORY: | |
2625 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2626 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
2627 | case KVM_CAP_INTERNAL_ERROR_DATA: |
2628 | #ifdef CONFIG_HAVE_KVM_MSI | |
2629 | case KVM_CAP_SIGNAL_MSI: | |
2630 | #endif | |
297e2105 | 2631 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 2632 | case KVM_CAP_IRQFD: |
92b591a4 AG |
2633 | case KVM_CAP_IRQFD_RESAMPLE: |
2634 | #endif | |
2635 | case KVM_CAP_CHECK_EXTENSION_VM: | |
2636 | return 1; | |
2637 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2638 | case KVM_CAP_IRQ_ROUTING: | |
2639 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
2640 | #endif |
2641 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
2642 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
2643 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 AG |
2644 | #endif |
2645 | default: | |
2646 | break; | |
2647 | } | |
2648 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2649 | } | |
2650 | ||
bccf2150 AK |
2651 | static long kvm_vm_ioctl(struct file *filp, |
2652 | unsigned int ioctl, unsigned long arg) | |
2653 | { | |
2654 | struct kvm *kvm = filp->private_data; | |
2655 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2656 | int r; |
bccf2150 | 2657 | |
6d4e4c4f AK |
2658 | if (kvm->mm != current->mm) |
2659 | return -EIO; | |
bccf2150 AK |
2660 | switch (ioctl) { |
2661 | case KVM_CREATE_VCPU: | |
2662 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2663 | break; |
6fc138d2 IE |
2664 | case KVM_SET_USER_MEMORY_REGION: { |
2665 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2666 | ||
2667 | r = -EFAULT; | |
2668 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 2669 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
2670 | goto out; |
2671 | ||
47ae31e2 | 2672 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2673 | break; |
2674 | } | |
2675 | case KVM_GET_DIRTY_LOG: { | |
2676 | struct kvm_dirty_log log; | |
2677 | ||
2678 | r = -EFAULT; | |
893bdbf1 | 2679 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 2680 | goto out; |
2c6f5df9 | 2681 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2682 | break; |
2683 | } | |
5f94c174 LV |
2684 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2685 | case KVM_REGISTER_COALESCED_MMIO: { | |
2686 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2687 | |
5f94c174 | 2688 | r = -EFAULT; |
893bdbf1 | 2689 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2690 | goto out; |
5f94c174 | 2691 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2692 | break; |
2693 | } | |
2694 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2695 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2696 | |
5f94c174 | 2697 | r = -EFAULT; |
893bdbf1 | 2698 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2699 | goto out; |
5f94c174 | 2700 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2701 | break; |
2702 | } | |
2703 | #endif | |
721eecbf GH |
2704 | case KVM_IRQFD: { |
2705 | struct kvm_irqfd data; | |
2706 | ||
2707 | r = -EFAULT; | |
893bdbf1 | 2708 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 2709 | goto out; |
d4db2935 | 2710 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2711 | break; |
2712 | } | |
d34e6b17 GH |
2713 | case KVM_IOEVENTFD: { |
2714 | struct kvm_ioeventfd data; | |
2715 | ||
2716 | r = -EFAULT; | |
893bdbf1 | 2717 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
2718 | goto out; |
2719 | r = kvm_ioeventfd(kvm, &data); | |
2720 | break; | |
2721 | } | |
07975ad3 JK |
2722 | #ifdef CONFIG_HAVE_KVM_MSI |
2723 | case KVM_SIGNAL_MSI: { | |
2724 | struct kvm_msi msi; | |
2725 | ||
2726 | r = -EFAULT; | |
893bdbf1 | 2727 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
2728 | goto out; |
2729 | r = kvm_send_userspace_msi(kvm, &msi); | |
2730 | break; | |
2731 | } | |
23d43cf9 CD |
2732 | #endif |
2733 | #ifdef __KVM_HAVE_IRQ_LINE | |
