]>
Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
9 | * |
10 | * Authors: | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * Yaniv Kamay <yaniv@qumranet.com> | |
13 | * | |
14 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
15 | * the COPYING file in the top-level directory. | |
16 | * | |
17 | */ | |
18 | ||
e2174021 | 19 | #include "iodev.h" |
6aa8b732 | 20 | |
edf88417 | 21 | #include <linux/kvm_host.h> |
6aa8b732 AK |
22 | #include <linux/kvm.h> |
23 | #include <linux/module.h> | |
24 | #include <linux/errno.h> | |
6aa8b732 | 25 | #include <linux/percpu.h> |
6aa8b732 AK |
26 | #include <linux/mm.h> |
27 | #include <linux/miscdevice.h> | |
28 | #include <linux/vmalloc.h> | |
6aa8b732 | 29 | #include <linux/reboot.h> |
6aa8b732 AK |
30 | #include <linux/debugfs.h> |
31 | #include <linux/highmem.h> | |
32 | #include <linux/file.h> | |
fb3600cc | 33 | #include <linux/syscore_ops.h> |
774c47f1 | 34 | #include <linux/cpu.h> |
e8edc6e0 | 35 | #include <linux/sched.h> |
d9e368d6 AK |
36 | #include <linux/cpumask.h> |
37 | #include <linux/smp.h> | |
d6d28168 | 38 | #include <linux/anon_inodes.h> |
04d2cc77 | 39 | #include <linux/profile.h> |
7aa81cc0 | 40 | #include <linux/kvm_para.h> |
6fc138d2 | 41 | #include <linux/pagemap.h> |
8d4e1288 | 42 | #include <linux/mman.h> |
35149e21 | 43 | #include <linux/swap.h> |
e56d532f | 44 | #include <linux/bitops.h> |
547de29e | 45 | #include <linux/spinlock.h> |
6ff5894c | 46 | #include <linux/compat.h> |
bc6678a3 | 47 | #include <linux/srcu.h> |
8f0b1ab6 | 48 | #include <linux/hugetlb.h> |
5a0e3ad6 | 49 | #include <linux/slab.h> |
743eeb0b SL |
50 | #include <linux/sort.h> |
51 | #include <linux/bsearch.h> | |
6aa8b732 | 52 | |
e495606d | 53 | #include <asm/processor.h> |
e495606d AK |
54 | #include <asm/io.h> |
55 | #include <asm/uaccess.h> | |
3e021bf5 | 56 | #include <asm/pgtable.h> |
6aa8b732 | 57 | |
5f94c174 | 58 | #include "coalesced_mmio.h" |
af585b92 | 59 | #include "async_pf.h" |
5f94c174 | 60 | |
229456fc MT |
61 | #define CREATE_TRACE_POINTS |
62 | #include <trace/events/kvm.h> | |
63 | ||
6aa8b732 AK |
64 | MODULE_AUTHOR("Qumranet"); |
65 | MODULE_LICENSE("GPL"); | |
66 | ||
fa40a821 MT |
67 | /* |
68 | * Ordering of locks: | |
69 | * | |
fae3a353 | 70 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
71 | */ |
72 | ||
2f303b74 | 73 | DEFINE_SPINLOCK(kvm_lock); |
4a937f96 | 74 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 75 | LIST_HEAD(vm_list); |
133de902 | 76 | |
7f59f492 | 77 | static cpumask_var_t cpus_hardware_enabled; |
10474ae8 AG |
78 | static int kvm_usage_count = 0; |
79 | static atomic_t hardware_enable_failed; | |
1b6c0168 | 80 | |
c16f862d RR |
81 | struct kmem_cache *kvm_vcpu_cache; |
82 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 83 | |
15ad7146 AK |
84 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
85 | ||
76f7c879 | 86 | struct dentry *kvm_debugfs_dir; |
6aa8b732 | 87 | |
bccf2150 AK |
88 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
89 | unsigned long arg); | |
1dda606c AG |
90 | #ifdef CONFIG_COMPAT |
91 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, | |
92 | unsigned long arg); | |
93 | #endif | |
10474ae8 AG |
94 | static int hardware_enable_all(void); |
95 | static void hardware_disable_all(void); | |
bccf2150 | 96 | |
e93f8a0f MT |
97 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
98 | ||
b7c4145b AK |
99 | bool kvm_rebooting; |
100 | EXPORT_SYMBOL_GPL(kvm_rebooting); | |
4ecac3fd | 101 | |
54dee993 MT |
102 | static bool largepages_enabled = true; |
103 | ||
a2766325 | 104 | bool kvm_is_mmio_pfn(pfn_t pfn) |
cbff90a7 | 105 | { |
11feeb49 AA |
106 | if (pfn_valid(pfn)) |
107 | return PageReserved(pfn_to_page(pfn)); | |
cbff90a7 BAY |
108 | |
109 | return true; | |
110 | } | |
111 | ||
bccf2150 AK |
112 | /* |
113 | * Switches to specified vcpu, until a matching vcpu_put() | |
114 | */ | |
9fc77441 | 115 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 116 | { |
15ad7146 AK |
117 | int cpu; |
118 | ||
9fc77441 MT |
119 | if (mutex_lock_killable(&vcpu->mutex)) |
120 | return -EINTR; | |
34bb10b7 RR |
121 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
122 | /* The thread running this VCPU changed. */ | |
123 | struct pid *oldpid = vcpu->pid; | |
124 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
125 | rcu_assign_pointer(vcpu->pid, newpid); | |
126 | synchronize_rcu(); | |
127 | put_pid(oldpid); | |
128 | } | |
15ad7146 AK |
129 | cpu = get_cpu(); |
130 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 131 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 132 | put_cpu(); |
9fc77441 | 133 | return 0; |
6aa8b732 AK |
134 | } |
135 | ||
313a3dc7 | 136 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 137 | { |
15ad7146 | 138 | preempt_disable(); |
313a3dc7 | 139 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
140 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
141 | preempt_enable(); | |
6aa8b732 AK |
142 | mutex_unlock(&vcpu->mutex); |
143 | } | |
144 | ||
d9e368d6 AK |
145 | static void ack_flush(void *_completed) |
146 | { | |
d9e368d6 AK |
147 | } |
148 | ||
49846896 | 149 | static bool make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 150 | { |
597a5f55 | 151 | int i, cpu, me; |
6ef7a1bc RR |
152 | cpumask_var_t cpus; |
153 | bool called = true; | |
d9e368d6 | 154 | struct kvm_vcpu *vcpu; |
d9e368d6 | 155 | |
79f55997 | 156 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 157 | |
3cba4130 | 158 | me = get_cpu(); |
988a2cae | 159 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 160 | kvm_make_request(req, vcpu); |
d9e368d6 | 161 | cpu = vcpu->cpu; |
6b7e2d09 XG |
162 | |
163 | /* Set ->requests bit before we read ->mode */ | |
164 | smp_mb(); | |
165 | ||
166 | if (cpus != NULL && cpu != -1 && cpu != me && | |
167 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 168 | cpumask_set_cpu(cpu, cpus); |
49846896 | 169 | } |
6ef7a1bc RR |
170 | if (unlikely(cpus == NULL)) |
171 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
172 | else if (!cpumask_empty(cpus)) | |
173 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
174 | else | |
175 | called = false; | |
3cba4130 | 176 | put_cpu(); |
6ef7a1bc | 177 | free_cpumask_var(cpus); |
49846896 | 178 | return called; |
d9e368d6 AK |
179 | } |
180 | ||
49846896 | 181 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 182 | { |
bec87d6e | 183 | long dirty_count = kvm->tlbs_dirty; |
a4ee1ca4 XG |
184 | |
185 | smp_mb(); | |
49846896 RR |
186 | if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
187 | ++kvm->stat.remote_tlb_flush; | |
a4ee1ca4 | 188 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a MT |
189 | } |
190 | ||
49846896 RR |
191 | void kvm_reload_remote_mmus(struct kvm *kvm) |
192 | { | |
193 | make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); | |
194 | } | |
2e53d63a | 195 | |
d828199e MT |
196 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
197 | { | |
198 | make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
199 | } | |
200 | ||
3d81bc7e | 201 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
c7c9c56c | 202 | { |
3d81bc7e | 203 | make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); |
c7c9c56c YZ |
204 | } |
205 | ||
fb3f0f51 RR |
206 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
207 | { | |
208 | struct page *page; | |
209 | int r; | |
210 | ||
211 | mutex_init(&vcpu->mutex); | |
212 | vcpu->cpu = -1; | |
fb3f0f51 RR |
213 | vcpu->kvm = kvm; |
214 | vcpu->vcpu_id = id; | |
34bb10b7 | 215 | vcpu->pid = NULL; |
b6958ce4 | 216 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 217 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
218 | |
219 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
220 | if (!page) { | |
221 | r = -ENOMEM; | |
222 | goto fail; | |
223 | } | |
224 | vcpu->run = page_address(page); | |
225 | ||
4c088493 R |
226 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
227 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 228 | vcpu->preempted = false; |
4c088493 | 229 | |
e9b11c17 | 230 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 231 | if (r < 0) |
e9b11c17 | 232 | goto fail_free_run; |
fb3f0f51 RR |
233 | return 0; |
234 | ||
fb3f0f51 RR |
235 | fail_free_run: |
236 | free_page((unsigned long)vcpu->run); | |
237 | fail: | |
76fafa5e | 238 | return r; |
fb3f0f51 RR |
239 | } |
240 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
241 | ||
242 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
243 | { | |
34bb10b7 | 244 | put_pid(vcpu->pid); |
e9b11c17 | 245 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
246 | free_page((unsigned long)vcpu->run); |
247 | } | |
248 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
249 | ||
e930bffe AA |
250 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
251 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
252 | { | |
253 | return container_of(mn, struct kvm, mmu_notifier); | |
254 | } | |
255 | ||
256 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
257 | struct mm_struct *mm, | |
258 | unsigned long address) | |
259 | { | |
260 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 261 | int need_tlb_flush, idx; |
e930bffe AA |
262 | |
263 | /* | |
264 | * When ->invalidate_page runs, the linux pte has been zapped | |
265 | * already but the page is still allocated until | |
266 | * ->invalidate_page returns. So if we increase the sequence | |
267 | * here the kvm page fault will notice if the spte can't be | |
268 | * established because the page is going to be freed. If | |
269 | * instead the kvm page fault establishes the spte before | |
270 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
271 | * before returning. | |
272 | * | |
273 | * The sequence increase only need to be seen at spin_unlock | |
274 | * time, and not at spin_lock time. | |
275 | * | |
276 | * Increasing the sequence after the spin_unlock would be | |
277 | * unsafe because the kvm page fault could then establish the | |
278 | * pte after kvm_unmap_hva returned, without noticing the page | |
279 | * is going to be freed. | |
280 | */ | |
bc6678a3 | 281 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 282 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 283 | |
e930bffe | 284 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 285 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
286 | /* we've to flush the tlb before the pages can be freed */ |
287 | if (need_tlb_flush) | |
288 | kvm_flush_remote_tlbs(kvm); | |
289 | ||
565f3be2 TY |
290 | spin_unlock(&kvm->mmu_lock); |
291 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
292 | } |
293 | ||
3da0dd43 IE |
294 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
295 | struct mm_struct *mm, | |
296 | unsigned long address, | |
297 | pte_t pte) | |
298 | { | |
299 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 300 | int idx; |
3da0dd43 | 301 | |
bc6678a3 | 302 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
303 | spin_lock(&kvm->mmu_lock); |
304 | kvm->mmu_notifier_seq++; | |
305 | kvm_set_spte_hva(kvm, address, pte); | |
306 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 307 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
308 | } |
309 | ||
e930bffe AA |
310 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
311 | struct mm_struct *mm, | |
312 | unsigned long start, | |
313 | unsigned long end) | |
314 | { | |
315 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 316 | int need_tlb_flush = 0, idx; |
e930bffe | 317 | |
bc6678a3 | 318 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
319 | spin_lock(&kvm->mmu_lock); |
320 | /* | |
321 | * The count increase must become visible at unlock time as no | |
322 | * spte can be established without taking the mmu_lock and | |
323 | * count is also read inside the mmu_lock critical section. | |
324 | */ | |
325 | kvm->mmu_notifier_count++; | |
b3ae2096 | 326 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 327 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
328 | /* we've to flush the tlb before the pages can be freed */ |
329 | if (need_tlb_flush) | |
330 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
331 | |
332 | spin_unlock(&kvm->mmu_lock); | |
333 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
334 | } |
335 | ||
336 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
337 | struct mm_struct *mm, | |
338 | unsigned long start, | |
339 | unsigned long end) | |
340 | { | |
341 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
342 | ||
343 | spin_lock(&kvm->mmu_lock); | |
344 | /* | |
345 | * This sequence increase will notify the kvm page fault that | |
346 | * the page that is going to be mapped in the spte could have | |
347 | * been freed. | |
348 | */ | |
349 | kvm->mmu_notifier_seq++; | |
a355aa54 | 350 | smp_wmb(); |
e930bffe AA |
351 | /* |
352 | * The above sequence increase must be visible before the | |
a355aa54 PM |
353 | * below count decrease, which is ensured by the smp_wmb above |
354 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
355 | */ |
356 | kvm->mmu_notifier_count--; | |
357 | spin_unlock(&kvm->mmu_lock); | |
358 | ||
359 | BUG_ON(kvm->mmu_notifier_count < 0); | |
360 | } | |
361 | ||
362 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
363 | struct mm_struct *mm, | |
364 | unsigned long address) | |
365 | { | |
366 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 367 | int young, idx; |
e930bffe | 368 | |
bc6678a3 | 369 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 370 | spin_lock(&kvm->mmu_lock); |
e930bffe | 371 | |
565f3be2 | 372 | young = kvm_age_hva(kvm, address); |
e930bffe AA |
373 | if (young) |
374 | kvm_flush_remote_tlbs(kvm); | |
375 | ||
565f3be2 TY |
376 | spin_unlock(&kvm->mmu_lock); |
377 | srcu_read_unlock(&kvm->srcu, idx); | |
378 | ||
e930bffe AA |
379 | return young; |
380 | } | |
381 | ||
8ee53820 AA |
382 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
383 | struct mm_struct *mm, | |
384 | unsigned long address) | |
385 | { | |
386 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
387 | int young, idx; | |
388 | ||
389 | idx = srcu_read_lock(&kvm->srcu); | |
390 | spin_lock(&kvm->mmu_lock); | |
391 | young = kvm_test_age_hva(kvm, address); | |
392 | spin_unlock(&kvm->mmu_lock); | |
393 | srcu_read_unlock(&kvm->srcu, idx); | |
394 | ||
395 | return young; | |
396 | } | |
397 | ||
85db06e5 MT |
