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" |
3c3c29fd | 60 | #include "vfio.h" |
5f94c174 | 61 | |
229456fc MT |
62 | #define CREATE_TRACE_POINTS |
63 | #include <trace/events/kvm.h> | |
64 | ||
6aa8b732 AK |
65 | MODULE_AUTHOR("Qumranet"); |
66 | MODULE_LICENSE("GPL"); | |
67 | ||
fa40a821 MT |
68 | /* |
69 | * Ordering of locks: | |
70 | * | |
fae3a353 | 71 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
72 | */ |
73 | ||
2f303b74 | 74 | DEFINE_SPINLOCK(kvm_lock); |
4a937f96 | 75 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 76 | LIST_HEAD(vm_list); |
133de902 | 77 | |
7f59f492 | 78 | static cpumask_var_t cpus_hardware_enabled; |
10474ae8 AG |
79 | static int kvm_usage_count = 0; |
80 | static atomic_t hardware_enable_failed; | |
1b6c0168 | 81 | |
c16f862d RR |
82 | struct kmem_cache *kvm_vcpu_cache; |
83 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 84 | |
15ad7146 AK |
85 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
86 | ||
76f7c879 | 87 | struct dentry *kvm_debugfs_dir; |
6aa8b732 | 88 | |
bccf2150 AK |
89 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
90 | unsigned long arg); | |
1dda606c AG |
91 | #ifdef CONFIG_COMPAT |
92 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, | |
93 | unsigned long arg); | |
94 | #endif | |
10474ae8 AG |
95 | static int hardware_enable_all(void); |
96 | static void hardware_disable_all(void); | |
bccf2150 | 97 | |
e93f8a0f | 98 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e SH |
99 | |
100 | static void kvm_release_pfn_dirty(pfn_t pfn); | |
101 | static void mark_page_dirty_in_slot(struct kvm *kvm, | |
102 | struct kvm_memory_slot *memslot, gfn_t gfn); | |
e93f8a0f | 103 | |
52480137 | 104 | __visible bool kvm_rebooting; |
b7c4145b | 105 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 106 | |
54dee993 MT |
107 | static bool largepages_enabled = true; |
108 | ||
a2766325 | 109 | bool kvm_is_mmio_pfn(pfn_t pfn) |
cbff90a7 | 110 | { |
11feeb49 AA |
111 | if (pfn_valid(pfn)) |
112 | return PageReserved(pfn_to_page(pfn)); | |
cbff90a7 BAY |
113 | |
114 | return true; | |
115 | } | |
116 | ||
bccf2150 AK |
117 | /* |
118 | * Switches to specified vcpu, until a matching vcpu_put() | |
119 | */ | |
9fc77441 | 120 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 121 | { |
15ad7146 AK |
122 | int cpu; |
123 | ||
9fc77441 MT |
124 | if (mutex_lock_killable(&vcpu->mutex)) |
125 | return -EINTR; | |
34bb10b7 RR |
126 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
127 | /* The thread running this VCPU changed. */ | |
128 | struct pid *oldpid = vcpu->pid; | |
129 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
130 | rcu_assign_pointer(vcpu->pid, newpid); | |
7103f60d CB |
131 | if (oldpid) |
132 | synchronize_rcu(); | |
34bb10b7 RR |
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 | 459 | if (r) |
719d93cd | 460 | goto out_err_no_disable; |
10474ae8 AG |
461 | |
462 | r = hardware_enable_all(); | |
463 | if (r) | |
719d93cd | 464 | goto out_err_no_disable; |
10474ae8 | 465 | |
75858a84 AK |
466 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
467 | INIT_HLIST_HEAD(&kvm->mask_notifier_list); | |
c77dcacb PB |
468 | #endif |
469 | #ifdef CONFIG_HAVE_KVM_IRQFD | |
136bdfee | 470 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 471 | #endif |
6aa8b732 | 472 | |
1e702d9a AW |
473 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
474 | ||
46a26bf5 MT |
475 | r = -ENOMEM; |
476 | kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); | |
477 | if (!kvm->memslots) | |
719d93cd | 478 | goto out_err_no_srcu; |
00f034a1 PB |
479 | |
480 | /* | |
481 | * Init kvm generation close to the maximum to easily test the | |
482 | * code of handling generation number wrap-around. | |
483 | */ | |
484 | kvm->memslots->generation = -150; | |
485 | ||
bf3e05bc | 486 | kvm_init_memslots_id(kvm); |
bc6678a3 | 487 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
488 | goto out_err_no_srcu; |
489 | if (init_srcu_struct(&kvm->irq_srcu)) | |
490 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
491 | for (i = 0; i < KVM_NR_BUSES; i++) { |
492 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
493 | GFP_KERNEL); | |
57e7fbee | 494 | if (!kvm->buses[i]) |
e93f8a0f | 495 | goto out_err; |
e93f8a0f | 496 | } |
e930bffe | 497 | |
74b5c5bf | 498 | spin_lock_init(&kvm->mmu_lock); |
6d4e4c4f AK |
499 | kvm->mm = current->mm; |
500 | atomic_inc(&kvm->mm->mm_count); | |
d34e6b17 | 501 | kvm_eventfd_init(kvm); |
11ec2804 | 502 | mutex_init(&kvm->lock); |
60eead79 | 503 | mutex_init(&kvm->irq_lock); |
79fac95e | 504 | mutex_init(&kvm->slots_lock); |
d39f13b0 | 505 | atomic_set(&kvm->users_count, 1); |
07f0a7bd | 506 | INIT_LIST_HEAD(&kvm->devices); |
74b5c5bf MW |
507 | |
508 | r = kvm_init_mmu_notifier(kvm); | |
509 | if (r) | |
510 | goto out_err; | |
511 | ||
2f303b74 | 512 | spin_lock(&kvm_lock); |
5e58cfe4 | 513 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 514 | spin_unlock(&kvm_lock); |
d89f5eff | 515 | |
f17abe9a | 516 | return kvm; |
10474ae8 AG |
517 | |
518 | out_err: | |
719d93cd CB |
519 | cleanup_srcu_struct(&kvm->irq_srcu); |
520 | out_err_no_irq_srcu: | |
57e7fbee | 521 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 522 | out_err_no_srcu: |
10474ae8 | 523 | hardware_disable_all(); |
719d93cd | 524 | out_err_no_disable: |
e93f8a0f MT |
525 | for (i = 0; i < KVM_NR_BUSES; i++) |
526 | kfree(kvm->buses[i]); | |
46a26bf5 | 527 | kfree(kvm->memslots); |
d89f5eff | 528 | kvm_arch_free_vm(kvm); |
10474ae8 | 529 | return ERR_PTR(r); |
f17abe9a AK |
530 | } |
531 | ||
92eca8fa TY |
532 | /* |
533 | * Avoid using vmalloc for a small buffer. | |
534 | * Should not be used when the size is statically known. | |
535 | */ | |
c1a7b32a | 536 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
537 | { |
538 | if (size > PAGE_SIZE) | |
539 | return vzalloc(size); | |
540 | else | |
541 | return kzalloc(size, GFP_KERNEL); | |
542 | } | |
543 | ||
c1a7b32a | 544 | void kvm_kvfree(const void *addr) |
92eca8fa TY |
545 | { |
546 | if (is_vmalloc_addr(addr)) | |
547 | vfree(addr); | |
548 | else | |
549 | kfree(addr); | |
550 | } | |
551 | ||
a36a57b1 TY |
552 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
553 | { | |
554 | if (!memslot->dirty_bitmap) | |
555 | return; | |
556 | ||
92eca8fa | 557 | kvm_kvfree(memslot->dirty_bitmap); |
a36a57b1 TY |
558 | memslot->dirty_bitmap = NULL; |
559 | } | |
560 | ||
6aa8b732 AK |
561 | /* |
562 | * Free any memory in @free but not in @dont. | |
563 | */ | |
5587027c | 564 | static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free, |
6aa8b732 AK |
565 | struct kvm_memory_slot *dont) |
566 | { | |
6aa8b732 | 567 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
a36a57b1 | 568 | kvm_destroy_dirty_bitmap(free); |
6aa8b732 | 569 | |
5587027c | 570 | kvm_arch_free_memslot(kvm, free, dont); |
05da4558 | 571 | |
6aa8b732 | 572 | free->npages = 0; |
6aa8b732 AK |
573 | } |
574 | ||
7940876e | 575 | static void kvm_free_physmem(struct kvm *kvm) |
6aa8b732 | 576 | { |
46a26bf5 | 577 | struct kvm_memslots *slots = kvm->memslots; |
be6ba0f0 | 578 | struct kvm_memory_slot *memslot; |
46a26bf5 | 579 | |
be6ba0f0 | 580 | kvm_for_each_memslot(memslot, slots) |
5587027c | 581 | kvm_free_physmem_slot(kvm, memslot, NULL); |
6aa8b732 | 582 | |
46a26bf5 | 583 | kfree(kvm->memslots); |
6aa8b732 AK |
584 | } |
585 | ||
07f0a7bd SW |
586 | static void kvm_destroy_devices(struct kvm *kvm) |
587 | { | |
588 | struct list_head *node, *tmp; | |
589 | ||
590 | list_for_each_safe(node, tmp, &kvm->devices) { | |
591 | struct kvm_device *dev = | |
592 | list_entry(node, struct kvm_device, vm_node); | |
593 | ||
594 | list_del(node); | |
595 | dev->ops->destroy(dev); | |
596 | } | |
597 | } | |
598 | ||
f17abe9a AK |
599 | static void kvm_destroy_vm(struct kvm *kvm) |
600 | { | |
e93f8a0f | 601 | int i; |
6d4e4c4f AK |
602 | struct mm_struct *mm = kvm->mm; |
603 | ||
ad8ba2cd | 604 | kvm_arch_sync_events(kvm); |
2f303b74 | 605 | spin_lock(&kvm_lock); |
133de902 | 606 | list_del(&kvm->vm_list); |
2f303b74 | 607 | spin_unlock(&kvm_lock); |
399ec807 | 608 | kvm_free_irq_routing(kvm); |
e93f8a0f MT |
609 | for (i = 0; i < KVM_NR_BUSES; i++) |
610 | kvm_io_bus_destroy(kvm->buses[i]); | |
980da6ce | 611 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
612 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
613 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 614 | #else |
2df72e9b | 615 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 616 | #endif |
d19a9cd2 | 617 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 618 | kvm_destroy_devices(kvm); |
d89f5eff | 619 | kvm_free_physmem(kvm); |
820b3fcd | 620 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
621 | cleanup_srcu_struct(&kvm->srcu); |
622 | kvm_arch_free_vm(kvm); | |
10474ae8 | 623 | hardware_disable_all(); |
6d4e4c4f | 624 | mmdrop(mm); |
f17abe9a AK |
625 | } |
626 | ||
d39f13b0 IE |
627 | void kvm_get_kvm(struct kvm *kvm) |
628 | { | |
629 | atomic_inc(&kvm->users_count); | |
630 | } | |
631 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
632 | ||
633 | void kvm_put_kvm(struct kvm *kvm) | |
634 | { | |
635 | if (atomic_dec_and_test(&kvm->users_count)) | |
636 | kvm_destroy_vm(kvm); | |
637 | } | |
638 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
639 | ||
640 | ||
f17abe9a AK |
641 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
642 | { | |
643 | struct kvm *kvm = filp->private_data; | |
644 | ||
721eecbf GH |
645 | kvm_irqfd_release(kvm); |
646 | ||
d39f13b0 | 647 | kvm_put_kvm(kvm); |
6aa8b732 AK |
648 | return 0; |
649 | } | |
650 | ||
515a0127 TY |
651 | /* |
652 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 653 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 654 | */ |
a36a57b1 TY |
655 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
656 | { | |
515a0127 | 657 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 658 | |
92eca8fa | 659 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
660 | if (!memslot->dirty_bitmap) |
661 | return -ENOMEM; | |
662 | ||
a36a57b1 TY |
663 | return 0; |
664 | } | |
665 | ||
bf3e05bc XG |
666 | static int cmp_memslot(const void *slot1, const void *slot2) |
667 | { | |
668 | struct kvm_memory_slot *s1, *s2; | |
669 | ||
670 | s1 = (struct kvm_memory_slot *)slot1; | |
671 | s2 = (struct kvm_memory_slot *)slot2; | |
672 | ||
673 | if (s1->npages < s2->npages) | |
674 | return 1; | |
675 | if (s1->npages > s2->npages) | |
676 | return -1; | |
677 | ||
678 | return 0; | |
679 | } | |
680 | ||
681 | /* | |
682 | * Sort the memslots base on its size, so the larger slots | |
683 | * will get better fit. | |
684 | */ | |
685 | static void sort_memslots(struct kvm_memslots *slots) | |
686 | { | |
f85e2cb5 XG |
687 | int i; |
688 | ||
bf3e05bc XG |
689 | sort(slots->memslots, KVM_MEM_SLOTS_NUM, |
690 | sizeof(struct kvm_memory_slot), cmp_memslot, NULL); | |
f85e2cb5 XG |
691 | |
692 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) | |
693 | slots->id_to_index[slots->memslots[i].id] = i; | |
bf3e05bc XG |
694 | } |
695 | ||
7940876e | 696 | static void update_memslots(struct kvm_memslots *slots, |
ee3d1570 | 697 | struct kvm_memory_slot *new) |
be593d62 XG |
698 | { |
699 | if (new) { | |
700 | int id = new->id; | |
28a37544 | 701 | struct kvm_memory_slot *old = id_to_memslot(slots, id); |
bf3e05bc | 702 | unsigned long npages = old->npages; |
be593d62 | 703 | |
28a37544 | 704 | *old = *new; |
bf3e05bc XG |
705 | if (new->npages != npages) |
706 | sort_memslots(slots); | |
be593d62 | 707 | } |
be593d62 XG |
708 | } |
709 | ||
a50d64d6 XG |
710 | static int check_memory_region_flags(struct kvm_userspace_memory_region *mem) |
711 | { | |
4d8b81ab XG |
712 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
713 | ||
0f8a4de3 | 714 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
715 | valid_flags |= KVM_MEM_READONLY; |
716 | #endif | |
717 | ||
718 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
719 | return -EINVAL; |
720 | ||
721 | return 0; | |
722 | } | |
723 | ||
7ec4fb44 GN |
724 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
725 | struct kvm_memslots *slots, struct kvm_memory_slot *new) | |
726 | { | |
727 | struct kvm_memslots *old_memslots = kvm->memslots; | |
728 | ||
ee3d1570 DM |
729 | /* |
730 | * Set the low bit in the generation, which disables SPTE caching | |
731 | * until the end of synchronize_srcu_expedited. | |
732 | */ | |
733 | WARN_ON(old_memslots->generation & 1); | |
734 | slots->generation = old_memslots->generation + 1; | |
735 | ||
736 | update_memslots(slots, new); | |
7ec4fb44 GN |
737 | rcu_assign_pointer(kvm->memslots, slots); |
738 | synchronize_srcu_expedited(&kvm->srcu); | |
e59dbe09 | 739 | |
ee3d1570 DM |
740 | /* |
741 | * Increment the new memslot generation a second time. This prevents | |
742 | * vm exits that race with memslot updates from caching a memslot | |