2734 | case KVM_IRQ_LINE_STATUS: | |
2735 | case KVM_IRQ_LINE: { | |
2736 | struct kvm_irq_level irq_event; | |
2737 | ||
2738 | r = -EFAULT; | |
893bdbf1 | 2739 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
2740 | goto out; |
2741 | ||
aa2fbe6d YZ |
2742 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2743 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2744 | if (r) |
2745 | goto out; | |
2746 | ||
2747 | r = -EFAULT; | |
2748 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 2749 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
2750 | goto out; |
2751 | } | |
2752 | ||
2753 | r = 0; | |
2754 | break; | |
2755 | } | |
73880c80 | 2756 | #endif |
aa8d5944 AG |
2757 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2758 | case KVM_SET_GSI_ROUTING: { | |
2759 | struct kvm_irq_routing routing; | |
2760 | struct kvm_irq_routing __user *urouting; | |
2761 | struct kvm_irq_routing_entry *entries; | |
2762 | ||
2763 | r = -EFAULT; | |
2764 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2765 | goto out; | |
2766 | r = -EINVAL; | |
2767 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2768 | goto out; | |
2769 | if (routing.flags) | |
2770 | goto out; | |
2771 | r = -ENOMEM; | |
2772 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2773 | if (!entries) | |
2774 | goto out; | |
2775 | r = -EFAULT; | |
2776 | urouting = argp; | |
2777 | if (copy_from_user(entries, urouting->entries, | |
2778 | routing.nr * sizeof(*entries))) | |
2779 | goto out_free_irq_routing; | |
2780 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2781 | routing.flags); | |
a642a175 | 2782 | out_free_irq_routing: |
aa8d5944 AG |
2783 | vfree(entries); |
2784 | break; | |
2785 | } | |
2786 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2787 | case KVM_CREATE_DEVICE: { |
2788 | struct kvm_create_device cd; | |
2789 | ||
2790 | r = -EFAULT; | |
2791 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2792 | goto out; | |
2793 | ||
2794 | r = kvm_ioctl_create_device(kvm, &cd); | |
2795 | if (r) | |
2796 | goto out; | |
2797 | ||
2798 | r = -EFAULT; | |
2799 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2800 | goto out; | |
2801 | ||
2802 | r = 0; | |
2803 | break; | |
2804 | } | |
92b591a4 AG |
2805 | case KVM_CHECK_EXTENSION: |
2806 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
2807 | break; | |
f17abe9a | 2808 | default: |
1fe779f8 | 2809 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
2810 | } |
2811 | out: | |
2812 | return r; | |
2813 | } | |
2814 | ||
de8e5d74 | 2815 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2816 | struct compat_kvm_dirty_log { |
2817 | __u32 slot; | |
2818 | __u32 padding1; | |
2819 | union { | |
2820 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2821 | __u64 padding2; | |
2822 | }; | |
2823 | }; | |
2824 | ||
2825 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2826 | unsigned int ioctl, unsigned long arg) | |
2827 | { | |
2828 | struct kvm *kvm = filp->private_data; | |
2829 | int r; | |
2830 | ||
2831 | if (kvm->mm != current->mm) | |
2832 | return -EIO; | |
2833 | switch (ioctl) { | |
2834 | case KVM_GET_DIRTY_LOG: { | |
2835 | struct compat_kvm_dirty_log compat_log; | |
2836 | struct kvm_dirty_log log; | |
2837 | ||
2838 | r = -EFAULT; | |
2839 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2840 | sizeof(compat_log))) | |
2841 | goto out; | |
2842 | log.slot = compat_log.slot; | |
2843 | log.padding1 = compat_log.padding1; | |
2844 | log.padding2 = compat_log.padding2; | |
2845 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2846 | ||
2847 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2848 | break; |
2849 | } | |
2850 | default: | |
2851 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2852 | } | |
2853 | ||
2854 | out: | |
2855 | return r; | |
2856 | } | |
2857 | #endif | |
2858 | ||
3d3aab1b | 2859 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2860 | .release = kvm_vm_release, |