398 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
399 | struct mm_struct *mm) | |
400 | { | |
401 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
402 | int idx; |
403 | ||
404 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 405 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 406 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
407 | } |
408 | ||
e930bffe AA |
409 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
410 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
411 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
412 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
413 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 414 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 415 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 416 | .release = kvm_mmu_notifier_release, |
e930bffe | 417 | }; |
4c07b0a4 AK |
418 | |
419 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
420 | { | |
421 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
422 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
423 | } | |
424 | ||
425 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
426 | ||
427 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
428 | { | |
429 | return 0; | |
430 | } | |
431 | ||
e930bffe AA |
432 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
433 | ||
bf3e05bc XG |
434 | static void kvm_init_memslots_id(struct kvm *kvm) |
435 | { | |
436 | int i; | |
437 | struct kvm_memslots *slots = kvm->memslots; | |
438 | ||
439 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
f85e2cb5 | 440 | slots->id_to_index[i] = slots->memslots[i].id = i; |
bf3e05bc XG |
441 | } |
442 | ||
e08b9637 | 443 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 444 | { |
d89f5eff JK |
445 | int r, i; |
446 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 447 | |
d89f5eff JK |
448 | if (!kvm) |
449 | return ERR_PTR(-ENOMEM); | |
450 | ||
e08b9637 | 451 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff JK |
452 | if (r) |
453 | goto out_err_nodisable; | |
10474ae8 AG |
454 | |
455 | r = hardware_enable_all(); | |
456 | if (r) | |
457 | goto out_err_nodisable; | |
458 | ||
75858a84 AK |
459 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
460 | INIT_HLIST_HEAD(&kvm->mask_notifier_list); | |
136bdfee | 461 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 462 | #endif |
6aa8b732 | 463 | |
1e702d9a AW |
464 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
465 | ||
46a26bf5 MT |
466 | r = -ENOMEM; |
467 | kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); | |
468 | if (!kvm->memslots) | |
57e7fbee | 469 | goto out_err_nosrcu; |
bf3e05bc | 470 | kvm_init_memslots_id(kvm); |
bc6678a3 | 471 | if (init_srcu_struct(&kvm->srcu)) |
57e7fbee | 472 | goto out_err_nosrcu; |
e93f8a0f MT |
473 | for (i = 0; i < KVM_NR_BUSES; i++) { |
474 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
475 | GFP_KERNEL); | |
57e7fbee | 476 | if (!kvm->buses[i]) |
e93f8a0f | 477 | goto out_err; |
e93f8a0f | 478 | } |
e930bffe | 479 | |
74b5c5bf | 480 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
481 | kvm->mm = current->mm; |
482 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 483 | kvm_eventfd_init(kvm); |
11ec2804 | 484 | mutex_init(&kvm->lock); |
60eead79 | 485 | mutex_init(&kvm->irq_lock); |
79fac95e | 486 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 487 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 488 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
489 | |
490 | r = kvm_init_mmu_notifier(kvm); | |
491 | if (r) | |
492 | goto out_err; | |
493 | ||
2f303b74 | 494 | spin_lock(&kvm_lock); |
5e58cfe4 | 495 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 496 | spin_unlock(&kvm_lock); |
d89f5eff | 497 | |
f17abe9a | 498 | return kvm; |
10474ae8 AG |
499 | |
500 | out_err: | |
57e7fbee JK |
501 | cleanup_srcu_struct(&kvm->srcu); |
502 | out_err_nosrcu: | |
10474ae8 AG |
503 | hardware_disable_all(); |
504 | out_err_nodisable: | |
e93f8a0f MT |
505 | for (i = 0; i < KVM_NR_BUSES; i++) |
506 | kfree(kvm->buses[i]); | |
46a26bf5 | 507 | kfree(kvm->memslots); |
d89f5eff | 508 | kvm_arch_free_vm(kvm); |
10474ae8 | 509 | return ERR_PTR(r); |
f17abe9a AK |
510 | } |
511 | ||
92eca8fa TY |
512 | /* |
513 | * Avoid using vmalloc for a small buffer. | |
514 | * Should not be used when the size is statically known. | |
515 | */ | |
c1a7b32a | 516 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
517 | { |
518 | if (size > PAGE_SIZE) | |
519 | return vzalloc(size); | |
520 | else | |
521 | return kzalloc(size, GFP_KERNEL); | |
522 | } | |
523 | ||
c1a7b32a | 524 | void kvm_kvfree(const void *addr) |
92eca8fa TY |
525 | { |
526 | if (is_vmalloc_addr(addr)) | |
527 | vfree(addr); | |
528 | else | |
529 | kfree(addr); | |
530 | } | |
531 | ||
a36a57b1 TY |
532 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
533 | { | |
534 | if (!memslot->dirty_bitmap) | |
535 | return; | |
536 | ||
92eca8fa | 537 | kvm_kvfree(memslot->dirty_bitmap); |
a36a57b1 TY |
538 | memslot->dirty_bitmap = NULL; |
539 | } | |
540 | ||
6aa8b732 AK |
541 | /* |
542 | * Free any memory in @free but not in @dont. | |
543 | */ | |
544 | static void kvm_free_physmem_slot(struct kvm_memory_slot *free, | |
545 | struct kvm_memory_slot *dont) | |
546 | { | |
6aa8b732 | 547 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
a36a57b1 | 548 | kvm_destroy_dirty_bitmap(free); |
6aa8b732 | 549 | |
db3fe4eb | 550 | kvm_arch_free_memslot(free, dont); |
05da4558 | 551 | |
6aa8b732 | 552 | free->npages = 0; |
6aa8b732 AK |
553 | } |
554 | ||
d19a9cd2 | 555 | void kvm_free_physmem(struct kvm *kvm) |
6aa8b732 | 556 | { |
46a26bf5 | 557 | struct kvm_memslots *slots = kvm->memslots; |
be6ba0f0 | 558 | struct kvm_memory_slot *memslot; |
46a26bf5 | 559 | |
be6ba0f0 XG |
560 | kvm_for_each_memslot(memslot, slots) |
561 | kvm_free_physmem_slot(memslot, NULL); | |
6aa8b732 | 562 | |
46a26bf5 | 563 | kfree(kvm->memslots); |
6aa8b732 AK |
564 | } |
565 | ||
07f0a7bd SW |
566 | static void kvm_destroy_devices(struct kvm *kvm) |
567 | { | |
568 | struct list_head *node, *tmp; | |
569 | ||
570 | list_for_each_safe(node, tmp, &kvm->devices) { | |
571 | struct kvm_device *dev = | |
572 | list_entry(node, struct kvm_device, vm_node); | |
573 | ||
574 | list_del(node); | |
575 | dev->ops->destroy(dev); | |
576 | } | |
577 | } | |
578 | ||
f17abe9a AK |
579 | static void kvm_destroy_vm(struct kvm *kvm) |
580 | { | |
e93f8a0f | 581 | int i; |
6d4e4c4f AK |
582 | struct mm_struct *mm = kvm->mm; |
583 | ||
ad8ba2cd | 584 | kvm_arch_sync_events(kvm); |
2f303b74 | 585 | spin_lock(&kvm_lock); |
133de902 | 586 | list_del(&kvm->vm_list); |
2f303b74 | 587 | spin_unlock(&kvm_lock); |
399ec807 | 588 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
589 | for (i = 0; i < KVM_NR_BUSES; i++) |
590 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 591 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
592 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
593 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 594 | #else |
2df72e9b | 595 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 596 | #endif |
d19a9cd2 | 597 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 598 | kvm_destroy_devices(kvm); |
d89f5eff JK |
599 | kvm_free_physmem(kvm); |
600 | cleanup_srcu_struct(&kvm->srcu); | |
601 | kvm_arch_free_vm(kvm); | |
10474ae8 | 602 | hardware_disable_all(); |
6d4e4c4f | 603 | mmdrop(mm); |
f17abe9a AK |
604 | } |
605 | ||
d39f13b0 IE |
606 | void kvm_get_kvm(struct kvm *kvm) |
607 | { | |
608 | atomic_inc(&kvm->users_count); | |
609 | } | |
610 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
611 | ||
612 | void kvm_put_kvm(struct kvm *kvm) | |
613 | { | |
614 | if (atomic_dec_and_test(&kvm->users_count)) | |
615 | kvm_destroy_vm(kvm); | |
616 | } | |
617 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
618 | ||
619 | ||
f17abe9a AK |
620 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
621 | { | |
622 | struct kvm *kvm = filp->private_data; | |
623 | ||
721eecbf GH |
624 | kvm_irqfd_release(kvm); |
625 | ||
d39f13b0 | 626 | kvm_put_kvm(kvm); |
6aa8b732 AK |
627 | return 0; |
628 | } | |
629 | ||
515a0127 TY |
630 | /* |
631 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 632 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 633 | */ |
a36a57b1 TY |
634 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
635 | { | |
189a2f7b | 636 | #ifndef CONFIG_S390 |
515a0127 | 637 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 638 | |
92eca8fa | 639 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
640 | if (!memslot->dirty_bitmap) |
641 | return -ENOMEM; | |
642 | ||
189a2f7b | 643 | #endif /* !CONFIG_S390 */ |
a36a57b1 TY |
644 | return 0; |
645 | } | |
646 | ||
bf3e05bc XG |
647 | static int cmp_memslot(const void *slot1, const void *slot2) |
648 | { | |
649 | struct kvm_memory_slot *s1, *s2; | |
650 | ||
651 | s1 = (struct kvm_memory_slot *)slot1; | |
652 | s2 = (struct kvm_memory_slot *)slot2; | |
653 | ||
654 | if (s1->npages < s2->npages) | |
655 | return 1; | |
656 | if (s1->npages > s2->npages) | |
657 | return -1; | |
658 | ||
659 | return 0; | |
660 | } | |
661 | ||
662 | /* | |
663 | * Sort the memslots base on its size, so the larger slots | |
664 | * will get better fit. | |
665 | */ | |
666 | static void sort_memslots(struct kvm_memslots *slots) | |
667 | { | |
f85e2cb5 XG |
668 | int i; |
669 | ||
bf3e05bc XG |
670 | sort(slots->memslots, KVM_MEM_SLOTS_NUM, |
671 | sizeof(struct kvm_memory_slot), cmp_memslot, NULL); | |
f85e2cb5 XG |
672 | |
673 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
674 | slots->id_to_index[slots->memslots[i].id] = i; | |
bf3e05bc XG |
675 | } |
676 | ||
116c14c0 AW |
677 | void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new, |
678 | u64 last_generation) | |
be593d62 XG |
679 | { |
680 | if (new) { | |
681 | int id = new->id; | |
28a37544 | 682 | struct kvm_memory_slot *old = id_to_memslot(slots, id); |
bf3e05bc | 683 | unsigned long npages = old->npages; |
be593d62 | 684 | |
28a37544 | 685 | *old = *new; |
bf3e05bc XG |
686 | if (new->npages != npages) |
687 | sort_memslots(slots); | |
be593d62 XG |
688 | } |
689 | ||
116c14c0 | 690 | slots->generation = last_generation + 1; |
be593d62 XG |
691 | } |
692 | ||
a50d64d6 XG |
693 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
694 | { | |
4d8b81ab XG |
695 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
696 | ||
697 | #ifdef KVM_CAP_READONLY_MEM | |
698 | valid_flags |= KVM_MEM_READONLY; | |
699 | #endif | |
700 | ||
701 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
702 | return -EINVAL; |
703 | ||
704 | return 0; | |
705 | } | |
706 | ||
7ec4fb44 GN |
707 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
708 | struct kvm_memslots *slots, struct kvm_memory_slot *new) | |
709 | { | |
710 | struct kvm_memslots *old_memslots = kvm->memslots; | |
711 | ||
712 | update_memslots(slots, new, kvm->memslots->generation); | |
713 | rcu_assign_pointer(kvm->memslots, slots); | |
714 | synchronize_srcu_expedited(&kvm->srcu); | |
e59dbe09 TY |
715 | |
716 | kvm_arch_memslots_updated(kvm); | |
717 | ||
718 | return old_memslots; | |
7ec4fb44 GN |
719 | } |
720 | ||
6aa8b732 AK |
721 | /* |
722 | * Allocate some memory and give it an address in the guest physical address | |
723 | * space. | |
724 | * | |
725 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 726 | * |
10589a46 | 727 | * Must be called holding mmap_sem for write. |
6aa8b732 | 728 | */ |
f78e0e2e | 729 | int __kvm_set_memory_region(struct kvm *kvm, |
47ae31e2 | 730 | struct kvm_userspace_memory_region *mem) |
6aa8b732 | 731 | { |
8234b22e | 732 | int r; |
6aa8b732 | 733 | gfn_t base_gfn; |
28bcb112 | 734 | unsigned long npages; |
a843fac2 | 735 | struct kvm_memory_slot *slot; |
6aa8b732 | 736 | struct kvm_memory_slot old, new; |
b7f69c55 | 737 | struct kvm_memslots *slots = NULL, *old_memslots; |
f64c0398 | 738 | enum kvm_mr_change change; |
6aa8b732 | 739 | |
a50d64d6 XG |
740 | r = check_memory_region_flags(mem); |
741 | if (r) | |
742 | goto out; | |
743 | ||
6aa8b732 AK |
744 | r = -EINVAL; |
745 | /* General sanity checks */ | |
746 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
747 | goto out; | |
748 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
749 | goto out; | |
fa3d315a | 750 | /* We can read the guest memory with __xxx_user() later on. */ |
47ae31e2 | 751 | if ((mem->slot < KVM_USER_MEM_SLOTS) && |
fa3d315a | 752 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
753 | !access_ok(VERIFY_WRITE, |
754 | (void __user *)(unsigned long)mem->userspace_addr, | |
755 | mem->memory_size))) | |
78749809 | 756 | goto out; |
93a5cef0 | 757 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
758 | goto out; |
759 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
760 | goto out; | |
761 | ||
a843fac2 | 762 | slot = id_to_memslot(kvm->memslots, mem->slot); |
6aa8b732 AK |
763 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
764 | npages = mem->memory_size >> PAGE_SHIFT; | |
765 | ||
660c22c4 TY |
766 | r = -EINVAL; |
767 | if (npages > KVM_MEM_MAX_NR_PAGES) | |
768 | goto out; | |
769 | ||
6aa8b732 AK |
770 | if (!npages) |
771 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
772 | ||
a843fac2 | 773 | new = old = *slot; |
6aa8b732 | 774 | |
e36d96f7 | 775 | new.id = mem->slot; |
6aa8b732 AK |
776 | new.base_gfn = base_gfn; |
777 | new.npages = npages; | |
778 | new.flags = mem->flags; | |
779 | ||
6aa8b732 | 780 | r = -EINVAL; |
f64c0398 TY |
781 | if (npages) { |
782 | if (!old.npages) | |
783 | change = KVM_MR_CREATE; | |
784 | else { /* Modify an existing slot. */ | |
785 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
786 | (npages != old.npages) || |
787 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
788 | goto out; |
789 | ||
790 | if (base_gfn != old.base_gfn) | |
791 | change = KVM_MR_MOVE; | |
792 | else if (new.flags != old.flags) | |
793 | change = KVM_MR_FLAGS_ONLY; | |
794 | else { /* Nothing to change. */ | |
795 | r = 0; | |
796 | goto out; | |
797 | } | |
798 | } | |
799 | } else if (old.npages) { | |
800 | change = KVM_MR_DELETE; | |
801 | } else /* Modify a non-existent slot: disallowed. */ | |
0ea75e1d | 802 | goto out; |
6aa8b732 | 803 | |
f64c0398 | 804 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
805 | /* Check for overlaps */ |
806 | r = -EEXIST; | |
