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 | 97 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e SH |
98 | static void update_memslots(struct kvm_memslots *slots, |
99 | struct kvm_memory_slot *new, u64 last_generation); | |
100 | ||
101 | static void kvm_release_pfn_dirty(pfn_t pfn); | |
102 | static void mark_page_dirty_in_slot(struct kvm *kvm, | |
103 | struct kvm_memory_slot *memslot, gfn_t gfn); | |
e93f8a0f | 104 | |
52480137 | 105 | __visible bool kvm_rebooting; |
b7c4145b | 106 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 107 | |
54dee993 MT |
108 | static bool largepages_enabled = true; |
109 | ||
a2766325 | 110 | bool kvm_is_mmio_pfn(pfn_t pfn) |
cbff90a7 | 111 | { |
11feeb49 AA |
112 | if (pfn_valid(pfn)) |
113 | return PageReserved(pfn_to_page(pfn)); | |
cbff90a7 BAY |
114 | |
115 | return true; | |
116 | } | |
117 | ||
bccf2150 AK |
118 | /* |
119 | * Switches to specified vcpu, until a matching vcpu_put() | |
120 | */ | |
9fc77441 | 121 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 122 | { |
15ad7146 AK |
123 | int cpu; |
124 | ||
9fc77441 MT |
125 | if (mutex_lock_killable(&vcpu->mutex)) |
126 | return -EINTR; | |
34bb10b7 RR |
127 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
128 | /* The thread running this VCPU changed. */ | |
129 | struct pid *oldpid = vcpu->pid; | |
130 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
131 | rcu_assign_pointer(vcpu->pid, newpid); | |
132 | synchronize_rcu(); | |
133 | put_pid(oldpid); | |
134 | } | |
15ad7146 AK |
135 | cpu = get_cpu(); |
136 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 137 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 138 | put_cpu(); |
9fc77441 | 139 | return 0; |
6aa8b732 AK |
140 | } |
141 | ||
313a3dc7 | 142 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 143 | { |
15ad7146 | 144 | preempt_disable(); |
313a3dc7 | 145 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
146 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
147 | preempt_enable(); | |
6aa8b732 AK |
148 | mutex_unlock(&vcpu->mutex); |
149 | } | |
150 | ||
d9e368d6 AK |
151 | static void ack_flush(void *_completed) |
152 | { | |
d9e368d6 AK |
153 | } |
154 | ||
49846896 | 155 | static bool make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 156 | { |
597a5f55 | 157 | int i, cpu, me; |
6ef7a1bc RR |
158 | cpumask_var_t cpus; |
159 | bool called = true; | |
d9e368d6 | 160 | struct kvm_vcpu *vcpu; |
d9e368d6 | 161 | |
79f55997 | 162 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 163 | |
3cba4130 | 164 | me = get_cpu(); |
988a2cae | 165 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 166 | kvm_make_request(req, vcpu); |
d9e368d6 | 167 | cpu = vcpu->cpu; |
6b7e2d09 XG |
168 | |
169 | /* Set ->requests bit before we read ->mode */ | |
170 | smp_mb(); | |
171 | ||
172 | if (cpus != NULL && cpu != -1 && cpu != me && | |
173 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 174 | cpumask_set_cpu(cpu, cpus); |
49846896 | 175 | } |
6ef7a1bc RR |
176 | if (unlikely(cpus == NULL)) |
177 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
178 | else if (!cpumask_empty(cpus)) | |
179 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
180 | else | |
181 | called = false; | |
3cba4130 | 182 | put_cpu(); |
6ef7a1bc | 183 | free_cpumask_var(cpus); |
49846896 | 184 | return called; |
d9e368d6 AK |
185 | } |
186 | ||
49846896 | 187 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 188 | { |
a086f6a1 XG |
189 | long dirty_count = kvm->tlbs_dirty; |
190 | ||
191 | smp_mb(); | |
49846896 RR |
192 | if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
193 | ++kvm->stat.remote_tlb_flush; | |
a086f6a1 | 194 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 195 | } |
2ba9f0d8 | 196 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
2e53d63a | 197 | |
49846896 RR |
198 | void kvm_reload_remote_mmus(struct kvm *kvm) |
199 | { | |
200 | make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); | |
201 | } | |
2e53d63a | 202 | |
d828199e MT |
203 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
204 | { | |
205 | make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
206 | } | |
207 | ||
3d81bc7e | 208 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
c7c9c56c | 209 | { |
3d81bc7e | 210 | make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); |
c7c9c56c YZ |
211 | } |
212 | ||
fb3f0f51 RR |
213 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
214 | { | |
215 | struct page *page; | |
216 | int r; | |
217 | ||
218 | mutex_init(&vcpu->mutex); | |
219 | vcpu->cpu = -1; | |
fb3f0f51 RR |
220 | vcpu->kvm = kvm; |
221 | vcpu->vcpu_id = id; | |
34bb10b7 | 222 | vcpu->pid = NULL; |
b6958ce4 | 223 | init_waitqueue_head(&vcpu->wq); |
af585b92 | 224 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 RR |
225 | |
226 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
227 | if (!page) { | |
228 | r = -ENOMEM; | |
229 | goto fail; | |
230 | } | |
231 | vcpu->run = page_address(page); | |
232 | ||
4c088493 R |
233 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
234 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 235 | vcpu->preempted = false; |
4c088493 | 236 | |
e9b11c17 | 237 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 238 | if (r < 0) |
e9b11c17 | 239 | goto fail_free_run; |
fb3f0f51 RR |
240 | return 0; |
241 | ||
fb3f0f51 RR |
242 | fail_free_run: |
243 | free_page((unsigned long)vcpu->run); | |
244 | fail: | |
76fafa5e | 245 | return r; |
fb3f0f51 RR |
246 | } |
247 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
248 | ||
249 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
250 | { | |
34bb10b7 | 251 | put_pid(vcpu->pid); |
e9b11c17 | 252 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
253 | free_page((unsigned long)vcpu->run); |
254 | } | |
255 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
256 | ||
e930bffe AA |
257 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
258 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
259 | { | |
260 | return container_of(mn, struct kvm, mmu_notifier); | |
261 | } | |
262 | ||
263 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
264 | struct mm_struct *mm, | |
265 | unsigned long address) | |
266 | { | |
267 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 268 | int need_tlb_flush, idx; |
e930bffe AA |
269 | |
270 | /* | |
271 | * When ->invalidate_page runs, the linux pte has been zapped | |
272 | * already but the page is still allocated until | |
273 | * ->invalidate_page returns. So if we increase the sequence | |
274 | * here the kvm page fault will notice if the spte can't be | |
275 | * established because the page is going to be freed. If | |
276 | * instead the kvm page fault establishes the spte before | |
277 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
278 | * before returning. | |
279 | * | |
280 | * The sequence increase only need to be seen at spin_unlock | |
281 | * time, and not at spin_lock time. | |
282 | * | |
283 | * Increasing the sequence after the spin_unlock would be | |
284 | * unsafe because the kvm page fault could then establish the | |
285 | * pte after kvm_unmap_hva returned, without noticing the page | |
286 | * is going to be freed. | |
287 | */ | |
bc6678a3 | 288 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 289 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 290 | |
e930bffe | 291 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 292 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
293 | /* we've to flush the tlb before the pages can be freed */ |
294 | if (need_tlb_flush) | |
295 | kvm_flush_remote_tlbs(kvm); | |
296 | ||
565f3be2 TY |
297 | spin_unlock(&kvm->mmu_lock); |
298 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
299 | } |
300 | ||
3da0dd43 IE |
301 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
302 | struct mm_struct *mm, | |
303 | unsigned long address, | |
304 | pte_t pte) | |
305 | { | |
306 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 307 | int idx; |
3da0dd43 | 308 | |
bc6678a3 | 309 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
310 | spin_lock(&kvm->mmu_lock); |
311 | kvm->mmu_notifier_seq++; | |
312 | kvm_set_spte_hva(kvm, address, pte); | |
313 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 314 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
315 | } |
316 | ||
e930bffe AA |
317 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
318 | struct mm_struct *mm, | |
319 | unsigned long start, | |
320 | unsigned long end) | |
321 | { | |
322 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 323 | int need_tlb_flush = 0, idx; |
e930bffe | 324 | |
bc6678a3 | 325 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
326 | spin_lock(&kvm->mmu_lock); |
327 | /* | |
328 | * The count increase must become visible at unlock time as no | |
329 | * spte can be established without taking the mmu_lock and | |
330 | * count is also read inside the mmu_lock critical section. | |
331 | */ | |
332 | kvm->mmu_notifier_count++; | |
b3ae2096 | 333 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 334 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
335 | /* we've to flush the tlb before the pages can be freed */ |
336 | if (need_tlb_flush) | |
337 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
338 | |
339 | spin_unlock(&kvm->mmu_lock); | |
340 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
341 | } |
342 | ||
343 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
344 | struct mm_struct *mm, | |
345 | unsigned long start, | |
346 | unsigned long end) | |
347 | { | |
348 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
349 | ||
350 | spin_lock(&kvm->mmu_lock); | |
351 | /* | |
352 | * This sequence increase will notify the kvm page fault that | |
353 | * the page that is going to be mapped in the spte could have | |
354 | * been freed. | |
355 | */ | |
356 | kvm->mmu_notifier_seq++; | |
a355aa54 | 357 | smp_wmb(); |
e930bffe AA |
358 | /* |
359 | * The above sequence increase must be visible before the | |
a355aa54 PM |
360 | * below count decrease, which is ensured by the smp_wmb above |
361 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
362 | */ |
363 | kvm->mmu_notifier_count--; | |
364 | spin_unlock(&kvm->mmu_lock); | |
365 | ||
366 | BUG_ON(kvm->mmu_notifier_count < 0); | |
367 | } | |
368 | ||
369 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
370 | struct mm_struct *mm, | |
371 | unsigned long address) | |
372 | { | |
373 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 374 | int young, idx; |
e930bffe | 375 | |
bc6678a3 | 376 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 377 | spin_lock(&kvm->mmu_lock); |
e930bffe | 378 | |
565f3be2 | 379 | young = kvm_age_hva(kvm, address); |
e930bffe AA |
380 | if (young) |
381 | kvm_flush_remote_tlbs(kvm); | |
382 | ||
565f3be2 TY |
383 | spin_unlock(&kvm->mmu_lock); |
384 | srcu_read_unlock(&kvm->srcu, idx); | |
385 | ||
e930bffe AA |
386 | return young; |
387 | } | |
388 | ||
8ee53820 AA |
389 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
390 | struct mm_struct *mm, | |
391 | unsigned long address) | |
392 | { | |
393 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
394 | int young, idx; | |
395 | ||
396 | idx = srcu_read_lock(&kvm->srcu); | |
397 | spin_lock(&kvm->mmu_lock); | |
398 | young = kvm_test_age_hva(kvm, address); | |
399 | spin_unlock(&kvm->mmu_lock); | |
400 | srcu_read_unlock(&kvm->srcu, idx); | |
401 | ||
402 | return young; | |
403 | } | |
404 | ||
85db06e5 MT |
405 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
406 | struct mm_struct *mm) | |
407 | { | |
408 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
409 | int idx; |
410 | ||
411 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 412 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 413 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
414 | } |
415 | ||
e930bffe AA |
416 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
417 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
418 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
419 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
420 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
8ee53820 | 421 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 422 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 423 | .release = kvm_mmu_notifier_release, |
e930bffe | 424 | }; |
4c07b0a4 AK |
425 | |
426 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
427 | { | |
428 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
429 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
430 | } | |
431 | ||
432 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
433 | ||
434 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
435 | { | |
436 | return 0; | |
437 | } | |
438 | ||
e930bffe AA |
439 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
440 | ||
bf3e05bc XG |
441 | static void kvm_init_memslots_id(struct kvm *kvm) |
442 | { | |
443 | int i; | |
444 | struct kvm_memslots *slots = kvm->memslots; | |
445 | ||
446 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
f85e2cb5 | 447 | slots->id_to_index[i] = slots->memslots[i].id = i; |
bf3e05bc XG |
448 | } |
449 | ||
e08b9637 | 450 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 451 | { |
d89f5eff JK |
452 | int r, i; |
453 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 454 | |
d89f5eff JK |
455 | if (!kvm) |
456 | return ERR_PTR(-ENOMEM); | |
457 | ||
e08b9637 | 458 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff JK |
459 | if (r) |
460 | goto out_err_nodisable; | |
10474ae8 AG |
461 | |
462 | r = hardware_enable_all(); | |
463 | if (r) | |
464 | goto out_err_nodisable; | |
465 | ||
75858a84 AK |
466 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
467 | INIT_HLIST_HEAD(&kvm->mask_notifier_list); | |