2861 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 2862 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
2863 | .compat_ioctl = kvm_vm_compat_ioctl, |
2864 | #endif | |
6038f373 | 2865 | .llseek = noop_llseek, |
f17abe9a AK |
2866 | }; |
2867 | ||
e08b9637 | 2868 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2869 | { |
aac87636 | 2870 | int r; |
f17abe9a AK |
2871 | struct kvm *kvm; |
2872 | ||
e08b9637 | 2873 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2874 | if (IS_ERR(kvm)) |
2875 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2876 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2877 | r = kvm_coalesced_mmio_init(kvm); | |
2878 | if (r < 0) { | |
2879 | kvm_put_kvm(kvm); | |
2880 | return r; | |
2881 | } | |
2882 | #endif | |
24009b05 | 2883 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2884 | if (r < 0) |
66c0b394 | 2885 | kvm_put_kvm(kvm); |
f17abe9a | 2886 | |
aac87636 | 2887 | return r; |
f17abe9a AK |
2888 | } |
2889 | ||
2890 | static long kvm_dev_ioctl(struct file *filp, | |
2891 | unsigned int ioctl, unsigned long arg) | |
2892 | { | |
07c45a36 | 2893 | long r = -EINVAL; |
f17abe9a AK |
2894 | |
2895 | switch (ioctl) { | |
2896 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2897 | if (arg) |
2898 | goto out; | |
f17abe9a AK |
2899 | r = KVM_API_VERSION; |
2900 | break; | |
2901 | case KVM_CREATE_VM: | |
e08b9637 | 2902 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2903 | break; |
018d00d2 | 2904 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 2905 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 2906 | break; |
07c45a36 | 2907 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
2908 | if (arg) |
2909 | goto out; | |
adb1ff46 AK |
2910 | r = PAGE_SIZE; /* struct kvm_run */ |
2911 | #ifdef CONFIG_X86 | |
2912 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2913 | #endif |
2914 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2915 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2916 | #endif |
07c45a36 | 2917 | break; |
d4c9ff2d FEL |
2918 | case KVM_TRACE_ENABLE: |
2919 | case KVM_TRACE_PAUSE: | |
2920 | case KVM_TRACE_DISABLE: | |
2023a29c | 2921 | r = -EOPNOTSUPP; |
d4c9ff2d | 2922 | break; |
6aa8b732 | 2923 | default: |
043405e1 | 2924 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2925 | } |
2926 | out: | |
2927 | return r; | |
2928 | } | |
2929 | ||
6aa8b732 | 2930 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2931 | .unlocked_ioctl = kvm_dev_ioctl, |
2932 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2933 | .llseek = noop_llseek, |
6aa8b732 AK |
2934 | }; |
2935 | ||
2936 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2937 | KVM_MINOR, |
6aa8b732 AK |
2938 | "kvm", |
2939 | &kvm_chardev_ops, | |
2940 | }; | |
2941 | ||
75b7127c | 2942 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2943 | { |
2944 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2945 | int r; |
1b6c0168 | 2946 | |
7f59f492 | 2947 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2948 | return; |
10474ae8 | 2949 | |
7f59f492 | 2950 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 2951 | |
13a34e06 | 2952 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
2953 | |
2954 | if (r) { | |
2955 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2956 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 2957 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 2958 | } |
1b6c0168 AK |
2959 | } |
2960 | ||
4fa92fb2 | 2961 | static void hardware_enable(void) |
75b7127c | 2962 | { |
4a937f96 | 2963 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2964 | if (kvm_usage_count) |
2965 | hardware_enable_nolock(NULL); | |
4a937f96 | 2966 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2967 | } |
2968 | ||
2969 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2970 | { |
2971 | int cpu = raw_smp_processor_id(); | |
2972 | ||