807 | kvm_for_each_memslot(slot, kvm->memslots) { | |
a843fac2 TY |
808 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
809 | (slot->id == mem->slot)) | |
0a706bee TY |
810 | continue; |
811 | if (!((base_gfn + npages <= slot->base_gfn) || | |
812 | (base_gfn >= slot->base_gfn + slot->npages))) | |
813 | goto out; | |
814 | } | |
6aa8b732 | 815 | } |
6aa8b732 | 816 | |
6aa8b732 AK |
817 | /* Free page dirty bitmap if unneeded */ |
818 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 819 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
820 | |
821 | r = -ENOMEM; | |
f64c0398 | 822 | if (change == KVM_MR_CREATE) { |
189a2f7b | 823 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 824 | |
db3fe4eb TY |
825 | if (kvm_arch_create_memslot(&new, npages)) |
826 | goto out_free; | |
6aa8b732 | 827 | } |
ec04b260 | 828 | |
6aa8b732 AK |
829 | /* Allocate page dirty bitmap if needed */ |
830 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 831 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 832 | goto out_free; |
6aa8b732 AK |
833 | } |
834 | ||
f64c0398 | 835 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
bc6678a3 | 836 | r = -ENOMEM; |
6da64fdb TM |
837 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), |
838 | GFP_KERNEL); | |
bc6678a3 MT |
839 | if (!slots) |
840 | goto out_free; | |
28a37544 XG |
841 | slot = id_to_memslot(slots, mem->slot); |
842 | slot->flags |= KVM_MEMSLOT_INVALID; | |
843 | ||
7ec4fb44 | 844 | old_memslots = install_new_memslots(kvm, slots, NULL); |
bc6678a3 | 845 | |
e40f193f AW |
846 | /* slot was deleted or moved, clear iommu mapping */ |
847 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
848 | /* From this point no new shadow pages pointing to a deleted, |
849 | * or moved, memslot will be created. | |
bc6678a3 MT |
850 | * |
851 | * validation of sp->gfn happens in: | |
852 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
853 | * - kvm_is_visible_gfn (mmu_check_roots) | |
854 | */ | |
2df72e9b | 855 | kvm_arch_flush_shadow_memslot(kvm, slot); |
b7f69c55 | 856 | slots = old_memslots; |
bc6678a3 | 857 | } |
34d4cb8f | 858 | |
7b6195a9 | 859 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 860 | if (r) |
b7f69c55 | 861 | goto out_slots; |
f7784b8e | 862 | |
bc6678a3 | 863 | r = -ENOMEM; |
b7f69c55 AW |
864 | /* |
865 | * We can re-use the old_memslots from above, the only difference | |
866 | * from the currently installed memslots is the invalid flag. This | |
867 | * will get overwritten by update_memslots anyway. | |
868 | */ | |
869 | if (!slots) { | |
870 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), | |
871 | GFP_KERNEL); | |
872 | if (!slots) | |
873 | goto out_free; | |
874 | } | |
bc6678a3 | 875 | |
261874b0 AW |
876 | /* |
877 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
878 | * un-mapped and re-mapped if their base changes. Since base change |
879 | * unmapping is handled above with slot deletion, mapping alone is | |
880 | * needed here. Anything else the iommu might care about for existing | |
881 | * slots (size changes, userspace addr changes and read-only flag | |
882 | * changes) is disallowed above, so any other attribute changes getting | |
883 | * here can be skipped. | |
261874b0 | 884 | */ |
75d61fbc TY |
885 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
886 | r = kvm_iommu_map_pages(kvm, &new); | |
887 | if (r) | |
888 | goto out_slots; | |
e40f193f AW |
889 | } |
890 | ||
bc6678a3 | 891 | /* actual memory is freed via old in kvm_free_physmem_slot below */ |
f64c0398 | 892 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 893 | new.dirty_bitmap = NULL; |
db3fe4eb | 894 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
895 | } |
896 | ||
7ec4fb44 | 897 | old_memslots = install_new_memslots(kvm, slots, &new); |
3ad82a7e | 898 | |
8482644a | 899 | kvm_arch_commit_memory_region(kvm, mem, &old, change); |
82ce2c96 | 900 | |
bc6678a3 MT |
901 | kvm_free_physmem_slot(&old, &new); |
902 | kfree(old_memslots); | |
903 | ||
6aa8b732 AK |
904 | return 0; |
905 | ||
e40f193f AW |
906 | out_slots: |
907 | kfree(slots); | |
f78e0e2e | 908 | out_free: |
6aa8b732 AK |
909 | kvm_free_physmem_slot(&new, &old); |
910 | out: | |
911 | return r; | |
210c7c4d | 912 | } |
f78e0e2e SY |
913 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
914 | ||
915 | int kvm_set_memory_region(struct kvm *kvm, | |
47ae31e2 | 916 | struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
917 | { |
918 | int r; | |
919 | ||
79fac95e | 920 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 921 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 922 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
923 | return r; |
924 | } | |
210c7c4d IE |
925 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
926 | ||
1fe779f8 | 927 | int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
47ae31e2 | 928 | struct kvm_userspace_memory_region *mem) |
210c7c4d | 929 | { |
bbacc0c1 | 930 | if (mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 931 | return -EINVAL; |
47ae31e2 | 932 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
933 | } |
934 | ||
5bb064dc ZX |
935 | int kvm_get_dirty_log(struct kvm *kvm, |
936 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 AK |
937 | { |
938 | struct kvm_memory_slot *memslot; | |
939 | int r, i; | |
87bf6e7d | 940 | unsigned long n; |
6aa8b732 AK |
941 | unsigned long any = 0; |
942 | ||
6aa8b732 | 943 | r = -EINVAL; |
bbacc0c1 | 944 | if (log->slot >= KVM_USER_MEM_SLOTS) |
6aa8b732 AK |
945 | goto out; |
946 | ||
28a37544 | 947 | memslot = id_to_memslot(kvm->memslots, log->slot); |
6aa8b732 AK |
948 | r = -ENOENT; |
949 | if (!memslot->dirty_bitmap) | |
950 | goto out; | |
951 | ||
87bf6e7d | 952 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 953 | |
cd1a4a98 | 954 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
955 | any = memslot->dirty_bitmap[i]; |
956 | ||
957 | r = -EFAULT; | |
958 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
959 | goto out; | |
960 | ||
5bb064dc ZX |
961 | if (any) |
962 | *is_dirty = 1; | |
6aa8b732 AK |
963 | |
964 | r = 0; | |
6aa8b732 | 965 | out: |
6aa8b732 AK |
966 | return r; |
967 | } | |
968 | ||
db3fe4eb TY |
969 | bool kvm_largepages_enabled(void) |
970 | { | |
971 | return largepages_enabled; | |
972 | } | |
973 | ||
54dee993 MT |
974 | void kvm_disable_largepages(void) |
975 | { | |
976 | largepages_enabled = false; | |
977 | } | |
978 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
979 | ||
49c7754c GN |
980 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
981 | { | |
982 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
983 | } | |
a1f4d395 | 984 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 985 | |
e0d62c7f IE |
986 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
987 | { | |
bf3e05bc | 988 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 989 | |
bbacc0c1 | 990 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
991 | memslot->flags & KVM_MEMSLOT_INVALID) |
992 | return 0; | |
e0d62c7f | 993 | |
bf3e05bc | 994 | return 1; |
e0d62c7f IE |
995 | } |
996 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
997 | ||
8f0b1ab6 JR |
998 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
999 | { | |
1000 | struct vm_area_struct *vma; | |
1001 | unsigned long addr, size; | |
1002 | ||
1003 | size = PAGE_SIZE; | |
1004 | ||
1005 | addr = gfn_to_hva(kvm, gfn); | |
1006 | if (kvm_is_error_hva(addr)) | |
1007 | return PAGE_SIZE; | |
1008 | ||
1009 | down_read(¤t->mm->mmap_sem); | |
1010 | vma = find_vma(current->mm, addr); | |
1011 | if (!vma) | |
1012 | goto out; | |
1013 | ||
1014 | size = vma_kernel_pagesize(vma); | |
1015 | ||
1016 | out: | |
1017 | up_read(¤t->mm->mmap_sem); | |
1018 | ||
1019 | return size; | |
1020 | } | |
1021 | ||
4d8b81ab XG |
1022 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1023 | { | |
1024 | return slot->flags & KVM_MEM_READONLY; | |
1025 | } | |
1026 | ||
4d8b81ab XG |
1027 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1028 | gfn_t *nr_pages, bool write) | |
539cb660 | 1029 | { |
bc6678a3 | 1030 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1031 | return KVM_HVA_ERR_BAD; |
48987781 | 1032 | |
4d8b81ab XG |
1033 | if (memslot_is_readonly(slot) && write) |
1034 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1035 | |
1036 | if (nr_pages) | |
1037 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1038 | ||
4d8b81ab | 1039 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1040 | } |
48987781 | 1041 | |
4d8b81ab XG |
1042 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1043 | gfn_t *nr_pages) | |
1044 | { | |
1045 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1046 | } |
48987781 | 1047 | |
4d8b81ab XG |
1048 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
1049 | gfn_t gfn) | |
1050 | { | |
1051 | return gfn_to_hva_many(slot, gfn, NULL); | |
1052 | } | |
1053 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1054 | ||
48987781 XG |
1055 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1056 | { | |
49c7754c | 1057 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1058 | } |
0d150298 | 1059 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1060 | |
86ab8cff | 1061 | /* |
ba6a3541 PB |
1062 | * If writable is set to false, the hva returned by this function is only |
1063 | * allowed to be read. | |
86ab8cff | 1064 | */ |
ba6a3541 | 1065 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
86ab8cff | 1066 | { |
ba6a3541 PB |
1067 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
1068 | if (writable) | |
1069 | *writable = !memslot_is_readonly(slot); | |
1070 | ||
4d8b81ab | 1071 | return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false); |
86ab8cff XG |
1072 | } |
1073 | ||
1074 | static int kvm_read_hva(void *data, void __user *hva, int len) | |
8030089f | 1075 | { |
86ab8cff XG |
1076 | return __copy_from_user(data, hva, len); |
1077 | } | |
1078 | ||
1079 | static int kvm_read_hva_atomic(void *data, void __user *hva, int len) | |
1080 | { | |
1081 | return __copy_from_user_inatomic(data, hva, len); | |
8030089f GN |
1082 | } |
1083 | ||
39369f7a | 1084 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1085 | unsigned long start, int write, struct page **page) |
1086 | { | |
1087 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1088 | ||
1089 | if (write) | |
1090 | flags |= FOLL_WRITE; | |
1091 | ||
1092 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1093 | } | |
1094 | ||
fafc3dba HY |
1095 | static inline int check_user_page_hwpoison(unsigned long addr) |
1096 | { | |
1097 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1098 | ||
1099 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1100 | flags, NULL, NULL, NULL); | |
1101 | return rc == -EHWPOISON; | |
1102 | } | |
1103 | ||
2fc84311 XG |
1104 | /* |
1105 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1106 | * true indicates success, otherwise false is returned. | |
1107 | */ | |
1108 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1109 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1110 | { |
8d4e1288 | 1111 | struct page *page[1]; |
2fc84311 | 1112 | int npages; |
954bbbc2 | 1113 | |
2fc84311 XG |
1114 | if (!(async || atomic)) |
1115 | return false; | |
af585b92 | 1116 | |
12ce13fe XG |
1117 | /* |
1118 | * Fast pin a writable pfn only if it is a write fault request | |
1119 | * or the caller allows to map a writable pfn for a read fault | |
1120 | * request. | |
1121 | */ | |
1122 | if (!(write_fault || writable)) | |
1123 | return false; | |
612819c3 | 1124 | |
2fc84311 XG |
1125 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1126 | if (npages == 1) { | |
1127 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1128 | |
2fc84311 XG |
1129 | if (writable) |
1130 | *writable = true; | |
1131 | return true; | |
1132 | } | |
af585b92 | 1133 | |
2fc84311 XG |
1134 | return false; |
1135 | } | |
612819c3 | 1136 | |
2fc84311 XG |
1137 | /* |
1138 | * The slow path to get the pfn of the specified host virtual address, | |
1139 | * 1 indicates success, -errno is returned if error is detected. | |
1140 | */ | |
1141 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1142 | bool *writable, pfn_t *pfn) | |
1143 | { | |
1144 | struct page *page[1]; | |
1145 | int npages = 0; | |
612819c3 | 1146 | |
2fc84311 XG |
1147 | might_sleep(); |
1148 | ||
1149 | if (writable) | |
1150 | *writable = write_fault; | |
1151 | ||
1152 | if (async) { | |
1153 | down_read(¤t->mm->mmap_sem); | |
1154 | npages = get_user_page_nowait(current, current->mm, | |
1155 | addr, write_fault, page); | |
1156 | up_read(¤t->mm->mmap_sem); | |
1157 | } else | |
1158 | npages = get_user_pages_fast(addr, 1, write_fault, | |
1159 | page); | |
1160 | if (npages != 1) | |
1161 | return npages; | |
1162 | ||
1163 | /* map read fault as writable if possible */ | |
12ce13fe | 1164 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1165 | struct page *wpage[1]; |
1166 | ||
1167 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1168 | if (npages == 1) { | |
1169 | *writable = true; | |
1170 | put_page(page[0]); | |
1171 | page[0] = wpage[0]; | |
612819c3 | 1172 | } |
2fc84311 XG |
1173 | |
1174 | npages = 1; | |
887c08ac | 1175 | } |
2fc84311 XG |
1176 | *pfn = page_to_pfn(page[0]); |
1177 | return npages; | |
1178 | } | |
539cb660 | 1179 | |
4d8b81ab XG |
1180 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1181 | { | |
1182 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1183 | return false; | |
2e2e3738 | 1184 | |
4d8b81ab XG |
1185 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1186 | return false; | |
887c08ac | 1187 | |
4d8b81ab XG |
1188 | return true; |
1189 | } | |
bf998156 | 1190 | |
12ce13fe XG |
1191 | /* |
1192 | * Pin guest page in memory and return its pfn. | |
1193 | * @addr: host virtual address which maps memory to the guest | |
1194 | * @atomic: whether this function can sleep | |
1195 | * @async: whether this function need to wait IO complete if the | |
1196 | * host page is not in the memory | |
1197 | * @write_fault: whether we should get a writable host page | |
1198 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1199 | * | |
1200 | * The function will map a writable host page for these two cases: | |
1201 | * 1): @write_fault = true | |
1202 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1203 | * whether the mapping is writable. | |