743 | * generation that will (potentially) be valid forever. | |
744 | */ | |
745 | slots->generation++; | |
746 | ||
e59dbe09 TY |
747 | kvm_arch_memslots_updated(kvm); |
748 | ||
749 | return old_memslots; | |
7ec4fb44 GN |
750 | } |
751 | ||
6aa8b732 AK |
752 | /* |
753 | * Allocate some memory and give it an address in the guest physical address | |
754 | * space. | |
755 | * | |
756 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 757 | * |
10589a46 | 758 | * Must be called holding mmap_sem for write. |
6aa8b732 | 759 | */ |
f78e0e2e | 760 | int __kvm_set_memory_region(struct kvm *kvm, |
47ae31e2 | 761 | struct kvm_userspace_memory_region *mem) |
6aa8b732 | 762 | { |
8234b22e | 763 | int r; |
6aa8b732 | 764 | gfn_t base_gfn; |
28bcb112 | 765 | unsigned long npages; |
a843fac2 | 766 | struct kvm_memory_slot *slot; |
6aa8b732 | 767 | struct kvm_memory_slot old, new; |
b7f69c55 | 768 | struct kvm_memslots *slots = NULL, *old_memslots; |
f64c0398 | 769 | enum kvm_mr_change change; |
6aa8b732 | 770 | |
a50d64d6 XG |
771 | r = check_memory_region_flags(mem); |
772 | if (r) | |
773 | goto out; | |
774 | ||
6aa8b732 AK |
775 | r = -EINVAL; |
776 | /* General sanity checks */ | |
777 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
778 | goto out; | |
779 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
780 | goto out; | |
fa3d315a | 781 | /* We can read the guest memory with __xxx_user() later on. */ |
47ae31e2 | 782 | if ((mem->slot < KVM_USER_MEM_SLOTS) && |
fa3d315a | 783 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
784 | !access_ok(VERIFY_WRITE, |
785 | (void __user *)(unsigned long)mem->userspace_addr, | |
786 | mem->memory_size))) | |
78749809 | 787 | goto out; |
93a5cef0 | 788 | if (mem->slot >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
789 | goto out; |
790 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
791 | goto out; | |
792 | ||
a843fac2 | 793 | slot = id_to_memslot(kvm->memslots, mem->slot); |
6aa8b732 AK |
794 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
795 | npages = mem->memory_size >> PAGE_SHIFT; | |
796 | ||
660c22c4 TY |
797 | if (npages > KVM_MEM_MAX_NR_PAGES) |
798 | goto out; | |
799 | ||
6aa8b732 AK |
800 | if (!npages) |
801 | mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
802 | ||
a843fac2 | 803 | new = old = *slot; |
6aa8b732 | 804 | |
e36d96f7 | 805 | new.id = mem->slot; |
6aa8b732 AK |
806 | new.base_gfn = base_gfn; |
807 | new.npages = npages; | |
808 | new.flags = mem->flags; | |
809 | ||
f64c0398 TY |
810 | if (npages) { |
811 | if (!old.npages) | |
812 | change = KVM_MR_CREATE; | |
813 | else { /* Modify an existing slot. */ | |
814 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
815 | (npages != old.npages) || |
816 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
817 | goto out; |
818 | ||
819 | if (base_gfn != old.base_gfn) | |
820 | change = KVM_MR_MOVE; | |
821 | else if (new.flags != old.flags) | |
822 | change = KVM_MR_FLAGS_ONLY; | |
823 | else { /* Nothing to change. */ | |
824 | r = 0; | |
825 | goto out; | |
826 | } | |
827 | } | |
828 | } else if (old.npages) { | |
829 | change = KVM_MR_DELETE; | |
830 | } else /* Modify a non-existent slot: disallowed. */ | |
0ea75e1d | 831 | goto out; |
6aa8b732 | 832 | |
f64c0398 | 833 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
834 | /* Check for overlaps */ |
835 | r = -EEXIST; | |
836 | kvm_for_each_memslot(slot, kvm->memslots) { | |
a843fac2 TY |
837 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
838 | (slot->id == mem->slot)) | |
0a706bee TY |
839 | continue; |
840 | if (!((base_gfn + npages <= slot->base_gfn) || | |
841 | (base_gfn >= slot->base_gfn + slot->npages))) | |
842 | goto out; | |
843 | } | |
6aa8b732 | 844 | } |
6aa8b732 | 845 | |
6aa8b732 AK |
846 | /* Free page dirty bitmap if unneeded */ |
847 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 848 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
849 | |
850 | r = -ENOMEM; | |
f64c0398 | 851 | if (change == KVM_MR_CREATE) { |
189a2f7b | 852 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 853 | |
5587027c | 854 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 855 | goto out_free; |
6aa8b732 | 856 | } |
ec04b260 | 857 | |
6aa8b732 AK |
858 | /* Allocate page dirty bitmap if needed */ |
859 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 860 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 861 | goto out_free; |
6aa8b732 AK |
862 | } |
863 | ||
f64c0398 | 864 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
6da64fdb TM |
865 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), |
866 | GFP_KERNEL); | |
bc6678a3 MT |
867 | if (!slots) |
868 | goto out_free; | |
28a37544 XG |
869 | slot = id_to_memslot(slots, mem->slot); |
870 | slot->flags |= KVM_MEMSLOT_INVALID; | |
871 | ||
7ec4fb44 | 872 | old_memslots = install_new_memslots(kvm, slots, NULL); |
bc6678a3 | 873 | |
e40f193f AW |
874 | /* slot was deleted or moved, clear iommu mapping */ |
875 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
876 | /* From this point no new shadow pages pointing to a deleted, |
877 | * or moved, memslot will be created. | |
bc6678a3 MT |
878 | * |
879 | * validation of sp->gfn happens in: | |
880 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
881 | * - kvm_is_visible_gfn (mmu_check_roots) | |
882 | */ | |
2df72e9b | 883 | kvm_arch_flush_shadow_memslot(kvm, slot); |
b7f69c55 | 884 | slots = old_memslots; |
bc6678a3 | 885 | } |
34d4cb8f | 886 | |
7b6195a9 | 887 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 888 | if (r) |
b7f69c55 | 889 | goto out_slots; |
f7784b8e | 890 | |
bc6678a3 | 891 | r = -ENOMEM; |
b7f69c55 AW |
892 | /* |
893 | * We can re-use the old_memslots from above, the only difference | |
894 | * from the currently installed memslots is the invalid flag. This | |
895 | * will get overwritten by update_memslots anyway. | |
896 | */ | |
897 | if (!slots) { | |
898 | slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots), | |
899 | GFP_KERNEL); | |
900 | if (!slots) | |
901 | goto out_free; | |
902 | } | |
bc6678a3 MT |
903 | |
904 | /* actual memory is freed via old in kvm_free_physmem_slot below */ | |
f64c0398 | 905 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 906 | new.dirty_bitmap = NULL; |
db3fe4eb | 907 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
908 | } |
909 | ||
7ec4fb44 | 910 | old_memslots = install_new_memslots(kvm, slots, &new); |
3ad82a7e | 911 | |
8482644a | 912 | kvm_arch_commit_memory_region(kvm, mem, &old, change); |
82ce2c96 | 913 | |
5587027c | 914 | kvm_free_physmem_slot(kvm, &old, &new); |
bc6678a3 MT |
915 | kfree(old_memslots); |
916 | ||
261874b0 AW |
917 | /* |
918 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
919 | * un-mapped and re-mapped if their base changes. Since base change |
920 | * unmapping is handled above with slot deletion, mapping alone is | |
921 | * needed here. Anything else the iommu might care about for existing | |
922 | * slots (size changes, userspace addr changes and read-only flag | |
923 | * changes) is disallowed above, so any other attribute changes getting | |
924 | * here can be skipped. | |
261874b0 | 925 | */ |
75d61fbc TY |
926 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
927 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 928 | return r; |
bc6678a3 MT |
929 | } |
930 | ||
6aa8b732 AK |
931 | return 0; |
932 | ||
e40f193f AW |
933 | out_slots: |
934 | kfree(slots); | |
f78e0e2e | 935 | out_free: |
5587027c | 936 | kvm_free_physmem_slot(kvm, &new, &old); |
6aa8b732 AK |
937 | out: |
938 | return r; | |
210c7c4d | 939 | } |
f78e0e2e SY |
940 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
941 | ||
942 | int kvm_set_memory_region(struct kvm *kvm, | |
47ae31e2 | 943 | struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
944 | { |
945 | int r; | |
946 | ||
79fac95e | 947 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 948 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 949 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
950 | return r; |
951 | } | |
210c7c4d IE |
952 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
953 | ||
7940876e SH |
954 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
955 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 956 | { |
bbacc0c1 | 957 | if (mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 958 | return -EINVAL; |
47ae31e2 | 959 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
960 | } |
961 | ||
5bb064dc ZX |
962 | int kvm_get_dirty_log(struct kvm *kvm, |
963 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 AK |
964 | { |
965 | struct kvm_memory_slot *memslot; | |
966 | int r, i; | |
87bf6e7d | 967 | unsigned long n; |
6aa8b732 AK |
968 | unsigned long any = 0; |
969 | ||
6aa8b732 | 970 | r = -EINVAL; |
bbacc0c1 | 971 | if (log->slot >= KVM_USER_MEM_SLOTS) |
6aa8b732 AK |
972 | goto out; |
973 | ||
28a37544 | 974 | memslot = id_to_memslot(kvm->memslots, log->slot); |
6aa8b732 AK |
975 | r = -ENOENT; |
976 | if (!memslot->dirty_bitmap) | |
977 | goto out; | |
978 | ||
87bf6e7d | 979 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 980 | |
cd1a4a98 | 981 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
982 | any = memslot->dirty_bitmap[i]; |
983 | ||
984 | r = -EFAULT; | |
985 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
986 | goto out; | |
987 | ||
5bb064dc ZX |
988 | if (any) |
989 | *is_dirty = 1; | |
6aa8b732 AK |
990 | |
991 | r = 0; | |
6aa8b732 | 992 | out: |
6aa8b732 AK |
993 | return r; |
994 | } | |
2ba9f0d8 | 995 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 996 | |
db3fe4eb TY |
997 | bool kvm_largepages_enabled(void) |
998 | { | |
999 | return largepages_enabled; | |
1000 | } | |
1001 | ||
54dee993 MT |
1002 | void kvm_disable_largepages(void) |
1003 | { | |
1004 | largepages_enabled = false; | |
1005 | } | |
1006 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1007 | ||
49c7754c GN |
1008 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1009 | { | |
1010 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1011 | } | |
a1f4d395 | 1012 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1013 | |
e0d62c7f IE |
1014 | int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
1015 | { | |
bf3e05bc | 1016 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1017 | |
bbacc0c1 | 1018 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc XG |
1019 | memslot->flags & KVM_MEMSLOT_INVALID) |
1020 | return 0; | |
e0d62c7f | 1021 | |
bf3e05bc | 1022 | return 1; |
e0d62c7f IE |
1023 | } |
1024 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1025 | ||
8f0b1ab6 JR |
1026 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1027 | { | |
1028 | struct vm_area_struct *vma; | |
1029 | unsigned long addr, size; | |
1030 | ||
1031 | size = PAGE_SIZE; | |
1032 | ||
1033 | addr = gfn_to_hva(kvm, gfn); | |
1034 | if (kvm_is_error_hva(addr)) | |
1035 | return PAGE_SIZE; | |
1036 | ||
1037 | down_read(¤t->mm->mmap_sem); | |
1038 | vma = find_vma(current->mm, addr); | |
1039 | if (!vma) | |
1040 | goto out; | |
1041 | ||
1042 | size = vma_kernel_pagesize(vma); | |
1043 | ||
1044 | out: | |
1045 | up_read(¤t->mm->mmap_sem); | |
1046 | ||
1047 | return size; | |
1048 | } | |
1049 | ||
4d8b81ab XG |
1050 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1051 | { | |
1052 | return slot->flags & KVM_MEM_READONLY; | |
1053 | } | |
1054 | ||
4d8b81ab XG |
1055 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1056 | gfn_t *nr_pages, bool write) | |
539cb660 | 1057 | { |
bc6678a3 | 1058 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1059 | return KVM_HVA_ERR_BAD; |
48987781 | 1060 | |
4d8b81ab XG |
1061 | if (memslot_is_readonly(slot) && write) |
1062 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1063 | |
1064 | if (nr_pages) | |
1065 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1066 | ||
4d8b81ab | 1067 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1068 | } |
48987781 | 1069 | |
4d8b81ab XG |
1070 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1071 | gfn_t *nr_pages) | |
1072 | { | |
1073 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1074 | } |
48987781 | 1075 | |
4d8b81ab | 1076 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1077 | gfn_t gfn) |
4d8b81ab XG |
1078 | { |
1079 | return gfn_to_hva_many(slot, gfn, NULL); | |
1080 | } | |
1081 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1082 | ||
48987781 XG |
1083 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1084 | { | |
49c7754c | 1085 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1086 | } |
0d150298 | 1087 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1088 | |
86ab8cff | 1089 | /* |
ba6a3541 PB |
1090 | * If writable is set to false, the hva returned by this function is only |
1091 | * allowed to be read. | |
86ab8cff | 1092 | */ |