136bdfee | 468 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 469 | #endif |
6aa8b732 | 470 | |
1e702d9a AW |
471 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
472 | ||
46a26bf5 MT |
473 | r = -ENOMEM; |
474 | kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); | |
475 | if (!kvm->memslots) | |
57e7fbee | 476 | goto out_err_nosrcu; |
bf3e05bc | 477 | kvm_init_memslots_id(kvm); |
bc6678a3 | 478 | if (init_srcu_struct(&kvm->srcu)) |
57e7fbee | 479 | goto out_err_nosrcu; |
e93f8a0f MT |
480 | for (i = 0; i < KVM_NR_BUSES; i++) { |
481 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
482 | GFP_KERNEL); | |
57e7fbee | 483 | if (!kvm->buses[i]) |
e93f8a0f | 484 | goto out_err; |
e93f8a0f | 485 | } |
e930bffe | 486 | |
74b5c5bf | 487 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
488 | kvm->mm = current->mm; |
489 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 490 | kvm_eventfd_init(kvm); |
11ec2804 | 491 | mutex_init(&kvm->lock); |
60eead79 | 492 | mutex_init(&kvm->irq_lock); |
79fac95e | 493 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 494 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 495 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
496 | |
497 | r = kvm_init_mmu_notifier(kvm); | |
498 | if (r) | |
499 | goto out_err; | |
500 | ||
2f303b74 | 501 | spin_lock(&kvm_lock); |
5e58cfe4 | 502 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 503 | spin_unlock(&kvm_lock); |
d89f5eff | 504 | |
f17abe9a | 505 | return kvm; |
10474ae8 AG |
506 | |
507 | out_err: | |
57e7fbee JK |
508 | cleanup_srcu_struct(&kvm->srcu); |
509 | out_err_nosrcu: | |
10474ae8 AG |
510 | hardware_disable_all(); |
511 | out_err_nodisable: | |
e93f8a0f MT |
512 | for (i = 0; i < KVM_NR_BUSES; i++) |
513 | kfree(kvm->buses[i]); | |
46a26bf5 | 514 | kfree(kvm->memslots); |
d89f5eff | 515 | kvm_arch_free_vm(kvm); |
10474ae8 | 516 | return ERR_PTR(r); |
f17abe9a AK |
517 | } |
518 | ||
92eca8fa TY |
519 | /* |
520 | * Avoid using vmalloc for a small buffer. | |
521 | * Should not be used when the size is statically known. | |
522 | */ | |
c1a7b32a | 523 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
524 | { |
525 | if (size > PAGE_SIZE) | |
526 | return vzalloc(size); | |
527 | else | |
528 | return kzalloc(size, GFP_KERNEL); | |
529 | } | |
530 | ||
c1a7b32a | 531 | void kvm_kvfree(const void *addr) |
92eca8fa TY |
532 | { |
533 | if (is_vmalloc_addr(addr)) | |
534 | vfree(addr); | |
535 | else | |
536 | kfree(addr); | |
537 | } | |
538 | ||
a36a57b1 TY |
539 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
540 | { | |
541 | if (!memslot->dirty_bitmap) | |
542 | return; | |
543 | ||
92eca8fa | 544 | kvm_kvfree(memslot->dirty_bitmap); |
a36a57b1 TY |
545 | memslot->dirty_bitmap = NULL; |
546 | } | |
547 | ||
6aa8b732 AK |
548 | /* |
549 | * Free any memory in @free but not in @dont. | |
550 | */ | |
5587027c | 551 | static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free, |
6aa8b732 AK |
552 | struct kvm_memory_slot *dont) |
553 | { | |
6aa8b732 | 554 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
a36a57b1 | 555 | kvm_destroy_dirty_bitmap(free); |
6aa8b732 | 556 | |
5587027c | 557 | kvm_arch_free_memslot(kvm, free, dont); |
05da4558 | 558 | |
6aa8b732 | 559 | free->npages = 0; |
6aa8b732 AK |
560 | } |
561 | ||
7940876e | 562 | static void kvm_free_physmem(struct kvm *kvm) |
6aa8b732 | 563 | { |
46a26bf5 | 564 | struct kvm_memslots *slots = kvm->memslots; |
be6ba0f0 | 565 | struct kvm_memory_slot *memslot; |
46a26bf5 | 566 | |
be6ba0f0 | 567 | kvm_for_each_memslot(memslot, slots) |
5587027c | 568 | kvm_free_physmem_slot(kvm, memslot, NULL); |
6aa8b732 | 569 | |
46a26bf5 | 570 | kfree(kvm->memslots); |
6aa8b732 AK |
571 | } |
572 | ||
07f0a7bd SW |
573 | static void kvm_destroy_devices(struct kvm *kvm) |
574 | { | |
575 | struct list_head *node, *tmp; | |
576 | ||
577 | list_for_each_safe(node, tmp, &kvm->devices) { | |
578 | struct kvm_device *dev = | |
579 | list_entry(node, struct kvm_device, vm_node); | |
580 | ||
581 | list_del(node); | |
582 | dev->ops->destroy(dev); | |
583 | } | |
584 | } | |
585 | ||
f17abe9a AK |
586 | static void kvm_destroy_vm(struct kvm *kvm) |
587 | { | |
e93f8a0f | 588 | int i; |
6d4e4c4f AK |
589 | struct mm_struct *mm = kvm->mm; |
590 | ||
ad8ba2cd | 591 | kvm_arch_sync_events(kvm); |
2f303b74 | 592 | spin_lock(&kvm_lock); |
133de902 | 593 | list_del(&kvm->vm_list); |
2f303b74 | 594 | spin_unlock(&kvm_lock); |
399ec807 | 595 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
596 | for (i = 0; i < KVM_NR_BUSES; i++) |
597 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 598 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
599 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
600 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 601 | #else |
2df72e9b | 602 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 603 | #endif |
d19a9cd2 | 604 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 605 | kvm_destroy_devices(kvm); |
d89f5eff JK |
606 | kvm_free_physmem(kvm); |
607 | cleanup_srcu_struct(&kvm->srcu); | |
608 | kvm_arch_free_vm(kvm); | |
10474ae8 | 609 | hardware_disable_all(); |
6d4e4c4f | 610 | mmdrop(mm); |
f17abe9a AK |
611 | } |
612 | ||
d39f13b0 IE |
613 | void kvm_get_kvm(struct kvm *kvm) |
614 | { | |
615 | atomic_inc(&kvm->users_count); | |
616 | } | |
617 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
618 | ||
619 | void kvm_put_kvm(struct kvm *kvm) | |
620 | { | |
621 | if (atomic_dec_and_test(&kvm->users_count)) | |
622 | kvm_destroy_vm(kvm); | |
623 | } | |
624 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
625 | ||
626 | ||
f17abe9a AK |
627 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
628 | { | |
629 | struct kvm *kvm = filp->private_data; | |
630 | ||
721eecbf GH |
631 | kvm_irqfd_release(kvm); |
632 | ||
d39f13b0 | 633 | kvm_put_kvm(kvm); |
6aa8b732 AK |
634 | return 0; |
635 | } | |
636 | ||
515a0127 TY |
637 | /* |
638 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 639 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 640 | */ |
a36a57b1 TY |
641 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
642 | { | |
515a0127 | 643 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 644 | |
92eca8fa | 645 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
646 | if (!memslot->dirty_bitmap) |
647 | return -ENOMEM; | |
648 | ||
a36a57b1 TY |
649 | return 0; |
650 | } | |
651 | ||
bf3e05bc XG |
652 | static int cmp_memslot(const void *slot1, const void *slot2) |
653 | { | |
654 | struct kvm_memory_slot *s1, *s2; | |
655 | ||
656 | s1 = (struct kvm_memory_slot *)slot1; | |
657 | s2 = (struct kvm_memory_slot *)slot2; | |
658 | ||
659 | if (s1->npages < s2->npages) | |
660 | return 1; | |
661 | if (s1->npages > s2->npages) | |
662 | return -1; | |
663 | ||
664 | return 0; | |
665 | } | |
666 | ||
667 | /* | |
668 | * Sort the memslots base on its size, so the larger slots | |
669 | * will get better fit. | |
670 | */ | |
671 | static void sort_memslots(struct kvm_memslots *slots) | |
672 | { | |
f85e2cb5 XG |
673 | int i; |
674 | ||
bf3e05bc XG |
675 | sort(slots->memslots, KVM_MEM_SLOTS_NUM, |
676 | sizeof(struct kvm_memory_slot), cmp_memslot, NULL); | |
f85e2cb5 XG |
677 | |
678 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
679 | slots->id_to_index[slots->memslots[i].id] = i; | |
bf3e05bc XG |
680 | } |
681 | ||
7940876e SH |
682 | static void update_memslots(struct kvm_memslots *slots, |
683 | struct kvm_memory_slot *new, | |
684 | u64 last_generation) | |
be593d62 XG |
685 | { |
686 | if (new) { | |
687 | int id = new->id; | |
28a37544 | 688 | struct kvm_memory_slot *old = id_to_memslot(slots, id); |
bf3e05bc | 689 | unsigned long npages = old->npages; |
be593d62 | 690 | |
28a37544 | 691 | *old = *new; |
bf3e05bc XG |
692 | if (new->npages != npages) |
693 | sort_memslots(slots); | |
be593d62 XG |
694 | } |
695 | ||
116c14c0 | 696 | slots->generation = last_generation + 1; |
be593d62 XG |
697 | } |
698 | ||
a50d64d6 XG |
699 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
700 | { | |
4d8b81ab XG |
701 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
702 | ||
703 | #ifdef KVM_CAP_READONLY_MEM | |
704 | valid_flags |= KVM_MEM_READONLY; | |
705 | #endif | |
706 | ||
707 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
708 | return -EINVAL; |
709 | ||
710 | return 0; | |
711 | } | |
712 | ||
7ec4fb44 GN |
713 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
714 | struct kvm_memslots *slots, struct kvm_memory_slot *new) | |
715 | { | |
716 | struct kvm_memslots *old_memslots = kvm->memslots; | |
717 | ||
718 | update_memslots(slots, new, kvm->memslots->generation); | |
719 | rcu_assign_pointer(kvm->memslots, slots); | |
720 | synchronize_srcu_expedited(&kvm->srcu); | |
e59dbe09 TY |
721 | |
722 | kvm_arch_memslots_updated(kvm); | |
723 | ||
724 | return old_memslots; | |
7ec4fb44 GN |
725 | } |
726 | ||
6aa8b732 AK |
727 | /* |
728 | * Allocate some memory and give it an address in the guest physical address | |
729 | * space. | |
730 | * | |
731 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 732 | * |
10589a46 | 733 | * Must be called holding mmap_sem for write. |
6aa8b732 | 734 | */ |
f78e0e2e | 735 | int __kvm_set_memory_region(struct kvm *kvm, |
47ae31e2 | 736 | struct kvm_userspace_memory_region *mem) |
6aa8b732 | 737 | { |
8234b22e | 738 | int r; |
6aa8b732 | 739 | gfn_t base_gfn; |
28bcb112 | 740 | unsigned long npages; |
a843fac2 | 741 | struct kvm_memory_slot *slot; |
6aa8b732 | 742 | struct kvm_memory_slot old, new; |
b7f69c55 | 743 | struct kvm_memslots *slots = NULL, *old_memslots; |
f64c0398 | 744 | enum kvm_mr_change change; |
6aa8b732 | 745 | |
a50d64d6 XG |
746 | r = check_memory_region_flags(mem); |
747 | if (r) | |
748 | goto out; | |
749 | ||
6aa8b732 AK |
750 | r = -EINVAL; |
751 | /* General sanity checks */ | |
752 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
753 | goto out; | |
754 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
755 | goto out; | |
fa3d315a | 756 | /* We can read the guest memory with __xxx_user() later on. */ |
47ae31e2 | 757 | if ((mem->slot < KVM_USER_MEM_SLOTS) && |
fa3d315a | 758 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
759 | !access_ok(VERIFY_WRITE, |
760 | (void __user *)(unsigned long)mem->userspace_addr, | |
761 | mem->memory_size))) | |
78749809 | 762 | goto out; |
93a5cef0 | 763 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
764 | goto out; |
765 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
766 | goto out; | |
767 | ||
a843fac2 | 768 | slot = id_to_memslot(kvm->memslots, mem->slot); |
6aa8b732 AK |
769 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
770 | npages = mem->memory_size >> PAGE_SHIFT; | |
771 | ||
660c22c4 TY |
772 | r = -EINVAL; |
773 | if (npages > KVM_MEM_MAX_NR_PAGES) | |
774 | goto out; | |
775 | ||
6aa8b732 AK |
776 | if (!npages) |
777 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
778 | ||
a843fac2 | 779 | new = old = *slot; |
6aa8b732 | 780 | |
e36d96f7 | 781 | new.id = mem->slot; |
6aa8b732 AK |
782 | new.base_gfn = base_gfn; |
783 | new.npages = npages; | |
784 | new.flags = mem->flags; | |
785 | ||
6aa8b732 | 786 | r = -EINVAL; |
f64c0398 TY |
787 | if (npages) { |
788 | if (!old.npages) | |
789 | change = KVM_MR_CREATE; | |
790 | else { /* Modify an existing slot. */ | |
791 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
792 | (npages != old.npages) || |
793 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
794 | goto out; |
795 | ||
796 | if (base_gfn != old.base_gfn) | |
797 | change = KVM_MR_MOVE; | |
798 | else if (new.flags != old.flags) | |
799 | change = KVM_MR_FLAGS_ONLY; | |
800 | else { /* Nothing to change. */ | |
801 | r = 0; | |
802 | goto out; | |
803 | } | |
804 | } | |
805 | } else if (old.npages) { | |
806 | change = KVM_MR_DELETE; | |
807 | } else /* Modify a non-existent slot: disallowed. */ | |
0ea75e1d | 808 | goto out; |
6aa8b732 | 809 | |
f64c0398 | 810 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
811 | /* Check for overlaps */ |
812 | r = -EEXIST; | |
813 | kvm_for_each_memslot(slot, kvm->memslots) { | |
a843fac2 TY |
814 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
815 | (slot->id == mem->slot)) | |
0a706bee TY |
816 | continue; |
817 | if (!((base_gfn + npages <= slot->base_gfn) || | |
818 | (base_gfn >= slot->base_gfn + slot->npages))) | |
819 | goto out; | |
820 | } | |
6aa8b732 | 821 | } |
6aa8b732 | 822 | |
6aa8b732 AK |
823 | /* Free page dirty bitmap if unneeded */ |
824 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 825 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
826 | |
827 | r = -ENOMEM; | |
f64c0398 | 828 | if (change == KVM_MR_CREATE) { |
189a2f7b | 829 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 830 | |
5587027c | 831 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 832 | goto out_free; |
6aa8b732 | 833 | } |
ec04b260 | 834 | |
6aa8b732 AK |
835 | /* Allocate page dirty bitmap if needed */ |
836 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 837 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 838 | goto out_free; |
6aa8b732 AK |
839 | } |
840 | ||
f64c0398 | 841 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
bc6678a3 | 842 | r = -ENOMEM; |
6da64fdb TM |
843 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), |
844 | GFP_KERNEL); | |
bc6678a3 MT |
845 | if (!slots) |
846 | goto out_free; | |
28a37544 XG |
847 | slot = id_to_memslot(slots, mem->slot); |
848 | slot->flags |= KVM_MEMSLOT_INVALID; | |
849 | ||
7ec4fb44 | 850 | old_memslots = install_new_memslots(kvm, slots, NULL); |
bc6678a3 | 851 | |
e40f193f AW |
852 | /* slot was deleted or moved, clear iommu mapping */ |
853 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
854 | /* From this point no new shadow pages pointing to a deleted, |