1204 | */ | |
2fc84311 XG |
1205 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1206 | bool write_fault, bool *writable) | |
1207 | { | |
1208 | struct vm_area_struct *vma; | |
1209 | pfn_t pfn = 0; | |
1210 | int npages; | |
2e2e3738 | 1211 | |
2fc84311 XG |
1212 | /* we can do it either atomically or asynchronously, not both */ |
1213 | BUG_ON(atomic && async); | |
8d4e1288 | 1214 | |
2fc84311 XG |
1215 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1216 | return pfn; | |
1217 | ||
1218 | if (atomic) | |
1219 | return KVM_PFN_ERR_FAULT; | |
1220 | ||
1221 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1222 | if (npages == 1) | |
1223 | return pfn; | |
8d4e1288 | 1224 | |
2fc84311 XG |
1225 | down_read(¤t->mm->mmap_sem); |
1226 | if (npages == -EHWPOISON || | |
1227 | (!async && check_user_page_hwpoison(addr))) { | |
1228 | pfn = KVM_PFN_ERR_HWPOISON; | |
1229 | goto exit; | |
1230 | } | |
1231 | ||
1232 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1233 | ||
1234 | if (vma == NULL) | |
1235 | pfn = KVM_PFN_ERR_FAULT; | |
1236 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1237 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1238 | vma->vm_pgoff; | |
1239 | BUG_ON(!kvm_is_mmio_pfn(pfn)); | |
1240 | } else { | |
4d8b81ab | 1241 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1242 | *async = true; |
1243 | pfn = KVM_PFN_ERR_FAULT; | |
1244 | } | |
1245 | exit: | |
1246 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1247 | return pfn; |
35149e21 AL |
1248 | } |
1249 | ||
4d8b81ab XG |
1250 | static pfn_t |
1251 | __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, | |
1252 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1253 | { |
4d8b81ab XG |
1254 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1255 | ||
1256 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1257 | return KVM_PFN_ERR_RO_FAULT; | |
1258 | ||
1259 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1260 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1261 | |
1262 | /* Do not map writable pfn in the readonly memslot. */ | |
1263 | if (writable && memslot_is_readonly(slot)) { | |
1264 | *writable = false; | |
1265 | writable = NULL; | |
1266 | } | |
1267 | ||
1268 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1269 | writable); | |
887c08ac | 1270 | } |
887c08ac | 1271 | |
612819c3 MT |
1272 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async, |
1273 | bool write_fault, bool *writable) | |
506f0d6f | 1274 | { |
4d8b81ab | 1275 | struct kvm_memory_slot *slot; |
506f0d6f | 1276 | |
af585b92 GN |
1277 | if (async) |
1278 | *async = false; | |
1279 | ||
4d8b81ab | 1280 | slot = gfn_to_memslot(kvm, gfn); |
506f0d6f | 1281 | |
4d8b81ab XG |
1282 | return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault, |
1283 | writable); | |
365fb3fd XG |
1284 | } |
1285 | ||
1286 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) | |
1287 | { | |
612819c3 | 1288 | return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL); |
365fb3fd XG |
1289 | } |
1290 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1291 | ||
612819c3 MT |
1292 | pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async, |
1293 | bool write_fault, bool *writable) | |
af585b92 | 1294 | { |
612819c3 | 1295 | return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable); |
af585b92 GN |
1296 | } |
1297 | EXPORT_SYMBOL_GPL(gfn_to_pfn_async); | |
1298 | ||
365fb3fd XG |
1299 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1300 | { | |
612819c3 | 1301 | return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL); |
506f0d6f | 1302 | } |
35149e21 AL |
1303 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
1304 | ||
612819c3 MT |
1305 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1306 | bool *writable) | |
1307 | { | |
1308 | return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable); | |
1309 | } | |
1310 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1311 | ||
d5661048 | 1312 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1313 | { |
4d8b81ab | 1314 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f MT |
1315 | } |
1316 | ||
037d92dc | 1317 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1318 | { |
4d8b81ab | 1319 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1320 | } |
037d92dc | 1321 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1322 | |
48987781 XG |
1323 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
1324 | int nr_pages) | |
1325 | { | |
1326 | unsigned long addr; | |
1327 | gfn_t entry; | |
1328 | ||
49c7754c | 1329 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
48987781 XG |
1330 | if (kvm_is_error_hva(addr)) |
1331 | return -1; | |
1332 | ||
1333 | if (entry < nr_pages) | |
1334 | return 0; | |
1335 | ||
1336 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1337 | } | |
1338 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1339 | ||
a2766325 XG |
1340 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1341 | { | |
81c52c56 | 1342 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1343 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1344 | |
cb9aaa30 XG |
1345 | if (kvm_is_mmio_pfn(pfn)) { |
1346 | WARN_ON(1); | |
6cede2e6 | 1347 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1348 | } |
a2766325 XG |
1349 | |
1350 | return pfn_to_page(pfn); | |
1351 | } | |
1352 | ||
35149e21 AL |
1353 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1354 | { | |
2e2e3738 AL |
1355 | pfn_t pfn; |
1356 | ||
1357 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1358 | |
a2766325 | 1359 | return kvm_pfn_to_page(pfn); |
954bbbc2 | 1360 | } |
aab61cc0 | 1361 | |
954bbbc2 AK |
1362 | EXPORT_SYMBOL_GPL(gfn_to_page); |
1363 | ||
b4231d61 IE |
1364 | void kvm_release_page_clean(struct page *page) |
1365 | { | |
32cad84f XG |
1366 | WARN_ON(is_error_page(page)); |
1367 | ||
35149e21 | 1368 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1369 | } |
1370 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1371 | ||
35149e21 AL |
1372 | void kvm_release_pfn_clean(pfn_t pfn) |
1373 | { | |
81c52c56 | 1374 | if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1375 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1376 | } |
1377 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1378 | ||
b4231d61 | 1379 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1380 | { |
a2766325 XG |
1381 | WARN_ON(is_error_page(page)); |
1382 | ||
35149e21 AL |
1383 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1384 | } | |
1385 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1386 | ||
1387 | void kvm_release_pfn_dirty(pfn_t pfn) | |
1388 | { | |
1389 | kvm_set_pfn_dirty(pfn); | |
1390 | kvm_release_pfn_clean(pfn); | |
1391 | } | |
1392 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); | |
1393 | ||
1394 | void kvm_set_page_dirty(struct page *page) | |
1395 | { | |
1396 | kvm_set_pfn_dirty(page_to_pfn(page)); | |
1397 | } | |
1398 | EXPORT_SYMBOL_GPL(kvm_set_page_dirty); | |
1399 | ||
1400 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1401 | { | |
c77fb9dc | 1402 | if (!kvm_is_mmio_pfn(pfn)) { |
2e2e3738 AL |
1403 | struct page *page = pfn_to_page(pfn); |
1404 | if (!PageReserved(page)) | |
1405 | SetPageDirty(page); | |
1406 | } | |
8a7ae055 | 1407 | } |
35149e21 AL |
1408 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1409 | ||
1410 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1411 | { | |
c77fb9dc | 1412 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1413 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1414 | } |
1415 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1416 | ||
1417 | void kvm_get_pfn(pfn_t pfn) | |
1418 | { | |
c77fb9dc | 1419 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1420 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1421 | } |
1422 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1423 | |
195aefde IE |
1424 | static int next_segment(unsigned long len, int offset) |
1425 | { | |
1426 | if (len > PAGE_SIZE - offset) | |
1427 | return PAGE_SIZE - offset; | |
1428 | else | |
1429 | return len; | |
1430 | } | |
1431 | ||
1432 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1433 | int len) | |
1434 | { | |
e0506bcb IE |
1435 | int r; |
1436 | unsigned long addr; | |
195aefde | 1437 | |
ba6a3541 | 1438 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
e0506bcb IE |
1439 | if (kvm_is_error_hva(addr)) |
1440 | return -EFAULT; | |
86ab8cff | 1441 | r = kvm_read_hva(data, (void __user *)addr + offset, len); |
e0506bcb | 1442 | if (r) |
195aefde | 1443 | return -EFAULT; |
195aefde IE |
1444 | return 0; |
1445 | } | |
1446 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); | |
1447 | ||
1448 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) | |
1449 | { | |
1450 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1451 | int seg; | |
1452 | int offset = offset_in_page(gpa); | |
1453 | int ret; | |
1454 | ||
1455 | while ((seg = next_segment(len, offset)) != 0) { | |
1456 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1457 | if (ret < 0) | |
1458 | return ret; | |
1459 | offset = 0; | |
1460 | len -= seg; | |
1461 | data += seg; | |
1462 | ++gfn; | |
1463 | } | |
1464 | return 0; | |
1465 | } | |
1466 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1467 | ||
7ec54588 MT |
1468 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1469 | unsigned long len) | |
1470 | { | |
1471 | int r; | |
1472 | unsigned long addr; | |
1473 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1474 | int offset = offset_in_page(gpa); | |
1475 | ||
ba6a3541 | 1476 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
7ec54588 MT |
1477 | if (kvm_is_error_hva(addr)) |
1478 | return -EFAULT; | |
0aac03f0 | 1479 | pagefault_disable(); |
86ab8cff | 1480 | r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1481 | pagefault_enable(); |
7ec54588 MT |
1482 | if (r) |
1483 | return -EFAULT; | |
1484 | return 0; | |
1485 | } | |
1486 | EXPORT_SYMBOL(kvm_read_guest_atomic); | |
1487 | ||
195aefde IE |
1488 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
1489 | int offset, int len) | |
1490 | { | |
e0506bcb IE |
1491 | int r; |
1492 | unsigned long addr; | |
195aefde | 1493 | |
e0506bcb IE |
1494 | addr = gfn_to_hva(kvm, gfn); |
1495 | if (kvm_is_error_hva(addr)) | |
1496 | return -EFAULT; | |
8b0cedff | 1497 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1498 | if (r) |
195aefde | 1499 | return -EFAULT; |
195aefde IE |
1500 | mark_page_dirty(kvm, gfn); |
1501 | return 0; | |
1502 | } | |
1503 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); | |
1504 | ||
1505 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, | |
1506 | unsigned long len) | |
1507 | { | |
1508 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1509 | int seg; | |
1510 | int offset = offset_in_page(gpa); | |
1511 | int ret; | |
1512 | ||
1513 | while ((seg = next_segment(len, offset)) != 0) { | |
1514 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1515 | if (ret < 0) | |
1516 | return ret; | |
1517 | offset = 0; | |
1518 | len -= seg; | |
1519 | data += seg; | |
1520 | ++gfn; | |
1521 | } | |
1522 | return 0; | |
1523 | } | |
1524 | ||
49c7754c | 1525 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1526 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1527 | { |
1528 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1529 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1530 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1531 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1532 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1533 | gfn_t nr_pages_avail; | |
49c7754c GN |
1534 | |
1535 | ghc->gpa = gpa; | |
1536 | ghc->generation = slots->generation; | |
8f964525 AH |
1537 | ghc->len = len; |
1538 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1539 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail); | |
1540 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) { | |
49c7754c | 1541 | ghc->hva += offset; |
8f964525 AH |
1542 | } else { |
1543 | /* | |
1544 | * If the requested region crosses two memslots, we still | |
1545 | * verify that the entire region is valid here. | |
1546 | */ | |
1547 | while (start_gfn <= end_gfn) { | |
1548 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1549 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1550 | &nr_pages_avail); | |
1551 | if (kvm_is_error_hva(ghc->hva)) | |
1552 | return -EFAULT; | |
1553 | start_gfn += nr_pages_avail; | |
1554 | } | |
1555 | /* Use the slow path for cross page reads and writes. */ | |
1556 | ghc->memslot = NULL; | |
1557 | } | |
49c7754c GN |
1558 | return 0; |
1559 | } | |
1560 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1561 | ||
1562 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1563 | void *data, unsigned long len) | |
1564 | { | |
1565 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1566 | int r; | |
1567 | ||
8f964525 AH |
1568 | BUG_ON(len > ghc->len); |
1569 | ||
49c7754c | 1570 | if (slots->generation != ghc->generation) |
8f964525 AH |
1571 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1572 | ||
1573 | if (unlikely(!ghc->memslot)) | |
1574 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1575 | |
1576 | if (kvm_is_error_hva(ghc->hva)) | |
1577 | return -EFAULT; | |
1578 | ||
8b0cedff | 1579 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1580 | if (r) |
1581 | return -EFAULT; | |
1582 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); | |
1583 | ||
1584 | return 0; | |
1585 | } | |
1586 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1587 | ||
e03b644f GN |
1588 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1589 | void *data, unsigned long len) | |
1590 | { | |
1591 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1592 | int r; | |
1593 | ||
8f964525 AH |
1594 | BUG_ON(len > ghc->len); |
1595 | ||
e03b644f | 1596 | if (slots->generation != ghc->generation) |
8f964525 AH |
1597 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1598 | ||
1599 | if (unlikely(!ghc->memslot)) | |
1600 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1601 | |
1602 | if (kvm_is_error_hva(ghc->hva)) | |
1603 | return -EFAULT; | |
1604 | ||
1605 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1606 | if (r) | |
1607 | return -EFAULT; | |
1608 | ||
1609 | return 0; | |
1610 | } | |