ba6a3541 | 1093 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
86ab8cff | 1094 | { |
ba6a3541 | 1095 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
a2ac07fe GN |
1096 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1097 | ||
1098 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1099 | *writable = !memslot_is_readonly(slot); |
1100 | ||
a2ac07fe | 1101 | return hva; |
86ab8cff XG |
1102 | } |
1103 | ||
1104 | static int kvm_read_hva(void *data, void __user *hva, int len) | |
8030089f | 1105 | { |
86ab8cff XG |
1106 | return __copy_from_user(data, hva, len); |
1107 | } | |
1108 | ||
1109 | static int kvm_read_hva_atomic(void *data, void __user *hva, int len) | |
1110 | { | |
1111 | return __copy_from_user_inatomic(data, hva, len); | |
8030089f GN |
1112 | } |
1113 | ||
39369f7a | 1114 | static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm, |
0857b9e9 GN |
1115 | unsigned long start, int write, struct page **page) |
1116 | { | |
1117 | int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET; | |
1118 | ||
1119 | if (write) | |
1120 | flags |= FOLL_WRITE; | |
1121 | ||
1122 | return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL); | |
1123 | } | |
1124 | ||
234b239b ALC |
1125 | int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm, |
1126 | unsigned long addr, bool write_fault, | |
1127 | struct page **pagep) | |
1128 | { | |
1129 | int npages; | |
1130 | int locked = 1; | |
1131 | int flags = FOLL_TOUCH | FOLL_HWPOISON | | |
1132 | (pagep ? FOLL_GET : 0) | | |
1133 | (write_fault ? FOLL_WRITE : 0); | |
1134 | ||
1135 | /* | |
1136 | * If retrying the fault, we get here *not* having allowed the filemap | |
1137 | * to wait on the page lock. We should now allow waiting on the IO with | |
1138 | * the mmap semaphore released. | |
1139 | */ | |
1140 | down_read(&mm->mmap_sem); | |
1141 | npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL, | |
1142 | &locked); | |
1143 | if (!locked) { | |
1144 | VM_BUG_ON(npages != -EBUSY); | |
1145 | ||
1146 | if (!pagep) | |
1147 | return 0; | |
1148 | ||
1149 | /* | |
1150 | * The previous call has now waited on the IO. Now we can | |
1151 | * retry and complete. Pass TRIED to ensure we do not re | |
1152 | * schedule async IO (see e.g. filemap_fault). | |
1153 | */ | |
1154 | down_read(&mm->mmap_sem); | |
1155 | npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED, | |
1156 | pagep, NULL, NULL); | |
1157 | } | |
1158 | up_read(&mm->mmap_sem); | |
1159 | return npages; | |
1160 | } | |
1161 | ||
fafc3dba HY |
1162 | static inline int check_user_page_hwpoison(unsigned long addr) |
1163 | { | |
1164 | int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE; | |
1165 | ||
1166 | rc = __get_user_pages(current, current->mm, addr, 1, | |
1167 | flags, NULL, NULL, NULL); | |
1168 | return rc == -EHWPOISON; | |
1169 | } | |
1170 | ||
2fc84311 XG |
1171 | /* |
1172 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1173 | * true indicates success, otherwise false is returned. | |
1174 | */ | |
1175 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
1176 | bool write_fault, bool *writable, pfn_t *pfn) | |
954bbbc2 | 1177 | { |
8d4e1288 | 1178 | struct page *page[1]; |
2fc84311 | 1179 | int npages; |
954bbbc2 | 1180 | |
2fc84311 XG |
1181 | if (!(async || atomic)) |
1182 | return false; | |
af585b92 | 1183 | |
12ce13fe XG |
1184 | /* |
1185 | * Fast pin a writable pfn only if it is a write fault request | |
1186 | * or the caller allows to map a writable pfn for a read fault | |
1187 | * request. | |
1188 | */ | |
1189 | if (!(write_fault || writable)) | |
1190 | return false; | |
612819c3 | 1191 | |
2fc84311 XG |
1192 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1193 | if (npages == 1) { | |
1194 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1195 | |
2fc84311 XG |
1196 | if (writable) |
1197 | *writable = true; | |
1198 | return true; | |
1199 | } | |
af585b92 | 1200 | |
2fc84311 XG |
1201 | return false; |
1202 | } | |
612819c3 | 1203 | |
2fc84311 XG |
1204 | /* |
1205 | * The slow path to get the pfn of the specified host virtual address, | |
1206 | * 1 indicates success, -errno is returned if error is detected. | |
1207 | */ | |
1208 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
1209 | bool *writable, pfn_t *pfn) | |
1210 | { | |
1211 | struct page *page[1]; | |
1212 | int npages = 0; | |
612819c3 | 1213 | |
2fc84311 XG |
1214 | might_sleep(); |
1215 | ||
1216 | if (writable) | |
1217 | *writable = write_fault; | |
1218 | ||
1219 | if (async) { | |
1220 | down_read(¤t->mm->mmap_sem); | |
1221 | npages = get_user_page_nowait(current, current->mm, | |
1222 | addr, write_fault, page); | |
1223 | up_read(¤t->mm->mmap_sem); | |
234b239b ALC |
1224 | } else { |
1225 | /* | |
1226 | * By now we have tried gup_fast, and possibly async_pf, and we | |
1227 | * are certainly not atomic. Time to retry the gup, allowing | |
1228 | * mmap semaphore to be relinquished in the case of IO. | |
1229 | */ | |
1230 | npages = kvm_get_user_page_io(current, current->mm, addr, | |
1231 | write_fault, page); | |
1232 | } | |
2fc84311 XG |
1233 | if (npages != 1) |
1234 | return npages; | |
1235 | ||
1236 | /* map read fault as writable if possible */ | |
12ce13fe | 1237 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1238 | struct page *wpage[1]; |
1239 | ||
1240 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1241 | if (npages == 1) { | |
1242 | *writable = true; | |
1243 | put_page(page[0]); | |
1244 | page[0] = wpage[0]; | |
612819c3 | 1245 | } |
2fc84311 XG |
1246 | |
1247 | npages = 1; | |
887c08ac | 1248 | } |
2fc84311 XG |
1249 | *pfn = page_to_pfn(page[0]); |
1250 | return npages; | |
1251 | } | |
539cb660 | 1252 | |
4d8b81ab XG |
1253 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1254 | { | |
1255 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1256 | return false; | |
2e2e3738 | 1257 | |
4d8b81ab XG |
1258 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1259 | return false; | |
887c08ac | 1260 | |
4d8b81ab XG |
1261 | return true; |
1262 | } | |
bf998156 | 1263 | |
12ce13fe XG |
1264 | /* |
1265 | * Pin guest page in memory and return its pfn. | |
1266 | * @addr: host virtual address which maps memory to the guest | |
1267 | * @atomic: whether this function can sleep | |
1268 | * @async: whether this function need to wait IO complete if the | |
1269 | * host page is not in the memory | |
1270 | * @write_fault: whether we should get a writable host page | |
1271 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1272 | * | |
1273 | * The function will map a writable host page for these two cases: | |
1274 | * 1): @write_fault = true | |
1275 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1276 | * whether the mapping is writable. | |
1277 | */ | |
2fc84311 XG |
1278 | static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
1279 | bool write_fault, bool *writable) | |
1280 | { | |
1281 | struct vm_area_struct *vma; | |
1282 | pfn_t pfn = 0; | |
1283 | int npages; | |
2e2e3738 | 1284 | |
2fc84311 XG |
1285 | /* we can do it either atomically or asynchronously, not both */ |
1286 | BUG_ON(atomic && async); | |
8d4e1288 | 1287 | |
2fc84311 XG |
1288 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1289 | return pfn; | |
1290 | ||
1291 | if (atomic) | |
1292 | return KVM_PFN_ERR_FAULT; | |
1293 | ||
1294 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1295 | if (npages == 1) | |
1296 | return pfn; | |
8d4e1288 | 1297 | |
2fc84311 XG |
1298 | down_read(¤t->mm->mmap_sem); |
1299 | if (npages == -EHWPOISON || | |
1300 | (!async && check_user_page_hwpoison(addr))) { | |
1301 | pfn = KVM_PFN_ERR_HWPOISON; | |
1302 | goto exit; | |
1303 | } | |
1304 | ||
1305 | vma = find_vma_intersection(current->mm, addr, addr + 1); | |
1306 | ||
1307 | if (vma == NULL) | |
1308 | pfn = KVM_PFN_ERR_FAULT; | |
1309 | else if ((vma->vm_flags & VM_PFNMAP)) { | |
1310 | pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
1311 | vma->vm_pgoff; | |
1312 | BUG_ON(!kvm_is_mmio_pfn(pfn)); | |
1313 | } else { | |
4d8b81ab | 1314 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1315 | *async = true; |
1316 | pfn = KVM_PFN_ERR_FAULT; | |
1317 | } | |
1318 | exit: | |
1319 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1320 | return pfn; |
35149e21 AL |
1321 | } |
1322 | ||
4d8b81ab XG |
1323 | static pfn_t |
1324 | __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic, | |
1325 | bool *async, bool write_fault, bool *writable) | |
887c08ac | 1326 | { |
4d8b81ab XG |
1327 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1328 | ||
1329 | if (addr == KVM_HVA_ERR_RO_BAD) | |
1330 | return KVM_PFN_ERR_RO_FAULT; | |
1331 | ||
1332 | if (kvm_is_error_hva(addr)) | |
81c52c56 | 1333 | return KVM_PFN_NOSLOT; |
4d8b81ab XG |
1334 | |
1335 | /* Do not map writable pfn in the readonly memslot. */ | |
1336 | if (writable && memslot_is_readonly(slot)) { | |
1337 | *writable = false; | |
1338 | writable = NULL; | |
1339 | } | |
1340 | ||
1341 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1342 | writable); | |
887c08ac | 1343 | } |
887c08ac | 1344 | |
612819c3 MT |
1345 | static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async, |
1346 | bool write_fault, bool *writable) | |
506f0d6f | 1347 | { |
4d8b81ab | 1348 | struct kvm_memory_slot *slot; |
506f0d6f | 1349 | |
af585b92 GN |
1350 | if (async) |
1351 | *async = false; | |
1352 | ||
4d8b81ab | 1353 | slot = gfn_to_memslot(kvm, gfn); |
506f0d6f | 1354 | |
4d8b81ab XG |
1355 | return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault, |
1356 | writable); | |
365fb3fd XG |
1357 | } |
1358 | ||
1359 | pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) | |
1360 | { | |
612819c3 | 1361 | return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL); |
365fb3fd XG |
1362 | } |
1363 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1364 | ||
612819c3 MT |
1365 | pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async, |
1366 | bool write_fault, bool *writable) | |
af585b92 | 1367 | { |
612819c3 | 1368 | return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable); |
af585b92 GN |
1369 | } |
1370 | EXPORT_SYMBOL_GPL(gfn_to_pfn_async); | |
1371 | ||
365fb3fd XG |
1372 | pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
1373 | { | |
612819c3 | 1374 | return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL); |
506f0d6f | 1375 | } |
35149e21 AL |
1376 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
1377 | ||
612819c3 MT |
1378 | pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
1379 | bool *writable) | |
1380 | { | |
1381 | return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable); | |
1382 | } | |
1383 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1384 | ||
d5661048 | 1385 | pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1386 | { |
4d8b81ab | 1387 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f MT |
1388 | } |
1389 | ||
037d92dc | 1390 | pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1391 | { |
4d8b81ab | 1392 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1393 | } |
037d92dc | 1394 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1395 | |
48987781 XG |
1396 | int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages, |
1397 | int nr_pages) | |
1398 | { | |
1399 | unsigned long addr; | |
1400 | gfn_t entry; | |
1401 | ||
49c7754c | 1402 | addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry); |
48987781 XG |
1403 | if (kvm_is_error_hva(addr)) |
1404 | return -1; | |
1405 | ||
1406 | if (entry < nr_pages) | |
1407 | return 0; | |
1408 | ||
1409 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1410 | } | |
1411 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1412 | ||
a2766325 XG |
1413 | static struct page *kvm_pfn_to_page(pfn_t pfn) |
1414 | { | |
81c52c56 | 1415 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1416 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1417 | |
cb9aaa30 XG |
1418 | if (kvm_is_mmio_pfn(pfn)) { |
1419 | WARN_ON(1); | |
6cede2e6 | 1420 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1421 | } |
a2766325 XG |
1422 | |
1423 | return pfn_to_page(pfn); | |
1424 | } | |
1425 | ||
35149e21 AL |
1426 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1427 | { | |
2e2e3738 AL |
1428 | pfn_t pfn; |
1429 | ||
1430 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1431 | |
a2766325 | 1432 | return kvm_pfn_to_page(pfn); |
954bbbc2 | 1433 | } |
aab61cc0 | 1434 | |
954bbbc2 AK |
1435 | EXPORT_SYMBOL_GPL(gfn_to_page); |
1436 | ||
b4231d61 IE |
1437 | void kvm_release_page_clean(struct page *page) |
1438 | { | |
32cad84f XG |
1439 | WARN_ON(is_error_page(page)); |
1440 | ||
35149e21 | 1441 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1442 | } |
1443 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1444 | ||
35149e21 AL |
1445 | void kvm_release_pfn_clean(pfn_t pfn) |
1446 | { | |