855 | * or moved, memslot will be created. | |
bc6678a3 MT |
856 | * |
857 | * validation of sp->gfn happens in: | |
858 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
859 | * - kvm_is_visible_gfn (mmu_check_roots) | |
860 | */ | |
2df72e9b | 861 | kvm_arch_flush_shadow_memslot(kvm, slot); |
b7f69c55 | 862 | slots = old_memslots; |
bc6678a3 | 863 | } |
34d4cb8f | 864 | |
7b6195a9 | 865 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 866 | if (r) |
b7f69c55 | 867 | goto out_slots; |
f7784b8e | 868 | |
bc6678a3 | 869 | r = -ENOMEM; |
b7f69c55 AW |
870 | /* |
871 | * We can re-use the old_memslots from above, the only difference | |
872 | * from the currently installed memslots is the invalid flag. This | |
873 | * will get overwritten by update_memslots anyway. | |
874 | */ | |
875 | if (!slots) { | |
876 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), | |
877 | GFP_KERNEL); | |
878 | if (!slots) | |
879 | goto out_free; | |
880 | } | |
bc6678a3 MT |
881 | |
882 | /* actual memory is freed via old in kvm_free_physmem_slot below */ | |
f64c0398 | 883 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 884 | new.dirty_bitmap = NULL; |
db3fe4eb | 885 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
886 | } |
887 | ||
7ec4fb44 | 888 | old_memslots = install_new_memslots(kvm, slots, &new); |
3ad82a7e | 889 | |
8482644a | 890 | kvm_arch_commit_memory_region(kvm, mem, &old, change); |
82ce2c96 | 891 | |
5587027c | 892 | kvm_free_physmem_slot(kvm, &old, &new); |
bc6678a3 MT |
893 | kfree(old_memslots); |
894 | ||
261874b0 AW |
895 | /* |
896 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
897 | * un-mapped and re-mapped if their base changes. Since base change |
898 | * unmapping is handled above with slot deletion, mapping alone is | |
899 | * needed here. Anything else the iommu might care about for existing | |
900 | * slots (size changes, userspace addr changes and read-only flag | |
901 | * changes) is disallowed above, so any other attribute changes getting | |
902 | * here can be skipped. | |
261874b0 | 903 | */ |
75d61fbc TY |
904 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
905 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 906 | return r; |
bc6678a3 MT |
907 | } |
908 | ||
6aa8b732 AK |
909 | return 0; |
910 | ||
e40f193f AW |
911 | out_slots: |
912 | kfree(slots); | |
f78e0e2e | 913 | out_free: |
5587027c | 914 | kvm_free_physmem_slot(kvm, &new, &old); |
6aa8b732 AK |
915 | out: |
916 | return r; | |
210c7c4d | 917 | } |
f78e0e2e SY |
918 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
919 | ||
920 | int kvm_set_memory_region(struct kvm *kvm, | |
47ae31e2 | 921 | struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
922 | { |
923 | int r; | |
924 | ||
79fac95e | 925 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 926 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 927 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
928 | return r; |
929 | } | |
210c7c4d IE |
930 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
931 | ||
7940876e SH |
932 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
933 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 934 | { |
bbacc0c1 | 935 | if (mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 936 | return -EINVAL; |
47ae31e2 | 937 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
938 | } |
939 | ||
5bb064dc ZX |
940 | int kvm_get_dirty_log(struct kvm *kvm, |
941 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 AK |
942 | { |
943 | struct kvm_memory_slot *memslot; | |
944 | int r, i; | |
87bf6e7d | 945 | unsigned long n; |
6aa8b732 AK |
946 | unsigned long any = 0; |
947 | ||
6aa8b732 | 948 | r = -EINVAL; |
bbacc0c1 | 949 | if (log->slot >= KVM_USER_MEM_SLOTS) |
6aa8b732 AK |
950 | goto out; |
951 | ||
28a37544 | 952 | memslot = id_to_memslot(kvm->memslots, log->slot); |
6aa8b732 AK |
953 | r = -ENOENT; |
954 | if (!memslot->dirty_bitmap) | |
955 | goto out; | |
956 | ||
87bf6e7d | 957 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 958 | |
cd1a4a98 | 959 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
960 | any = memslot->dirty_bitmap[i]; |
961 | ||
962 | r = -EFAULT; | |
963 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
964 | goto out; | |
965 | ||
5bb064dc ZX |
966 | if (any) |
967 | *is_dirty = 1; | |
6aa8b732 AK |
968 | |
969 | r = 0; | |
6aa8b732 | 970 | out: |
6aa8b732 AK |
971 | return r; |
972 | } | |
2ba9f0d8 | 973 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 974 | |
db3fe4eb TY |
975 | bool kvm_largepages_enabled(void) |
976 | { | |
977 | return largepages_enabled; | |
978 | } | |
979 | ||
54dee993 MT |
980 | void kvm_disable_largepages(void) |
981 | { | |
982 | largepages_enabled = false; | |
983 | } | |
984 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
985 | ||
49c7754c GN |
986 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
987 | { | |
988 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
989 | } | |
a1f4d395 | 990 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 991 | |
e0d62c7f IE |
992 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
993 | { | |
bf3e05bc | 994 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 995 | |
bbacc0c1 | 996 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
997 | memslot->flags & KVM_MEMSLOT_INVALID) |
998 | return 0; | |
e0d62c7f | 999 | |
bf3e05bc | 1000 | return 1; |
e0d62c7f IE |
1001 | } |
1002 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1003 | ||
8f0b1ab6 JR |
1004 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1005 | { | |
1006 | struct vm_area_struct *vma; | |
1007 | unsigned long addr, size; | |
1008 | ||
1009 | size = PAGE_SIZE; | |
1010 | ||
1011 | addr = gfn_to_hva(kvm, gfn); | |
1012 | if (kvm_is_error_hva(addr)) | |
1013 | return PAGE_SIZE; | |
1014 | ||
1015 | down_read(¤t->mm->mmap_sem); | |
1016 | vma = find_vma(current->mm, addr); | |
1017 | if (!vma) | |
1018 | goto out; | |
1019 | ||
1020 | size = vma_kernel_pagesize(vma); | |
1021 | ||
1022 | out: | |
1023 | up_read(¤t->mm->mmap_sem); | |
1024 | ||
1025 | return size; | |
1026 | } | |
1027 | ||
4d8b81ab XG |
1028 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1029 | { | |
1030 | return slot->flags & KVM_MEM_READONLY; | |
1031 | } | |
1032 | ||
4d8b81ab XG |
1033 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1034 | gfn_t *nr_pages, bool write) | |
539cb660 | 1035 | { |
bc6678a3 | 1036 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1037 | return KVM_HVA_ERR_BAD; |
48987781 | 1038 | |
4d8b81ab XG |
1039 | if (memslot_is_readonly(slot) && write) |
1040 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1041 | |
1042 | if (nr_pages) | |
1043 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1044 | ||
4d8b81ab | 1045 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1046 | } |
48987781 | 1047 | |
4d8b81ab XG |
1048 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1049 | gfn_t *nr_pages) | |
1050 | { | |
1051 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1052 | } |
48987781 | 1053 | |
4d8b81ab | 1054 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1055 | gfn_t gfn) |
4d8b81ab XG |
1056 | { |
1057 | return gfn_to_hva_many(slot, gfn, NULL); | |
1058 | } | |
1059 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1060 | ||
48987781 XG |
1061 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1062 | { | |
49c7754c | 1063 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1064 | } |
0d150298 | 1065 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1066 | |
86ab8cff | 1067 | /* |
ba6a3541 PB |
1068 | * If writable is set to false, the hva returned by this function is only |
1069 | * allowed to be read. | |
86ab8cff | 1070 | */ |
ba6a3541 | 1071 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
86ab8cff | 1072 | { |
ba6a3541 | 1073 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
a2ac07fe GN |
1074 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1075 | ||
1076 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1077 | *writable = !memslot_is_readonly(slot); |
1078 | ||
a2ac07fe | 1079 | return hva; |
86ab8cff XG |
1080 | } |
1081 | ||
1082 | static int kvm_read_hva(void *data, void __user *hva, int len) | |
8030089f | 1083 | { |
86ab8cff XG |
1084 | return __copy_from_user(data, hva, len); |
1085 | } | |
1086 | ||
1087 | static int kvm_read_hva_atomic(void *data, void __user *hva, int len) | |
1088 | { | |
1089 | return __copy_from_user_inatomic(data, hva, len); | |
8030089f GN |
1090 | } |
1091 | ||
39369f7a | 1092 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1093 | unsigned long start, int write, struct page **page) |
1094 | { | |
1095 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1096 | ||
1097 | if (write) | |
1098 | flags |= FOLL_WRITE; | |
1099 | ||
1100 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1101 | } | |
1102 | ||
fafc3dba HY |
1103 | static inline int check_user_page_hwpoison(unsigned long addr) |
1104 | { | |
1105 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1106 | ||
1107 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1108 | flags, NULL, NULL, NULL); | |
1109 | return rc == -EHWPOISON; | |
1110 | } | |
1111 | ||
2fc84311 XG |
1112 | /* |
1113 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1114 | * true indicates success, otherwise false is returned. | |
1115 | */ | |
1116 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1117 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1118 | { |
8d4e1288 | 1119 | struct page *page[1]; |
2fc84311 | 1120 | int npages; |
954bbbc2 | 1121 | |
2fc84311 XG |
1122 | if (!(async || atomic)) |
1123 | return false; | |
af585b92 | 1124 | |
12ce13fe XG |
1125 | /* |
1126 | * Fast pin a writable pfn only if it is a write fault request | |
1127 | * or the caller allows to map a writable pfn for a read fault | |
1128 | * request. | |
1129 | */ | |
1130 | if (!(write_fault || writable)) | |
1131 | return false; | |
612819c3 | 1132 | |
2fc84311 XG |
1133 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1134 | if (npages == 1) { | |
1135 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1136 | |
2fc84311 XG |
1137 | if (writable) |
1138 | *writable = true; | |
1139 | return true; | |
1140 | } | |
af585b92 | 1141 | |
2fc84311 XG |
1142 | return false; |
1143 | } | |
612819c3 | 1144 | |
2fc84311 XG |
1145 | /* |
1146 | * The slow path to get the pfn of the specified host virtual address, | |
1147 | * 1 indicates success, -errno is returned if error is detected. | |
1148 | */ | |
1149 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1150 | bool *writable, pfn_t *pfn) | |
1151 | { | |
1152 | struct page *page[1]; | |
1153 | int npages = 0; | |
612819c3 | 1154 | |
2fc84311 XG |
1155 | might_sleep(); |
1156 | ||
1157 | if (writable) | |
1158 | *writable = write_fault; | |
1159 | ||
1160 | if (async) { | |
1161 | down_read(¤t->mm->mmap_sem); | |
1162 | npages = get_user_page_nowait(current, current->mm, | |
1163 | addr, write_fault, page); | |
1164 | up_read(¤t->mm->mmap_sem); | |
1165 | } else | |
1166 | npages = get_user_pages_fast(addr, 1, write_fault, | |
1167 | page); | |
1168 | if (npages != 1) | |
1169 | return npages; | |
1170 | ||
1171 | /* map read fault as writable if possible */ | |
12ce13fe | 1172 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1173 | struct page *wpage[1]; |
1174 | ||
1175 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1176 | if (npages == 1) { | |
1177 | *writable = true; | |
1178 | put_page(page[0]); | |
1179 | page[0] = wpage[0]; | |
612819c3 | 1180 | } |
2fc84311 XG |
1181 | |
1182 | npages = 1; | |
887c08ac | 1183 | } |
2fc84311 XG |
1184 | *pfn = page_to_pfn(page[0]); |
1185 | return npages; | |
1186 | } | |
539cb660 | 1187 | |
4d8b81ab XG |
1188 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1189 | { | |
1190 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1191 | return false; | |
2e2e3738 | 1192 | |
4d8b81ab XG |
1193 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1194 | return false; | |
887c08ac | 1195 | |
4d8b81ab XG |
1196 | return true; |
1197 | } | |
bf998156 | 1198 | |
12ce13fe XG |
1199 | /* |
1200 | * Pin guest page in memory and return its pfn. | |
1201 | * @addr: host virtual address which maps memory to the guest | |
1202 | * @atomic: whether this function can sleep | |
1203 | * @async: whether this function need to wait IO complete if the | |
1204 | * host page is not in the memory | |
1205 | * @write_fault: whether we should get a writable host page | |
1206 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1207 | * | |
1208 | * The function will map a writable host page for these two cases: | |
1209 | * 1): @write_fault = true | |
1210 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1211 | * whether the mapping is writable. | |
1212 | */ | |
2fc84311 XG |
1213 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1214 | bool write_fault, bool *writable) | |
1215 | { | |
1216 | struct vm_area_struct *vma; | |
1217 | pfn_t pfn = 0; | |
1218 | int npages; | |
2e2e3738 | 1219 | |
2fc84311 XG |
1220 | /* we can do it either atomically or asynchronously, not both */ |
1221 | BUG_ON(atomic && async); | |
8d4e1288 | 1222 | |
2fc84311 XG |
1223 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1224 | return pfn; | |
1225 | ||
1226 | if (atomic) | |
1227 | return KVM_PFN_ERR_FAULT; | |
1228 | ||
1229 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1230 | if (npages == 1) | |
1231 | return pfn; | |
8d4e1288 | 1232 | |
2fc84311 XG |
1233 | down_read(¤t->mm->mmap_sem); |
1234 | if (npages == -EHWPOISON || | |
1235 | (!async && check_user_page_hwpoison(addr))) { | |
1236 | pfn = KVM_PFN_ERR_HWPOISON; | |
1237 | goto exit; | |
1238 | } | |
1239 | ||
1240 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1241 | ||
1242 | if (vma == NULL) | |
1243 | pfn = KVM_PFN_ERR_FAULT; | |