1611 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1612 | ||
195aefde IE |
1613 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1614 | { | |
3bcc8a8c HC |
1615 | return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page, |
1616 | offset, len); | |
195aefde IE |
1617 | } |
1618 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1619 | ||
1620 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1621 | { | |
1622 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1623 | int seg; | |
1624 | int offset = offset_in_page(gpa); | |
1625 | int ret; | |
1626 | ||
1627 | while ((seg = next_segment(len, offset)) != 0) { | |
1628 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); | |
1629 | if (ret < 0) | |
1630 | return ret; | |
1631 | offset = 0; | |
1632 | len -= seg; | |
1633 | ++gfn; | |
1634 | } | |
1635 | return 0; | |
1636 | } | |
1637 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1638 | ||
49c7754c GN |
1639 | void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot, |
1640 | gfn_t gfn) | |
6aa8b732 | 1641 | { |
7e9d619d RR |
1642 | if (memslot && memslot->dirty_bitmap) { |
1643 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1644 | |
b74ca3b3 | 1645 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1646 | } |
1647 | } | |
1648 | ||
49c7754c GN |
1649 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1650 | { | |
1651 | struct kvm_memory_slot *memslot; | |
1652 | ||
1653 | memslot = gfn_to_memslot(kvm, gfn); | |
1654 | mark_page_dirty_in_slot(kvm, memslot, gfn); | |
1655 | } | |
1656 | ||
b6958ce4 ED |
1657 | /* |
1658 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1659 | */ | |
8776e519 | 1660 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1661 | { |
e5c239cf MT |
1662 | DEFINE_WAIT(wait); |
1663 | ||
1664 | for (;;) { | |
1665 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1666 | ||
a1b37100 | 1667 | if (kvm_arch_vcpu_runnable(vcpu)) { |
a8eeb04a | 1668 | kvm_make_request(KVM_REQ_UNHALT, vcpu); |
e5c239cf | 1669 | break; |
d7690175 | 1670 | } |
09cec754 GN |
1671 | if (kvm_cpu_has_pending_timer(vcpu)) |
1672 | break; | |
e5c239cf MT |
1673 | if (signal_pending(current)) |
1674 | break; | |
1675 | ||
b6958ce4 | 1676 | schedule(); |
b6958ce4 | 1677 | } |
d3bef15f | 1678 | |
e5c239cf | 1679 | finish_wait(&vcpu->wq, &wait); |
b6958ce4 ED |
1680 | } |
1681 | ||
8c84780d | 1682 | #ifndef CONFIG_S390 |
b6d33834 CD |
1683 | /* |
1684 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1685 | */ | |
1686 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1687 | { | |
1688 | int me; | |
1689 | int cpu = vcpu->cpu; | |
1690 | wait_queue_head_t *wqp; | |
1691 | ||
1692 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1693 | if (waitqueue_active(wqp)) { | |
1694 | wake_up_interruptible(wqp); | |
1695 | ++vcpu->stat.halt_wakeup; | |
1696 | } | |
1697 | ||
1698 | me = get_cpu(); | |
1699 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1700 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1701 | smp_send_reschedule(cpu); | |
1702 | put_cpu(); | |
1703 | } | |
a20ed54d | 1704 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1705 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1706 | |
6aa8b732 AK |
1707 | void kvm_resched(struct kvm_vcpu *vcpu) |
1708 | { | |
3fca0365 YD |
1709 | if (!need_resched()) |
1710 | return; | |
6aa8b732 | 1711 | cond_resched(); |
6aa8b732 AK |
1712 | } |
1713 | EXPORT_SYMBOL_GPL(kvm_resched); | |
1714 | ||
41628d33 KW |
1715 | bool kvm_vcpu_yield_to(struct kvm_vcpu *target) |
1716 | { | |
1717 | struct pid *pid; | |
1718 | struct task_struct *task = NULL; | |
c45c528e | 1719 | bool ret = false; |
41628d33 KW |
1720 | |
1721 | rcu_read_lock(); | |
1722 | pid = rcu_dereference(target->pid); | |
1723 | if (pid) | |
1724 | task = get_pid_task(target->pid, PIDTYPE_PID); | |
1725 | rcu_read_unlock(); | |
1726 | if (!task) | |
c45c528e | 1727 | return ret; |
41628d33 KW |
1728 | if (task->flags & PF_VCPU) { |
1729 | put_task_struct(task); | |
c45c528e | 1730 | return ret; |
41628d33 | 1731 | } |
c45c528e | 1732 | ret = yield_to(task, 1); |
41628d33 | 1733 | put_task_struct(task); |
c45c528e R |
1734 | |
1735 | return ret; | |
41628d33 KW |
1736 | } |
1737 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
1738 | ||
06e48c51 R |
1739 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
1740 | /* | |
1741 | * Helper that checks whether a VCPU is eligible for directed yield. | |
1742 | * Most eligible candidate to yield is decided by following heuristics: | |
1743 | * | |
1744 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
1745 | * (preempted lock holder), indicated by @in_spin_loop. | |
1746 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
1747 | * | |
1748 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
1749 | * chance last time (mostly it has become eligible now since we have probably | |
1750 | * yielded to lockholder in last iteration. This is done by toggling | |
1751 | * @dy_eligible each time a VCPU checked for eligibility.) | |
1752 | * | |
1753 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
1754 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
1755 | * burning. Giving priority for a potential lock-holder increases lock | |
1756 | * progress. | |
1757 | * | |
1758 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
1759 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
1760 | * and continue with next VCPU and so on. | |
1761 | */ | |
1762 | bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) | |
1763 | { | |
1764 | bool eligible; | |
1765 | ||
1766 | eligible = !vcpu->spin_loop.in_spin_loop || | |
1767 | (vcpu->spin_loop.in_spin_loop && | |
1768 | vcpu->spin_loop.dy_eligible); | |
1769 | ||
1770 | if (vcpu->spin_loop.in_spin_loop) | |
1771 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
1772 | ||
1773 | return eligible; | |
1774 | } | |
1775 | #endif | |
c45c528e | 1776 | |
217ece61 | 1777 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 1778 | { |
217ece61 RR |
1779 | struct kvm *kvm = me->kvm; |
1780 | struct kvm_vcpu *vcpu; | |
1781 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
1782 | int yielded = 0; | |
c45c528e | 1783 | int try = 3; |
217ece61 RR |
1784 | int pass; |
1785 | int i; | |
d255f4f2 | 1786 | |
4c088493 | 1787 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
1788 | /* |
1789 | * We boost the priority of a VCPU that is runnable but not | |
1790 | * currently running, because it got preempted by something | |
1791 | * else and called schedule in __vcpu_run. Hopefully that | |
1792 | * VCPU is holding the lock that we need and will release it. | |
1793 | * We approximate round-robin by starting at the last boosted VCPU. | |
1794 | */ | |
c45c528e | 1795 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 1796 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 1797 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
1798 | i = last_boosted_vcpu; |
1799 | continue; | |
1800 | } else if (pass && i > last_boosted_vcpu) | |
1801 | break; | |
7bc7ae25 R |
1802 | if (!ACCESS_ONCE(vcpu->preempted)) |
1803 | continue; | |
217ece61 RR |
1804 | if (vcpu == me) |
1805 | continue; | |
1806 | if (waitqueue_active(&vcpu->wq)) | |
1807 | continue; | |
06e48c51 R |
1808 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
1809 | continue; | |
c45c528e R |
1810 | |
1811 | yielded = kvm_vcpu_yield_to(vcpu); | |
1812 | if (yielded > 0) { | |
217ece61 | 1813 | kvm->last_boosted_vcpu = i; |
217ece61 | 1814 | break; |
c45c528e R |
1815 | } else if (yielded < 0) { |
1816 | try--; | |
1817 | if (!try) | |
1818 | break; | |
217ece61 | 1819 | } |
217ece61 RR |
1820 | } |
1821 | } | |
4c088493 | 1822 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
1823 | |
1824 | /* Ensure vcpu is not eligible during next spinloop */ | |
1825 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
1826 | } |
1827 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
1828 | ||
e4a533a4 | 1829 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
1830 | { |
1831 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
1832 | struct page *page; |
1833 | ||
e4a533a4 | 1834 | if (vmf->pgoff == 0) |
039576c0 | 1835 | page = virt_to_page(vcpu->run); |
09566765 | 1836 | #ifdef CONFIG_X86 |
e4a533a4 | 1837 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 1838 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
1839 | #endif |
1840 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
1841 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
1842 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 1843 | #endif |
039576c0 | 1844 | else |
5b1c1493 | 1845 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 1846 | get_page(page); |
e4a533a4 NP |
1847 | vmf->page = page; |
1848 | return 0; | |
9a2bb7f4 AK |
1849 | } |
1850 | ||
f0f37e2f | 1851 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 1852 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
1853 | }; |
1854 | ||
1855 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
1856 | { | |
1857 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
1858 | return 0; | |
1859 | } | |
1860 | ||
bccf2150 AK |
1861 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
1862 | { | |
1863 | struct kvm_vcpu *vcpu = filp->private_data; | |
1864 | ||
66c0b394 | 1865 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
1866 | return 0; |
1867 | } | |
1868 | ||
3d3aab1b | 1869 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
1870 | .release = kvm_vcpu_release, |
1871 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
1dda606c AG |
1872 | #ifdef CONFIG_COMPAT |
1873 | .compat_ioctl = kvm_vcpu_compat_ioctl, | |
1874 | #endif | |
9a2bb7f4 | 1875 | .mmap = kvm_vcpu_mmap, |
6038f373 | 1876 | .llseek = noop_llseek, |
bccf2150 AK |
1877 | }; |
1878 | ||
1879 | /* | |
1880 | * Allocates an inode for the vcpu. | |
1881 | */ | |
1882 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
1883 | { | |
24009b05 | 1884 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
1885 | } |
1886 | ||
c5ea7660 AK |
1887 | /* |
1888 | * Creates some virtual cpus. Good luck creating more than one. | |
1889 | */ | |
73880c80 | 1890 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
1891 | { |
1892 | int r; | |
988a2cae | 1893 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 1894 | |
73880c80 | 1895 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
1896 | if (IS_ERR(vcpu)) |
1897 | return PTR_ERR(vcpu); | |
c5ea7660 | 1898 | |
15ad7146 AK |
1899 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
1900 | ||
26e5215f AK |
1901 | r = kvm_arch_vcpu_setup(vcpu); |
1902 | if (r) | |
d780592b | 1903 | goto vcpu_destroy; |
26e5215f | 1904 | |
11ec2804 | 1905 | mutex_lock(&kvm->lock); |
3e515705 AK |
1906 | if (!kvm_vcpu_compatible(vcpu)) { |
1907 | r = -EINVAL; | |
1908 | goto unlock_vcpu_destroy; | |
1909 | } | |
73880c80 GN |
1910 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
1911 | r = -EINVAL; | |
d780592b | 1912 | goto unlock_vcpu_destroy; |
fb3f0f51 | 1913 | } |
73880c80 | 1914 | |
988a2cae GN |
1915 | kvm_for_each_vcpu(r, v, kvm) |
1916 | if (v->vcpu_id == id) { | |
73880c80 | 1917 | r = -EEXIST; |
d780592b | 1918 | goto unlock_vcpu_destroy; |
73880c80 GN |
1919 | } |
1920 | ||
1921 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 1922 | |
fb3f0f51 | 1923 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 1924 | kvm_get_kvm(kvm); |
bccf2150 | 1925 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
1926 | if (r < 0) { |
1927 | kvm_put_kvm(kvm); | |
d780592b | 1928 | goto unlock_vcpu_destroy; |
73880c80 GN |
1929 | } |
1930 | ||
1931 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
1932 | smp_wmb(); | |
1933 | atomic_inc(&kvm->online_vcpus); | |
1934 | ||
73880c80 | 1935 | mutex_unlock(&kvm->lock); |
42897d86 | 1936 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 1937 | return r; |
39c3b86e | 1938 | |
d780592b | 1939 | unlock_vcpu_destroy: |
7d8fece6 | 1940 | mutex_unlock(&kvm->lock); |
d780592b | 1941 | vcpu_destroy: |
d40ccc62 | 1942 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
1943 | return r; |
1944 | } | |
1945 | ||
1961d276 AK |
1946 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
1947 | { | |
1948 | if (sigset) { | |
1949 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1950 | vcpu->sigset_active = 1; | |
1951 | vcpu->sigset = *sigset; | |
1952 | } else | |
1953 | vcpu->sigset_active = 0; | |
1954 | return 0; | |
1955 | } | |
1956 | ||
bccf2150 AK |
1957 | static long kvm_vcpu_ioctl(struct file *filp, |
1958 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 1959 | { |
bccf2150 | 1960 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 1961 | void __user *argp = (void __user *)arg; |
313a3dc7 | 1962 | int r; |
fa3795a7 DH |
1963 | struct kvm_fpu *fpu = NULL; |
1964 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 1965 | |
6d4e4c4f AK |
1966 | if (vcpu->kvm->mm != current->mm) |
1967 | return -EIO; | |
2122ff5e | 1968 | |
2f4d9b54 | 1969 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
1970 | /* |
1971 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
1972 | * so vcpu_load() would break it. | |
1973 | */ | |
1974 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT) | |
1975 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); | |
1976 | #endif | |
1977 | ||
1978 | ||
9fc77441 MT |
1979 | r = vcpu_load(vcpu); |
1980 | if (r) | |
1981 | return r; | |
6aa8b732 | 1982 | switch (ioctl) { |
9a2bb7f4 | 1983 | case KVM_RUN: |
f0fe5108 AK |
1984 | r = -EINVAL; |
1985 | if (arg) | |
1986 | goto out; | |
b6c7a5dc | 1987 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 1988 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 1989 | break; |
6aa8b732 | 1990 | case KVM_GET_REGS: { |
3e4bb3ac | 1991 | struct kvm_regs *kvm_regs; |
6aa8b732 | 1992 | |
3e4bb3ac XZ |
1993 | r = -ENOMEM; |
1994 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
1995 | if (!kvm_regs) | |
6aa8b732 | 1996 | goto out; |
3e4bb3ac XZ |
1997 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
1998 | if (r) | |
1999 | goto out_free1; | |
6aa8b732 | 2000 | r = -EFAULT; |
3e4bb3ac XZ |
2001 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2002 | goto out_free1; | |
6aa8b732 | 2003 | r = 0; |
3e4bb3ac XZ |
2004 | out_free1: |
2005 | kfree(kvm_regs); | |
6aa8b732 AK |
2006 | break; |
2007 | } | |
2008 | case KVM_SET_REGS: { | |
3e4bb3ac | 2009 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2010 | |
