]>
Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
9 | * |
10 | * Authors: | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * Yaniv Kamay <yaniv@qumranet.com> | |
13 | * | |
14 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
15 | * the COPYING file in the top-level directory. | |
16 | * | |
17 | */ | |
18 | ||
af669ac6 | 19 | #include <kvm/iodev.h> |
6aa8b732 | 20 | |
edf88417 | 21 | #include <linux/kvm_host.h> |
6aa8b732 AK |
22 | #include <linux/kvm.h> |
23 | #include <linux/module.h> | |
24 | #include <linux/errno.h> | |
6aa8b732 | 25 | #include <linux/percpu.h> |
6aa8b732 AK |
26 | #include <linux/mm.h> |
27 | #include <linux/miscdevice.h> | |
28 | #include <linux/vmalloc.h> | |
6aa8b732 | 29 | #include <linux/reboot.h> |
6aa8b732 AK |
30 | #include <linux/debugfs.h> |
31 | #include <linux/highmem.h> | |
32 | #include <linux/file.h> | |
fb3600cc | 33 | #include <linux/syscore_ops.h> |
774c47f1 | 34 | #include <linux/cpu.h> |
174cd4b1 | 35 | #include <linux/sched/signal.h> |
6e84f315 | 36 | #include <linux/sched/mm.h> |
03441a34 | 37 | #include <linux/sched/stat.h> |
d9e368d6 AK |
38 | #include <linux/cpumask.h> |
39 | #include <linux/smp.h> | |
d6d28168 | 40 | #include <linux/anon_inodes.h> |
04d2cc77 | 41 | #include <linux/profile.h> |
7aa81cc0 | 42 | #include <linux/kvm_para.h> |
6fc138d2 | 43 | #include <linux/pagemap.h> |
8d4e1288 | 44 | #include <linux/mman.h> |
35149e21 | 45 | #include <linux/swap.h> |
e56d532f | 46 | #include <linux/bitops.h> |
547de29e | 47 | #include <linux/spinlock.h> |
6ff5894c | 48 | #include <linux/compat.h> |
bc6678a3 | 49 | #include <linux/srcu.h> |
8f0b1ab6 | 50 | #include <linux/hugetlb.h> |
5a0e3ad6 | 51 | #include <linux/slab.h> |
743eeb0b SL |
52 | #include <linux/sort.h> |
53 | #include <linux/bsearch.h> | |
6aa8b732 | 54 | |
e495606d | 55 | #include <asm/processor.h> |
e495606d | 56 | #include <asm/io.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
3e021bf5 | 59 | #include <asm/pgtable.h> |
6aa8b732 | 60 | |
5f94c174 | 61 | #include "coalesced_mmio.h" |
af585b92 | 62 | #include "async_pf.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
229456fc MT |
65 | #define CREATE_TRACE_POINTS |
66 | #include <trace/events/kvm.h> | |
67 | ||
536a6f88 JF |
68 | /* Worst case buffer size needed for holding an integer. */ |
69 | #define ITOA_MAX_LEN 12 | |
70 | ||
6aa8b732 AK |
71 | MODULE_AUTHOR("Qumranet"); |
72 | MODULE_LICENSE("GPL"); | |
73 | ||
920552b2 | 74 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 75 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
f7819512 | 76 | module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR); |
ec76d819 | 77 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 78 | |
aca6ff29 | 79 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 80 | unsigned int halt_poll_ns_grow = 2; |
6b6de68c | 81 | module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR); |
ec76d819 | 82 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 WL |
83 | |
84 | /* Default resets per-vcpu halt_poll_ns . */ | |
ec76d819 | 85 | unsigned int halt_poll_ns_shrink; |
6b6de68c | 86 | module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR); |
ec76d819 | 87 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 88 | |
fa40a821 MT |
89 | /* |
90 | * Ordering of locks: | |
91 | * | |
b7d409de | 92 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
93 | */ |
94 | ||
2f303b74 | 95 | DEFINE_SPINLOCK(kvm_lock); |
4a937f96 | 96 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 97 | LIST_HEAD(vm_list); |
133de902 | 98 | |
7f59f492 | 99 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 100 | static int kvm_usage_count; |
10474ae8 | 101 | static atomic_t hardware_enable_failed; |
1b6c0168 | 102 | |
c16f862d RR |
103 | struct kmem_cache *kvm_vcpu_cache; |
104 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 105 | |
15ad7146 AK |
106 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
107 | ||
76f7c879 | 108 | struct dentry *kvm_debugfs_dir; |
e23a808b | 109 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 110 | |
536a6f88 JF |
111 | static int kvm_debugfs_num_entries; |
112 | static const struct file_operations *stat_fops_per_vm[]; | |
113 | ||
bccf2150 AK |
114 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
115 | unsigned long arg); | |
de8e5d74 | 116 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
117 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
118 | unsigned long arg); | |
119 | #endif | |
10474ae8 AG |