1244 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1245 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1246 | vma->vm_pgoff; | |
1247 | BUG_ON(!kvm_is_mmio_pfn(pfn)); | |
1248 | } else { | |
4d8b81ab | 1249 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1250 | *async = true; |
1251 | pfn = KVM_PFN_ERR_FAULT; | |
1252 | } | |
1253 | exit: | |
1254 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1255 | return pfn; |
35149e21 AL |
1256 | } |
1257 | ||
4d8b81ab XG |
1258 | static pfn_t |
1259 | __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, | |
1260 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1261 | { |
4d8b81ab XG |
1262 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1263 | ||
1264 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1265 | return KVM_PFN_ERR_RO_FAULT; | |
1266 | ||
1267 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1268 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1269 | |
1270 | /* Do not map writable pfn in the readonly memslot. */ | |
1271 | if (writable && memslot_is_readonly(slot)) { | |
1272 | *writable = false; | |
1273 | writable = NULL; | |
1274 | } | |
1275 | ||
1276 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1277 | writable); | |
887c08ac | 1278 | } |
887c08ac | 1279 | |
612819c3 MT |
1280 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async, |
1281 | bool write_fault, bool *writable) | |
506f0d6f | 1282 | { |
4d8b81ab | 1283 | struct kvm_memory_slot *slot; |
506f0d6f | 1284 | |
af585b92 GN |
1285 | if (async) |
1286 | *async = false; | |
1287 | ||
4d8b81ab | 1288 | slot = gfn_to_memslot(kvm, gfn); |
506f0d6f | 1289 | |
4d8b81ab XG |
1290 | return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault, |
1291 | writable); | |
365fb3fd XG |
1292 | } |
1293 | ||
1294 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) | |
1295 | { | |
612819c3 | 1296 | return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL); |
365fb3fd XG |
1297 | } |
1298 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1299 | ||
612819c3 MT |
1300 | pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async, |
1301 | bool write_fault, bool *writable) | |
af585b92 | 1302 | { |
612819c3 | 1303 | return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable); |
af585b92 GN |
1304 | } |
1305 | EXPORT_SYMBOL_GPL(gfn_to_pfn_async); | |
1306 | ||
365fb3fd XG |
1307 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1308 | { | |
612819c3 | 1309 | return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL); |
506f0d6f | 1310 | } |
35149e21 AL |
1311 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
1312 | ||
612819c3 MT |
1313 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1314 | bool *writable) | |
1315 | { | |
1316 | return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable); | |
1317 | } | |
1318 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1319 | ||
d5661048 | 1320 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1321 | { |
4d8b81ab | 1322 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f MT |
1323 | } |
1324 | ||
037d92dc | 1325 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1326 | { |
4d8b81ab | 1327 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1328 | } |
037d92dc | 1329 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1330 | |
48987781 XG |
1331 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
1332 | int nr_pages) | |
1333 | { | |
1334 | unsigned long addr; | |
1335 | gfn_t entry; | |
1336 | ||
49c7754c | 1337 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
48987781 XG |
1338 | if (kvm_is_error_hva(addr)) |
1339 | return -1; | |
1340 | ||
1341 | if (entry < nr_pages) | |
1342 | return 0; | |
1343 | ||
1344 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1345 | } | |
1346 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1347 | ||
a2766325 XG |
1348 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1349 | { | |
81c52c56 | 1350 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1351 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1352 | |
cb9aaa30 XG |
1353 | if (kvm_is_mmio_pfn(pfn)) { |
1354 | WARN_ON(1); | |
6cede2e6 | 1355 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1356 | } |
a2766325 XG |
1357 | |
1358 | return pfn_to_page(pfn); | |
1359 | } | |
1360 | ||
35149e21 AL |
1361 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1362 | { | |
2e2e3738 AL |
1363 | pfn_t pfn; |
1364 | ||
1365 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1366 | |
a2766325 | 1367 | return kvm_pfn_to_page(pfn); |
954bbbc2 | 1368 | } |
aab61cc0 | 1369 | |
954bbbc2 AK |
1370 | EXPORT_SYMBOL_GPL(gfn_to_page); |
1371 | ||
b4231d61 IE |
1372 | void kvm_release_page_clean(struct page *page) |
1373 | { | |
32cad84f XG |
1374 | WARN_ON(is_error_page(page)); |
1375 | ||
35149e21 | 1376 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1377 | } |
1378 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1379 | ||
35149e21 AL |
1380 | void kvm_release_pfn_clean(pfn_t pfn) |
1381 | { | |
81c52c56 | 1382 | if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1383 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1384 | } |
1385 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1386 | ||
b4231d61 | 1387 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1388 | { |
a2766325 XG |
1389 | WARN_ON(is_error_page(page)); |
1390 | ||
35149e21 AL |
1391 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1392 | } | |
1393 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1394 | ||
7940876e | 1395 | static void kvm_release_pfn_dirty(pfn_t pfn) |
35149e21 AL |
1396 | { |
1397 | kvm_set_pfn_dirty(pfn); | |
1398 | kvm_release_pfn_clean(pfn); | |
1399 | } | |
35149e21 AL |
1400 | |
1401 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1402 | { | |
c77fb9dc | 1403 | if (!kvm_is_mmio_pfn(pfn)) { |
2e2e3738 AL |
1404 | struct page *page = pfn_to_page(pfn); |
1405 | if (!PageReserved(page)) | |
1406 | SetPageDirty(page); | |
1407 | } | |
8a7ae055 | 1408 | } |
35149e21 AL |
1409 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1410 | ||
1411 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1412 | { | |
c77fb9dc | 1413 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1414 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1415 | } |
1416 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1417 | ||
1418 | void kvm_get_pfn(pfn_t pfn) | |
1419 | { | |
c77fb9dc | 1420 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1421 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1422 | } |
1423 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1424 | |
195aefde IE |
1425 | static int next_segment(unsigned long len, int offset) |
1426 | { | |
1427 | if (len > PAGE_SIZE - offset) | |
1428 | return PAGE_SIZE - offset; | |
1429 | else | |
1430 | return len; | |
1431 | } | |
1432 | ||
1433 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1434 | int len) | |
1435 | { | |
e0506bcb IE |
1436 | int r; |
1437 | unsigned long addr; | |
195aefde | 1438 | |
ba6a3541 | 1439 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
e0506bcb IE |
1440 | if (kvm_is_error_hva(addr)) |
1441 | return -EFAULT; | |
86ab8cff | 1442 | r = kvm_read_hva(data, (void __user *)addr + offset, len); |
e0506bcb | 1443 | if (r) |
195aefde | 1444 | return -EFAULT; |
195aefde IE |
1445 | return 0; |
1446 | } | |
1447 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); | |
1448 | ||
1449 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) | |
1450 | { | |
1451 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1452 | int seg; | |
1453 | int offset = offset_in_page(gpa); | |
1454 | int ret; | |
1455 | ||
1456 | while ((seg = next_segment(len, offset)) != 0) { | |
1457 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1458 | if (ret < 0) | |
1459 | return ret; | |
1460 | offset = 0; | |
1461 | len -= seg; | |
1462 | data += seg; | |
1463 | ++gfn; | |
1464 | } | |
1465 | return 0; | |
1466 | } | |
1467 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1468 | ||
7ec54588 MT |
1469 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1470 | unsigned long len) | |
1471 | { | |
1472 | int r; | |
1473 | unsigned long addr; | |
1474 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1475 | int offset = offset_in_page(gpa); | |
1476 | ||
ba6a3541 | 1477 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
7ec54588 MT |
1478 | if (kvm_is_error_hva(addr)) |
1479 | return -EFAULT; | |
0aac03f0 | 1480 | pagefault_disable(); |
86ab8cff | 1481 | r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1482 | pagefault_enable(); |
7ec54588 MT |
1483 | if (r) |
1484 | return -EFAULT; | |
1485 | return 0; | |
1486 | } | |
1487 | EXPORT_SYMBOL(kvm_read_guest_atomic); | |
1488 | ||
195aefde IE |
1489 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
1490 | int offset, int len) | |
1491 | { | |
e0506bcb IE |
1492 | int r; |
1493 | unsigned long addr; | |
195aefde | 1494 | |
e0506bcb IE |
1495 | addr = gfn_to_hva(kvm, gfn); |
1496 | if (kvm_is_error_hva(addr)) | |
1497 | return -EFAULT; | |
8b0cedff | 1498 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1499 | if (r) |
195aefde | 1500 | return -EFAULT; |
195aefde IE |
1501 | mark_page_dirty(kvm, gfn); |
1502 | return 0; | |
1503 | } | |
1504 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); | |
1505 | ||
1506 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, | |
1507 | unsigned long len) | |
1508 | { | |
1509 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1510 | int seg; | |
1511 | int offset = offset_in_page(gpa); | |
1512 | int ret; | |
1513 | ||
1514 | while ((seg = next_segment(len, offset)) != 0) { | |
1515 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1516 | if (ret < 0) | |
1517 | return ret; | |
1518 | offset = 0; | |
1519 | len -= seg; | |
1520 | data += seg; | |
1521 | ++gfn; | |
1522 | } | |
1523 | return 0; | |
1524 | } | |
1525 | ||
49c7754c | 1526 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1527 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1528 | { |
1529 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1530 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1531 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1532 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1533 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1534 | gfn_t nr_pages_avail; | |
49c7754c GN |
1535 | |
1536 | ghc->gpa = gpa; | |
1537 | ghc->generation = slots->generation; | |
8f964525 AH |
1538 | ghc->len = len; |
1539 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1540 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail); | |
1541 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) { | |
49c7754c | 1542 | ghc->hva += offset; |
8f964525 AH |
1543 | } else { |
1544 | /* | |
1545 | * If the requested region crosses two memslots, we still | |
1546 | * verify that the entire region is valid here. | |
1547 | */ | |
1548 | while (start_gfn <= end_gfn) { | |
1549 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1550 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1551 | &nr_pages_avail); | |
1552 | if (kvm_is_error_hva(ghc->hva)) | |
1553 | return -EFAULT; | |
1554 | start_gfn += nr_pages_avail; | |
1555 | } | |
1556 | /* Use the slow path for cross page reads and writes. */ | |
1557 | ghc->memslot = NULL; | |
1558 | } | |
49c7754c GN |
1559 | return 0; |
1560 | } | |
1561 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1562 | ||
1563 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1564 | void *data, unsigned long len) | |
1565 | { | |
1566 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1567 | int r; | |
1568 | ||
8f964525 AH |
1569 | BUG_ON(len > ghc->len); |
1570 | ||
49c7754c | 1571 | if (slots->generation != ghc->generation) |
8f964525 AH |
1572 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1573 | ||
1574 | if (unlikely(!ghc->memslot)) | |
1575 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1576 | |
1577 | if (kvm_is_error_hva(ghc->hva)) | |
1578 | return -EFAULT; | |
1579 | ||
8b0cedff | 1580 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1581 | if (r) |
1582 | return -EFAULT; | |
1583 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); | |
1584 | ||
1585 | return 0; | |
1586 | } | |
1587 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1588 | ||
e03b644f GN |
1589 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1590 | void *data, unsigned long len) | |
1591 | { | |
1592 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1593 | int r; | |
1594 | ||
8f964525 AH |
1595 | BUG_ON(len > ghc->len); |
1596 | ||
e03b644f | 1597 | if (slots->generation != ghc->generation) |
8f964525 AH |
1598 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1599 | ||
1600 | if (unlikely(!ghc->memslot)) | |
1601 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1602 | |
1603 | if (kvm_is_error_hva(ghc->hva)) | |
1604 | return -EFAULT; | |
1605 | ||
1606 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1607 | if (r) | |
1608 | return -EFAULT; | |
1609 | ||
1610 | return 0; | |
1611 | } | |
1612 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1613 | ||
195aefde IE |
1614 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1615 | { | |
8a3caa6d HC |
1616 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
1617 | ||
1618 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
1619 | } |
1620 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1621 | ||
1622 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1623 | { | |
1624 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1625 | int seg; | |
1626 | int offset = offset_in_page(gpa); | |
1627 | int ret; | |
1628 | ||
1629 | while ((seg = next_segment(len, offset)) != 0) { | |
1630 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); | |
1631 | if (ret < 0) | |