3e4bb3ac | 2011 | r = -ENOMEM; |
ff5c2c03 SL |
2012 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2013 | if (IS_ERR(kvm_regs)) { | |
2014 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2015 | goto out; |
ff5c2c03 | 2016 | } |
3e4bb3ac | 2017 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2018 | kfree(kvm_regs); |
6aa8b732 AK |
2019 | break; |
2020 | } | |
2021 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2022 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2023 | r = -ENOMEM; | |
2024 | if (!kvm_sregs) | |
2025 | goto out; | |
2026 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2027 | if (r) |
2028 | goto out; | |
2029 | r = -EFAULT; | |
fa3795a7 | 2030 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2031 | goto out; |
2032 | r = 0; | |
2033 | break; | |
2034 | } | |
2035 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2036 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2037 | if (IS_ERR(kvm_sregs)) { | |
2038 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2039 | kvm_sregs = NULL; |
6aa8b732 | 2040 | goto out; |
ff5c2c03 | 2041 | } |
fa3795a7 | 2042 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2043 | break; |
2044 | } | |
62d9f0db MT |
2045 | case KVM_GET_MP_STATE: { |
2046 | struct kvm_mp_state mp_state; | |
2047 | ||
2048 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2049 | if (r) | |
2050 | goto out; | |
2051 | r = -EFAULT; | |
2052 | if (copy_to_user(argp, &mp_state, sizeof mp_state)) | |
2053 | goto out; | |
2054 | r = 0; | |
2055 | break; | |
2056 | } | |
2057 | case KVM_SET_MP_STATE: { | |
2058 | struct kvm_mp_state mp_state; | |
2059 | ||
2060 | r = -EFAULT; | |
2061 | if (copy_from_user(&mp_state, argp, sizeof mp_state)) | |
2062 | goto out; | |
2063 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2064 | break; |
2065 | } | |
6aa8b732 AK |
2066 | case KVM_TRANSLATE: { |
2067 | struct kvm_translation tr; | |
2068 | ||
2069 | r = -EFAULT; | |
2f366987 | 2070 | if (copy_from_user(&tr, argp, sizeof tr)) |
6aa8b732 | 2071 | goto out; |
8b006791 | 2072 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2073 | if (r) |
2074 | goto out; | |
2075 | r = -EFAULT; | |
2f366987 | 2076 | if (copy_to_user(argp, &tr, sizeof tr)) |
6aa8b732 AK |
2077 | goto out; |
2078 | r = 0; | |
2079 | break; | |
2080 | } | |
d0bfb940 JK |
2081 | case KVM_SET_GUEST_DEBUG: { |
2082 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2083 | |
2084 | r = -EFAULT; | |
2f366987 | 2085 | if (copy_from_user(&dbg, argp, sizeof dbg)) |
6aa8b732 | 2086 | goto out; |
d0bfb940 | 2087 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2088 | break; |
2089 | } | |
1961d276 AK |
2090 | case KVM_SET_SIGNAL_MASK: { |
2091 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2092 | struct kvm_signal_mask kvm_sigmask; | |
2093 | sigset_t sigset, *p; | |
2094 | ||
2095 | p = NULL; | |
2096 | if (argp) { | |
2097 | r = -EFAULT; | |
2098 | if (copy_from_user(&kvm_sigmask, argp, | |
2099 | sizeof kvm_sigmask)) | |
2100 | goto out; | |
2101 | r = -EINVAL; | |
2102 | if (kvm_sigmask.len != sizeof sigset) | |
2103 | goto out; | |
2104 | r = -EFAULT; | |
2105 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
2106 | sizeof sigset)) | |
2107 | goto out; | |
2108 | p = &sigset; | |
2109 | } | |
376d41ff | 2110 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2111 | break; |
2112 | } | |
b8836737 | 2113 | case KVM_GET_FPU: { |
fa3795a7 DH |
2114 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2115 | r = -ENOMEM; | |
2116 | if (!fpu) | |
2117 | goto out; | |
2118 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2119 | if (r) |
2120 | goto out; | |
2121 | r = -EFAULT; | |
fa3795a7 | 2122 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2123 | goto out; |
2124 | r = 0; | |
2125 | break; | |
2126 | } | |
2127 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2128 | fpu = memdup_user(argp, sizeof(*fpu)); |
2129 | if (IS_ERR(fpu)) { | |
2130 | r = PTR_ERR(fpu); | |
18595411 | 2131 | fpu = NULL; |
b8836737 | 2132 | goto out; |
ff5c2c03 | 2133 | } |
fa3795a7 | 2134 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2135 | break; |
2136 | } | |
bccf2150 | 2137 | default: |
313a3dc7 | 2138 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2139 | } |
2140 | out: | |
2122ff5e | 2141 | vcpu_put(vcpu); |
fa3795a7 DH |
2142 | kfree(fpu); |
2143 | kfree(kvm_sregs); | |
bccf2150 AK |
2144 | return r; |
2145 | } | |
2146 | ||
1dda606c AG |
2147 | #ifdef CONFIG_COMPAT |
2148 | static long kvm_vcpu_compat_ioctl(struct file *filp, | |
2149 | unsigned int ioctl, unsigned long arg) | |
2150 | { | |
2151 | struct kvm_vcpu *vcpu = filp->private_data; | |
2152 | void __user *argp = compat_ptr(arg); | |
2153 | int r; | |
2154 | ||
2155 | if (vcpu->kvm->mm != current->mm) | |
2156 | return -EIO; | |
2157 | ||
2158 | switch (ioctl) { | |
2159 | case KVM_SET_SIGNAL_MASK: { | |
2160 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2161 | struct kvm_signal_mask kvm_sigmask; | |
2162 | compat_sigset_t csigset; | |
2163 | sigset_t sigset; | |
2164 | ||
2165 | if (argp) { | |
2166 | r = -EFAULT; | |
2167 | if (copy_from_user(&kvm_sigmask, argp, | |
2168 | sizeof kvm_sigmask)) | |
2169 | goto out; | |
2170 | r = -EINVAL; | |
2171 | if (kvm_sigmask.len != sizeof csigset) | |
2172 | goto out; | |
2173 | r = -EFAULT; | |
2174 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
2175 | sizeof csigset)) | |
2176 | goto out; | |
760a9a30 AC |
2177 | sigset_from_compat(&sigset, &csigset); |
2178 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2179 | } else | |
2180 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2181 | break; |
2182 | } | |
2183 | default: | |
2184 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2185 | } | |
2186 | ||
2187 | out: | |
2188 | return r; | |
2189 | } | |
2190 | #endif | |
2191 | ||
852b6d57 SW |
2192 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2193 | int (*accessor)(struct kvm_device *dev, | |
2194 | struct kvm_device_attr *attr), | |
2195 | unsigned long arg) | |
2196 | { | |
2197 | struct kvm_device_attr attr; | |
2198 | ||
2199 | if (!accessor) | |
2200 | return -EPERM; | |
2201 | ||
2202 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2203 | return -EFAULT; | |
2204 | ||
2205 | return accessor(dev, &attr); | |
2206 | } | |
2207 | ||
2208 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2209 | unsigned long arg) | |
2210 | { | |
2211 | struct kvm_device *dev = filp->private_data; | |
2212 | ||
2213 | switch (ioctl) { | |
2214 | case KVM_SET_DEVICE_ATTR: | |
2215 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2216 | case KVM_GET_DEVICE_ATTR: | |
2217 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2218 | case KVM_HAS_DEVICE_ATTR: | |
2219 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2220 | default: | |
2221 | if (dev->ops->ioctl) | |
2222 | return dev->ops->ioctl(dev, ioctl, arg); | |
2223 | ||
2224 | return -ENOTTY; | |
2225 | } | |
2226 | } | |
2227 | ||
852b6d57 SW |
2228 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2229 | { | |
2230 | struct kvm_device *dev = filp->private_data; | |
2231 | struct kvm *kvm = dev->kvm; | |
2232 | ||
852b6d57 SW |
2233 | kvm_put_kvm(kvm); |
2234 | return 0; | |
2235 | } | |
2236 | ||
2237 | static const struct file_operations kvm_device_fops = { | |
2238 | .unlocked_ioctl = kvm_device_ioctl, | |
db6ae615 SW |
2239 | #ifdef CONFIG_COMPAT |
2240 | .compat_ioctl = kvm_device_ioctl, | |
2241 | #endif | |
852b6d57 SW |
2242 | .release = kvm_device_release, |
2243 | }; | |
2244 | ||
2245 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2246 | { | |
2247 | if (filp->f_op != &kvm_device_fops) | |
2248 | return NULL; | |
2249 | ||
2250 | return filp->private_data; | |
2251 | } | |
2252 | ||
2253 | static int kvm_ioctl_create_device(struct kvm *kvm, | |
2254 | struct kvm_create_device *cd) | |
2255 | { | |
2256 | struct kvm_device_ops *ops = NULL; | |
2257 | struct kvm_device *dev; | |
2258 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2259 | int ret; | |
2260 | ||
2261 | switch (cd->type) { | |
5df554ad SW |
2262 | #ifdef CONFIG_KVM_MPIC |
2263 | case KVM_DEV_TYPE_FSL_MPIC_20: | |
2264 | case KVM_DEV_TYPE_FSL_MPIC_42: | |
2265 | ops = &kvm_mpic_ops; | |
2266 | break; | |
5975a2e0 PM |
2267 | #endif |
2268 | #ifdef CONFIG_KVM_XICS | |
2269 | case KVM_DEV_TYPE_XICS: | |
2270 | ops = &kvm_xics_ops; | |
2271 | break; | |
5df554ad | 2272 | #endif |
852b6d57 SW |
2273 | default: |
2274 | return -ENODEV; | |
2275 | } | |
2276 | ||
2277 | if (test) | |
2278 | return 0; | |
2279 | ||
2280 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2281 | if (!dev) | |
2282 | return -ENOMEM; | |
2283 | ||
2284 | dev->ops = ops; | |
2285 | dev->kvm = kvm; | |
852b6d57 SW |
2286 | |
2287 | ret = ops->create(dev, cd->type); | |
2288 | if (ret < 0) { | |
2289 | kfree(dev); | |
2290 | return ret; | |
2291 | } | |
2292 | ||
24009b05 | 2293 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2294 | if (ret < 0) { |
2295 | ops->destroy(dev); | |
2296 | return ret; | |
2297 | } | |
2298 | ||
07f0a7bd | 2299 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2300 | kvm_get_kvm(kvm); |
2301 | cd->fd = ret; | |
2302 | return 0; | |
2303 | } | |
2304 | ||
bccf2150 AK |
2305 | static long kvm_vm_ioctl(struct file *filp, |
2306 | unsigned int ioctl, unsigned long arg) | |
2307 | { | |
2308 | struct kvm *kvm = filp->private_data; | |
2309 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2310 | int r; |
bccf2150 | 2311 | |
6d4e4c4f AK |
2312 | if (kvm->mm != current->mm) |
2313 | return -EIO; | |
bccf2150 AK |
2314 | switch (ioctl) { |
2315 | case KVM_CREATE_VCPU: | |
2316 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2317 | break; |
6fc138d2 IE |
2318 | case KVM_SET_USER_MEMORY_REGION: { |
2319 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2320 | ||
2321 | r = -EFAULT; | |
2322 | if (copy_from_user(&kvm_userspace_mem, argp, | |
2323 | sizeof kvm_userspace_mem)) | |
2324 | goto out; | |
2325 | ||
47ae31e2 | 2326 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2327 | break; |
2328 | } | |
2329 | case KVM_GET_DIRTY_LOG: { | |
2330 | struct kvm_dirty_log log; | |
2331 | ||
2332 | r = -EFAULT; | |
2f366987 | 2333 | if (copy_from_user(&log, argp, sizeof log)) |
6aa8b732 | 2334 | goto out; |
2c6f5df9 | 2335 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2336 | break; |
2337 | } | |
5f94c174 LV |
2338 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2339 | case KVM_REGISTER_COALESCED_MMIO: { | |
2340 | struct kvm_coalesced_mmio_zone zone; | |
2341 | r = -EFAULT; | |
2342 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2343 | goto out; | |
5f94c174 | 2344 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2345 | break; |
2346 | } | |
2347 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2348 | struct kvm_coalesced_mmio_zone zone; | |
2349 | r = -EFAULT; | |
2350 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2351 | goto out; | |
5f94c174 | 2352 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2353 | break; |
2354 | } | |
2355 | #endif | |
721eecbf GH |
2356 | case KVM_IRQFD: { |
2357 | struct kvm_irqfd data; | |
2358 | ||
2359 | r = -EFAULT; | |
2360 | if (copy_from_user(&data, argp, sizeof data)) | |
2361 | goto out; | |
d4db2935 | 2362 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2363 | break; |
2364 | } | |
d34e6b17 GH |
2365 | case KVM_IOEVENTFD: { |
2366 | struct kvm_ioeventfd data; | |
2367 | ||
2368 | r = -EFAULT; | |
2369 | if (copy_from_user(&data, argp, sizeof data)) | |
2370 | goto out; | |
2371 | r = kvm_ioeventfd(kvm, &data); | |
2372 | break; | |
2373 | } | |
73880c80 GN |
2374 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2375 | case KVM_SET_BOOT_CPU_ID: | |
2376 | r = 0; | |
894a9c55 | 2377 | mutex_lock(&kvm->lock); |
73880c80 GN |
2378 | if (atomic_read(&kvm->online_vcpus) != 0) |
2379 | r = -EBUSY; | |
2380 | else | |
2381 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2382 | mutex_unlock(&kvm->lock); |
73880c80 | 2383 | break; |
07975ad3 JK |
2384 | #endif |
2385 | #ifdef CONFIG_HAVE_KVM_MSI | |
2386 | case KVM_SIGNAL_MSI: { | |
2387 | struct kvm_msi msi; | |
2388 | ||
2389 | r = -EFAULT; | |
2390 | if (copy_from_user(&msi, argp, sizeof msi)) | |
2391 | goto out; | |
2392 | r = kvm_send_userspace_msi(kvm, &msi); | |
2393 | break; | |
2394 | } | |
23d43cf9 CD |
2395 | #endif |
2396 | #ifdef __KVM_HAVE_IRQ_LINE | |
2397 | case KVM_IRQ_LINE_STATUS: | |
2398 | case KVM_IRQ_LINE: { | |
2399 | struct kvm_irq_level irq_event; | |
2400 | ||
2401 | r = -EFAULT; | |
2402 | if (copy_from_user(&irq_event, argp, sizeof irq_event)) | |
2403 | goto out; | |
2404 | ||
aa2fbe6d YZ |
2405 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2406 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2407 | if (r) |
2408 | goto out; | |
2409 | ||
2410 | r = -EFAULT; | |
2411 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
2412 | if (copy_to_user(argp, &irq_event, sizeof irq_event)) | |
2413 | goto out; | |
2414 | } | |
2415 | ||
2416 | r = 0; | |
2417 | break; | |
2418 | } | |
73880c80 | 2419 | #endif |
aa8d5944 AG |
2420 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2421 | case KVM_SET_GSI_ROUTING: { | |
2422 | struct kvm_irq_routing routing; | |
2423 | struct kvm_irq_routing __user *urouting; | |
2424 | struct kvm_irq_routing_entry *entries; | |
2425 | ||
2426 | r = -EFAULT; | |
2427 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2428 | goto out; | |
2429 | r = -EINVAL; | |
2430 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2431 | goto out; | |
2432 | if (routing.flags) | |
2433 | goto out; | |
2434 | r = -ENOMEM; | |
2435 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2436 | if (!entries) | |
2437 | goto out; | |
2438 | r = -EFAULT; | |
2439 | urouting = argp; | |
2440 | if (copy_from_user(entries, urouting->entries, | |
2441 | routing.nr * sizeof(*entries))) | |
2442 | goto out_free_irq_routing; | |
2443 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2444 | routing.flags); | |
2445 | out_free_irq_routing: | |
2446 | vfree(entries); | |
2447 | break; | |
2448 | } | |
2449 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2450 | case KVM_CREATE_DEVICE: { |
2451 | struct kvm_create_device cd; | |
2452 | ||
2453 | r = -EFAULT; | |
2454 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2455 | goto out; | |