120 | static int hardware_enable_all(void); |
121 | static void hardware_disable_all(void); | |
bccf2150 | 122 | |
e93f8a0f | 123 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 124 | |
ba049e93 | 125 | static void kvm_release_pfn_dirty(kvm_pfn_t pfn); |
bc009e43 | 126 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 127 | |
52480137 | 128 | __visible bool kvm_rebooting; |
b7c4145b | 129 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 130 | |
54dee993 MT |
131 | static bool largepages_enabled = true; |
132 | ||
ba049e93 | 133 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 134 | { |
11feeb49 | 135 | if (pfn_valid(pfn)) |
bf4bea8e | 136 | return PageReserved(pfn_to_page(pfn)); |
cbff90a7 BAY |
137 | |
138 | return true; | |
139 | } | |
140 | ||
bccf2150 AK |
141 | /* |
142 | * Switches to specified vcpu, until a matching vcpu_put() | |
143 | */ | |
9fc77441 | 144 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 145 | { |
15ad7146 AK |
146 | int cpu; |
147 | ||
9fc77441 MT |
148 | if (mutex_lock_killable(&vcpu->mutex)) |
149 | return -EINTR; | |
15ad7146 AK |
150 | cpu = get_cpu(); |
151 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 152 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 153 | put_cpu(); |
9fc77441 | 154 | return 0; |
6aa8b732 | 155 | } |
2f1fe811 | 156 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 157 | |
313a3dc7 | 158 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 159 | { |
15ad7146 | 160 | preempt_disable(); |
313a3dc7 | 161 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
162 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
163 | preempt_enable(); | |
6aa8b732 AK |
164 | mutex_unlock(&vcpu->mutex); |
165 | } | |
2f1fe811 | 166 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 167 | |
d9e368d6 AK |
168 | static void ack_flush(void *_completed) |
169 | { | |
d9e368d6 AK |
170 | } |
171 | ||
445b8236 | 172 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 173 | { |
597a5f55 | 174 | int i, cpu, me; |
6ef7a1bc RR |
175 | cpumask_var_t cpus; |
176 | bool called = true; | |
d9e368d6 | 177 | struct kvm_vcpu *vcpu; |
d9e368d6 | 178 | |
79f55997 | 179 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 180 | |
3cba4130 | 181 | me = get_cpu(); |
988a2cae | 182 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 183 | kvm_make_request(req, vcpu); |
d9e368d6 | 184 | cpu = vcpu->cpu; |
6b7e2d09 | 185 | |
a30a0509 LT |
186 | /* Set ->requests bit before we read ->mode. */ |
187 | smp_mb__after_atomic(); | |
6b7e2d09 XG |
188 | |
189 | if (cpus != NULL && cpu != -1 && cpu != me && | |
190 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 191 | cpumask_set_cpu(cpu, cpus); |
49846896 | 192 | } |
6ef7a1bc RR |
193 | if (unlikely(cpus == NULL)) |
194 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
195 | else if (!cpumask_empty(cpus)) | |
196 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
197 | else | |
198 | called = false; | |
3cba4130 | 199 | put_cpu(); |
6ef7a1bc | 200 | free_cpumask_var(cpus); |
49846896 | 201 | return called; |
d9e368d6 AK |
202 | } |
203 | ||
a6d51016 | 204 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 205 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 206 | { |
4ae3cb3a LT |
207 | /* |
208 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
209 | * kvm_make_all_cpus_request. | |
210 | */ | |
211 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
212 | ||
213 | /* | |
214 | * We want to publish modifications to the page tables before reading | |
215 | * mode. Pairs with a memory barrier in arch-specific code. | |
216 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
217 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
218 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
219 | * | |
220 | * There is already an smp_mb__after_atomic() before | |
221 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
222 | * barrier here. | |
223 | */ | |
445b8236 | 224 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
49846896 | 225 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 226 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 227 | } |
2ba9f0d8 | 228 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 229 | #endif |
2e53d63a | 230 | |
49846896 RR |
231 | void kvm_reload_remote_mmus(struct kvm *kvm) |
232 | { | |
445b8236 | 233 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 234 | } |
2e53d63a | 235 | |
fb3f0f51 RR |
236 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
237 | { | |
238 | struct page *page; | |