7f59f492 | 2973 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2974 | return; |
7f59f492 | 2975 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 2976 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
2977 | } |
2978 | ||
4fa92fb2 | 2979 | static void hardware_disable(void) |
75b7127c | 2980 | { |
4a937f96 | 2981 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2982 | if (kvm_usage_count) |
2983 | hardware_disable_nolock(NULL); | |
4a937f96 | 2984 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2985 | } |
2986 | ||
10474ae8 AG |
2987 | static void hardware_disable_all_nolock(void) |
2988 | { | |
2989 | BUG_ON(!kvm_usage_count); | |
2990 | ||
2991 | kvm_usage_count--; | |
2992 | if (!kvm_usage_count) | |
75b7127c | 2993 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2994 | } |
2995 | ||
2996 | static void hardware_disable_all(void) | |
2997 | { | |
4a937f96 | 2998 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 2999 | hardware_disable_all_nolock(); |
4a937f96 | 3000 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3001 | } |
3002 | ||
3003 | static int hardware_enable_all(void) | |
3004 | { | |
3005 | int r = 0; | |
3006 | ||
4a937f96 | 3007 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3008 | |
3009 | kvm_usage_count++; | |
3010 | if (kvm_usage_count == 1) { | |
3011 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3012 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3013 | |
3014 | if (atomic_read(&hardware_enable_failed)) { | |
3015 | hardware_disable_all_nolock(); | |
3016 | r = -EBUSY; | |
3017 | } | |
3018 | } | |
3019 | ||
4a937f96 | 3020 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3021 | |
3022 | return r; | |
3023 | } | |
3024 | ||
774c47f1 AK |
3025 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
3026 | void *v) | |
3027 | { | |
1a6f4d7f | 3028 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 3029 | switch (val) { |
cec9ad27 | 3030 | case CPU_DYING: |
4fa92fb2 | 3031 | hardware_disable(); |
6ec8a856 | 3032 | break; |
da908f2f | 3033 | case CPU_STARTING: |
4fa92fb2 | 3034 | hardware_enable(); |
774c47f1 AK |
3035 | break; |
3036 | } | |
3037 | return NOTIFY_OK; | |
3038 | } | |
3039 | ||
9a2b85c6 | 3040 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3041 | void *v) |
9a2b85c6 | 3042 | { |
8e1c1815 SY |
3043 | /* |
3044 | * Some (well, at least mine) BIOSes hang on reboot if | |
3045 | * in vmx root mode. | |
3046 | * | |
3047 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3048 | */ | |
1170adc6 | 3049 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3050 | kvm_rebooting = true; |
75b7127c | 3051 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3052 | return NOTIFY_OK; |
3053 | } | |
3054 | ||
3055 | static struct notifier_block kvm_reboot_notifier = { | |
3056 | .notifier_call = kvm_reboot, | |
3057 | .priority = 0, | |
3058 | }; | |
3059 | ||
e93f8a0f | 3060 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3061 | { |
3062 | int i; | |
3063 | ||
3064 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3065 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3066 | |
3067 | kvm_iodevice_destructor(pos); | |
3068 | } | |
e93f8a0f | 3069 | kfree(bus); |
2eeb2e94 GH |
3070 | } |
3071 | ||
c21fbff1 | 3072 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3073 | const struct kvm_io_range *r2) |
743eeb0b | 3074 | { |
743eeb0b SL |
3075 | if (r1->addr < r2->addr) |
3076 | return -1; | |
3077 | if (r1->addr + r1->len > r2->addr + r2->len) | |
3078 | return 1; | |
3079 | return 0; | |
3080 | } | |
3081 | ||
a343c9b7 PB |
3082 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3083 | { | |
c21fbff1 | 3084 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3085 | } |
3086 | ||
39369f7a | 3087 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
3088 | gpa_t addr, int len) |
3089 | { | |
743eeb0b SL |
3090 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