1632 | return ret; | |
1633 | offset = 0; | |
1634 | len -= seg; | |
1635 | ++gfn; | |
1636 | } | |
1637 | return 0; | |
1638 | } | |
1639 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1640 | ||
7940876e SH |
1641 | static void mark_page_dirty_in_slot(struct kvm *kvm, |
1642 | struct kvm_memory_slot *memslot, | |
1643 | gfn_t gfn) | |
6aa8b732 | 1644 | { |
7e9d619d RR |
1645 | if (memslot && memslot->dirty_bitmap) { |
1646 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1647 | |
b74ca3b3 | 1648 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1649 | } |
1650 | } | |
1651 | ||
49c7754c GN |
1652 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1653 | { | |
1654 | struct kvm_memory_slot *memslot; | |
1655 | ||
1656 | memslot = gfn_to_memslot(kvm, gfn); | |
1657 | mark_page_dirty_in_slot(kvm, memslot, gfn); | |
1658 | } | |
2ba9f0d8 | 1659 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 1660 | |
b6958ce4 ED |
1661 | /* |
1662 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1663 | */ | |
8776e519 | 1664 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1665 | { |
e5c239cf MT |
1666 | DEFINE_WAIT(wait); |
1667 | ||
1668 | for (;;) { | |
1669 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1670 | ||
a1b37100 | 1671 | if (kvm_arch_vcpu_runnable(vcpu)) { |
a8eeb04a | 1672 | kvm_make_request(KVM_REQ_UNHALT, vcpu); |
e5c239cf | 1673 | break; |
d7690175 | 1674 | } |
09cec754 GN |
1675 | if (kvm_cpu_has_pending_timer(vcpu)) |
1676 | break; | |
e5c239cf MT |
1677 | if (signal_pending(current)) |
1678 | break; | |
1679 | ||
b6958ce4 | 1680 | schedule(); |
b6958ce4 | 1681 | } |
d3bef15f | 1682 | |
e5c239cf | 1683 | finish_wait(&vcpu->wq, &wait); |
b6958ce4 | 1684 | } |
2ba9f0d8 | 1685 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 1686 | |
8c84780d | 1687 | #ifndef CONFIG_S390 |
b6d33834 CD |
1688 | /* |
1689 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1690 | */ | |
1691 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1692 | { | |
1693 | int me; | |
1694 | int cpu = vcpu->cpu; | |
1695 | wait_queue_head_t *wqp; | |
1696 | ||
1697 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1698 | if (waitqueue_active(wqp)) { | |
1699 | wake_up_interruptible(wqp); | |
1700 | ++vcpu->stat.halt_wakeup; | |
1701 | } | |
1702 | ||
1703 | me = get_cpu(); | |
1704 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1705 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1706 | smp_send_reschedule(cpu); | |
1707 | put_cpu(); | |
1708 | } | |
a20ed54d | 1709 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1710 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1711 | |
41628d33 KW |
1712 | bool kvm_vcpu_yield_to(struct kvm_vcpu *target) |
1713 | { | |
1714 | struct pid *pid; | |
1715 | struct task_struct *task = NULL; | |
c45c528e | 1716 | bool ret = false; |
41628d33 KW |
1717 | |
1718 | rcu_read_lock(); | |
1719 | pid = rcu_dereference(target->pid); | |
1720 | if (pid) | |
1721 | task = get_pid_task(target->pid, PIDTYPE_PID); | |
1722 | rcu_read_unlock(); | |
1723 | if (!task) | |
c45c528e | 1724 | return ret; |
41628d33 KW |
1725 | if (task->flags & PF_VCPU) { |
1726 | put_task_struct(task); | |
c45c528e | 1727 | return ret; |
41628d33 | 1728 | } |
c45c528e | 1729 | ret = yield_to(task, 1); |
41628d33 | 1730 | put_task_struct(task); |
c45c528e R |
1731 | |
1732 | return ret; | |
41628d33 KW |
1733 | } |
1734 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
1735 | ||
06e48c51 R |
1736 | /* |
1737 | * Helper that checks whether a VCPU is eligible for directed yield. | |
1738 | * Most eligible candidate to yield is decided by following heuristics: | |
1739 | * | |
1740 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
1741 | * (preempted lock holder), indicated by @in_spin_loop. | |
1742 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
1743 | * | |
1744 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
1745 | * chance last time (mostly it has become eligible now since we have probably | |
1746 | * yielded to lockholder in last iteration. This is done by toggling | |
1747 | * @dy_eligible each time a VCPU checked for eligibility.) | |
1748 | * | |
1749 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
1750 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
1751 | * burning. Giving priority for a potential lock-holder increases lock | |
1752 | * progress. | |
1753 | * | |
1754 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
1755 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
1756 | * and continue with next VCPU and so on. | |
1757 | */ | |
7940876e | 1758 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 1759 | { |
4a55dd72 | 1760 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
1761 | bool eligible; |
1762 | ||
1763 | eligible = !vcpu->spin_loop.in_spin_loop || | |
1764 | (vcpu->spin_loop.in_spin_loop && | |
1765 | vcpu->spin_loop.dy_eligible); | |
1766 | ||
1767 | if (vcpu->spin_loop.in_spin_loop) | |
1768 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
1769 | ||
1770 | return eligible; | |
4a55dd72 SW |
1771 | #else |
1772 | return true; | |
06e48c51 | 1773 | #endif |
4a55dd72 | 1774 | } |
c45c528e | 1775 | |
217ece61 | 1776 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 1777 | { |
217ece61 RR |
1778 | struct kvm *kvm = me->kvm; |
1779 | struct kvm_vcpu *vcpu; | |
1780 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
1781 | int yielded = 0; | |
c45c528e | 1782 | int try = 3; |
217ece61 RR |
1783 | int pass; |
1784 | int i; | |
d255f4f2 | 1785 | |
4c088493 | 1786 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
1787 | /* |
1788 | * We boost the priority of a VCPU that is runnable but not | |
1789 | * currently running, because it got preempted by something | |
1790 | * else and called schedule in __vcpu_run. Hopefully that | |
1791 | * VCPU is holding the lock that we need and will release it. | |
1792 | * We approximate round-robin by starting at the last boosted VCPU. | |
1793 | */ | |
c45c528e | 1794 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 1795 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 1796 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
1797 | i = last_boosted_vcpu; |
1798 | continue; | |
1799 | } else if (pass && i > last_boosted_vcpu) | |
1800 | break; | |
7bc7ae25 R |
1801 | if (!ACCESS_ONCE(vcpu->preempted)) |
1802 | continue; | |
217ece61 RR |
1803 | if (vcpu == me) |
1804 | continue; | |
98f4a146 | 1805 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 1806 | continue; |
06e48c51 R |
1807 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
1808 | continue; | |
c45c528e R |
1809 | |
1810 | yielded = kvm_vcpu_yield_to(vcpu); | |
1811 | if (yielded > 0) { | |
217ece61 | 1812 | kvm->last_boosted_vcpu = i; |
217ece61 | 1813 | break; |
c45c528e R |
1814 | } else if (yielded < 0) { |
1815 | try--; | |
1816 | if (!try) | |
1817 | break; | |
217ece61 | 1818 | } |
217ece61 RR |
1819 | } |
1820 | } | |
4c088493 | 1821 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
1822 | |
1823 | /* Ensure vcpu is not eligible during next spinloop */ | |
1824 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
1825 | } |
1826 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
1827 | ||
e4a533a4 | 1828 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
1829 | { |
1830 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
1831 | struct page *page; |
1832 | ||
e4a533a4 | 1833 | if (vmf->pgoff == 0) |
039576c0 | 1834 | page = virt_to_page(vcpu->run); |
09566765 | 1835 | #ifdef CONFIG_X86 |
e4a533a4 | 1836 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 1837 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
1838 | #endif |
1839 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
1840 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
1841 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 1842 | #endif |
039576c0 | 1843 | else |
5b1c1493 | 1844 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 1845 | get_page(page); |
e4a533a4 | 1846 | vmf->page = page; |
1847 | return 0; | |
9a2bb7f4 AK |
1848 | } |
1849 | ||
f0f37e2f | 1850 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 1851 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
1852 | }; |
1853 | ||
1854 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
1855 | { | |
1856 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
1857 | return 0; | |
1858 | } | |
1859 | ||
bccf2150 AK |
1860 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
1861 | { | |
1862 | struct kvm_vcpu *vcpu = filp->private_data; | |
1863 | ||
66c0b394 | 1864 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
1865 | return 0; |
1866 | } | |
1867 | ||
3d3aab1b | 1868 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
1869 | .release = kvm_vcpu_release, |
1870 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
1dda606c AG |
1871 | #ifdef CONFIG_COMPAT |
1872 | .compat_ioctl = kvm_vcpu_compat_ioctl, | |
1873 | #endif | |
9a2bb7f4 | 1874 | .mmap = kvm_vcpu_mmap, |
6038f373 | 1875 | .llseek = noop_llseek, |
bccf2150 AK |
1876 | }; |
1877 | ||
1878 | /* | |
1879 | * Allocates an inode for the vcpu. | |
1880 | */ | |
1881 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
1882 | { | |
24009b05 | 1883 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
1884 | } |
1885 | ||
c5ea7660 AK |
1886 | /* |
1887 | * Creates some virtual cpus. Good luck creating more than one. | |
1888 | */ | |
73880c80 | 1889 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
1890 | { |
1891 | int r; | |
988a2cae | 1892 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 1893 | |
338c7dba AH |
1894 | if (id >= KVM_MAX_VCPUS) |
1895 | return -EINVAL; | |
1896 | ||
73880c80 | 1897 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
1898 | if (IS_ERR(vcpu)) |
1899 | return PTR_ERR(vcpu); | |
c5ea7660 | 1900 | |
15ad7146 AK |
1901 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
1902 | ||
26e5215f AK |
1903 | r = kvm_arch_vcpu_setup(vcpu); |
1904 | if (r) | |
d780592b | 1905 | goto vcpu_destroy; |
26e5215f | 1906 | |
11ec2804 | 1907 | mutex_lock(&kvm->lock); |
3e515705 AK |
1908 | if (!kvm_vcpu_compatible(vcpu)) { |
1909 | r = -EINVAL; | |
1910 | goto unlock_vcpu_destroy; | |
1911 | } | |
73880c80 GN |
1912 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
1913 | r = -EINVAL; | |
d780592b | 1914 | goto unlock_vcpu_destroy; |
fb3f0f51 | 1915 | } |
73880c80 | 1916 | |
988a2cae GN |
1917 | kvm_for_each_vcpu(r, v, kvm) |
1918 | if (v->vcpu_id == id) { | |
73880c80 | 1919 | r = -EEXIST; |
d780592b | 1920 | goto unlock_vcpu_destroy; |
73880c80 GN |
1921 | } |
1922 | ||
1923 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 1924 | |
fb3f0f51 | 1925 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 1926 | kvm_get_kvm(kvm); |
bccf2150 | 1927 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
1928 | if (r < 0) { |
1929 | kvm_put_kvm(kvm); | |
d780592b | 1930 | goto unlock_vcpu_destroy; |
73880c80 GN |
1931 | } |
1932 | ||
1933 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
1934 | smp_wmb(); | |
1935 | atomic_inc(&kvm->online_vcpus); | |
1936 | ||
73880c80 | 1937 | mutex_unlock(&kvm->lock); |
42897d86 | 1938 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 1939 | return r; |
39c3b86e | 1940 | |
d780592b | 1941 | unlock_vcpu_destroy: |
7d8fece6 | 1942 | mutex_unlock(&kvm->lock); |
d780592b | 1943 | vcpu_destroy: |
d40ccc62 | 1944 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
1945 | return r; |
1946 | } | |
1947 | ||
1961d276 AK |
1948 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
1949 | { | |
1950 | if (sigset) { | |
1951 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
1952 | vcpu->sigset_active = 1; | |
1953 | vcpu->sigset = *sigset; | |
1954 | } else | |
1955 | vcpu->sigset_active = 0; | |
1956 | return 0; | |
1957 | } | |
1958 | ||
bccf2150 AK |
1959 | static long kvm_vcpu_ioctl(struct file *filp, |
1960 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 1961 | { |
bccf2150 | 1962 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 1963 | void __user *argp = (void __user *)arg; |
313a3dc7 | 1964 | int r; |
fa3795a7 DH |
1965 | struct kvm_fpu *fpu = NULL; |
1966 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 1967 | |
6d4e4c4f AK |
1968 | if (vcpu->kvm->mm != current->mm) |
1969 | return -EIO; | |
2122ff5e | 1970 | |
2f4d9b54 | 1971 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
1972 | /* |
1973 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
1974 | * so vcpu_load() would break it. | |
1975 | */ | |
1976 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT) | |
1977 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); | |
1978 | #endif | |
1979 | ||
1980 | ||
9fc77441 MT |
1981 | r = vcpu_load(vcpu); |
1982 | if (r) | |
1983 | return r; | |
6aa8b732 | 1984 | switch (ioctl) { |
9a2bb7f4 | 1985 | case KVM_RUN: |
f0fe5108 AK |
1986 | r = -EINVAL; |
1987 | if (arg) | |
1988 | goto out; | |
b6c7a5dc | 1989 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 1990 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 1991 | break; |
6aa8b732 | 1992 | case KVM_GET_REGS: { |
3e4bb3ac | 1993 | struct kvm_regs *kvm_regs; |
6aa8b732 | 1994 | |
3e4bb3ac XZ |
1995 | r = -ENOMEM; |
1996 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
1997 | if (!kvm_regs) | |
6aa8b732 | 1998 | goto out; |
3e4bb3ac XZ |
1999 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2000 | if (r) | |
2001 | goto out_free1; | |
6aa8b732 | 2002 | r = -EFAULT; |
3e4bb3ac XZ |
2003 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2004 | goto out_free1; | |
6aa8b732 | 2005 | r = 0; |
3e4bb3ac XZ |
2006 | out_free1: |
2007 | kfree(kvm_regs); | |
6aa8b732 AK |
2008 | break; |
2009 | } | |
2010 | case KVM_SET_REGS: { | |
3e4bb3ac | 2011 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2012 | |