2456 | ||
2457 | r = kvm_ioctl_create_device(kvm, &cd); | |
2458 | if (r) | |
2459 | goto out; | |
2460 | ||
2461 | r = -EFAULT; | |
2462 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2463 | goto out; | |
2464 | ||
2465 | r = 0; | |
2466 | break; | |
2467 | } | |
f17abe9a | 2468 | default: |
1fe779f8 | 2469 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
bfd99ff5 AK |
2470 | if (r == -ENOTTY) |
2471 | r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg); | |
f17abe9a AK |
2472 | } |
2473 | out: | |
2474 | return r; | |
2475 | } | |
2476 | ||
6ff5894c AB |
2477 | #ifdef CONFIG_COMPAT |
2478 | struct compat_kvm_dirty_log { | |
2479 | __u32 slot; | |
2480 | __u32 padding1; | |
2481 | union { | |
2482 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2483 | __u64 padding2; | |
2484 | }; | |
2485 | }; | |
2486 | ||
2487 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2488 | unsigned int ioctl, unsigned long arg) | |
2489 | { | |
2490 | struct kvm *kvm = filp->private_data; | |
2491 | int r; | |
2492 | ||
2493 | if (kvm->mm != current->mm) | |
2494 | return -EIO; | |
2495 | switch (ioctl) { | |
2496 | case KVM_GET_DIRTY_LOG: { | |
2497 | struct compat_kvm_dirty_log compat_log; | |
2498 | struct kvm_dirty_log log; | |
2499 | ||
2500 | r = -EFAULT; | |
2501 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2502 | sizeof(compat_log))) | |
2503 | goto out; | |
2504 | log.slot = compat_log.slot; | |
2505 | log.padding1 = compat_log.padding1; | |
2506 | log.padding2 = compat_log.padding2; | |
2507 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2508 | ||
2509 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2510 | break; |
2511 | } | |
2512 | default: | |
2513 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2514 | } | |
2515 | ||
2516 | out: | |
2517 | return r; | |
2518 | } | |
2519 | #endif | |
2520 | ||
e4a533a4 | 2521 | static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
f17abe9a | 2522 | { |
777b3f49 MT |
2523 | struct page *page[1]; |
2524 | unsigned long addr; | |
2525 | int npages; | |
2526 | gfn_t gfn = vmf->pgoff; | |
f17abe9a | 2527 | struct kvm *kvm = vma->vm_file->private_data; |
f17abe9a | 2528 | |
777b3f49 MT |
2529 | addr = gfn_to_hva(kvm, gfn); |
2530 | if (kvm_is_error_hva(addr)) | |
e4a533a4 | 2531 | return VM_FAULT_SIGBUS; |
777b3f49 MT |
2532 | |
2533 | npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page, | |
2534 | NULL); | |
2535 | if (unlikely(npages != 1)) | |
e4a533a4 | 2536 | return VM_FAULT_SIGBUS; |
777b3f49 MT |
2537 | |
2538 | vmf->page = page[0]; | |
e4a533a4 | 2539 | return 0; |
f17abe9a AK |
2540 | } |
2541 | ||
f0f37e2f | 2542 | static const struct vm_operations_struct kvm_vm_vm_ops = { |
e4a533a4 | 2543 | .fault = kvm_vm_fault, |
f17abe9a AK |
2544 | }; |
2545 | ||
2546 | static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma) | |
2547 | { | |
2548 | vma->vm_ops = &kvm_vm_vm_ops; | |
2549 | return 0; | |
2550 | } | |
2551 | ||
3d3aab1b | 2552 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2553 | .release = kvm_vm_release, |
2554 | .unlocked_ioctl = kvm_vm_ioctl, | |
6ff5894c AB |
2555 | #ifdef CONFIG_COMPAT |
2556 | .compat_ioctl = kvm_vm_compat_ioctl, | |
2557 | #endif | |
f17abe9a | 2558 | .mmap = kvm_vm_mmap, |
6038f373 | 2559 | .llseek = noop_llseek, |
f17abe9a AK |
2560 | }; |
2561 | ||
e08b9637 | 2562 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2563 | { |
aac87636 | 2564 | int r; |
f17abe9a AK |
2565 | struct kvm *kvm; |
2566 | ||
e08b9637 | 2567 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2568 | if (IS_ERR(kvm)) |
2569 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2570 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2571 | r = kvm_coalesced_mmio_init(kvm); | |
2572 | if (r < 0) { | |
2573 | kvm_put_kvm(kvm); | |
2574 | return r; | |
2575 | } | |
2576 | #endif | |
24009b05 | 2577 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2578 | if (r < 0) |
66c0b394 | 2579 | kvm_put_kvm(kvm); |
f17abe9a | 2580 | |
aac87636 | 2581 | return r; |
f17abe9a AK |
2582 | } |
2583 | ||
1a811b61 AK |
2584 | static long kvm_dev_ioctl_check_extension_generic(long arg) |
2585 | { | |
2586 | switch (arg) { | |
ca9edaee | 2587 | case KVM_CAP_USER_MEMORY: |
1a811b61 | 2588 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
4cd481f6 | 2589 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: |
73880c80 GN |
2590 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2591 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2592 | #endif | |
a9c7399d | 2593 | case KVM_CAP_INTERNAL_ERROR_DATA: |
07975ad3 JK |
2594 | #ifdef CONFIG_HAVE_KVM_MSI |
2595 | case KVM_CAP_SIGNAL_MSI: | |
7df35f54 AG |
2596 | #endif |
2597 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2598 | case KVM_CAP_IRQFD_RESAMPLE: | |
07975ad3 | 2599 | #endif |
1a811b61 | 2600 | return 1; |
a725d56a | 2601 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
399ec807 | 2602 | case KVM_CAP_IRQ_ROUTING: |
36463146 | 2603 | return KVM_MAX_IRQ_ROUTES; |
399ec807 | 2604 | #endif |
1a811b61 AK |
2605 | default: |
2606 | break; | |
2607 | } | |
2608 | return kvm_dev_ioctl_check_extension(arg); | |
2609 | } | |
2610 | ||
f17abe9a AK |
2611 | static long kvm_dev_ioctl(struct file *filp, |
2612 | unsigned int ioctl, unsigned long arg) | |
2613 | { | |
07c45a36 | 2614 | long r = -EINVAL; |
f17abe9a AK |
2615 | |
2616 | switch (ioctl) { | |
2617 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2618 | r = -EINVAL; |
2619 | if (arg) | |
2620 | goto out; | |
f17abe9a AK |
2621 | r = KVM_API_VERSION; |
2622 | break; | |
2623 | case KVM_CREATE_VM: | |
e08b9637 | 2624 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2625 | break; |
018d00d2 | 2626 | case KVM_CHECK_EXTENSION: |
1a811b61 | 2627 | r = kvm_dev_ioctl_check_extension_generic(arg); |
5d308f45 | 2628 | break; |
07c45a36 AK |
2629 | case KVM_GET_VCPU_MMAP_SIZE: |
2630 | r = -EINVAL; | |
2631 | if (arg) | |
2632 | goto out; | |
adb1ff46 AK |
2633 | r = PAGE_SIZE; /* struct kvm_run */ |
2634 | #ifdef CONFIG_X86 | |
2635 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2636 | #endif |
2637 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2638 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2639 | #endif |
07c45a36 | 2640 | break; |
d4c9ff2d FEL |
2641 | case KVM_TRACE_ENABLE: |
2642 | case KVM_TRACE_PAUSE: | |
2643 | case KVM_TRACE_DISABLE: | |
2023a29c | 2644 | r = -EOPNOTSUPP; |
d4c9ff2d | 2645 | break; |
6aa8b732 | 2646 | default: |
043405e1 | 2647 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2648 | } |
2649 | out: | |
2650 | return r; | |
2651 | } | |
2652 | ||
6aa8b732 | 2653 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2654 | .unlocked_ioctl = kvm_dev_ioctl, |
2655 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2656 | .llseek = noop_llseek, |
6aa8b732 AK |
2657 | }; |
2658 | ||
2659 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2660 | KVM_MINOR, |
6aa8b732 AK |
2661 | "kvm", |
2662 | &kvm_chardev_ops, | |
2663 | }; | |
2664 | ||
75b7127c | 2665 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2666 | { |
2667 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2668 | int r; |
1b6c0168 | 2669 | |
7f59f492 | 2670 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2671 | return; |
10474ae8 | 2672 | |
7f59f492 | 2673 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 AG |
2674 | |
2675 | r = kvm_arch_hardware_enable(NULL); | |
2676 | ||
2677 | if (r) { | |
2678 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2679 | atomic_inc(&hardware_enable_failed); | |
2680 | printk(KERN_INFO "kvm: enabling virtualization on " | |
2681 | "CPU%d failed\n", cpu); | |
2682 | } | |
1b6c0168 AK |
2683 | } |
2684 | ||
4fa92fb2 | 2685 | static void hardware_enable(void) |
75b7127c | 2686 | { |
4a937f96 | 2687 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2688 | if (kvm_usage_count) |
2689 | hardware_enable_nolock(NULL); | |
4a937f96 | 2690 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2691 | } |
2692 | ||
2693 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2694 | { |
2695 | int cpu = raw_smp_processor_id(); | |
2696 | ||
7f59f492 | 2697 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2698 | return; |
7f59f492 | 2699 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
e9b11c17 | 2700 | kvm_arch_hardware_disable(NULL); |
1b6c0168 AK |
2701 | } |
2702 | ||
4fa92fb2 | 2703 | static void hardware_disable(void) |
75b7127c | 2704 | { |
4a937f96 | 2705 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2706 | if (kvm_usage_count) |
2707 | hardware_disable_nolock(NULL); | |
4a937f96 | 2708 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2709 | } |
2710 | ||
10474ae8 AG |
2711 | static void hardware_disable_all_nolock(void) |
2712 | { | |
2713 | BUG_ON(!kvm_usage_count); | |
2714 | ||
2715 | kvm_usage_count--; | |
2716 | if (!kvm_usage_count) | |
75b7127c | 2717 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2718 | } |
2719 | ||
2720 | static void hardware_disable_all(void) | |
2721 | { | |
4a937f96 | 2722 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 2723 | hardware_disable_all_nolock(); |
4a937f96 | 2724 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2725 | } |
2726 | ||
2727 | static int hardware_enable_all(void) | |
2728 | { | |
2729 | int r = 0; | |
2730 | ||
4a937f96 | 2731 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
2732 | |
2733 | kvm_usage_count++; | |
2734 | if (kvm_usage_count == 1) { | |
2735 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 2736 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
2737 | |
2738 | if (atomic_read(&hardware_enable_failed)) { | |
2739 | hardware_disable_all_nolock(); | |
2740 | r = -EBUSY; | |
2741 | } | |
2742 | } | |
2743 | ||
4a937f96 | 2744 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2745 | |
2746 | return r; | |
2747 | } | |
2748 | ||
774c47f1 AK |
2749 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
2750 | void *v) | |
2751 | { | |
2752 | int cpu = (long)v; | |
2753 | ||
1a6f4d7f | 2754 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 2755 | switch (val) { |
cec9ad27 | 2756 | case CPU_DYING: |
6ec8a856 AK |
2757 | printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n", |
2758 | cpu); | |
4fa92fb2 | 2759 | hardware_disable(); |
6ec8a856 | 2760 | break; |
da908f2f | 2761 | case CPU_STARTING: |
43934a38 JK |
2762 | printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n", |
2763 | cpu); | |
4fa92fb2 | 2764 | hardware_enable(); |
774c47f1 AK |
2765 | break; |
2766 | } | |
2767 | return NOTIFY_OK; | |
2768 | } | |
2769 | ||
9a2b85c6 | 2770 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 2771 | void *v) |
9a2b85c6 | 2772 | { |
8e1c1815 SY |
2773 | /* |
2774 | * Some (well, at least mine) BIOSes hang on reboot if | |
2775 | * in vmx root mode. | |
2776 | * | |
2777 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
2778 | */ | |
2779 | printk(KERN_INFO "kvm: exiting hardware virtualization\n"); | |
2780 | kvm_rebooting = true; | |
75b7127c | 2781 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
2782 | return NOTIFY_OK; |
2783 | } | |
2784 | ||
2785 | static struct notifier_block kvm_reboot_notifier = { | |
2786 | .notifier_call = kvm_reboot, | |
2787 | .priority = 0, | |
2788 | }; | |
2789 | ||
e93f8a0f | 2790 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
2791 | { |
2792 | int i; | |
2793 | ||
2794 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 2795 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
2796 | |
2797 | kvm_iodevice_destructor(pos); | |
2798 | } | |
e93f8a0f | 2799 | kfree(bus); |
2eeb2e94 GH |
2800 | } |
2801 | ||
c21fbff1 PB |
2802 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
2803 | const struct kvm_io_range *r2) | |
743eeb0b | 2804 | { |
743eeb0b SL |
2805 | if (r1->addr < r2->addr) |
2806 | return -1; | |
2807 | if (r1->addr + r1->len > r2->addr + r2->len) | |
2808 | return 1; | |
2809 | return 0; | |
2810 | } | |
2811 | ||
a343c9b7 PB |
2812 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
2813 | { | |
c21fbff1 | 2814 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
2815 | } |
2816 | ||
39369f7a | 2817 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
2818 | gpa_t addr, int len) |
2819 | { | |
743eeb0b SL |
2820 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
2821 | .addr = addr, | |
2822 | .len = len, | |
2823 | .dev = dev, | |
2824 | }; | |
2825 | ||
2826 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
2827 | kvm_io_bus_sort_cmp, NULL); | |
2828 | ||
2829 | return 0; | |
2830 | } | |
2831 | ||
39369f7a | 2832 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
2833 | gpa_t addr, int len) |
2834 | { | |
2835 | struct kvm_io_range *range, key; | |
2836 | int off; | |
2837 | ||
2838 | key = (struct kvm_io_range) { | |
2839 | .addr = addr, | |
2840 | .len = len, | |
2841 | }; | |
2842 | ||
2843 | range = bsearch(&key, bus->range, bus->dev_count, | |
2844 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
2845 | if (range == NULL) | |
2846 | return -ENOENT; | |
2847 | ||
2848 | off = range - bus->range; | |
2849 | ||
c21fbff1 | 2850 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
2851 | off--; |
2852 | ||
2853 | return off; | |
2854 | } | |
2855 | ||
126a5af5 CH |
2856 | static int __kvm_io_bus_write(struct kvm_io_bus *bus, |
2857 | struct kvm_io_range *range, const void *val) | |
2858 | { | |
2859 | int idx; | |
2860 | ||
2861 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
2862 | if (idx < 0) | |
2863 | return -EOPNOTSUPP; | |
2864 | ||
2865 | while (idx < bus->dev_count && | |
c21fbff1 | 2866 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
2867 | if (!kvm_iodevice_write(bus->range[idx].dev, range->addr, |
2868 | range->len, val)) | |
2869 | return idx; | |
2870 | idx++; | |
2871 | } | |
2872 | ||
2873 | return -EOPNOTSUPP; | |
2874 | } | |
2875 | ||
bda9020e | 2876 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e93f8a0f | 2877 | int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 2878 | int len, const void *val) |
2eeb2e94 | 2879 | { |
90d83dc3 | 2880 | struct kvm_io_bus *bus; |
743eeb0b | 2881 | struct kvm_io_range range; |
126a5af5 | 2882 | int r; |
743eeb0b SL |
2883 | |
2884 | range = (struct kvm_io_range) { | |
2885 | .addr = addr, | |
2886 | .len = len, | |
2887 | }; | |
90d83dc3 LJ |
2888 | |
2889 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