3091 | .addr = addr, | |
3092 | .len = len, | |
3093 | .dev = dev, | |
3094 | }; | |
3095 | ||
3096 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
3097 | kvm_io_bus_sort_cmp, NULL); | |
3098 | ||
3099 | return 0; | |
3100 | } | |
3101 | ||
39369f7a | 3102 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3103 | gpa_t addr, int len) |
3104 | { | |
3105 | struct kvm_io_range *range, key; | |
3106 | int off; | |
3107 | ||
3108 | key = (struct kvm_io_range) { | |
3109 | .addr = addr, | |
3110 | .len = len, | |
3111 | }; | |
3112 | ||
3113 | range = bsearch(&key, bus->range, bus->dev_count, | |
3114 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3115 | if (range == NULL) | |
3116 | return -ENOENT; | |
3117 | ||
3118 | off = range - bus->range; | |
3119 | ||
c21fbff1 | 3120 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3121 | off--; |
3122 | ||
3123 | return off; | |
3124 | } | |
3125 | ||
e32edf4f | 3126 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3127 | struct kvm_io_range *range, const void *val) |
3128 | { | |
3129 | int idx; | |
3130 | ||
3131 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3132 | if (idx < 0) | |
3133 | return -EOPNOTSUPP; | |
3134 | ||
3135 | while (idx < bus->dev_count && | |
c21fbff1 | 3136 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3137 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3138 | range->len, val)) |
3139 | return idx; | |
3140 | idx++; | |
3141 | } | |
3142 | ||
3143 | return -EOPNOTSUPP; | |
3144 | } | |
3145 | ||
bda9020e | 3146 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3147 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3148 | int len, const void *val) |
2eeb2e94 | 3149 | { |
90d83dc3 | 3150 | struct kvm_io_bus *bus; |
743eeb0b | 3151 | struct kvm_io_range range; |
126a5af5 | 3152 | int r; |
743eeb0b SL |
3153 | |
3154 | range = (struct kvm_io_range) { | |
3155 | .addr = addr, | |
3156 | .len = len, | |
3157 | }; | |
90d83dc3 | 3158 | |
e32edf4f NN |
3159 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3160 | r = __kvm_io_bus_write(vcpu, bus, &range, val); | |
126a5af5 CH |
3161 | return r < 0 ? r : 0; |
3162 | } | |
3163 | ||
3164 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3165 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3166 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3167 | { |
3168 | struct kvm_io_bus *bus; | |
3169 | struct kvm_io_range range; | |
3170 | ||
3171 | range = (struct kvm_io_range) { | |
3172 | .addr = addr, | |
3173 | .len = len, | |
3174 | }; | |
3175 | ||
e32edf4f | 3176 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
126a5af5 CH |
3177 | |
3178 | /* First try the device referenced by cookie. */ | |
3179 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3180 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3181 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3182 | val)) |
3183 | return cookie; | |
3184 | ||
3185 | /* | |
3186 | * cookie contained garbage; fall back to search and return the | |
3187 | * correct cookie value. | |
3188 | */ | |
e32edf4f | 3189 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3190 | } |
3191 | ||
e32edf4f NN |
3192 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3193 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3194 | { |
3195 | int idx; | |
3196 | ||
3197 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3198 | if (idx < 0) |
3199 | return -EOPNOTSUPP; | |
3200 | ||
3201 | while (idx < bus->dev_count && | |
c21fbff1 | 3202 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3203 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3204 | range->len, val)) |
3205 | return idx; | |
743eeb0b SL |
3206 | idx++; |
3207 | } | |
3208 | ||
bda9020e MT |
3209 | return -EOPNOTSUPP; |
3210 | } | |
68c3b4d1 | 3211 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3212 | |