3e4bb3ac | 2013 | r = -ENOMEM; |
ff5c2c03 SL |
2014 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2015 | if (IS_ERR(kvm_regs)) { | |
2016 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2017 | goto out; |
ff5c2c03 | 2018 | } |
3e4bb3ac | 2019 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2020 | kfree(kvm_regs); |
6aa8b732 AK |
2021 | break; |
2022 | } | |
2023 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2024 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2025 | r = -ENOMEM; | |
2026 | if (!kvm_sregs) | |
2027 | goto out; | |
2028 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2029 | if (r) |
2030 | goto out; | |
2031 | r = -EFAULT; | |
fa3795a7 | 2032 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2033 | goto out; |
2034 | r = 0; | |
2035 | break; | |
2036 | } | |
2037 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2038 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2039 | if (IS_ERR(kvm_sregs)) { | |
2040 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2041 | kvm_sregs = NULL; |
6aa8b732 | 2042 | goto out; |
ff5c2c03 | 2043 | } |
fa3795a7 | 2044 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2045 | break; |
2046 | } | |
62d9f0db MT |
2047 | case KVM_GET_MP_STATE: { |
2048 | struct kvm_mp_state mp_state; | |
2049 | ||
2050 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2051 | if (r) | |
2052 | goto out; | |
2053 | r = -EFAULT; | |
2054 | if (copy_to_user(argp, &mp_state, sizeof mp_state)) | |
2055 | goto out; | |
2056 | r = 0; | |
2057 | break; | |
2058 | } | |
2059 | case KVM_SET_MP_STATE: { | |
2060 | struct kvm_mp_state mp_state; | |
2061 | ||
2062 | r = -EFAULT; | |
2063 | if (copy_from_user(&mp_state, argp, sizeof mp_state)) | |
2064 | goto out; | |
2065 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2066 | break; |
2067 | } | |
6aa8b732 AK |
2068 | case KVM_TRANSLATE: { |
2069 | struct kvm_translation tr; | |
2070 | ||
2071 | r = -EFAULT; | |
2f366987 | 2072 | if (copy_from_user(&tr, argp, sizeof tr)) |
6aa8b732 | 2073 | goto out; |
8b006791 | 2074 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2075 | if (r) |
2076 | goto out; | |
2077 | r = -EFAULT; | |
2f366987 | 2078 | if (copy_to_user(argp, &tr, sizeof tr)) |
6aa8b732 AK |
2079 | goto out; |
2080 | r = 0; | |
2081 | break; | |
2082 | } | |
d0bfb940 JK |
2083 | case KVM_SET_GUEST_DEBUG: { |
2084 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2085 | |
2086 | r = -EFAULT; | |
2f366987 | 2087 | if (copy_from_user(&dbg, argp, sizeof dbg)) |
6aa8b732 | 2088 | goto out; |
d0bfb940 | 2089 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2090 | break; |
2091 | } | |
1961d276 AK |
2092 | case KVM_SET_SIGNAL_MASK: { |
2093 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2094 | struct kvm_signal_mask kvm_sigmask; | |
2095 | sigset_t sigset, *p; | |
2096 | ||
2097 | p = NULL; | |
2098 | if (argp) { | |
2099 | r = -EFAULT; | |
2100 | if (copy_from_user(&kvm_sigmask, argp, | |
2101 | sizeof kvm_sigmask)) | |
2102 | goto out; | |
2103 | r = -EINVAL; | |
2104 | if (kvm_sigmask.len != sizeof sigset) | |
2105 | goto out; | |
2106 | r = -EFAULT; | |
2107 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
2108 | sizeof sigset)) | |
2109 | goto out; | |
2110 | p = &sigset; | |
2111 | } | |
376d41ff | 2112 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2113 | break; |
2114 | } | |
b8836737 | 2115 | case KVM_GET_FPU: { |
fa3795a7 DH |
2116 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2117 | r = -ENOMEM; | |
2118 | if (!fpu) | |
2119 | goto out; | |
2120 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2121 | if (r) |
2122 | goto out; | |
2123 | r = -EFAULT; | |
fa3795a7 | 2124 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2125 | goto out; |
2126 | r = 0; | |
2127 | break; | |
2128 | } | |
2129 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2130 | fpu = memdup_user(argp, sizeof(*fpu)); |
2131 | if (IS_ERR(fpu)) { | |
2132 | r = PTR_ERR(fpu); | |
18595411 | 2133 | fpu = NULL; |
b8836737 | 2134 | goto out; |
ff5c2c03 | 2135 | } |
fa3795a7 | 2136 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2137 | break; |
2138 | } | |
bccf2150 | 2139 | default: |
313a3dc7 | 2140 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2141 | } |
2142 | out: | |
2122ff5e | 2143 | vcpu_put(vcpu); |
fa3795a7 DH |
2144 | kfree(fpu); |
2145 | kfree(kvm_sregs); | |
bccf2150 AK |
2146 | return r; |
2147 | } | |
2148 | ||
1dda606c AG |
2149 | #ifdef CONFIG_COMPAT |
2150 | static long kvm_vcpu_compat_ioctl(struct file *filp, | |
2151 | unsigned int ioctl, unsigned long arg) | |
2152 | { | |
2153 | struct kvm_vcpu *vcpu = filp->private_data; | |
2154 | void __user *argp = compat_ptr(arg); | |
2155 | int r; | |
2156 | ||
2157 | if (vcpu->kvm->mm != current->mm) | |
2158 | return -EIO; | |
2159 | ||
2160 | switch (ioctl) { | |
2161 | case KVM_SET_SIGNAL_MASK: { | |
2162 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2163 | struct kvm_signal_mask kvm_sigmask; | |
2164 | compat_sigset_t csigset; | |
2165 | sigset_t sigset; | |
2166 | ||
2167 | if (argp) { | |
2168 | r = -EFAULT; | |
2169 | if (copy_from_user(&kvm_sigmask, argp, | |
2170 | sizeof kvm_sigmask)) | |
2171 | goto out; | |
2172 | r = -EINVAL; | |
2173 | if (kvm_sigmask.len != sizeof csigset) | |
2174 | goto out; | |
2175 | r = -EFAULT; | |
2176 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
2177 | sizeof csigset)) | |
2178 | goto out; | |
760a9a30 AC |
2179 | sigset_from_compat(&sigset, &csigset); |
2180 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2181 | } else | |
2182 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2183 | break; |
2184 | } | |
2185 | default: | |
2186 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2187 | } | |
2188 | ||
2189 | out: | |
2190 | return r; | |
2191 | } | |
2192 | #endif | |
2193 | ||
852b6d57 SW |
2194 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2195 | int (*accessor)(struct kvm_device *dev, | |
2196 | struct kvm_device_attr *attr), | |
2197 | unsigned long arg) | |
2198 | { | |
2199 | struct kvm_device_attr attr; | |
2200 | ||
2201 | if (!accessor) | |
2202 | return -EPERM; | |
2203 | ||
2204 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2205 | return -EFAULT; | |
2206 | ||
2207 | return accessor(dev, &attr); | |
2208 | } | |
2209 | ||
2210 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2211 | unsigned long arg) | |
2212 | { | |
2213 | struct kvm_device *dev = filp->private_data; | |
2214 | ||
2215 | switch (ioctl) { | |
2216 | case KVM_SET_DEVICE_ATTR: | |
2217 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2218 | case KVM_GET_DEVICE_ATTR: | |
2219 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2220 | case KVM_HAS_DEVICE_ATTR: | |
2221 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2222 | default: | |
2223 | if (dev->ops->ioctl) | |
2224 | return dev->ops->ioctl(dev, ioctl, arg); | |
2225 | ||
2226 | return -ENOTTY; | |
2227 | } | |
2228 | } | |
2229 | ||
852b6d57 SW |
2230 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2231 | { | |
2232 | struct kvm_device *dev = filp->private_data; | |
2233 | struct kvm *kvm = dev->kvm; | |
2234 | ||
852b6d57 SW |
2235 | kvm_put_kvm(kvm); |
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | static const struct file_operations kvm_device_fops = { | |
2240 | .unlocked_ioctl = kvm_device_ioctl, | |
db6ae615 SW |
2241 | #ifdef CONFIG_COMPAT |
2242 | .compat_ioctl = kvm_device_ioctl, | |
2243 | #endif | |
852b6d57 SW |
2244 | .release = kvm_device_release, |
2245 | }; | |
2246 | ||
2247 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2248 | { | |
2249 | if (filp->f_op != &kvm_device_fops) | |
2250 | return NULL; | |
2251 | ||
2252 | return filp->private_data; | |
2253 | } | |
2254 | ||
2255 | static int kvm_ioctl_create_device(struct kvm *kvm, | |
2256 | struct kvm_create_device *cd) | |
2257 | { | |
2258 | struct kvm_device_ops *ops = NULL; | |
2259 | struct kvm_device *dev; | |
2260 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2261 | int ret; | |
2262 | ||
2263 | switch (cd->type) { | |
5df554ad SW |
2264 | #ifdef CONFIG_KVM_MPIC |
2265 | case KVM_DEV_TYPE_FSL_MPIC_20: | |
2266 | case KVM_DEV_TYPE_FSL_MPIC_42: | |
2267 | ops = &kvm_mpic_ops; | |
2268 | break; | |
5975a2e0 PM |
2269 | #endif |
2270 | #ifdef CONFIG_KVM_XICS | |
2271 | case KVM_DEV_TYPE_XICS: | |
2272 | ops = &kvm_xics_ops; | |
2273 | break; | |
ec53500f AW |
2274 | #endif |
2275 | #ifdef CONFIG_KVM_VFIO | |
2276 | case KVM_DEV_TYPE_VFIO: | |
2277 | ops = &kvm_vfio_ops; | |
2278 | break; | |
7330672b CD |
2279 | #endif |
2280 | #ifdef CONFIG_KVM_ARM_VGIC | |
2281 | case KVM_DEV_TYPE_ARM_VGIC_V2: | |
2282 | ops = &kvm_arm_vgic_v2_ops; | |
2283 | break; | |
c05c4186 JF |
2284 | #endif |
2285 | #ifdef CONFIG_S390 | |
2286 | case KVM_DEV_TYPE_FLIC: | |
2287 | ops = &kvm_flic_ops; | |
2288 | break; | |
5df554ad | 2289 | #endif |
852b6d57 SW |
2290 | default: |
2291 | return -ENODEV; | |
2292 | } | |
2293 | ||
2294 | if (test) | |
2295 | return 0; | |
2296 | ||
2297 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2298 | if (!dev) | |
2299 | return -ENOMEM; | |
2300 | ||
2301 | dev->ops = ops; | |
2302 | dev->kvm = kvm; | |
852b6d57 SW |
2303 | |
2304 | ret = ops->create(dev, cd->type); | |
2305 | if (ret < 0) { | |
2306 | kfree(dev); | |
2307 | return ret; | |
2308 | } | |
2309 | ||
24009b05 | 2310 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2311 | if (ret < 0) { |
2312 | ops->destroy(dev); | |
2313 | return ret; | |
2314 | } | |
2315 | ||
07f0a7bd | 2316 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2317 | kvm_get_kvm(kvm); |
2318 | cd->fd = ret; | |
2319 | return 0; | |
2320 | } | |
2321 | ||
bccf2150 AK |
2322 | static long kvm_vm_ioctl(struct file *filp, |
2323 | unsigned int ioctl, unsigned long arg) | |
2324 | { | |
2325 | struct kvm *kvm = filp->private_data; | |
2326 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2327 | int r; |
bccf2150 | 2328 | |
6d4e4c4f AK |
2329 | if (kvm->mm != current->mm) |
2330 | return -EIO; | |
bccf2150 AK |
2331 | switch (ioctl) { |
2332 | case KVM_CREATE_VCPU: | |
2333 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2334 | break; |
6fc138d2 IE |
2335 | case KVM_SET_USER_MEMORY_REGION: { |
2336 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2337 | ||
2338 | r = -EFAULT; | |
2339 | if (copy_from_user(&kvm_userspace_mem, argp, | |
2340 | sizeof kvm_userspace_mem)) | |
2341 | goto out; | |
2342 | ||
47ae31e2 | 2343 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2344 | break; |
2345 | } | |
2346 | case KVM_GET_DIRTY_LOG: { | |
2347 | struct kvm_dirty_log log; | |
2348 | ||
2349 | r = -EFAULT; | |
2f366987 | 2350 | if (copy_from_user(&log, argp, sizeof log)) |
6aa8b732 | 2351 | goto out; |
2c6f5df9 | 2352 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2353 | break; |
2354 | } | |
5f94c174 LV |
2355 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2356 | case KVM_REGISTER_COALESCED_MMIO: { | |
2357 | struct kvm_coalesced_mmio_zone zone; | |
2358 | r = -EFAULT; | |
2359 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2360 | goto out; | |
5f94c174 | 2361 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2362 | break; |
2363 | } | |
2364 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2365 | struct kvm_coalesced_mmio_zone zone; | |
2366 | r = -EFAULT; | |
2367 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2368 | goto out; | |
5f94c174 | 2369 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2370 | break; |
2371 | } | |
2372 | #endif | |
721eecbf GH |
2373 | case KVM_IRQFD: { |
2374 | struct kvm_irqfd data; | |
2375 | ||
2376 | r = -EFAULT; | |
2377 | if (copy_from_user(&data, argp, sizeof data)) | |
2378 | goto out; | |
d4db2935 | 2379 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2380 | break; |
2381 | } | |
d34e6b17 GH |
2382 | case KVM_IOEVENTFD: { |
2383 | struct kvm_ioeventfd data; | |
2384 | ||
2385 | r = -EFAULT; | |
2386 | if (copy_from_user(&data, argp, sizeof data)) | |
2387 | goto out; | |
2388 | r = kvm_ioeventfd(kvm, &data); | |
2389 | break; | |
2390 | } | |
73880c80 GN |
2391 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2392 | case KVM_SET_BOOT_CPU_ID: | |
2393 | r = 0; | |
894a9c55 | 2394 | mutex_lock(&kvm->lock); |
73880c80 GN |
2395 | if (atomic_read(&kvm->online_vcpus) != 0) |
2396 | r = -EBUSY; | |
2397 | else | |
2398 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2399 | mutex_unlock(&kvm->lock); |
73880c80 | 2400 | break; |
07975ad3 JK |
2401 | #endif |
2402 | #ifdef CONFIG_HAVE_KVM_MSI | |
2403 | case KVM_SIGNAL_MSI: { | |
2404 | struct kvm_msi msi; | |
2405 | ||
2406 | r = -EFAULT; | |
2407 | if (copy_from_user(&msi, argp, sizeof msi)) | |
2408 | goto out; | |
2409 | r = kvm_send_userspace_msi(kvm, &msi); | |
2410 | break; | |
2411 | } | |
23d43cf9 CD |
2412 | #endif |
2413 | #ifdef __KVM_HAVE_IRQ_LINE | |
2414 | case KVM_IRQ_LINE_STATUS: | |
2415 | case KVM_IRQ_LINE: { | |
2416 | struct kvm_irq_level irq_event; | |
2417 | ||
2418 | r = -EFAULT; | |
2419 | if (copy_from_user(&irq_event, argp, sizeof irq_event)) | |
2420 | goto out; | |
2421 | ||
aa2fbe6d YZ |
2422 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2423 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2424 | if (r) |
2425 | goto out; | |
2426 | ||
2427 | r = -EFAULT; | |
2428 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
2429 | if (copy_to_user(argp, &irq_event, sizeof irq_event)) | |
2430 | goto out; | |
2431 | } | |
2432 | ||
2433 | r = 0; | |
2434 | break; | |
2435 | } | |
73880c80 | 2436 | #endif |
aa8d5944 AG |
2437 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2438 | case KVM_SET_GSI_ROUTING: { | |
2439 | struct kvm_irq_routing routing; | |
2440 | struct kvm_irq_routing __user *urouting; | |
2441 | struct kvm_irq_routing_entry *entries; | |
2442 | ||