81c52c56 | 1447 | if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1448 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1449 | } |
1450 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1451 | ||
b4231d61 | 1452 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1453 | { |
a2766325 XG |
1454 | WARN_ON(is_error_page(page)); |
1455 | ||
35149e21 AL |
1456 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1457 | } | |
1458 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1459 | ||
7940876e | 1460 | static void kvm_release_pfn_dirty(pfn_t pfn) |
35149e21 AL |
1461 | { |
1462 | kvm_set_pfn_dirty(pfn); | |
1463 | kvm_release_pfn_clean(pfn); | |
1464 | } | |
35149e21 AL |
1465 | |
1466 | void kvm_set_pfn_dirty(pfn_t pfn) | |
1467 | { | |
c77fb9dc | 1468 | if (!kvm_is_mmio_pfn(pfn)) { |
2e2e3738 AL |
1469 | struct page *page = pfn_to_page(pfn); |
1470 | if (!PageReserved(page)) | |
1471 | SetPageDirty(page); | |
1472 | } | |
8a7ae055 | 1473 | } |
35149e21 AL |
1474 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1475 | ||
1476 | void kvm_set_pfn_accessed(pfn_t pfn) | |
1477 | { | |
c77fb9dc | 1478 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1479 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1480 | } |
1481 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1482 | ||
1483 | void kvm_get_pfn(pfn_t pfn) | |
1484 | { | |
c77fb9dc | 1485 | if (!kvm_is_mmio_pfn(pfn)) |
2e2e3738 | 1486 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1487 | } |
1488 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1489 | |
195aefde IE |
1490 | static int next_segment(unsigned long len, int offset) |
1491 | { | |
1492 | if (len > PAGE_SIZE - offset) | |
1493 | return PAGE_SIZE - offset; | |
1494 | else | |
1495 | return len; | |
1496 | } | |
1497 | ||
1498 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1499 | int len) | |
1500 | { | |
e0506bcb IE |
1501 | int r; |
1502 | unsigned long addr; | |
195aefde | 1503 | |
ba6a3541 | 1504 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
e0506bcb IE |
1505 | if (kvm_is_error_hva(addr)) |
1506 | return -EFAULT; | |
86ab8cff | 1507 | r = kvm_read_hva(data, (void __user *)addr + offset, len); |
e0506bcb | 1508 | if (r) |
195aefde | 1509 | return -EFAULT; |
195aefde IE |
1510 | return 0; |
1511 | } | |
1512 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); | |
1513 | ||
1514 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) | |
1515 | { | |
1516 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1517 | int seg; | |
1518 | int offset = offset_in_page(gpa); | |
1519 | int ret; | |
1520 | ||
1521 | while ((seg = next_segment(len, offset)) != 0) { | |
1522 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1523 | if (ret < 0) | |
1524 | return ret; | |
1525 | offset = 0; | |
1526 | len -= seg; | |
1527 | data += seg; | |
1528 | ++gfn; | |
1529 | } | |
1530 | return 0; | |
1531 | } | |
1532 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1533 | ||
7ec54588 MT |
1534 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1535 | unsigned long len) | |
1536 | { | |
1537 | int r; | |
1538 | unsigned long addr; | |
1539 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1540 | int offset = offset_in_page(gpa); | |
1541 | ||
ba6a3541 | 1542 | addr = gfn_to_hva_prot(kvm, gfn, NULL); |
7ec54588 MT |
1543 | if (kvm_is_error_hva(addr)) |
1544 | return -EFAULT; | |
0aac03f0 | 1545 | pagefault_disable(); |
86ab8cff | 1546 | r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1547 | pagefault_enable(); |
7ec54588 MT |
1548 | if (r) |
1549 | return -EFAULT; | |
1550 | return 0; | |
1551 | } | |
1552 | EXPORT_SYMBOL(kvm_read_guest_atomic); | |
1553 | ||
195aefde IE |
1554 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data, |
1555 | int offset, int len) | |
1556 | { | |
e0506bcb IE |
1557 | int r; |
1558 | unsigned long addr; | |
195aefde | 1559 | |
e0506bcb IE |
1560 | addr = gfn_to_hva(kvm, gfn); |
1561 | if (kvm_is_error_hva(addr)) | |
1562 | return -EFAULT; | |
8b0cedff | 1563 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1564 | if (r) |
195aefde | 1565 | return -EFAULT; |
195aefde IE |
1566 | mark_page_dirty(kvm, gfn); |
1567 | return 0; | |
1568 | } | |
1569 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); | |
1570 | ||
1571 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, | |
1572 | unsigned long len) | |
1573 | { | |
1574 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1575 | int seg; | |
1576 | int offset = offset_in_page(gpa); | |
1577 | int ret; | |
1578 | ||
1579 | while ((seg = next_segment(len, offset)) != 0) { | |
1580 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1581 | if (ret < 0) | |
1582 | return ret; | |
1583 | offset = 0; | |
1584 | len -= seg; | |
1585 | data += seg; | |
1586 | ++gfn; | |
1587 | } | |
1588 | return 0; | |
1589 | } | |
1590 | ||
49c7754c | 1591 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1592 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1593 | { |
1594 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1595 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1596 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1597 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1598 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1599 | gfn_t nr_pages_avail; | |
49c7754c GN |
1600 | |
1601 | ghc->gpa = gpa; | |
1602 | ghc->generation = slots->generation; | |
8f964525 AH |
1603 | ghc->len = len; |
1604 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1605 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail); | |
1606 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) { | |
49c7754c | 1607 | ghc->hva += offset; |
8f964525 AH |
1608 | } else { |
1609 | /* | |
1610 | * If the requested region crosses two memslots, we still | |
1611 | * verify that the entire region is valid here. | |
1612 | */ | |
1613 | while (start_gfn <= end_gfn) { | |
1614 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1615 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1616 | &nr_pages_avail); | |
1617 | if (kvm_is_error_hva(ghc->hva)) | |
1618 | return -EFAULT; | |
1619 | start_gfn += nr_pages_avail; | |
1620 | } | |
1621 | /* Use the slow path for cross page reads and writes. */ | |
1622 | ghc->memslot = NULL; | |
1623 | } | |
49c7754c GN |
1624 | return 0; |
1625 | } | |
1626 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1627 | ||
1628 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
1629 | void *data, unsigned long len) | |
1630 | { | |
1631 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1632 | int r; | |
1633 | ||
8f964525 AH |
1634 | BUG_ON(len > ghc->len); |
1635 | ||
49c7754c | 1636 | if (slots->generation != ghc->generation) |
8f964525 AH |
1637 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1638 | ||
1639 | if (unlikely(!ghc->memslot)) | |
1640 | return kvm_write_guest(kvm, ghc->gpa, data, len); | |
49c7754c GN |
1641 | |
1642 | if (kvm_is_error_hva(ghc->hva)) | |
1643 | return -EFAULT; | |
1644 | ||
8b0cedff | 1645 | r = __copy_to_user((void __user *)ghc->hva, data, len); |
49c7754c GN |
1646 | if (r) |
1647 | return -EFAULT; | |
1648 | mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT); | |
1649 | ||
1650 | return 0; | |
1651 | } | |
1652 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); | |
1653 | ||
e03b644f GN |
1654 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1655 | void *data, unsigned long len) | |
1656 | { | |
1657 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1658 | int r; | |
1659 | ||
8f964525 AH |
1660 | BUG_ON(len > ghc->len); |
1661 | ||
e03b644f | 1662 | if (slots->generation != ghc->generation) |
8f964525 AH |
1663 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1664 | ||
1665 | if (unlikely(!ghc->memslot)) | |
1666 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
1667 | |
1668 | if (kvm_is_error_hva(ghc->hva)) | |
1669 | return -EFAULT; | |
1670 | ||
1671 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
1672 | if (r) | |
1673 | return -EFAULT; | |
1674 | ||
1675 | return 0; | |
1676 | } | |
1677 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
1678 | ||
195aefde IE |
1679 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
1680 | { | |
8a3caa6d HC |
1681 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
1682 | ||
1683 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
1684 | } |
1685 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
1686 | ||
1687 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
1688 | { | |
1689 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1690 | int seg; | |
1691 | int offset = offset_in_page(gpa); | |
1692 | int ret; | |
1693 | ||
1694 | while ((seg = next_segment(len, offset)) != 0) { | |
1695 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); | |
1696 | if (ret < 0) | |
1697 | return ret; | |
1698 | offset = 0; | |
1699 | len -= seg; | |
1700 | ++gfn; | |
1701 | } | |
1702 | return 0; | |
1703 | } | |
1704 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
1705 | ||
7940876e SH |
1706 | static void mark_page_dirty_in_slot(struct kvm *kvm, |
1707 | struct kvm_memory_slot *memslot, | |
1708 | gfn_t gfn) | |
6aa8b732 | 1709 | { |
7e9d619d RR |
1710 | if (memslot && memslot->dirty_bitmap) { |
1711 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 1712 | |
b74ca3b3 | 1713 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
1714 | } |
1715 | } | |
1716 | ||
49c7754c GN |
1717 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
1718 | { | |
1719 | struct kvm_memory_slot *memslot; | |
1720 | ||
1721 | memslot = gfn_to_memslot(kvm, gfn); | |
1722 | mark_page_dirty_in_slot(kvm, memslot, gfn); | |
1723 | } | |
2ba9f0d8 | 1724 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 1725 | |
b6958ce4 ED |
1726 | /* |
1727 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
1728 | */ | |
8776e519 | 1729 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 1730 | { |
e5c239cf MT |
1731 | DEFINE_WAIT(wait); |
1732 | ||
1733 | for (;;) { | |
1734 | prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); | |
1735 | ||
a1b37100 | 1736 | if (kvm_arch_vcpu_runnable(vcpu)) { |
a8eeb04a | 1737 | kvm_make_request(KVM_REQ_UNHALT, vcpu); |
e5c239cf | 1738 | break; |
d7690175 | 1739 | } |
09cec754 GN |
1740 | if (kvm_cpu_has_pending_timer(vcpu)) |
1741 | break; | |
e5c239cf MT |
1742 | if (signal_pending(current)) |
1743 | break; | |
1744 | ||
b6958ce4 | 1745 | schedule(); |
b6958ce4 | 1746 | } |
d3bef15f | 1747 | |
e5c239cf | 1748 | finish_wait(&vcpu->wq, &wait); |
b6958ce4 | 1749 | } |
2ba9f0d8 | 1750 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 1751 | |
8c84780d | 1752 | #ifndef CONFIG_S390 |
b6d33834 CD |
1753 | /* |
1754 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
1755 | */ | |
1756 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
1757 | { | |
1758 | int me; | |
1759 | int cpu = vcpu->cpu; | |
1760 | wait_queue_head_t *wqp; | |
1761 | ||
1762 | wqp = kvm_arch_vcpu_wq(vcpu); | |
1763 | if (waitqueue_active(wqp)) { | |
1764 | wake_up_interruptible(wqp); | |
1765 | ++vcpu->stat.halt_wakeup; | |
1766 | } | |
1767 | ||
1768 | me = get_cpu(); | |
1769 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
1770 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
1771 | smp_send_reschedule(cpu); | |
1772 | put_cpu(); | |
1773 | } | |
a20ed54d | 1774 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 1775 | #endif /* !CONFIG_S390 */ |
b6d33834 | 1776 | |
fa93384f | 1777 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
1778 | { |
1779 | struct pid *pid; | |
1780 | struct task_struct *task = NULL; | |
fa93384f | 1781 | int ret = 0; |
41628d33 KW |
1782 | |
1783 | rcu_read_lock(); | |
1784 | pid = rcu_dereference(target->pid); | |
1785 | if (pid) | |
1786 | task = get_pid_task(target->pid, PIDTYPE_PID); | |
1787 | rcu_read_unlock(); | |
1788 | if (!task) | |
c45c528e | 1789 | return ret; |
41628d33 KW |
1790 | if (task->flags & PF_VCPU) { |
1791 | put_task_struct(task); | |
c45c528e | 1792 | return ret; |
41628d33 | 1793 | } |
c45c528e | 1794 | ret = yield_to(task, 1); |
41628d33 | 1795 | put_task_struct(task); |
c45c528e R |
1796 | |
1797 | return ret; | |
41628d33 KW |
1798 | } |
1799 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
1800 | ||
06e48c51 R |
1801 | /* |
1802 | * Helper that checks whether a VCPU is eligible for directed yield. | |
1803 | * Most eligible candidate to yield is decided by following heuristics: | |
1804 | * | |
1805 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
1806 | * (preempted lock holder), indicated by @in_spin_loop. | |
1807 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
1808 | * | |
1809 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
1810 | * chance last time (mostly it has become eligible now since we have probably | |
1811 | * yielded to lockholder in last iteration. This is done by toggling | |