bda9020e | 3213 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3214 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3215 | int len, void *val) |
bda9020e | 3216 | { |
90d83dc3 | 3217 | struct kvm_io_bus *bus; |
743eeb0b | 3218 | struct kvm_io_range range; |
126a5af5 | 3219 | int r; |
743eeb0b SL |
3220 | |
3221 | range = (struct kvm_io_range) { | |
3222 | .addr = addr, | |
3223 | .len = len, | |
3224 | }; | |
e93f8a0f | 3225 | |
e32edf4f NN |
3226 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3227 | r = __kvm_io_bus_read(vcpu, bus, &range, val); | |
126a5af5 CH |
3228 | return r < 0 ? r : 0; |
3229 | } | |
743eeb0b | 3230 | |
2eeb2e94 | 3231 | |
79fac95e | 3232 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3233 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3234 | int len, struct kvm_io_device *dev) | |
6c474694 | 3235 | { |
e93f8a0f | 3236 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3237 | |
e93f8a0f | 3238 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3239 | /* exclude ioeventfd which is limited by maximum fd */ |
3240 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3241 | return -ENOSPC; |
2eeb2e94 | 3242 | |
a1300716 AK |
3243 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
3244 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
3245 | if (!new_bus) |
3246 | return -ENOMEM; | |
a1300716 AK |
3247 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3248 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3249 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3250 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3251 | synchronize_srcu_expedited(&kvm->srcu); | |
3252 | kfree(bus); | |
090b7aff GH |
3253 | |
3254 | return 0; | |
3255 | } | |
3256 | ||
79fac95e | 3257 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3258 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3259 | struct kvm_io_device *dev) | |
090b7aff | 3260 | { |
e93f8a0f MT |
3261 | int i, r; |
3262 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3263 | |
cdfca7b3 | 3264 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 3265 | r = -ENOENT; |
a1300716 AK |
3266 | for (i = 0; i < bus->dev_count; i++) |
3267 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3268 | r = 0; |
090b7aff GH |
3269 | break; |
3270 | } | |
e93f8a0f | 3271 | |
a1300716 | 3272 | if (r) |
e93f8a0f | 3273 | return r; |
a1300716 AK |
3274 | |
3275 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
3276 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
3277 | if (!new_bus) | |
3278 | return -ENOMEM; | |
3279 | ||
3280 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3281 | new_bus->dev_count--; | |
3282 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3283 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3284 | |
3285 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3286 | synchronize_srcu_expedited(&kvm->srcu); | |
3287 | kfree(bus); | |
3288 | return r; | |
2eeb2e94 GH |
3289 | } |
3290 | ||
774c47f1 AK |
3291 | static struct notifier_block kvm_cpu_notifier = { |
3292 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3293 | }; |
3294 | ||
8b88b099 | 3295 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3296 | { |
3297 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3298 | struct kvm *kvm; |
3299 | ||
8b88b099 | 3300 | *val = 0; |
2f303b74 | 3301 | spin_lock(&kvm_lock); |
ba1389b7 | 3302 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3303 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3304 | spin_unlock(&kvm_lock); |
8b88b099 | 3305 | return 0; |
ba1389b7 AK |
3306 | } |
3307 | ||
3308 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3309 | ||
8b88b099 | 3310 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3311 | { |
3312 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3313 | struct kvm *kvm; |
3314 | struct kvm_vcpu *vcpu; | |
3315 | int i; | |
3316 | ||
8b88b099 | 3317 | *val = 0; |
2f303b74 | 3318 | spin_lock(&kvm_lock); |