239 | int r; | |
240 | ||
241 | mutex_init(&vcpu->mutex); | |
242 | vcpu->cpu = -1; | |
fb3f0f51 RR |
243 | vcpu->kvm = kvm; |
244 | vcpu->vcpu_id = id; | |
34bb10b7 | 245 | vcpu->pid = NULL; |
8577370f | 246 | init_swait_queue_head(&vcpu->wq); |
af585b92 | 247 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 248 | |
bf9f6ac8 FW |
249 | vcpu->pre_pcpu = -1; |
250 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
251 | ||
fb3f0f51 RR |
252 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
253 | if (!page) { | |
254 | r = -ENOMEM; | |
255 | goto fail; | |
256 | } | |
257 | vcpu->run = page_address(page); | |
258 | ||
4c088493 R |
259 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
260 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 261 | vcpu->preempted = false; |
4c088493 | 262 | |
e9b11c17 | 263 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 264 | if (r < 0) |
e9b11c17 | 265 | goto fail_free_run; |
fb3f0f51 RR |
266 | return 0; |
267 | ||
fb3f0f51 RR |
268 | fail_free_run: |
269 | free_page((unsigned long)vcpu->run); | |
270 | fail: | |
76fafa5e | 271 | return r; |
fb3f0f51 RR |
272 | } |
273 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
274 | ||
275 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
276 | { | |
34bb10b7 | 277 | put_pid(vcpu->pid); |
e9b11c17 | 278 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
279 | free_page((unsigned long)vcpu->run); |
280 | } | |
281 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
282 | ||
e930bffe AA |
283 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
284 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
285 | { | |
286 | return container_of(mn, struct kvm, mmu_notifier); | |
287 | } | |
288 | ||
289 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
290 | struct mm_struct *mm, | |
291 | unsigned long address) | |
292 | { | |
293 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 294 | int need_tlb_flush, idx; |
e930bffe AA |
295 | |
296 | /* | |
297 | * When ->invalidate_page runs, the linux pte has been zapped | |
298 | * already but the page is still allocated until | |
299 | * ->invalidate_page returns. So if we increase the sequence | |
300 | * here the kvm page fault will notice if the spte can't be | |
301 | * established because the page is going to be freed. If | |
302 | * instead the kvm page fault establishes the spte before | |
303 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
304 | * before returning. | |
305 | * | |
306 | * The sequence increase only need to be seen at spin_unlock | |
307 | * time, and not at spin_lock time. | |
308 | * | |
309 | * Increasing the sequence after the spin_unlock would be | |
310 | * unsafe because the kvm page fault could then establish the | |
311 | * pte after kvm_unmap_hva returned, without noticing the page | |
312 | * is going to be freed. | |
313 | */ | |
bc6678a3 | 314 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 315 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 316 | |
e930bffe | 317 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 318 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
319 | /* we've to flush the tlb before the pages can be freed */ |
320 | if (need_tlb_flush) | |
321 | kvm_flush_remote_tlbs(kvm); | |
322 | ||
565f3be2 | 323 | spin_unlock(&kvm->mmu_lock); |
fe71557a TC |
324 | |
325 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); | |
326 | ||
565f3be2 | 327 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
328 | } |
329 | ||
3da0dd43 IE |
330 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
331 | struct mm_struct *mm, | |
332 | unsigned long address, | |
333 | pte_t pte) | |
334 | { | |
335 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 336 | int idx; |
3da0dd43 | 337 | |
bc6678a3 | 338 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
339 | spin_lock(&kvm->mmu_lock); |
340 | kvm->mmu_notifier_seq++; | |
341 | kvm_set_spte_hva(kvm, address, pte); | |
342 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 343 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
344 | } |
345 | ||
e930bffe AA |
346 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
347 | struct mm_struct *mm, | |
348 | unsigned long start, | |
349 | unsigned long end) | |
350 | { | |
351 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 352 | int need_tlb_flush = 0, idx; |
e930bffe | 353 | |
bc6678a3 | 354 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
355 | spin_lock(&kvm->mmu_lock); |
356 | /* | |