2443 | r = -EFAULT; | |
2444 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2445 | goto out; | |
2446 | r = -EINVAL; | |
2447 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2448 | goto out; | |
2449 | if (routing.flags) | |
2450 | goto out; | |
2451 | r = -ENOMEM; | |
2452 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2453 | if (!entries) | |
2454 | goto out; | |
2455 | r = -EFAULT; | |
2456 | urouting = argp; | |
2457 | if (copy_from_user(entries, urouting->entries, | |
2458 | routing.nr * sizeof(*entries))) | |
2459 | goto out_free_irq_routing; | |
2460 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2461 | routing.flags); | |
2462 | out_free_irq_routing: | |
2463 | vfree(entries); | |
2464 | break; | |
2465 | } | |
2466 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2467 | case KVM_CREATE_DEVICE: { |
2468 | struct kvm_create_device cd; | |
2469 | ||
2470 | r = -EFAULT; | |
2471 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2472 | goto out; | |
2473 | ||
2474 | r = kvm_ioctl_create_device(kvm, &cd); | |
2475 | if (r) | |
2476 | goto out; | |
2477 | ||
2478 | r = -EFAULT; | |
2479 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2480 | goto out; | |
2481 | ||
2482 | r = 0; | |
2483 | break; | |
2484 | } | |
f17abe9a | 2485 | default: |
1fe779f8 | 2486 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
bfd99ff5 AK |
2487 | if (r == -ENOTTY) |
2488 | r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg); | |
f17abe9a AK |
2489 | } |
2490 | out: | |
2491 | return r; | |
2492 | } | |
2493 | ||
6ff5894c AB |
2494 | #ifdef CONFIG_COMPAT |
2495 | struct compat_kvm_dirty_log { | |
2496 | __u32 slot; | |
2497 | __u32 padding1; | |
2498 | union { | |
2499 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2500 | __u64 padding2; | |
2501 | }; | |
2502 | }; | |
2503 | ||
2504 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2505 | unsigned int ioctl, unsigned long arg) | |
2506 | { | |
2507 | struct kvm *kvm = filp->private_data; | |
2508 | int r; | |
2509 | ||
2510 | if (kvm->mm != current->mm) | |
2511 | return -EIO; | |
2512 | switch (ioctl) { | |
2513 | case KVM_GET_DIRTY_LOG: { | |
2514 | struct compat_kvm_dirty_log compat_log; | |
2515 | struct kvm_dirty_log log; | |
2516 | ||
2517 | r = -EFAULT; | |
2518 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2519 | sizeof(compat_log))) | |
2520 | goto out; | |
2521 | log.slot = compat_log.slot; | |
2522 | log.padding1 = compat_log.padding1; | |
2523 | log.padding2 = compat_log.padding2; | |
2524 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2525 | ||
2526 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2527 | break; |
2528 | } | |
2529 | default: | |
2530 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2531 | } | |
2532 | ||
2533 | out: | |
2534 | return r; | |
2535 | } | |
2536 | #endif | |
2537 | ||
3d3aab1b | 2538 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2539 | .release = kvm_vm_release, |
2540 | .unlocked_ioctl = kvm_vm_ioctl, | |
6ff5894c AB |
2541 | #ifdef CONFIG_COMPAT |
2542 | .compat_ioctl = kvm_vm_compat_ioctl, | |
2543 | #endif | |
6038f373 | 2544 | .llseek = noop_llseek, |
f17abe9a AK |
2545 | }; |
2546 | ||
e08b9637 | 2547 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2548 | { |
aac87636 | 2549 | int r; |
f17abe9a AK |
2550 | struct kvm *kvm; |
2551 | ||
e08b9637 | 2552 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2553 | if (IS_ERR(kvm)) |
2554 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2555 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2556 | r = kvm_coalesced_mmio_init(kvm); | |
2557 | if (r < 0) { | |
2558 | kvm_put_kvm(kvm); | |
2559 | return r; | |
2560 | } | |
2561 | #endif | |
24009b05 | 2562 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2563 | if (r < 0) |
66c0b394 | 2564 | kvm_put_kvm(kvm); |
f17abe9a | 2565 | |
aac87636 | 2566 | return r; |
f17abe9a AK |
2567 | } |
2568 | ||
1a811b61 AK |
2569 | static long kvm_dev_ioctl_check_extension_generic(long arg) |
2570 | { | |
2571 | switch (arg) { | |
ca9edaee | 2572 | case KVM_CAP_USER_MEMORY: |
1a811b61 | 2573 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
4cd481f6 | 2574 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: |
73880c80 GN |
2575 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2576 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2577 | #endif | |
a9c7399d | 2578 | case KVM_CAP_INTERNAL_ERROR_DATA: |
07975ad3 JK |
2579 | #ifdef CONFIG_HAVE_KVM_MSI |
2580 | case KVM_CAP_SIGNAL_MSI: | |
7df35f54 AG |
2581 | #endif |
2582 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2583 | case KVM_CAP_IRQFD_RESAMPLE: | |
07975ad3 | 2584 | #endif |
1a811b61 | 2585 | return 1; |
a725d56a | 2586 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
399ec807 | 2587 | case KVM_CAP_IRQ_ROUTING: |
36463146 | 2588 | return KVM_MAX_IRQ_ROUTES; |
399ec807 | 2589 | #endif |
1a811b61 AK |
2590 | default: |
2591 | break; | |
2592 | } | |
2593 | return kvm_dev_ioctl_check_extension(arg); | |
2594 | } | |
2595 | ||
f17abe9a AK |
2596 | static long kvm_dev_ioctl(struct file *filp, |
2597 | unsigned int ioctl, unsigned long arg) | |
2598 | { | |
07c45a36 | 2599 | long r = -EINVAL; |
f17abe9a AK |
2600 | |
2601 | switch (ioctl) { | |
2602 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2603 | r = -EINVAL; |
2604 | if (arg) | |
2605 | goto out; | |
f17abe9a AK |
2606 | r = KVM_API_VERSION; |
2607 | break; | |
2608 | case KVM_CREATE_VM: | |
e08b9637 | 2609 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2610 | break; |
018d00d2 | 2611 | case KVM_CHECK_EXTENSION: |
1a811b61 | 2612 | r = kvm_dev_ioctl_check_extension_generic(arg); |
5d308f45 | 2613 | break; |
07c45a36 AK |
2614 | case KVM_GET_VCPU_MMAP_SIZE: |
2615 | r = -EINVAL; | |
2616 | if (arg) | |
2617 | goto out; | |
adb1ff46 AK |
2618 | r = PAGE_SIZE; /* struct kvm_run */ |
2619 | #ifdef CONFIG_X86 | |
2620 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2621 | #endif |
2622 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2623 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2624 | #endif |
07c45a36 | 2625 | break; |
d4c9ff2d FEL |
2626 | case KVM_TRACE_ENABLE: |
2627 | case KVM_TRACE_PAUSE: | |
2628 | case KVM_TRACE_DISABLE: | |
2023a29c | 2629 | r = -EOPNOTSUPP; |
d4c9ff2d | 2630 | break; |
6aa8b732 | 2631 | default: |
043405e1 | 2632 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2633 | } |
2634 | out: | |
2635 | return r; | |
2636 | } | |
2637 | ||
6aa8b732 | 2638 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2639 | .unlocked_ioctl = kvm_dev_ioctl, |
2640 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2641 | .llseek = noop_llseek, |
6aa8b732 AK |
2642 | }; |
2643 | ||
2644 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2645 | KVM_MINOR, |
6aa8b732 AK |
2646 | "kvm", |
2647 | &kvm_chardev_ops, | |
2648 | }; | |
2649 | ||
75b7127c | 2650 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2651 | { |
2652 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2653 | int r; |
1b6c0168 | 2654 | |
7f59f492 | 2655 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2656 | return; |
10474ae8 | 2657 | |
7f59f492 | 2658 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 AG |
2659 | |
2660 | r = kvm_arch_hardware_enable(NULL); | |
2661 | ||
2662 | if (r) { | |
2663 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2664 | atomic_inc(&hardware_enable_failed); | |
2665 | printk(KERN_INFO "kvm: enabling virtualization on " | |
2666 | "CPU%d failed\n", cpu); | |
2667 | } | |
1b6c0168 AK |
2668 | } |
2669 | ||
4fa92fb2 | 2670 | static void hardware_enable(void) |
75b7127c | 2671 | { |
4a937f96 | 2672 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2673 | if (kvm_usage_count) |
2674 | hardware_enable_nolock(NULL); | |
4a937f96 | 2675 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2676 | } |
2677 | ||
2678 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2679 | { |
2680 | int cpu = raw_smp_processor_id(); | |
2681 | ||
7f59f492 | 2682 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2683 | return; |
7f59f492 | 2684 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
e9b11c17 | 2685 | kvm_arch_hardware_disable(NULL); |
1b6c0168 AK |
2686 | } |
2687 | ||
4fa92fb2 | 2688 | static void hardware_disable(void) |
75b7127c | 2689 | { |
4a937f96 | 2690 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2691 | if (kvm_usage_count) |
2692 | hardware_disable_nolock(NULL); | |
4a937f96 | 2693 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2694 | } |
2695 | ||
10474ae8 AG |
2696 | static void hardware_disable_all_nolock(void) |
2697 | { | |
2698 | BUG_ON(!kvm_usage_count); | |
2699 | ||
2700 | kvm_usage_count--; | |
2701 | if (!kvm_usage_count) | |
75b7127c | 2702 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2703 | } |
2704 | ||
2705 | static void hardware_disable_all(void) | |
2706 | { | |
4a937f96 | 2707 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 2708 | hardware_disable_all_nolock(); |
4a937f96 | 2709 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2710 | } |
2711 | ||
2712 | static int hardware_enable_all(void) | |
2713 | { | |
2714 | int r = 0; | |
2715 | ||
4a937f96 | 2716 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
2717 | |
2718 | kvm_usage_count++; | |
2719 | if (kvm_usage_count == 1) { | |
2720 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 2721 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
2722 | |
2723 | if (atomic_read(&hardware_enable_failed)) { | |
2724 | hardware_disable_all_nolock(); | |
2725 | r = -EBUSY; | |
2726 | } | |
2727 | } | |
2728 | ||
4a937f96 | 2729 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2730 | |
2731 | return r; | |
2732 | } | |
2733 | ||
774c47f1 AK |
2734 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
2735 | void *v) | |
2736 | { | |
2737 | int cpu = (long)v; | |
2738 | ||
1a6f4d7f | 2739 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 2740 | switch (val) { |
cec9ad27 | 2741 | case CPU_DYING: |
6ec8a856 AK |
2742 | printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n", |
2743 | cpu); | |
4fa92fb2 | 2744 | hardware_disable(); |
6ec8a856 | 2745 | break; |
da908f2f | 2746 | case CPU_STARTING: |
43934a38 JK |
2747 | printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n", |
2748 | cpu); | |
4fa92fb2 | 2749 | hardware_enable(); |
774c47f1 AK |
2750 | break; |
2751 | } | |
2752 | return NOTIFY_OK; | |
2753 | } | |
2754 | ||
9a2b85c6 | 2755 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 2756 | void *v) |
9a2b85c6 | 2757 | { |
8e1c1815 SY |
2758 | /* |
2759 | * Some (well, at least mine) BIOSes hang on reboot if | |
2760 | * in vmx root mode. | |
2761 | * | |
2762 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
2763 | */ | |
2764 | printk(KERN_INFO "kvm: exiting hardware virtualization\n"); | |
2765 | kvm_rebooting = true; | |
75b7127c | 2766 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
2767 | return NOTIFY_OK; |
2768 | } | |
2769 | ||
2770 | static struct notifier_block kvm_reboot_notifier = { | |
2771 | .notifier_call = kvm_reboot, | |
2772 | .priority = 0, | |
2773 | }; | |
2774 | ||
e93f8a0f | 2775 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
2776 | { |
2777 | int i; | |
2778 | ||
2779 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 2780 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
2781 | |
2782 | kvm_iodevice_destructor(pos); | |
2783 | } | |
e93f8a0f | 2784 | kfree(bus); |
2eeb2e94 GH |
2785 | } |
2786 | ||
c21fbff1 PB |
2787 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
2788 | const struct kvm_io_range *r2) | |
743eeb0b | 2789 | { |
743eeb0b SL |
2790 | if (r1->addr < r2->addr) |
2791 | return -1; | |
2792 | if (r1->addr + r1->len > r2->addr + r2->len) | |
2793 | return 1; | |
2794 | return 0; | |
2795 | } | |
2796 | ||
a343c9b7 PB |
2797 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
2798 | { | |
c21fbff1 | 2799 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
2800 | } |
2801 | ||
39369f7a | 2802 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
2803 | gpa_t addr, int len) |
2804 | { | |
743eeb0b SL |
2805 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
2806 | .addr = addr, | |
2807 | .len = len, | |
2808 | .dev = dev, | |
2809 | }; | |
2810 | ||
2811 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
2812 | kvm_io_bus_sort_cmp, NULL); | |
2813 | ||
2814 | return 0; | |
2815 | } | |
2816 | ||
39369f7a | 2817 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
2818 | gpa_t addr, int len) |
2819 | { | |
2820 | struct kvm_io_range *range, key; | |
2821 | int off; | |
2822 | ||
2823 | key = (struct kvm_io_range) { | |
2824 | .addr = addr, | |
2825 | .len = len, | |
2826 | }; | |
2827 | ||
2828 | range = bsearch(&key, bus->range, bus->dev_count, | |
2829 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
2830 | if (range == NULL) | |
2831 | return -ENOENT; | |
2832 | ||
2833 | off = range - bus->range; | |
2834 | ||
c21fbff1 | 2835 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
2836 | off--; |
2837 | ||
2838 | return off; | |
2839 | } | |
2840 | ||
126a5af5 CH |
2841 | static int __kvm_io_bus_write(struct kvm_io_bus *bus, |
2842 | struct kvm_io_range *range, const void *val) | |
2843 | { | |
2844 | int idx; | |
2845 | ||
2846 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
2847 | if (idx < 0) | |
2848 | return -EOPNOTSUPP; | |
2849 | ||
2850 | while (idx < bus->dev_count && | |
c21fbff1 | 2851 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
2852 | if (!kvm_iodevice_write(bus->range[idx].dev, range->addr, |
2853 | range->len, val)) | |
2854 | return idx; | |
2855 | idx++; | |
2856 | } | |
2857 | ||
2858 | return -EOPNOTSUPP; | |
2859 | } | |
2860 | ||
bda9020e | 2861 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e93f8a0f | 2862 | int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 2863 | int len, const void *val) |