1165f5fe | 3319 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3320 | kvm_for_each_vcpu(i, vcpu, kvm) |
3321 | *val += *(u32 *)((void *)vcpu + offset); | |
3322 | ||
2f303b74 | 3323 | spin_unlock(&kvm_lock); |
8b88b099 | 3324 | return 0; |
1165f5fe AK |
3325 | } |
3326 | ||
ba1389b7 AK |
3327 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3328 | ||
828c0950 | 3329 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3330 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3331 | [KVM_STAT_VM] = &vm_stat_fops, | |
3332 | }; | |
1165f5fe | 3333 | |
4f69b680 | 3334 | static int kvm_init_debug(void) |
6aa8b732 | 3335 | { |
0c8eb04a | 3336 | int r = -EEXIST; |
6aa8b732 AK |
3337 | struct kvm_stats_debugfs_item *p; |
3338 | ||
76f7c879 | 3339 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3340 | if (kvm_debugfs_dir == NULL) |
3341 | goto out; | |
3342 | ||
3343 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3344 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3345 | (void *)(long)p->offset, |
ba1389b7 | 3346 | stat_fops[p->kind]); |
4f69b680 H |
3347 | if (p->dentry == NULL) |
3348 | goto out_dir; | |
3349 | } | |
3350 | ||
3351 | return 0; | |
3352 | ||
3353 | out_dir: | |
3354 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3355 | out: | |
3356 | return r; | |
6aa8b732 AK |
3357 | } |
3358 | ||
3359 | static void kvm_exit_debug(void) | |
3360 | { | |
3361 | struct kvm_stats_debugfs_item *p; | |
3362 | ||
3363 | for (p = debugfs_entries; p->name; ++p) | |
3364 | debugfs_remove(p->dentry); | |
76f7c879 | 3365 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3366 | } |
3367 | ||
fb3600cc | 3368 | static int kvm_suspend(void) |
59ae6c6b | 3369 | { |
10474ae8 | 3370 | if (kvm_usage_count) |
75b7127c | 3371 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3372 | return 0; |
3373 | } | |
3374 | ||
fb3600cc | 3375 | static void kvm_resume(void) |
59ae6c6b | 3376 | { |
ca84d1a2 | 3377 | if (kvm_usage_count) { |
4a937f96 | 3378 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3379 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3380 | } |
59ae6c6b AK |
3381 | } |
3382 | ||
fb3600cc | 3383 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3384 | .suspend = kvm_suspend, |
3385 | .resume = kvm_resume, | |
3386 | }; | |
3387 | ||
15ad7146 AK |
3388 | static inline |
3389 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3390 | { | |
3391 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3392 | } | |
3393 | ||
3394 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3395 | { | |
3396 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 3397 | |
3a08a8f9 R |
3398 | if (vcpu->preempted) |
3399 | vcpu->preempted = false; | |
15ad7146 | 3400 | |
e790d9ef RK |
3401 | kvm_arch_sched_in(vcpu, cpu); |
3402 | ||
e9b11c17 | 3403 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3404 | } |
3405 | ||
3406 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3407 | struct task_struct *next) | |
3408 | { | |
3409 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3410 | ||
3a08a8f9 R |
3411 | if (current->state == TASK_RUNNING) |
3412 | vcpu->preempted = true; | |
e9b11c17 | 3413 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3414 | } |
3415 | ||
0ee75bea | 3416 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3417 | struct module *module) |
6aa8b732 AK |
3418 | { |
3419 | int r; | |
002c7f7c | 3420 | int cpu; |
6aa8b732 | 3421 | |
f8c16bba ZX |
3422 | r = kvm_arch_init(opaque); |
3423 | if (r) | |
d2308784 | 3424 | goto out_fail; |
cb498ea2 | 3425 | |
7dac16c3 AH |
3426 | /* |
3427 | * kvm_arch_init makes sure there's at most one caller | |
3428 | * for architectures that support multiple implementations, | |
3429 | * like intel and amd on x86. | |
3430 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3431 | * conflicts in case kvm is already setup for another implementation. | |
3432 | */ | |
3433 | r = kvm_irqfd_init(); | |