357 | * The count increase must become visible at unlock time as no | |
358 | * spte can be established without taking the mmu_lock and | |
359 | * count is also read inside the mmu_lock critical section. | |
360 | */ | |
361 | kvm->mmu_notifier_count++; | |
b3ae2096 | 362 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 363 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
364 | /* we've to flush the tlb before the pages can be freed */ |
365 | if (need_tlb_flush) | |
366 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
367 | |
368 | spin_unlock(&kvm->mmu_lock); | |
369 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
370 | } |
371 | ||
372 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
373 | struct mm_struct *mm, | |
374 | unsigned long start, | |
375 | unsigned long end) | |
376 | { | |
377 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
378 | ||
379 | spin_lock(&kvm->mmu_lock); | |
380 | /* | |
381 | * This sequence increase will notify the kvm page fault that | |
382 | * the page that is going to be mapped in the spte could have | |
383 | * been freed. | |
384 | */ | |
385 | kvm->mmu_notifier_seq++; | |
a355aa54 | 386 | smp_wmb(); |
e930bffe AA |
387 | /* |
388 | * The above sequence increase must be visible before the | |
a355aa54 PM |
389 | * below count decrease, which is ensured by the smp_wmb above |
390 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
391 | */ |
392 | kvm->mmu_notifier_count--; | |
393 | spin_unlock(&kvm->mmu_lock); | |
394 | ||
395 | BUG_ON(kvm->mmu_notifier_count < 0); | |
396 | } | |
397 | ||
398 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
399 | struct mm_struct *mm, | |
57128468 ALC |
400 | unsigned long start, |
401 | unsigned long end) | |
e930bffe AA |
402 | { |
403 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 404 | int young, idx; |
e930bffe | 405 | |
bc6678a3 | 406 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 407 | spin_lock(&kvm->mmu_lock); |
e930bffe | 408 | |
57128468 | 409 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
410 | if (young) |
411 | kvm_flush_remote_tlbs(kvm); | |
412 | ||
565f3be2 TY |
413 | spin_unlock(&kvm->mmu_lock); |
414 | srcu_read_unlock(&kvm->srcu, idx); | |
415 | ||
e930bffe AA |
416 | return young; |
417 | } | |
418 | ||
1d7715c6 VD |
419 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
420 | struct mm_struct *mm, | |
421 | unsigned long start, | |
422 | unsigned long end) | |
423 | { | |
424 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
425 | int young, idx; | |
426 | ||
427 | idx = srcu_read_lock(&kvm->srcu); | |
428 | spin_lock(&kvm->mmu_lock); | |
429 | /* | |
430 | * Even though we do not flush TLB, this will still adversely | |
431 | * affect performance on pre-Haswell Intel EPT, where there is | |
432 | * no EPT Access Bit to clear so that we have to tear down EPT | |
433 | * tables instead. If we find this unacceptable, we can always | |
434 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
435 | * do anything on clear_young. | |
436 | * | |
437 | * Also note that currently we never issue secondary TLB flushes | |
438 | * from clear_young, leaving this job up to the regular system | |
439 | * cadence. If we find this inaccurate, we might come up with a | |
440 | * more sophisticated heuristic later. | |
441 | */ | |
442 | young = kvm_age_hva(kvm, start, end); | |
443 | spin_unlock(&kvm->mmu_lock); | |
444 | srcu_read_unlock(&kvm->srcu, idx); | |
445 | ||
446 | return young; | |
447 | } | |
448 | ||
8ee53820 AA |
449 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
450 | struct mm_struct *mm, | |
451 | unsigned long address) | |
452 | { | |
453 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
454 | int young, idx; | |
455 | ||
456 | idx = srcu_read_lock(&kvm->srcu); | |
457 | spin_lock(&kvm->mmu_lock); | |
458 | young = kvm_test_age_hva(kvm, address); | |
459 | spin_unlock(&kvm->mmu_lock); | |
460 | srcu_read_unlock(&kvm->srcu, idx); | |
461 | ||
462 | return young; | |
463 | } | |
464 | ||
85db06e5 MT |
465 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
466 | struct mm_struct *mm) | |
467 | { | |
468 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
469 | int idx; |
470 | ||
471 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 472 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 473 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
474 | } |
475 | ||
e930bffe AA |
476 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
477 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