2eeb2e94 | 2864 | { |
90d83dc3 | 2865 | struct kvm_io_bus *bus; |
743eeb0b | 2866 | struct kvm_io_range range; |
126a5af5 | 2867 | int r; |
743eeb0b SL |
2868 | |
2869 | range = (struct kvm_io_range) { | |
2870 | .addr = addr, | |
2871 | .len = len, | |
2872 | }; | |
90d83dc3 LJ |
2873 | |
2874 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
126a5af5 CH |
2875 | r = __kvm_io_bus_write(bus, &range, val); |
2876 | return r < 0 ? r : 0; | |
2877 | } | |
2878 | ||
2879 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
2880 | int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, | |
2881 | int len, const void *val, long cookie) | |
2882 | { | |
2883 | struct kvm_io_bus *bus; | |
2884 | struct kvm_io_range range; | |
2885 | ||
2886 | range = (struct kvm_io_range) { | |
2887 | .addr = addr, | |
2888 | .len = len, | |
2889 | }; | |
2890 | ||
2891 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
2892 | ||
2893 | /* First try the device referenced by cookie. */ | |
2894 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 2895 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
126a5af5 CH |
2896 | if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len, |
2897 | val)) | |
2898 | return cookie; | |
2899 | ||
2900 | /* | |
2901 | * cookie contained garbage; fall back to search and return the | |
2902 | * correct cookie value. | |
2903 | */ | |
2904 | return __kvm_io_bus_write(bus, &range, val); | |
2905 | } | |
2906 | ||
2907 | static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range, | |
2908 | void *val) | |
2909 | { | |
2910 | int idx; | |
2911 | ||
2912 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
2913 | if (idx < 0) |
2914 | return -EOPNOTSUPP; | |
2915 | ||
2916 | while (idx < bus->dev_count && | |
c21fbff1 | 2917 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
2918 | if (!kvm_iodevice_read(bus->range[idx].dev, range->addr, |
2919 | range->len, val)) | |
2920 | return idx; | |
743eeb0b SL |
2921 | idx++; |
2922 | } | |
2923 | ||
bda9020e MT |
2924 | return -EOPNOTSUPP; |
2925 | } | |
68c3b4d1 | 2926 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 2927 | |
bda9020e | 2928 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e93f8a0f MT |
2929 | int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2930 | int len, void *val) | |
bda9020e | 2931 | { |
90d83dc3 | 2932 | struct kvm_io_bus *bus; |
743eeb0b | 2933 | struct kvm_io_range range; |
126a5af5 | 2934 | int r; |
743eeb0b SL |
2935 | |
2936 | range = (struct kvm_io_range) { | |
2937 | .addr = addr, | |
2938 | .len = len, | |
2939 | }; | |
e93f8a0f | 2940 | |
90d83dc3 | 2941 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); |
126a5af5 CH |
2942 | r = __kvm_io_bus_read(bus, &range, val); |
2943 | return r < 0 ? r : 0; | |
2944 | } | |
743eeb0b | 2945 | |
2eeb2e94 | 2946 | |
79fac95e | 2947 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
2948 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2949 | int len, struct kvm_io_device *dev) | |
6c474694 | 2950 | { |
e93f8a0f | 2951 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 2952 | |
e93f8a0f | 2953 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
2954 | /* exclude ioeventfd which is limited by maximum fd */ |
2955 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 2956 | return -ENOSPC; |
2eeb2e94 | 2957 | |
a1300716 AK |
2958 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
2959 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
2960 | if (!new_bus) |
2961 | return -ENOMEM; | |
a1300716 AK |
2962 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
2963 | sizeof(struct kvm_io_range))); | |
743eeb0b | 2964 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
2965 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
2966 | synchronize_srcu_expedited(&kvm->srcu); | |
2967 | kfree(bus); | |
090b7aff GH |
2968 | |
2969 | return 0; | |
2970 | } | |
2971 | ||
79fac95e | 2972 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
2973 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
2974 | struct kvm_io_device *dev) | |
090b7aff | 2975 | { |
e93f8a0f MT |
2976 | int i, r; |
2977 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 2978 | |
cdfca7b3 | 2979 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 2980 | r = -ENOENT; |
a1300716 AK |
2981 | for (i = 0; i < bus->dev_count; i++) |
2982 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 2983 | r = 0; |
090b7aff GH |
2984 | break; |
2985 | } | |
e93f8a0f | 2986 | |
a1300716 | 2987 | if (r) |
e93f8a0f | 2988 | return r; |
a1300716 AK |
2989 | |
2990 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
2991 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
2992 | if (!new_bus) | |
2993 | return -ENOMEM; | |
2994 | ||
2995 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
2996 | new_bus->dev_count--; | |
2997 | memcpy(new_bus->range + i, bus->range + i + 1, | |
2998 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
2999 | |
3000 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3001 | synchronize_srcu_expedited(&kvm->srcu); | |
3002 | kfree(bus); | |
3003 | return r; | |
2eeb2e94 GH |
3004 | } |
3005 | ||
774c47f1 AK |
3006 | static struct notifier_block kvm_cpu_notifier = { |
3007 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3008 | }; |
3009 | ||
8b88b099 | 3010 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3011 | { |
3012 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3013 | struct kvm *kvm; |
3014 | ||
8b88b099 | 3015 | *val = 0; |
2f303b74 | 3016 | spin_lock(&kvm_lock); |
ba1389b7 | 3017 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3018 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3019 | spin_unlock(&kvm_lock); |
8b88b099 | 3020 | return 0; |
ba1389b7 AK |
3021 | } |
3022 | ||
3023 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3024 | ||
8b88b099 | 3025 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3026 | { |
3027 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3028 | struct kvm *kvm; |
3029 | struct kvm_vcpu *vcpu; | |
3030 | int i; | |
3031 | ||
8b88b099 | 3032 | *val = 0; |
2f303b74 | 3033 | spin_lock(&kvm_lock); |
1165f5fe | 3034 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3035 | kvm_for_each_vcpu(i, vcpu, kvm) |
3036 | *val += *(u32 *)((void *)vcpu + offset); | |
3037 | ||
2f303b74 | 3038 | spin_unlock(&kvm_lock); |
8b88b099 | 3039 | return 0; |
1165f5fe AK |
3040 | } |
3041 | ||
ba1389b7 AK |
3042 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3043 | ||
828c0950 | 3044 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3045 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3046 | [KVM_STAT_VM] = &vm_stat_fops, | |
3047 | }; | |
1165f5fe | 3048 | |
4f69b680 | 3049 | static int kvm_init_debug(void) |
6aa8b732 | 3050 | { |
0c8eb04a | 3051 | int r = -EEXIST; |
6aa8b732 AK |
3052 | struct kvm_stats_debugfs_item *p; |
3053 | ||
76f7c879 | 3054 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3055 | if (kvm_debugfs_dir == NULL) |
3056 | goto out; | |
3057 | ||
3058 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3059 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3060 | (void *)(long)p->offset, |
ba1389b7 | 3061 | stat_fops[p->kind]); |
4f69b680 H |
3062 | if (p->dentry == NULL) |
3063 | goto out_dir; | |
3064 | } | |
3065 | ||
3066 | return 0; | |
3067 | ||
3068 | out_dir: | |
3069 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3070 | out: | |
3071 | return r; | |
6aa8b732 AK |
3072 | } |
3073 | ||
3074 | static void kvm_exit_debug(void) | |
3075 | { | |
3076 | struct kvm_stats_debugfs_item *p; | |
3077 | ||
3078 | for (p = debugfs_entries; p->name; ++p) | |
3079 | debugfs_remove(p->dentry); | |
76f7c879 | 3080 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3081 | } |
3082 | ||
fb3600cc | 3083 | static int kvm_suspend(void) |
59ae6c6b | 3084 | { |
10474ae8 | 3085 | if (kvm_usage_count) |
75b7127c | 3086 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3087 | return 0; |
3088 | } | |
3089 | ||
fb3600cc | 3090 | static void kvm_resume(void) |
59ae6c6b | 3091 | { |
ca84d1a2 | 3092 | if (kvm_usage_count) { |
4a937f96 | 3093 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3094 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3095 | } |
59ae6c6b AK |
3096 | } |
3097 | ||
fb3600cc | 3098 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3099 | .suspend = kvm_suspend, |
3100 | .resume = kvm_resume, | |
3101 | }; | |
3102 | ||
15ad7146 AK |
3103 | static inline |
3104 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3105 | { | |
3106 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3107 | } | |
3108 | ||
3109 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3110 | { | |
3111 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3a08a8f9 R |
3112 | if (vcpu->preempted) |
3113 | vcpu->preempted = false; | |
15ad7146 | 3114 | |
e9b11c17 | 3115 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3116 | } |
3117 | ||
3118 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3119 | struct task_struct *next) | |
3120 | { | |
3121 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3122 | ||
3a08a8f9 R |
3123 | if (current->state == TASK_RUNNING) |
3124 | vcpu->preempted = true; | |
e9b11c17 | 3125 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3126 | } |
3127 | ||
0ee75bea | 3128 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3129 | struct module *module) |
6aa8b732 AK |
3130 | { |
3131 | int r; | |
002c7f7c | 3132 | int cpu; |
6aa8b732 | 3133 | |
f8c16bba ZX |
3134 | r = kvm_arch_init(opaque); |
3135 | if (r) | |
d2308784 | 3136 | goto out_fail; |
cb498ea2 | 3137 | |
7dac16c3 AH |
3138 | /* |
3139 | * kvm_arch_init makes sure there's at most one caller | |
3140 | * for architectures that support multiple implementations, | |
3141 | * like intel and amd on x86. | |
3142 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3143 | * conflicts in case kvm is already setup for another implementation. | |
3144 | */ | |
3145 | r = kvm_irqfd_init(); | |
3146 | if (r) | |
3147 | goto out_irqfd; | |
3148 | ||
8437a617 | 3149 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3150 | r = -ENOMEM; |
3151 | goto out_free_0; | |
3152 | } | |
3153 | ||
e9b11c17 | 3154 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3155 | if (r < 0) |
7f59f492 | 3156 | goto out_free_0a; |
6aa8b732 | 3157 | |
002c7f7c YS |
3158 | for_each_online_cpu(cpu) { |
3159 | smp_call_function_single(cpu, | |
e9b11c17 | 3160 | kvm_arch_check_processor_compat, |
8691e5a8 | 3161 | &r, 1); |
002c7f7c | 3162 | if (r < 0) |
d2308784 | 3163 | goto out_free_1; |
002c7f7c YS |
3164 | } |
3165 | ||
774c47f1 AK |
3166 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3167 | if (r) | |
d2308784 | 3168 | goto out_free_2; |
6aa8b732 AK |
3169 | register_reboot_notifier(&kvm_reboot_notifier); |
3170 | ||
c16f862d | 3171 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3172 | if (!vcpu_align) |
3173 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3174 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3175 | 0, NULL); |
c16f862d RR |
3176 | if (!kvm_vcpu_cache) { |
3177 | r = -ENOMEM; | |
fb3600cc | 3178 | goto out_free_3; |
c16f862d RR |
3179 | } |
3180 | ||
af585b92 GN |
3181 | r = kvm_async_pf_init(); |
3182 | if (r) | |
3183 | goto out_free; | |
3184 | ||
6aa8b732 | 3185 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3186 | kvm_vm_fops.owner = module; |
3187 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3188 | |
3189 | r = misc_register(&kvm_dev); | |
3190 | if (r) { | |
d77c26fc | 3191 | printk(KERN_ERR "kvm: misc device register failed\n"); |
af585b92 | 3192 | goto out_unreg; |
6aa8b732 AK |
3193 | } |
3194 | ||
fb3600cc RW |
3195 | register_syscore_ops(&kvm_syscore_ops); |
3196 | ||
15ad7146 AK |
3197 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3198 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3199 | ||
4f69b680 H |
3200 | r = kvm_init_debug(); |
3201 | if (r) { | |
3202 | printk(KERN_ERR "kvm: create debugfs files failed\n"); | |
3203 | goto out_undebugfs; | |
3204 | } | |
0ea4ed8e | 3205 | |
c7addb90 | 3206 | return 0; |
6aa8b732 | 3207 | |
4f69b680 H |
3208 | out_undebugfs: |
3209 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3210 | misc_deregister(&kvm_dev); |
af585b92 GN |
3211 | out_unreg: |
3212 | kvm_async_pf_deinit(); | |
6aa8b732 | 3213 | out_free: |
c16f862d | 3214 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3215 | out_free_3: |
6aa8b732 | 3216 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3217 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3218 | out_free_2: |
d2308784 | 3219 | out_free_1: |
e9b11c17 | 3220 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3221 | out_free_0a: |
3222 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3223 | out_free_0: |
a0f155e9 CH |
3224 | kvm_irqfd_exit(); |
3225 | out_irqfd: | |
7dac16c3 AH |
3226 | kvm_arch_exit(); |
3227 | out_fail: | |
6aa8b732 AK |
3228 | return r; |
3229 | } | |
cb498ea2 | 3230 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3231 | |
cb498ea2 | 3232 | void kvm_exit(void) |
6aa8b732 | 3233 | { |
0ea4ed8e | 3234 | kvm_exit_debug(); |
6aa8b732 | 3235 | misc_deregister(&kvm_dev); |
c16f862d | 3236 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3237 | kvm_async_pf_deinit(); |
fb3600cc | 3238 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3239 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3240 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3241 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3242 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3243 | kvm_arch_exit(); |
a0f155e9 | 3244 | kvm_irqfd_exit(); |
7f59f492 | 3245 | free_cpumask_var(cpus_hardware_enabled); |
6aa8b732 | 3246 | } |
cb498ea2 | 3247 | EXPORT_SYMBOL_GPL(kvm_exit); |