1812 | * @dy_eligible each time a VCPU checked for eligibility.) | |
1813 | * | |
1814 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
1815 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
1816 | * burning. Giving priority for a potential lock-holder increases lock | |
1817 | * progress. | |
1818 | * | |
1819 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
1820 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
1821 | * and continue with next VCPU and so on. | |
1822 | */ | |
7940876e | 1823 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 1824 | { |
4a55dd72 | 1825 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
1826 | bool eligible; |
1827 | ||
1828 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 1829 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
1830 | |
1831 | if (vcpu->spin_loop.in_spin_loop) | |
1832 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
1833 | ||
1834 | return eligible; | |
4a55dd72 SW |
1835 | #else |
1836 | return true; | |
06e48c51 | 1837 | #endif |
4a55dd72 | 1838 | } |
c45c528e | 1839 | |
217ece61 | 1840 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 1841 | { |
217ece61 RR |
1842 | struct kvm *kvm = me->kvm; |
1843 | struct kvm_vcpu *vcpu; | |
1844 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
1845 | int yielded = 0; | |
c45c528e | 1846 | int try = 3; |
217ece61 RR |
1847 | int pass; |
1848 | int i; | |
d255f4f2 | 1849 | |
4c088493 | 1850 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
1851 | /* |
1852 | * We boost the priority of a VCPU that is runnable but not | |
1853 | * currently running, because it got preempted by something | |
1854 | * else and called schedule in __vcpu_run. Hopefully that | |
1855 | * VCPU is holding the lock that we need and will release it. | |
1856 | * We approximate round-robin by starting at the last boosted VCPU. | |
1857 | */ | |
c45c528e | 1858 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 1859 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 1860 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
1861 | i = last_boosted_vcpu; |
1862 | continue; | |
1863 | } else if (pass && i > last_boosted_vcpu) | |
1864 | break; | |
7bc7ae25 R |
1865 | if (!ACCESS_ONCE(vcpu->preempted)) |
1866 | continue; | |
217ece61 RR |
1867 | if (vcpu == me) |
1868 | continue; | |
98f4a146 | 1869 | if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 1870 | continue; |
06e48c51 R |
1871 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
1872 | continue; | |
c45c528e R |
1873 | |
1874 | yielded = kvm_vcpu_yield_to(vcpu); | |
1875 | if (yielded > 0) { | |
217ece61 | 1876 | kvm->last_boosted_vcpu = i; |
217ece61 | 1877 | break; |
c45c528e R |
1878 | } else if (yielded < 0) { |
1879 | try--; | |
1880 | if (!try) | |
1881 | break; | |
217ece61 | 1882 | } |
217ece61 RR |
1883 | } |
1884 | } | |
4c088493 | 1885 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
1886 | |
1887 | /* Ensure vcpu is not eligible during next spinloop */ | |
1888 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
1889 | } |
1890 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
1891 | ||
e4a533a4 | 1892 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
1893 | { |
1894 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
1895 | struct page *page; |
1896 | ||
e4a533a4 | 1897 | if (vmf->pgoff == 0) |
039576c0 | 1898 | page = virt_to_page(vcpu->run); |
09566765 | 1899 | #ifdef CONFIG_X86 |
e4a533a4 | 1900 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 1901 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
1902 | #endif |
1903 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
1904 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
1905 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 1906 | #endif |
039576c0 | 1907 | else |
5b1c1493 | 1908 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 1909 | get_page(page); |
e4a533a4 | 1910 | vmf->page = page; |
1911 | return 0; | |
9a2bb7f4 AK |
1912 | } |
1913 | ||
f0f37e2f | 1914 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 1915 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
1916 | }; |
1917 | ||
1918 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
1919 | { | |
1920 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
1921 | return 0; | |
1922 | } | |
1923 | ||
bccf2150 AK |
1924 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
1925 | { | |
1926 | struct kvm_vcpu *vcpu = filp->private_data; | |
1927 | ||
66c0b394 | 1928 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
1929 | return 0; |
1930 | } | |
1931 | ||
3d3aab1b | 1932 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
1933 | .release = kvm_vcpu_release, |
1934 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
1dda606c AG |
1935 | #ifdef CONFIG_COMPAT |
1936 | .compat_ioctl = kvm_vcpu_compat_ioctl, | |
1937 | #endif | |
9a2bb7f4 | 1938 | .mmap = kvm_vcpu_mmap, |
6038f373 | 1939 | .llseek = noop_llseek, |
bccf2150 AK |
1940 | }; |
1941 | ||
1942 | /* | |
1943 | * Allocates an inode for the vcpu. | |
1944 | */ | |
1945 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
1946 | { | |
24009b05 | 1947 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
1948 | } |
1949 | ||
c5ea7660 AK |
1950 | /* |
1951 | * Creates some virtual cpus. Good luck creating more than one. | |
1952 | */ | |
73880c80 | 1953 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
1954 | { |
1955 | int r; | |
988a2cae | 1956 | struct kvm_vcpu *vcpu, *v; |
c5ea7660 | 1957 | |
338c7dba AH |
1958 | if (id >= KVM_MAX_VCPUS) |
1959 | return -EINVAL; | |
1960 | ||
73880c80 | 1961 | vcpu = kvm_arch_vcpu_create(kvm, id); |
fb3f0f51 RR |
1962 | if (IS_ERR(vcpu)) |
1963 | return PTR_ERR(vcpu); | |
c5ea7660 | 1964 | |
15ad7146 AK |
1965 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
1966 | ||
26e5215f AK |
1967 | r = kvm_arch_vcpu_setup(vcpu); |
1968 | if (r) | |
d780592b | 1969 | goto vcpu_destroy; |
26e5215f | 1970 | |
11ec2804 | 1971 | mutex_lock(&kvm->lock); |
3e515705 AK |
1972 | if (!kvm_vcpu_compatible(vcpu)) { |
1973 | r = -EINVAL; | |
1974 | goto unlock_vcpu_destroy; | |
1975 | } | |
73880c80 GN |
1976 | if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) { |
1977 | r = -EINVAL; | |
d780592b | 1978 | goto unlock_vcpu_destroy; |
fb3f0f51 | 1979 | } |
73880c80 | 1980 | |
988a2cae GN |
1981 | kvm_for_each_vcpu(r, v, kvm) |
1982 | if (v->vcpu_id == id) { | |
73880c80 | 1983 | r = -EEXIST; |
d780592b | 1984 | goto unlock_vcpu_destroy; |
73880c80 GN |
1985 | } |
1986 | ||
1987 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 1988 | |
fb3f0f51 | 1989 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 1990 | kvm_get_kvm(kvm); |
bccf2150 | 1991 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
1992 | if (r < 0) { |
1993 | kvm_put_kvm(kvm); | |
d780592b | 1994 | goto unlock_vcpu_destroy; |
73880c80 GN |
1995 | } |
1996 | ||
1997 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
1998 | smp_wmb(); | |
1999 | atomic_inc(&kvm->online_vcpus); | |
2000 | ||
73880c80 | 2001 | mutex_unlock(&kvm->lock); |
42897d86 | 2002 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2003 | return r; |
39c3b86e | 2004 | |
d780592b | 2005 | unlock_vcpu_destroy: |
7d8fece6 | 2006 | mutex_unlock(&kvm->lock); |
d780592b | 2007 | vcpu_destroy: |
d40ccc62 | 2008 | kvm_arch_vcpu_destroy(vcpu); |
c5ea7660 AK |
2009 | return r; |
2010 | } | |
2011 | ||
1961d276 AK |
2012 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2013 | { | |
2014 | if (sigset) { | |
2015 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2016 | vcpu->sigset_active = 1; | |
2017 | vcpu->sigset = *sigset; | |
2018 | } else | |
2019 | vcpu->sigset_active = 0; | |
2020 | return 0; | |
2021 | } | |
2022 | ||
bccf2150 AK |
2023 | static long kvm_vcpu_ioctl(struct file *filp, |
2024 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2025 | { |
bccf2150 | 2026 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2027 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2028 | int r; |
fa3795a7 DH |
2029 | struct kvm_fpu *fpu = NULL; |
2030 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2031 | |
6d4e4c4f AK |
2032 | if (vcpu->kvm->mm != current->mm) |
2033 | return -EIO; | |
2122ff5e | 2034 | |
2f4d9b54 | 2035 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2036 | /* |
2037 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2038 | * so vcpu_load() would break it. | |
2039 | */ | |
2040 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT) | |
2041 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); | |
2042 | #endif | |
2043 | ||
2044 | ||
9fc77441 MT |
2045 | r = vcpu_load(vcpu); |
2046 | if (r) | |
2047 | return r; | |
6aa8b732 | 2048 | switch (ioctl) { |
9a2bb7f4 | 2049 | case KVM_RUN: |
f0fe5108 AK |
2050 | r = -EINVAL; |
2051 | if (arg) | |
2052 | goto out; | |
b6c7a5dc | 2053 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2054 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2055 | break; |
6aa8b732 | 2056 | case KVM_GET_REGS: { |
3e4bb3ac | 2057 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2058 | |
3e4bb3ac XZ |
2059 | r = -ENOMEM; |
2060 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2061 | if (!kvm_regs) | |
6aa8b732 | 2062 | goto out; |
3e4bb3ac XZ |
2063 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2064 | if (r) | |
2065 | goto out_free1; | |
6aa8b732 | 2066 | r = -EFAULT; |
3e4bb3ac XZ |
2067 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2068 | goto out_free1; | |
6aa8b732 | 2069 | r = 0; |
3e4bb3ac XZ |
2070 | out_free1: |
2071 | kfree(kvm_regs); | |
6aa8b732 AK |
2072 | break; |
2073 | } | |
2074 | case KVM_SET_REGS: { | |
3e4bb3ac | 2075 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2076 | |
3e4bb3ac | 2077 | r = -ENOMEM; |
ff5c2c03 SL |
2078 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2079 | if (IS_ERR(kvm_regs)) { | |
2080 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2081 | goto out; |
ff5c2c03 | 2082 | } |
3e4bb3ac | 2083 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2084 | kfree(kvm_regs); |
6aa8b732 AK |
2085 | break; |
2086 | } | |
2087 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2088 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2089 | r = -ENOMEM; | |
2090 | if (!kvm_sregs) | |
2091 | goto out; | |
2092 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2093 | if (r) |
2094 | goto out; | |
2095 | r = -EFAULT; | |
fa3795a7 | 2096 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2097 | goto out; |
2098 | r = 0; | |
2099 | break; | |
2100 | } | |
2101 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2102 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2103 | if (IS_ERR(kvm_sregs)) { | |
2104 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2105 | kvm_sregs = NULL; |
6aa8b732 | 2106 | goto out; |
ff5c2c03 | 2107 | } |
fa3795a7 | 2108 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2109 | break; |
2110 | } | |
62d9f0db MT |
2111 | case KVM_GET_MP_STATE: { |
2112 | struct kvm_mp_state mp_state; | |
2113 | ||
2114 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2115 | if (r) | |
2116 | goto out; | |
2117 | r = -EFAULT; | |
2118 | if (copy_to_user(argp, &mp_state, sizeof mp_state)) | |
2119 | goto out; | |
2120 | r = 0; | |
2121 | break; | |
2122 | } | |
2123 | case KVM_SET_MP_STATE: { | |
2124 | struct kvm_mp_state mp_state; | |
2125 | ||
2126 | r = -EFAULT; | |
2127 | if (copy_from_user(&mp_state, argp, sizeof mp_state)) | |
2128 | goto out; | |
2129 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2130 | break; |
2131 | } | |
6aa8b732 AK |
2132 | case KVM_TRANSLATE: { |
2133 | struct kvm_translation tr; | |
2134 | ||
2135 | r = -EFAULT; | |
2f366987 | 2136 | if (copy_from_user(&tr, argp, sizeof tr)) |
6aa8b732 | 2137 | goto out; |
8b006791 | 2138 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2139 | if (r) |
2140 | goto out; | |
2141 | r = -EFAULT; | |
2f366987 | 2142 | if (copy_to_user(argp, &tr, sizeof tr)) |
6aa8b732 AK |
2143 | goto out; |
2144 | r = 0; | |
2145 | break; | |
2146 | } | |
d0bfb940 JK |
2147 | case KVM_SET_GUEST_DEBUG: { |
2148 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2149 | |
2150 | r = -EFAULT; | |
2f366987 | 2151 | if (copy_from_user(&dbg, argp, sizeof dbg)) |
6aa8b732 | 2152 | goto out; |
d0bfb940 | 2153 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2154 | break; |
2155 | } | |
1961d276 AK |
2156 | case KVM_SET_SIGNAL_MASK: { |
2157 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2158 | struct kvm_signal_mask kvm_sigmask; | |
2159 | sigset_t sigset, *p; | |
2160 | ||
2161 | p = NULL; | |
2162 | if (argp) { | |
2163 | r = -EFAULT; | |
2164 | if (copy_from_user(&kvm_sigmask, argp, | |
2165 | sizeof kvm_sigmask)) | |
2166 | goto out; | |
2167 | r = -EINVAL; | |
2168 | if (kvm_sigmask.len != sizeof sigset) | |
2169 | goto out; | |
2170 | r = -EFAULT; | |
2171 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