126a5af5 CH |
2890 | r = __kvm_io_bus_write(bus, &range, val); |
2891 | return r < 0 ? r : 0; | |
2892 | } | |
2893 | ||
2894 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
2895 | int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, | |
2896 | int len, const void *val, long cookie) | |
2897 | { | |
2898 | struct kvm_io_bus *bus; | |
2899 | struct kvm_io_range range; | |
2900 | ||
2901 | range = (struct kvm_io_range) { | |
2902 | .addr = addr, | |
2903 | .len = len, | |
2904 | }; | |
2905 | ||
2906 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
2907 | ||
2908 | /* First try the device referenced by cookie. */ | |
2909 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 2910 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
126a5af5 CH |
2911 | if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len, |
2912 | val)) | |
2913 | return cookie; | |
2914 | ||
2915 | /* | |
2916 | * cookie contained garbage; fall back to search and return the | |
2917 | * correct cookie value. | |
2918 | */ | |
2919 | return __kvm_io_bus_write(bus, &range, val); | |
2920 | } | |
2921 | ||
2922 | static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range, | |
2923 | void *val) | |
2924 | { | |
2925 | int idx; | |
2926 | ||
2927 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
2928 | if (idx < 0) |
2929 | return -EOPNOTSUPP; | |
2930 | ||
2931 | while (idx < bus->dev_count && | |
c21fbff1 | 2932 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
2933 | if (!kvm_iodevice_read(bus->range[idx].dev, range->addr, |
2934 | range->len, val)) | |
2935 | return idx; | |
743eeb0b SL |
2936 | idx++; |
2937 | } | |
2938 | ||
bda9020e MT |
2939 | return -EOPNOTSUPP; |
2940 | } | |
2eeb2e94 | 2941 | |
bda9020e | 2942 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e93f8a0f MT |
2943 | int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2944 | int len, void *val) | |
bda9020e | 2945 | { |
90d83dc3 | 2946 | struct kvm_io_bus *bus; |
743eeb0b | 2947 | struct kvm_io_range range; |
126a5af5 | 2948 | int r; |
743eeb0b SL |
2949 | |
2950 | range = (struct kvm_io_range) { | |
2951 | .addr = addr, | |
2952 | .len = len, | |
2953 | }; | |
e93f8a0f | 2954 | |
90d83dc3 | 2955 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); |
126a5af5 CH |
2956 | r = __kvm_io_bus_read(bus, &range, val); |
2957 | return r < 0 ? r : 0; | |
2958 | } | |
743eeb0b | 2959 | |
126a5af5 CH |
2960 | /* kvm_io_bus_read_cookie - called under kvm->slots_lock */ |
2961 | int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, | |
2962 | int len, void *val, long cookie) | |
2963 | { | |
2964 | struct kvm_io_bus *bus; | |
2965 | struct kvm_io_range range; | |
743eeb0b | 2966 | |
126a5af5 CH |
2967 | range = (struct kvm_io_range) { |
2968 | .addr = addr, | |
2969 | .len = len, | |
2970 | }; | |
2971 | ||
2972 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
2973 | ||
2974 | /* First try the device referenced by cookie. */ | |
2975 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 2976 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
126a5af5 CH |
2977 | if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len, |
2978 | val)) | |
2979 | return cookie; | |
2980 | ||
2981 | /* | |
2982 | * cookie contained garbage; fall back to search and return the | |
2983 | * correct cookie value. | |
2984 | */ | |
2985 | return __kvm_io_bus_read(bus, &range, val); | |
2eeb2e94 GH |
2986 | } |
2987 | ||
79fac95e | 2988 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
2989 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2990 | int len, struct kvm_io_device *dev) | |
6c474694 | 2991 | { |
e93f8a0f | 2992 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 2993 | |
e93f8a0f | 2994 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
2995 | /* exclude ioeventfd which is limited by maximum fd */ |
2996 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 2997 | return -ENOSPC; |
2eeb2e94 | 2998 | |
a1300716 AK |
2999 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
3000 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
3001 | if (!new_bus) |
3002 | return -ENOMEM; | |
a1300716 AK |
3003 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3004 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3005 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3006 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3007 | synchronize_srcu_expedited(&kvm->srcu); | |
3008 | kfree(bus); | |
090b7aff GH |
3009 | |
3010 | return 0; | |
3011 | } | |
3012 | ||
79fac95e | 3013 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3014 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3015 | struct kvm_io_device *dev) | |
090b7aff | 3016 | { |
e93f8a0f MT |
3017 | int i, r; |
3018 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3019 | |
cdfca7b3 | 3020 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 3021 | r = -ENOENT; |
a1300716 AK |
3022 | for (i = 0; i < bus->dev_count; i++) |
3023 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3024 | r = 0; |
090b7aff GH |
3025 | break; |
3026 | } | |
e93f8a0f | 3027 | |
a1300716 | 3028 | if (r) |
e93f8a0f | 3029 | return r; |
a1300716 AK |
3030 | |
3031 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
3032 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
3033 | if (!new_bus) | |
3034 | return -ENOMEM; | |
3035 | ||
3036 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3037 | new_bus->dev_count--; | |
3038 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3039 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3040 | |
3041 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3042 | synchronize_srcu_expedited(&kvm->srcu); | |
3043 | kfree(bus); | |
3044 | return r; | |
2eeb2e94 GH |
3045 | } |
3046 | ||
774c47f1 AK |
3047 | static struct notifier_block kvm_cpu_notifier = { |
3048 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3049 | }; |
3050 | ||
8b88b099 | 3051 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3052 | { |
3053 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3054 | struct kvm *kvm; |
3055 | ||
8b88b099 | 3056 | *val = 0; |
2f303b74 | 3057 | spin_lock(&kvm_lock); |
ba1389b7 | 3058 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3059 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3060 | spin_unlock(&kvm_lock); |
8b88b099 | 3061 | return 0; |
ba1389b7 AK |
3062 | } |
3063 | ||
3064 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3065 | ||
8b88b099 | 3066 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3067 | { |
3068 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3069 | struct kvm *kvm; |
3070 | struct kvm_vcpu *vcpu; | |
3071 | int i; | |
3072 | ||
8b88b099 | 3073 | *val = 0; |
2f303b74 | 3074 | spin_lock(&kvm_lock); |
1165f5fe | 3075 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3076 | kvm_for_each_vcpu(i, vcpu, kvm) |
3077 | *val += *(u32 *)((void *)vcpu + offset); | |
3078 | ||
2f303b74 | 3079 | spin_unlock(&kvm_lock); |
8b88b099 | 3080 | return 0; |
1165f5fe AK |
3081 | } |
3082 | ||
ba1389b7 AK |
3083 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3084 | ||
828c0950 | 3085 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3086 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3087 | [KVM_STAT_VM] = &vm_stat_fops, | |
3088 | }; | |
1165f5fe | 3089 | |
4f69b680 | 3090 | static int kvm_init_debug(void) |
6aa8b732 | 3091 | { |
4f69b680 | 3092 | int r = -EFAULT; |
6aa8b732 AK |
3093 | struct kvm_stats_debugfs_item *p; |
3094 | ||
76f7c879 | 3095 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3096 | if (kvm_debugfs_dir == NULL) |
3097 | goto out; | |
3098 | ||
3099 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3100 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3101 | (void *)(long)p->offset, |
ba1389b7 | 3102 | stat_fops[p->kind]); |
4f69b680 H |
3103 | if (p->dentry == NULL) |
3104 | goto out_dir; | |
3105 | } | |
3106 | ||
3107 | return 0; | |
3108 | ||
3109 | out_dir: | |
3110 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3111 | out: | |
3112 | return r; | |
6aa8b732 AK |
3113 | } |
3114 | ||
3115 | static void kvm_exit_debug(void) | |
3116 | { | |
3117 | struct kvm_stats_debugfs_item *p; | |
3118 | ||
3119 | for (p = debugfs_entries; p->name; ++p) | |
3120 | debugfs_remove(p->dentry); | |
76f7c879 | 3121 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3122 | } |
3123 | ||
fb3600cc | 3124 | static int kvm_suspend(void) |
59ae6c6b | 3125 | { |
10474ae8 | 3126 | if (kvm_usage_count) |
75b7127c | 3127 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3128 | return 0; |
3129 | } | |
3130 | ||
fb3600cc | 3131 | static void kvm_resume(void) |
59ae6c6b | 3132 | { |
ca84d1a2 | 3133 | if (kvm_usage_count) { |
4a937f96 | 3134 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3135 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3136 | } |
59ae6c6b AK |
3137 | } |
3138 | ||
fb3600cc | 3139 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3140 | .suspend = kvm_suspend, |
3141 | .resume = kvm_resume, | |
3142 | }; | |
3143 | ||
15ad7146 AK |
3144 | static inline |
3145 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3146 | { | |
3147 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3148 | } | |
3149 | ||
3150 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3151 | { | |
3152 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3a08a8f9 R |
3153 | if (vcpu->preempted) |
3154 | vcpu->preempted = false; | |
15ad7146 | 3155 | |
e9b11c17 | 3156 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3157 | } |
3158 | ||
3159 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3160 | struct task_struct *next) | |
3161 | { | |
3162 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3163 | ||
3a08a8f9 R |
3164 | if (current->state == TASK_RUNNING) |
3165 | vcpu->preempted = true; | |
e9b11c17 | 3166 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3167 | } |
3168 | ||
0ee75bea | 3169 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3170 | struct module *module) |
6aa8b732 AK |
3171 | { |
3172 | int r; | |
002c7f7c | 3173 | int cpu; |
6aa8b732 | 3174 | |
f8c16bba ZX |
3175 | r = kvm_arch_init(opaque); |
3176 | if (r) | |
d2308784 | 3177 | goto out_fail; |
cb498ea2 | 3178 | |
7dac16c3 AH |
3179 | /* |
3180 | * kvm_arch_init makes sure there's at most one caller | |
3181 | * for architectures that support multiple implementations, | |
3182 | * like intel and amd on x86. | |
3183 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3184 | * conflicts in case kvm is already setup for another implementation. | |
3185 | */ | |
3186 | r = kvm_irqfd_init(); | |
3187 | if (r) | |
3188 | goto out_irqfd; | |
3189 | ||
8437a617 | 3190 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3191 | r = -ENOMEM; |
3192 | goto out_free_0; | |
3193 | } | |
3194 | ||
e9b11c17 | 3195 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3196 | if (r < 0) |
7f59f492 | 3197 | goto out_free_0a; |
6aa8b732 | 3198 | |
002c7f7c YS |
3199 | for_each_online_cpu(cpu) { |
3200 | smp_call_function_single(cpu, | |
e9b11c17 | 3201 | kvm_arch_check_processor_compat, |
8691e5a8 | 3202 | &r, 1); |
002c7f7c | 3203 | if (r < 0) |
d2308784 | 3204 | goto out_free_1; |
002c7f7c YS |
3205 | } |
3206 | ||
774c47f1 AK |
3207 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3208 | if (r) | |
d2308784 | 3209 | goto out_free_2; |
6aa8b732 AK |
3210 | register_reboot_notifier(&kvm_reboot_notifier); |
3211 | ||
c16f862d | 3212 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3213 | if (!vcpu_align) |
3214 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3215 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3216 | 0, NULL); |
c16f862d RR |
3217 | if (!kvm_vcpu_cache) { |
3218 | r = -ENOMEM; | |
fb3600cc | 3219 | goto out_free_3; |
c16f862d RR |
3220 | } |
3221 | ||
af585b92 GN |
3222 | r = kvm_async_pf_init(); |
3223 | if (r) | |
3224 | goto out_free; | |
3225 | ||
6aa8b732 | 3226 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3227 | kvm_vm_fops.owner = module; |
3228 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3229 | |
3230 | r = misc_register(&kvm_dev); | |
3231 | if (r) { | |
d77c26fc | 3232 | printk(KERN_ERR "kvm: misc device register failed\n"); |
af585b92 | 3233 | goto out_unreg; |
6aa8b732 AK |
3234 | } |
3235 | ||
fb3600cc RW |
3236 | register_syscore_ops(&kvm_syscore_ops); |
3237 | ||
15ad7146 AK |
3238 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3239 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3240 | ||
4f69b680 H |
3241 | r = kvm_init_debug(); |
3242 | if (r) { | |
3243 | printk(KERN_ERR "kvm: create debugfs files failed\n"); | |
3244 | goto out_undebugfs; | |
3245 | } | |
0ea4ed8e | 3246 | |
c7addb90 | 3247 | return 0; |
6aa8b732 | 3248 | |
4f69b680 H |
3249 | out_undebugfs: |
3250 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3251 | misc_deregister(&kvm_dev); |
af585b92 GN |
3252 | out_unreg: |
3253 | kvm_async_pf_deinit(); | |
6aa8b732 | 3254 | out_free: |
c16f862d | 3255 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3256 | out_free_3: |
6aa8b732 | 3257 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3258 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3259 | out_free_2: |
d2308784 | 3260 | out_free_1: |
e9b11c17 | 3261 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3262 | out_free_0a: |
3263 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3264 | out_free_0: |
a0f155e9 CH |
3265 | kvm_irqfd_exit(); |
3266 | out_irqfd: | |
7dac16c3 AH |
3267 | kvm_arch_exit(); |
3268 | out_fail: | |
6aa8b732 AK |
3269 | return r; |
3270 | } | |
cb498ea2 | 3271 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3272 | |
cb498ea2 | 3273 | void kvm_exit(void) |
6aa8b732 | 3274 | { |
0ea4ed8e | 3275 | kvm_exit_debug(); |
6aa8b732 | 3276 | misc_deregister(&kvm_dev); |
c16f862d | 3277 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3278 | kvm_async_pf_deinit(); |
fb3600cc | 3279 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3280 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3281 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3282 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3283 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3284 | kvm_arch_exit(); |
a0f155e9 | 3285 | kvm_irqfd_exit(); |
7f59f492 | 3286 | free_cpumask_var(cpus_hardware_enabled); |
6aa8b732 | 3287 | } |
cb498ea2 | 3288 | EXPORT_SYMBOL_GPL(kvm_exit); |