3434 | if (r) | |
3435 | goto out_irqfd; | |
3436 | ||
8437a617 | 3437 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3438 | r = -ENOMEM; |
3439 | goto out_free_0; | |
3440 | } | |
3441 | ||
e9b11c17 | 3442 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3443 | if (r < 0) |
7f59f492 | 3444 | goto out_free_0a; |
6aa8b732 | 3445 | |
002c7f7c YS |
3446 | for_each_online_cpu(cpu) { |
3447 | smp_call_function_single(cpu, | |
e9b11c17 | 3448 | kvm_arch_check_processor_compat, |
8691e5a8 | 3449 | &r, 1); |
002c7f7c | 3450 | if (r < 0) |
d2308784 | 3451 | goto out_free_1; |
002c7f7c YS |
3452 | } |
3453 | ||
774c47f1 AK |
3454 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3455 | if (r) | |
d2308784 | 3456 | goto out_free_2; |
6aa8b732 AK |
3457 | register_reboot_notifier(&kvm_reboot_notifier); |
3458 | ||
c16f862d | 3459 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3460 | if (!vcpu_align) |
3461 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3462 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3463 | 0, NULL); |
c16f862d RR |
3464 | if (!kvm_vcpu_cache) { |
3465 | r = -ENOMEM; | |
fb3600cc | 3466 | goto out_free_3; |
c16f862d RR |
3467 | } |
3468 | ||
af585b92 GN |
3469 | r = kvm_async_pf_init(); |
3470 | if (r) | |
3471 | goto out_free; | |
3472 | ||
6aa8b732 | 3473 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3474 | kvm_vm_fops.owner = module; |
3475 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3476 | |
3477 | r = misc_register(&kvm_dev); | |
3478 | if (r) { | |
1170adc6 | 3479 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 3480 | goto out_unreg; |
6aa8b732 AK |
3481 | } |
3482 | ||
fb3600cc RW |
3483 | register_syscore_ops(&kvm_syscore_ops); |
3484 | ||
15ad7146 AK |
3485 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3486 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3487 | ||
4f69b680 H |
3488 | r = kvm_init_debug(); |
3489 | if (r) { | |
1170adc6 | 3490 | pr_err("kvm: create debugfs files failed\n"); |
4f69b680 H |
3491 | goto out_undebugfs; |
3492 | } | |
0ea4ed8e | 3493 | |
3c3c29fd PB |
3494 | r = kvm_vfio_ops_init(); |
3495 | WARN_ON(r); | |
3496 | ||
c7addb90 | 3497 | return 0; |
6aa8b732 | 3498 | |
4f69b680 H |
3499 | out_undebugfs: |
3500 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3501 | misc_deregister(&kvm_dev); |
af585b92 GN |
3502 | out_unreg: |
3503 | kvm_async_pf_deinit(); | |
6aa8b732 | 3504 | out_free: |
c16f862d | 3505 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3506 | out_free_3: |
6aa8b732 | 3507 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3508 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3509 | out_free_2: |
d2308784 | 3510 | out_free_1: |
e9b11c17 | 3511 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3512 | out_free_0a: |
3513 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3514 | out_free_0: |
a0f155e9 CH |
3515 | kvm_irqfd_exit(); |
3516 | out_irqfd: | |
7dac16c3 AH |
3517 | kvm_arch_exit(); |
3518 | out_fail: | |
6aa8b732 AK |
3519 | return r; |
3520 | } | |
cb498ea2 | 3521 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3522 | |
cb498ea2 | 3523 | void kvm_exit(void) |
6aa8b732 | 3524 | { |
0ea4ed8e | 3525 | kvm_exit_debug(); |
6aa8b732 | 3526 | misc_deregister(&kvm_dev); |
c16f862d | 3527 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3528 | kvm_async_pf_deinit(); |
fb3600cc | 3529 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3530 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3531 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3532 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3533 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3534 | kvm_arch_exit(); |
a0f155e9 | 3535 | kvm_irqfd_exit(); |
7f59f492 | 3536 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 3537 | kvm_vfio_ops_exit(); |
6aa8b732 | 3538 | } |
cb498ea2 | 3539 | EXPORT_SYMBOL_GPL(kvm_exit); |