2172 | sizeof sigset)) | |
2173 | goto out; | |
2174 | p = &sigset; | |
2175 | } | |
376d41ff | 2176 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2177 | break; |
2178 | } | |
b8836737 | 2179 | case KVM_GET_FPU: { |
fa3795a7 DH |
2180 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2181 | r = -ENOMEM; | |
2182 | if (!fpu) | |
2183 | goto out; | |
2184 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2185 | if (r) |
2186 | goto out; | |
2187 | r = -EFAULT; | |
fa3795a7 | 2188 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2189 | goto out; |
2190 | r = 0; | |
2191 | break; | |
2192 | } | |
2193 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2194 | fpu = memdup_user(argp, sizeof(*fpu)); |
2195 | if (IS_ERR(fpu)) { | |
2196 | r = PTR_ERR(fpu); | |
18595411 | 2197 | fpu = NULL; |
b8836737 | 2198 | goto out; |
ff5c2c03 | 2199 | } |
fa3795a7 | 2200 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2201 | break; |
2202 | } | |
bccf2150 | 2203 | default: |
313a3dc7 | 2204 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2205 | } |
2206 | out: | |
2122ff5e | 2207 | vcpu_put(vcpu); |
fa3795a7 DH |
2208 | kfree(fpu); |
2209 | kfree(kvm_sregs); | |
bccf2150 AK |
2210 | return r; |
2211 | } | |
2212 | ||
1dda606c AG |
2213 | #ifdef CONFIG_COMPAT |
2214 | static long kvm_vcpu_compat_ioctl(struct file *filp, | |
2215 | unsigned int ioctl, unsigned long arg) | |
2216 | { | |
2217 | struct kvm_vcpu *vcpu = filp->private_data; | |
2218 | void __user *argp = compat_ptr(arg); | |
2219 | int r; | |
2220 | ||
2221 | if (vcpu->kvm->mm != current->mm) | |
2222 | return -EIO; | |
2223 | ||
2224 | switch (ioctl) { | |
2225 | case KVM_SET_SIGNAL_MASK: { | |
2226 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2227 | struct kvm_signal_mask kvm_sigmask; | |
2228 | compat_sigset_t csigset; | |
2229 | sigset_t sigset; | |
2230 | ||
2231 | if (argp) { | |
2232 | r = -EFAULT; | |
2233 | if (copy_from_user(&kvm_sigmask, argp, | |
2234 | sizeof kvm_sigmask)) | |
2235 | goto out; | |
2236 | r = -EINVAL; | |
2237 | if (kvm_sigmask.len != sizeof csigset) | |
2238 | goto out; | |
2239 | r = -EFAULT; | |
2240 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
2241 | sizeof csigset)) | |
2242 | goto out; | |
760a9a30 AC |
2243 | sigset_from_compat(&sigset, &csigset); |
2244 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2245 | } else | |
2246 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2247 | break; |
2248 | } | |
2249 | default: | |
2250 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2251 | } | |
2252 | ||
2253 | out: | |
2254 | return r; | |
2255 | } | |
2256 | #endif | |
2257 | ||
852b6d57 SW |
2258 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2259 | int (*accessor)(struct kvm_device *dev, | |
2260 | struct kvm_device_attr *attr), | |
2261 | unsigned long arg) | |
2262 | { | |
2263 | struct kvm_device_attr attr; | |
2264 | ||
2265 | if (!accessor) | |
2266 | return -EPERM; | |
2267 | ||
2268 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2269 | return -EFAULT; | |
2270 | ||
2271 | return accessor(dev, &attr); | |
2272 | } | |
2273 | ||
2274 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2275 | unsigned long arg) | |
2276 | { | |
2277 | struct kvm_device *dev = filp->private_data; | |
2278 | ||
2279 | switch (ioctl) { | |
2280 | case KVM_SET_DEVICE_ATTR: | |
2281 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2282 | case KVM_GET_DEVICE_ATTR: | |
2283 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2284 | case KVM_HAS_DEVICE_ATTR: | |
2285 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2286 | default: | |
2287 | if (dev->ops->ioctl) | |
2288 | return dev->ops->ioctl(dev, ioctl, arg); | |
2289 | ||
2290 | return -ENOTTY; | |
2291 | } | |
2292 | } | |
2293 | ||
852b6d57 SW |
2294 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2295 | { | |
2296 | struct kvm_device *dev = filp->private_data; | |
2297 | struct kvm *kvm = dev->kvm; | |
2298 | ||
852b6d57 SW |
2299 | kvm_put_kvm(kvm); |
2300 | return 0; | |
2301 | } | |
2302 | ||
2303 | static const struct file_operations kvm_device_fops = { | |
2304 | .unlocked_ioctl = kvm_device_ioctl, | |
db6ae615 SW |
2305 | #ifdef CONFIG_COMPAT |
2306 | .compat_ioctl = kvm_device_ioctl, | |
2307 | #endif | |
852b6d57 SW |
2308 | .release = kvm_device_release, |
2309 | }; | |
2310 | ||
2311 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2312 | { | |
2313 | if (filp->f_op != &kvm_device_fops) | |
2314 | return NULL; | |
2315 | ||
2316 | return filp->private_data; | |
2317 | } | |
2318 | ||
d60eacb0 | 2319 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2320 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2321 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2322 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2323 | #endif |
d60eacb0 | 2324 | |
5975a2e0 | 2325 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2326 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2327 | #endif |
d60eacb0 WD |
2328 | }; |
2329 | ||
2330 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2331 | { | |
2332 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2333 | return -ENOSPC; | |
2334 | ||
2335 | if (kvm_device_ops_table[type] != NULL) | |
2336 | return -EEXIST; | |
2337 | ||
2338 | kvm_device_ops_table[type] = ops; | |
2339 | return 0; | |
2340 | } | |
2341 | ||
2342 | static int kvm_ioctl_create_device(struct kvm *kvm, | |
2343 | struct kvm_create_device *cd) | |
2344 | { | |
2345 | struct kvm_device_ops *ops = NULL; | |
2346 | struct kvm_device *dev; | |
2347 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2348 | int ret; | |
2349 | ||
2350 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2351 | return -ENODEV; | |
2352 | ||
2353 | ops = kvm_device_ops_table[cd->type]; | |
2354 | if (ops == NULL) | |
852b6d57 | 2355 | return -ENODEV; |
852b6d57 SW |
2356 | |
2357 | if (test) | |
2358 | return 0; | |
2359 | ||
2360 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2361 | if (!dev) | |
2362 | return -ENOMEM; | |
2363 | ||
2364 | dev->ops = ops; | |
2365 | dev->kvm = kvm; | |
852b6d57 SW |
2366 | |
2367 | ret = ops->create(dev, cd->type); | |
2368 | if (ret < 0) { | |
2369 | kfree(dev); | |
2370 | return ret; | |
2371 | } | |
2372 | ||
24009b05 | 2373 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 SW |
2374 | if (ret < 0) { |
2375 | ops->destroy(dev); | |
2376 | return ret; | |
2377 | } | |
2378 | ||
07f0a7bd | 2379 | list_add(&dev->vm_node, &kvm->devices); |
852b6d57 SW |
2380 | kvm_get_kvm(kvm); |
2381 | cd->fd = ret; | |
2382 | return 0; | |
2383 | } | |
2384 | ||
92b591a4 AG |
2385 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2386 | { | |
2387 | switch (arg) { | |
2388 | case KVM_CAP_USER_MEMORY: | |
2389 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2390 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
2391 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE | |
2392 | case KVM_CAP_SET_BOOT_CPU_ID: | |
2393 | #endif | |
2394 | case KVM_CAP_INTERNAL_ERROR_DATA: | |
2395 | #ifdef CONFIG_HAVE_KVM_MSI | |
2396 | case KVM_CAP_SIGNAL_MSI: | |
2397 | #endif | |
297e2105 | 2398 | #ifdef CONFIG_HAVE_KVM_IRQFD |
92b591a4 AG |
2399 | case KVM_CAP_IRQFD_RESAMPLE: |
2400 | #endif | |
2401 | case KVM_CAP_CHECK_EXTENSION_VM: | |
2402 | return 1; | |
2403 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2404 | case KVM_CAP_IRQ_ROUTING: | |
2405 | return KVM_MAX_IRQ_ROUTES; | |
2406 | #endif | |
2407 | default: | |
2408 | break; | |
2409 | } | |
2410 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2411 | } | |
2412 | ||
bccf2150 AK |
2413 | static long kvm_vm_ioctl(struct file *filp, |
2414 | unsigned int ioctl, unsigned long arg) | |
2415 | { | |
2416 | struct kvm *kvm = filp->private_data; | |
2417 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2418 | int r; |
bccf2150 | 2419 | |
6d4e4c4f AK |
2420 | if (kvm->mm != current->mm) |
2421 | return -EIO; | |
bccf2150 AK |
2422 | switch (ioctl) { |
2423 | case KVM_CREATE_VCPU: | |
2424 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2425 | break; |
6fc138d2 IE |
2426 | case KVM_SET_USER_MEMORY_REGION: { |
2427 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2428 | ||
2429 | r = -EFAULT; | |
2430 | if (copy_from_user(&kvm_userspace_mem, argp, | |
2431 | sizeof kvm_userspace_mem)) | |
2432 | goto out; | |
2433 | ||
47ae31e2 | 2434 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2435 | break; |
2436 | } | |
2437 | case KVM_GET_DIRTY_LOG: { | |
2438 | struct kvm_dirty_log log; | |
2439 | ||
2440 | r = -EFAULT; | |
2f366987 | 2441 | if (copy_from_user(&log, argp, sizeof log)) |
6aa8b732 | 2442 | goto out; |
2c6f5df9 | 2443 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2444 | break; |
2445 | } | |
5f94c174 LV |
2446 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2447 | case KVM_REGISTER_COALESCED_MMIO: { | |
2448 | struct kvm_coalesced_mmio_zone zone; | |
2449 | r = -EFAULT; | |
2450 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2451 | goto out; | |
5f94c174 | 2452 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2453 | break; |
2454 | } | |
2455 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2456 | struct kvm_coalesced_mmio_zone zone; | |
2457 | r = -EFAULT; | |
2458 | if (copy_from_user(&zone, argp, sizeof zone)) | |
2459 | goto out; | |
5f94c174 | 2460 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2461 | break; |
2462 | } | |
2463 | #endif | |
721eecbf GH |
2464 | case KVM_IRQFD: { |
2465 | struct kvm_irqfd data; | |
2466 | ||
2467 | r = -EFAULT; | |
2468 | if (copy_from_user(&data, argp, sizeof data)) | |
2469 | goto out; | |
d4db2935 | 2470 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
2471 | break; |
2472 | } | |
d34e6b17 GH |
2473 | case KVM_IOEVENTFD: { |
2474 | struct kvm_ioeventfd data; | |
2475 | ||
2476 | r = -EFAULT; | |
2477 | if (copy_from_user(&data, argp, sizeof data)) | |
2478 | goto out; | |
2479 | r = kvm_ioeventfd(kvm, &data); | |
2480 | break; | |
2481 | } | |
73880c80 GN |
2482 | #ifdef CONFIG_KVM_APIC_ARCHITECTURE |
2483 | case KVM_SET_BOOT_CPU_ID: | |
2484 | r = 0; | |
894a9c55 | 2485 | mutex_lock(&kvm->lock); |
73880c80 GN |
2486 | if (atomic_read(&kvm->online_vcpus) != 0) |
2487 | r = -EBUSY; | |
2488 | else | |
2489 | kvm->bsp_vcpu_id = arg; | |
894a9c55 | 2490 | mutex_unlock(&kvm->lock); |
73880c80 | 2491 | break; |
07975ad3 JK |
2492 | #endif |
2493 | #ifdef CONFIG_HAVE_KVM_MSI | |
2494 | case KVM_SIGNAL_MSI: { | |
2495 | struct kvm_msi msi; | |
2496 | ||
2497 | r = -EFAULT; | |
2498 | if (copy_from_user(&msi, argp, sizeof msi)) | |
2499 | goto out; | |
2500 | r = kvm_send_userspace_msi(kvm, &msi); | |
2501 | break; | |
2502 | } | |
23d43cf9 CD |
2503 | #endif |
2504 | #ifdef __KVM_HAVE_IRQ_LINE | |
2505 | case KVM_IRQ_LINE_STATUS: | |
2506 | case KVM_IRQ_LINE: { | |
2507 | struct kvm_irq_level irq_event; | |
2508 | ||
2509 | r = -EFAULT; | |
2510 | if (copy_from_user(&irq_event, argp, sizeof irq_event)) | |
2511 | goto out; | |
2512 | ||
aa2fbe6d YZ |
2513 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
2514 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
2515 | if (r) |
2516 | goto out; | |
2517 | ||
2518 | r = -EFAULT; | |
2519 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
2520 | if (copy_to_user(argp, &irq_event, sizeof irq_event)) | |
2521 | goto out; | |
2522 | } | |
2523 | ||
2524 | r = 0; | |
2525 | break; | |
2526 | } | |
73880c80 | 2527 | #endif |
aa8d5944 AG |
2528 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2529 | case KVM_SET_GSI_ROUTING: { | |
2530 | struct kvm_irq_routing routing; | |
2531 | struct kvm_irq_routing __user *urouting; | |
2532 | struct kvm_irq_routing_entry *entries; | |
2533 | ||
2534 | r = -EFAULT; | |
2535 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
2536 | goto out; | |
2537 | r = -EINVAL; | |
2538 | if (routing.nr >= KVM_MAX_IRQ_ROUTES) | |
2539 | goto out; | |
2540 | if (routing.flags) | |
2541 | goto out; | |
2542 | r = -ENOMEM; | |
2543 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
2544 | if (!entries) | |
2545 | goto out; | |
2546 | r = -EFAULT; | |
2547 | urouting = argp; | |
2548 | if (copy_from_user(entries, urouting->entries, | |
2549 | routing.nr * sizeof(*entries))) | |
2550 | goto out_free_irq_routing; | |
2551 | r = kvm_set_irq_routing(kvm, entries, routing.nr, | |
2552 | routing.flags); | |
2553 | out_free_irq_routing: | |
2554 | vfree(entries); | |
2555 | break; | |
2556 | } | |
2557 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
2558 | case KVM_CREATE_DEVICE: { |
2559 | struct kvm_create_device cd; | |
2560 | ||
2561 | r = -EFAULT; | |
2562 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
2563 | goto out; | |
2564 | ||
2565 | r = kvm_ioctl_create_device(kvm, &cd); | |
2566 | if (r) | |
2567 | goto out; | |
2568 | ||
2569 | r = -EFAULT; | |
2570 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
2571 | goto out; | |
2572 | ||
2573 | r = 0; | |
2574 | break; | |
2575 | } | |
92b591a4 AG |
2576 | case KVM_CHECK_EXTENSION: |
2577 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
2578 | break; | |
f17abe9a | 2579 | default: |
1fe779f8 | 2580 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
bfd99ff5 AK |
2581 | if (r == -ENOTTY) |
2582 | r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg); | |
f17abe9a AK |
2583 | } |
2584 | out: | |
2585 | return r; | |
2586 | } | |
2587 | ||
6ff5894c AB |
2588 | #ifdef CONFIG_COMPAT |
2589 | struct compat_kvm_dirty_log { | |
2590 | __u32 slot; | |
2591 | __u32 padding1; | |
2592 | union { | |
2593 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
2594 | __u64 padding2; | |
2595 | }; | |
2596 | }; | |
2597 | ||
2598 | static long kvm_vm_compat_ioctl(struct file *filp, | |
2599 | unsigned int ioctl, unsigned long arg) | |
2600 | { | |
2601 | struct kvm *kvm = filp->private_data; | |
2602 | int r; | |
2603 | ||
2604 | if (kvm->mm != current->mm) | |
2605 | return -EIO; | |
2606 | switch (ioctl) { | |
2607 | case KVM_GET_DIRTY_LOG: { | |
2608 | struct compat_kvm_dirty_log compat_log; | |
2609 | struct kvm_dirty_log log; | |
2610 | ||
2611 | r = -EFAULT; | |
2612 | if (copy_from_user(&compat_log, (void __user *)arg, | |
2613 | sizeof(compat_log))) | |
2614 | goto out; | |
2615 | log.slot = compat_log.slot; | |
2616 | log.padding1 = compat_log.padding1; | |
2617 | log.padding2 = compat_log.padding2; | |
2618 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
2619 | ||
2620 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
2621 | break; |
2622 | } | |
2623 | default: | |
2624 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
2625 | } | |
2626 | ||
2627 | out: | |
2628 | return r; | |
2629 | } | |
2630 | #endif | |
2631 | ||
3d3aab1b | 2632 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
2633 | .release = kvm_vm_release, |
2634 | .unlocked_ioctl = kvm_vm_ioctl, | |
6ff5894c AB |
2635 | #ifdef CONFIG_COMPAT |
2636 | .compat_ioctl = kvm_vm_compat_ioctl, | |
2637 | #endif | |
6038f373 | 2638 | .llseek = noop_llseek, |
f17abe9a AK |
2639 | }; |
2640 | ||
e08b9637 | 2641 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 2642 | { |
aac87636 | 2643 | int r; |
f17abe9a AK |
2644 | struct kvm *kvm; |
2645 | ||
e08b9637 | 2646 | kvm = kvm_create_vm(type); |
d6d28168 AK |
2647 | if (IS_ERR(kvm)) |
2648 | return PTR_ERR(kvm); | |
6ce5a090 TY |
2649 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2650 | r = kvm_coalesced_mmio_init(kvm); | |
2651 | if (r < 0) { | |
2652 | kvm_put_kvm(kvm); | |
2653 | return r; | |
2654 | } | |
2655 | #endif | |
24009b05 | 2656 | r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC); |
aac87636 | 2657 | if (r < 0) |
66c0b394 | 2658 | kvm_put_kvm(kvm); |
f17abe9a | 2659 | |
aac87636 | 2660 | return r; |
f17abe9a AK |
2661 | } |
2662 | ||
2663 | static long kvm_dev_ioctl(struct file *filp, | |
2664 | unsigned int ioctl, unsigned long arg) | |
2665 | { | |
07c45a36 | 2666 | long r = -EINVAL; |
f17abe9a AK |
2667 | |
2668 | switch (ioctl) { | |
2669 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
2670 | if (arg) |
2671 | goto out; | |
f17abe9a AK |
2672 | r = KVM_API_VERSION; |
2673 | break; | |
2674 | case KVM_CREATE_VM: | |
e08b9637 | 2675 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 2676 | break; |
018d00d2 | 2677 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 2678 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 2679 | break; |
07c45a36 | 2680 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
2681 | if (arg) |
2682 | goto out; | |
adb1ff46 AK |
2683 | r = PAGE_SIZE; /* struct kvm_run */ |
2684 | #ifdef CONFIG_X86 | |
2685 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
2686 | #endif |
2687 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2688 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 2689 | #endif |
07c45a36 | 2690 | break; |
d4c9ff2d FEL |
2691 | case KVM_TRACE_ENABLE: |
2692 | case KVM_TRACE_PAUSE: | |
2693 | case KVM_TRACE_DISABLE: | |
2023a29c | 2694 | r = -EOPNOTSUPP; |
d4c9ff2d | 2695 | break; |
6aa8b732 | 2696 | default: |
043405e1 | 2697 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
2698 | } |
2699 | out: | |
2700 | return r; | |
2701 | } | |
2702 | ||
6aa8b732 | 2703 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
2704 | .unlocked_ioctl = kvm_dev_ioctl, |
2705 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 2706 | .llseek = noop_llseek, |
6aa8b732 AK |
2707 | }; |
2708 | ||
2709 | static struct miscdevice kvm_dev = { | |
bbe4432e | 2710 | KVM_MINOR, |
6aa8b732 AK |
2711 | "kvm", |
2712 | &kvm_chardev_ops, | |
2713 | }; | |
2714 | ||
75b7127c | 2715 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
2716 | { |
2717 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 2718 | int r; |
1b6c0168 | 2719 | |
7f59f492 | 2720 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2721 | return; |
10474ae8 | 2722 | |
7f59f492 | 2723 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 2724 | |
13a34e06 | 2725 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
2726 | |
2727 | if (r) { | |
2728 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
2729 | atomic_inc(&hardware_enable_failed); | |
2730 | printk(KERN_INFO "kvm: enabling virtualization on " | |
2731 | "CPU%d failed\n", cpu); | |
2732 | } | |
1b6c0168 AK |
2733 | } |
2734 | ||
4fa92fb2 | 2735 | static void hardware_enable(void) |
75b7127c | 2736 | { |
4a937f96 | 2737 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2738 | if (kvm_usage_count) |
2739 | hardware_enable_nolock(NULL); | |
4a937f96 | 2740 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2741 | } |
2742 | ||
2743 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
2744 | { |
2745 | int cpu = raw_smp_processor_id(); | |
2746 | ||
7f59f492 | 2747 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 2748 | return; |
7f59f492 | 2749 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 2750 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
2751 | } |
2752 | ||
4fa92fb2 | 2753 | static void hardware_disable(void) |
75b7127c | 2754 | { |
4a937f96 | 2755 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
2756 | if (kvm_usage_count) |
2757 | hardware_disable_nolock(NULL); | |
4a937f96 | 2758 | raw_spin_unlock(&kvm_count_lock); |
75b7127c TY |
2759 | } |
2760 | ||
10474ae8 AG |
2761 | static void hardware_disable_all_nolock(void) |
2762 | { | |
2763 | BUG_ON(!kvm_usage_count); | |
2764 | ||
2765 | kvm_usage_count--; | |
2766 | if (!kvm_usage_count) | |
75b7127c | 2767 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
2768 | } |
2769 | ||
2770 | static void hardware_disable_all(void) | |
2771 | { | |
4a937f96 | 2772 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 2773 | hardware_disable_all_nolock(); |
4a937f96 | 2774 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2775 | } |
2776 | ||
2777 | static int hardware_enable_all(void) | |
2778 | { | |
2779 | int r = 0; | |
2780 | ||
4a937f96 | 2781 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
2782 | |
2783 | kvm_usage_count++; | |
2784 | if (kvm_usage_count == 1) { | |
2785 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 2786 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
2787 | |
2788 | if (atomic_read(&hardware_enable_failed)) { | |
2789 | hardware_disable_all_nolock(); | |
2790 | r = -EBUSY; | |
2791 | } | |
2792 | } | |
2793 | ||
4a937f96 | 2794 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
2795 | |
2796 | return r; | |
2797 | } | |
2798 | ||
774c47f1 AK |
2799 | static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, |
2800 | void *v) | |
2801 | { | |
2802 | int cpu = (long)v; | |
2803 | ||
1a6f4d7f | 2804 | val &= ~CPU_TASKS_FROZEN; |
774c47f1 | 2805 | switch (val) { |
cec9ad27 | 2806 | case CPU_DYING: |
6ec8a856 AK |
2807 | printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n", |
2808 | cpu); | |
4fa92fb2 | 2809 | hardware_disable(); |
6ec8a856 | 2810 | break; |
da908f2f | 2811 | case CPU_STARTING: |
43934a38 JK |
2812 | printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n", |
2813 | cpu); | |
4fa92fb2 | 2814 | hardware_enable(); |
774c47f1 AK |
2815 | break; |
2816 | } | |
2817 | return NOTIFY_OK; | |
2818 | } | |
2819 | ||
9a2b85c6 | 2820 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 2821 | void *v) |
9a2b85c6 | 2822 | { |
8e1c1815 SY |
2823 | /* |
2824 | * Some (well, at least mine) BIOSes hang on reboot if | |
2825 | * in vmx root mode. | |
2826 | * | |
2827 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
2828 | */ | |
2829 | printk(KERN_INFO "kvm: exiting hardware virtualization\n"); | |
2830 | kvm_rebooting = true; | |
75b7127c | 2831 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
2832 | return NOTIFY_OK; |
2833 | } | |
2834 | ||
2835 | static struct notifier_block kvm_reboot_notifier = { | |
2836 | .notifier_call = kvm_reboot, | |
2837 | .priority = 0, | |
2838 | }; | |
2839 | ||
e93f8a0f | 2840 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
2841 | { |
2842 | int i; | |
2843 | ||
2844 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 2845 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
2846 | |
2847 | kvm_iodevice_destructor(pos); | |
2848 | } | |
e93f8a0f | 2849 | kfree(bus); |
2eeb2e94 GH |
2850 | } |
2851 | ||
c21fbff1 PB |
2852 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
2853 | const struct kvm_io_range *r2) | |
743eeb0b | 2854 | { |
743eeb0b SL |
2855 | if (r1->addr < r2->addr) |
2856 | return -1; | |
2857 | if (r1->addr + r1->len > r2->addr + r2->len) | |
2858 | return 1; | |
2859 | return 0; | |
2860 | } | |
2861 | ||
a343c9b7 PB |
2862 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
2863 | { | |
c21fbff1 | 2864 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
2865 | } |
2866 | ||
39369f7a | 2867 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
2868 | gpa_t addr, int len) |
2869 | { | |
743eeb0b SL |
2870 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
2871 | .addr = addr, | |
2872 | .len = len, | |
2873 | .dev = dev, | |
2874 | }; | |
2875 | ||
2876 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
2877 | kvm_io_bus_sort_cmp, NULL); | |
2878 | ||
2879 | return 0; | |
2880 | } | |
2881 | ||
39369f7a | 2882 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
2883 | gpa_t addr, int len) |
2884 | { | |
2885 | struct kvm_io_range *range, key; | |
2886 | int off; | |
2887 | ||
2888 | key = (struct kvm_io_range) { | |
2889 | .addr = addr, | |
2890 | .len = len, | |
2891 | }; | |
2892 | ||
2893 | range = bsearch(&key, bus->range, bus->dev_count, | |
2894 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
2895 | if (range == NULL) | |
2896 | return -ENOENT; | |
2897 | ||
2898 | off = range - bus->range; | |
2899 | ||
c21fbff1 | 2900 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
2901 | off--; |
2902 | ||
2903 | return off; | |
2904 | } | |
2905 | ||
126a5af5 CH |
2906 | static int __kvm_io_bus_write(struct kvm_io_bus *bus, |
2907 | struct kvm_io_range *range, const void *val) | |
2908 | { | |
2909 | int idx; | |
2910 | ||
2911 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
2912 | if (idx < 0) | |
2913 | return -EOPNOTSUPP; | |
2914 | ||
2915 | while (idx < bus->dev_count && | |
c21fbff1 | 2916 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
2917 | if (!kvm_iodevice_write(bus->range[idx].dev, range->addr, |
2918 | range->len, val)) | |
2919 | return idx; | |
2920 | idx++; | |
2921 | } | |
2922 | ||
2923 | return -EOPNOTSUPP; | |
2924 | } | |
2925 | ||
bda9020e | 2926 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e93f8a0f | 2927 | int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 2928 | int len, const void *val) |
2eeb2e94 | 2929 | { |
90d83dc3 | 2930 | struct kvm_io_bus *bus; |
743eeb0b | 2931 | struct kvm_io_range range; |
126a5af5 | 2932 | int r; |
743eeb0b SL |
2933 | |
2934 | range = (struct kvm_io_range) { | |
2935 | .addr = addr, | |
2936 | .len = len, | |
2937 | }; | |
90d83dc3 LJ |
2938 | |
2939 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
126a5af5 CH |
2940 | r = __kvm_io_bus_write(bus, &range, val); |
2941 | return r < 0 ? r : 0; | |
2942 | } | |
2943 | ||
2944 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
2945 | int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, | |
2946 | int len, const void *val, long cookie) | |
2947 | { | |
2948 | struct kvm_io_bus *bus; | |
2949 | struct kvm_io_range range; | |
2950 | ||
2951 | range = (struct kvm_io_range) { | |
2952 | .addr = addr, | |
2953 | .len = len, | |
2954 | }; | |
2955 | ||
2956 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
2957 | ||
2958 | /* First try the device referenced by cookie. */ | |
2959 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 2960 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
126a5af5 CH |
2961 | if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len, |
2962 | val)) | |
2963 | return cookie; | |
2964 | ||
2965 | /* | |
2966 | * cookie contained garbage; fall back to search and return the | |
2967 | * correct cookie value. | |
2968 | */ | |
2969 | return __kvm_io_bus_write(bus, &range, val); | |
2970 | } | |
2971 | ||
2972 | static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range, | |
2973 | void *val) | |
2974 | { | |
2975 | int idx; | |
2976 | ||
2977 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
2978 | if (idx < 0) |
2979 | return -EOPNOTSUPP; | |
2980 | ||
2981 | while (idx < bus->dev_count && | |
c21fbff1 | 2982 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
126a5af5 CH |
2983 | if (!kvm_iodevice_read(bus->range[idx].dev, range->addr, |
2984 | range->len, val)) | |
2985 | return idx; | |
743eeb0b SL |
2986 | idx++; |
2987 | } | |
2988 | ||
bda9020e MT |
2989 | return -EOPNOTSUPP; |
2990 | } | |
68c3b4d1 | 2991 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 2992 | |
bda9020e | 2993 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e93f8a0f MT |
2994 | int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
2995 | int len, void *val) | |
bda9020e | 2996 | { |
90d83dc3 | 2997 | struct kvm_io_bus *bus; |
743eeb0b | 2998 | struct kvm_io_range range; |
126a5af5 | 2999 | int r; |
743eeb0b SL |
3000 | |
3001 | range = (struct kvm_io_range) { | |
3002 | .addr = addr, | |
3003 | .len = len, | |
3004 | }; | |
e93f8a0f | 3005 | |
90d83dc3 | 3006 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); |
126a5af5 CH |
3007 | r = __kvm_io_bus_read(bus, &range, val); |
3008 | return r < 0 ? r : 0; | |
3009 | } | |
743eeb0b | 3010 | |
2eeb2e94 | 3011 | |
79fac95e | 3012 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3013 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3014 | int len, struct kvm_io_device *dev) | |
6c474694 | 3015 | { |
e93f8a0f | 3016 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3017 | |
e93f8a0f | 3018 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3019 | /* exclude ioeventfd which is limited by maximum fd */ |
3020 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3021 | return -ENOSPC; |
2eeb2e94 | 3022 | |
a1300716 AK |
3023 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
3024 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
e93f8a0f MT |
3025 | if (!new_bus) |
3026 | return -ENOMEM; | |
a1300716 AK |
3027 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3028 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3029 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3030 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3031 | synchronize_srcu_expedited(&kvm->srcu); | |
3032 | kfree(bus); | |
090b7aff GH |
3033 | |
3034 | return 0; | |
3035 | } | |
3036 | ||
79fac95e | 3037 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3038 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3039 | struct kvm_io_device *dev) | |
090b7aff | 3040 | { |
e93f8a0f MT |
3041 | int i, r; |
3042 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3043 | |
cdfca7b3 | 3044 | bus = kvm->buses[bus_idx]; |
e93f8a0f | 3045 | r = -ENOENT; |
a1300716 AK |
3046 | for (i = 0; i < bus->dev_count; i++) |
3047 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3048 | r = 0; |
090b7aff GH |
3049 | break; |
3050 | } | |
e93f8a0f | 3051 | |
a1300716 | 3052 | if (r) |
e93f8a0f | 3053 | return r; |
a1300716 AK |
3054 | |
3055 | new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) * | |
3056 | sizeof(struct kvm_io_range)), GFP_KERNEL); | |
3057 | if (!new_bus) | |
3058 | return -ENOMEM; | |
3059 | ||
3060 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3061 | new_bus->dev_count--; | |
3062 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3063 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3064 | |
3065 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3066 | synchronize_srcu_expedited(&kvm->srcu); | |
3067 | kfree(bus); | |
3068 | return r; | |
2eeb2e94 GH |
3069 | } |
3070 | ||
774c47f1 AK |
3071 | static struct notifier_block kvm_cpu_notifier = { |
3072 | .notifier_call = kvm_cpu_hotplug, | |
774c47f1 AK |
3073 | }; |
3074 | ||
8b88b099 | 3075 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3076 | { |
3077 | unsigned offset = (long)_offset; | |
ba1389b7 AK |
3078 | struct kvm *kvm; |
3079 | ||
8b88b099 | 3080 | *val = 0; |
2f303b74 | 3081 | spin_lock(&kvm_lock); |
ba1389b7 | 3082 | list_for_each_entry(kvm, &vm_list, vm_list) |
8b88b099 | 3083 | *val += *(u32 *)((void *)kvm + offset); |
2f303b74 | 3084 | spin_unlock(&kvm_lock); |
8b88b099 | 3085 | return 0; |
ba1389b7 AK |
3086 | } |
3087 | ||
3088 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n"); | |
3089 | ||
8b88b099 | 3090 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3091 | { |
3092 | unsigned offset = (long)_offset; | |
1165f5fe AK |
3093 | struct kvm *kvm; |
3094 | struct kvm_vcpu *vcpu; | |
3095 | int i; | |
3096 | ||
8b88b099 | 3097 | *val = 0; |
2f303b74 | 3098 | spin_lock(&kvm_lock); |
1165f5fe | 3099 | list_for_each_entry(kvm, &vm_list, vm_list) |
988a2cae GN |
3100 | kvm_for_each_vcpu(i, vcpu, kvm) |
3101 | *val += *(u32 *)((void *)vcpu + offset); | |
3102 | ||
2f303b74 | 3103 | spin_unlock(&kvm_lock); |
8b88b099 | 3104 | return 0; |
1165f5fe AK |
3105 | } |
3106 | ||
ba1389b7 AK |
3107 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); |
3108 | ||
828c0950 | 3109 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3110 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3111 | [KVM_STAT_VM] = &vm_stat_fops, | |
3112 | }; | |
1165f5fe | 3113 | |
4f69b680 | 3114 | static int kvm_init_debug(void) |
6aa8b732 | 3115 | { |
0c8eb04a | 3116 | int r = -EEXIST; |
6aa8b732 AK |
3117 | struct kvm_stats_debugfs_item *p; |
3118 | ||
76f7c879 | 3119 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3120 | if (kvm_debugfs_dir == NULL) |
3121 | goto out; | |
3122 | ||
3123 | for (p = debugfs_entries; p->name; ++p) { | |
76f7c879 | 3124 | p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir, |
1165f5fe | 3125 | (void *)(long)p->offset, |
ba1389b7 | 3126 | stat_fops[p->kind]); |
4f69b680 H |
3127 | if (p->dentry == NULL) |
3128 | goto out_dir; | |
3129 | } | |
3130 | ||
3131 | return 0; | |
3132 | ||
3133 | out_dir: | |
3134 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3135 | out: | |
3136 | return r; | |
6aa8b732 AK |
3137 | } |
3138 | ||
3139 | static void kvm_exit_debug(void) | |
3140 | { | |
3141 | struct kvm_stats_debugfs_item *p; | |
3142 | ||
3143 | for (p = debugfs_entries; p->name; ++p) | |
3144 | debugfs_remove(p->dentry); | |
76f7c879 | 3145 | debugfs_remove(kvm_debugfs_dir); |
6aa8b732 AK |
3146 | } |
3147 | ||
fb3600cc | 3148 | static int kvm_suspend(void) |
59ae6c6b | 3149 | { |
10474ae8 | 3150 | if (kvm_usage_count) |
75b7127c | 3151 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3152 | return 0; |
3153 | } | |
3154 | ||
fb3600cc | 3155 | static void kvm_resume(void) |
59ae6c6b | 3156 | { |
ca84d1a2 | 3157 | if (kvm_usage_count) { |
4a937f96 | 3158 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3159 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3160 | } |
59ae6c6b AK |
3161 | } |
3162 | ||
fb3600cc | 3163 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3164 | .suspend = kvm_suspend, |
3165 | .resume = kvm_resume, | |
3166 | }; | |
3167 | ||
15ad7146 AK |
3168 | static inline |
3169 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3170 | { | |
3171 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3172 | } | |
3173 | ||
3174 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3175 | { | |
3176 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3a08a8f9 R |
3177 | if (vcpu->preempted) |
3178 | vcpu->preempted = false; | |
15ad7146 | 3179 | |
e790d9ef RK |
3180 | kvm_arch_sched_in(vcpu, cpu); |
3181 | ||
e9b11c17 | 3182 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3183 | } |
3184 | ||
3185 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3186 | struct task_struct *next) | |
3187 | { | |
3188 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3189 | ||
3a08a8f9 R |
3190 | if (current->state == TASK_RUNNING) |
3191 | vcpu->preempted = true; | |
e9b11c17 | 3192 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3193 | } |
3194 | ||
0ee75bea | 3195 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3196 | struct module *module) |
6aa8b732 AK |
3197 | { |
3198 | int r; | |
002c7f7c | 3199 | int cpu; |
6aa8b732 | 3200 | |
f8c16bba ZX |
3201 | r = kvm_arch_init(opaque); |
3202 | if (r) | |
d2308784 | 3203 | goto out_fail; |
cb498ea2 | 3204 | |
7dac16c3 AH |
3205 | /* |
3206 | * kvm_arch_init makes sure there's at most one caller | |
3207 | * for architectures that support multiple implementations, | |
3208 | * like intel and amd on x86. | |
3209 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating | |
3210 | * conflicts in case kvm is already setup for another implementation. | |
3211 | */ | |
3212 | r = kvm_irqfd_init(); | |
3213 | if (r) | |
3214 | goto out_irqfd; | |
3215 | ||
8437a617 | 3216 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3217 | r = -ENOMEM; |
3218 | goto out_free_0; | |
3219 | } | |
3220 | ||
e9b11c17 | 3221 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3222 | if (r < 0) |
7f59f492 | 3223 | goto out_free_0a; |
6aa8b732 | 3224 | |
002c7f7c YS |
3225 | for_each_online_cpu(cpu) { |
3226 | smp_call_function_single(cpu, | |
e9b11c17 | 3227 | kvm_arch_check_processor_compat, |
8691e5a8 | 3228 | &r, 1); |
002c7f7c | 3229 | if (r < 0) |
d2308784 | 3230 | goto out_free_1; |
002c7f7c YS |
3231 | } |
3232 | ||
774c47f1 AK |
3233 | r = register_cpu_notifier(&kvm_cpu_notifier); |
3234 | if (r) | |
d2308784 | 3235 | goto out_free_2; |
6aa8b732 AK |
3236 | register_reboot_notifier(&kvm_reboot_notifier); |
3237 | ||
c16f862d | 3238 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3239 | if (!vcpu_align) |
3240 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3241 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3242 | 0, NULL); |
c16f862d RR |
3243 | if (!kvm_vcpu_cache) { |
3244 | r = -ENOMEM; | |
fb3600cc | 3245 | goto out_free_3; |
c16f862d RR |
3246 | } |
3247 | ||
af585b92 GN |
3248 | r = kvm_async_pf_init(); |
3249 | if (r) | |
3250 | goto out_free; | |
3251 | ||
6aa8b732 | 3252 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3253 | kvm_vm_fops.owner = module; |
3254 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3255 | |
3256 | r = misc_register(&kvm_dev); | |
3257 | if (r) { | |
d77c26fc | 3258 | printk(KERN_ERR "kvm: misc device register failed\n"); |
af585b92 | 3259 | goto out_unreg; |
6aa8b732 AK |
3260 | } |
3261 | ||
fb3600cc RW |
3262 | register_syscore_ops(&kvm_syscore_ops); |
3263 | ||
15ad7146 AK |
3264 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3265 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3266 | ||
4f69b680 H |
3267 | r = kvm_init_debug(); |
3268 | if (r) { | |
3269 | printk(KERN_ERR "kvm: create debugfs files failed\n"); | |
3270 | goto out_undebugfs; | |
3271 | } | |
0ea4ed8e | 3272 | |
3c3c29fd PB |
3273 | r = kvm_vfio_ops_init(); |
3274 | WARN_ON(r); | |
3275 | ||
c7addb90 | 3276 | return 0; |
6aa8b732 | 3277 | |
4f69b680 H |
3278 | out_undebugfs: |
3279 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 3280 | misc_deregister(&kvm_dev); |
af585b92 GN |
3281 | out_unreg: |
3282 | kvm_async_pf_deinit(); | |
6aa8b732 | 3283 | out_free: |
c16f862d | 3284 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 3285 | out_free_3: |
6aa8b732 | 3286 | unregister_reboot_notifier(&kvm_reboot_notifier); |
774c47f1 | 3287 | unregister_cpu_notifier(&kvm_cpu_notifier); |
d2308784 | 3288 | out_free_2: |
d2308784 | 3289 | out_free_1: |
e9b11c17 | 3290 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
3291 | out_free_0a: |
3292 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 3293 | out_free_0: |
a0f155e9 CH |
3294 | kvm_irqfd_exit(); |
3295 | out_irqfd: | |
7dac16c3 AH |
3296 | kvm_arch_exit(); |
3297 | out_fail: | |
6aa8b732 AK |
3298 | return r; |
3299 | } | |
cb498ea2 | 3300 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 3301 | |
cb498ea2 | 3302 | void kvm_exit(void) |
6aa8b732 | 3303 | { |
0ea4ed8e | 3304 | kvm_exit_debug(); |
6aa8b732 | 3305 | misc_deregister(&kvm_dev); |
c16f862d | 3306 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 3307 | kvm_async_pf_deinit(); |
fb3600cc | 3308 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 3309 | unregister_reboot_notifier(&kvm_reboot_notifier); |
59ae6c6b | 3310 | unregister_cpu_notifier(&kvm_cpu_notifier); |
75b7127c | 3311 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 3312 | kvm_arch_hardware_unsetup(); |
f8c16bba | 3313 | kvm_arch_exit(); |
a0f155e9 | 3314 | kvm_irqfd_exit(); |
7f59f492 | 3315 | free_cpumask_var(cpus_hardware_enabled); |
6aa8b732 | 3316 | } |
cb498ea2 | 3317 | EXPORT_SYMBOL_GPL(kvm_exit); |