478 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
479 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
480 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 481 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 482 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 483 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 484 | .release = kvm_mmu_notifier_release, |
e930bffe | 485 | }; |
4c07b0a4 AK |
486 | |
487 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
488 | { | |
489 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
490 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
491 | } | |
492 | ||
493 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
494 | ||
495 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
496 | { | |
497 | return 0; | |
498 | } | |
499 | ||
e930bffe AA |
500 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
501 | ||
a47d2b07 | 502 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
503 | { |
504 | int i; | |
a47d2b07 | 505 | struct kvm_memslots *slots; |
bf3e05bc | 506 | |
a47d2b07 PB |
507 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
508 | if (!slots) | |
509 | return NULL; | |
510 | ||
bf3e05bc | 511 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 512 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
513 | |
514 | return slots; | |
515 | } | |
516 | ||
517 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
518 | { | |
519 | if (!memslot->dirty_bitmap) | |
520 | return; | |
521 | ||
522 | kvfree(memslot->dirty_bitmap); | |
523 | memslot->dirty_bitmap = NULL; | |
524 | } | |
525 | ||
526 | /* | |
527 | * Free any memory in @free but not in @dont. | |
528 | */ | |
529 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
530 | struct kvm_memory_slot *dont) | |
531 | { | |
532 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
533 | kvm_destroy_dirty_bitmap(free); | |
534 | ||
535 | kvm_arch_free_memslot(kvm, free, dont); | |
536 | ||
537 | free->npages = 0; | |
538 | } | |
539 | ||
540 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
541 | { | |
542 | struct kvm_memory_slot *memslot; | |
543 | ||
544 | if (!slots) | |
545 | return; | |
546 | ||
547 | kvm_for_each_memslot(memslot, slots) | |
548 | kvm_free_memslot(kvm, memslot, NULL); | |
549 | ||
550 | kvfree(slots); | |
bf3e05bc XG |
551 | } |
552 | ||
536a6f88 JF |
553 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
554 | { | |
555 | int i; | |
556 | ||
557 | if (!kvm->debugfs_dentry) | |
558 | return; | |
559 | ||
560 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
561 | ||
9d5a1dce LC |
562 | if (kvm->debugfs_stat_data) { |
563 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
564 | kfree(kvm->debugfs_stat_data[i]); | |
565 | kfree(kvm->debugfs_stat_data); | |
566 | } | |
536a6f88 JF |
567 | } |
568 | ||
569 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
570 | { | |
571 | char dir_name[ITOA_MAX_LEN * 2]; | |
572 | struct kvm_stat_data *stat_data; | |
573 | struct kvm_stats_debugfs_item *p; | |
574 | ||
575 | if (!debugfs_initialized()) | |
576 | return 0; | |
577 | ||
578 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
579 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, | |
580 | kvm_debugfs_dir); | |
581 | if (!kvm->debugfs_dentry) | |
582 | return -ENOMEM; | |
583 | ||
584 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
585 | sizeof(*kvm->debugfs_stat_data), | |
586 | GFP_KERNEL); | |
587 | if (!kvm->debugfs_stat_data) | |
588 | return -ENOMEM; | |
589 | ||
590 | for (p = debugfs_entries; p->name; p++) { | |
591 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL); | |
592 | if (!stat_data) | |
593 | return -ENOMEM; | |
594 | ||
595 | stat_data->kvm = kvm; | |
596 | stat_data->offset = p->offset; | |
597 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; | |
ce35ef27 | 598 | if (!debugfs_create_file(p->name, 0644, |
536a6f88 JF |
599 | kvm->debugfs_dentry, |
600 | stat_data, | |
601 | stat_fops_per_vm[p->kind])) | |
602 | return -ENOMEM; | |
603 | } | |
604 | return 0; | |
605 | } | |
606 | ||
e08b9637 | 607 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 608 | { |
d89f5eff JK |
609 | int r, i; |
610 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 611 | |
d89f5eff JK |
612 | if (!kvm) |
613 | return ERR_PTR(-ENOMEM); | |
614 | ||
e9ad4ec8 | 615 | spin_lock_init(&kvm->mmu_lock); |
f1f10076 | 616 | mmgrab(current->mm); |
e9ad4ec8 PB |
617 | kvm->mm = current->mm; |
618 | kvm_eventfd_init(kvm); | |
619 | mutex_init(&kvm->lock); | |
620 | mutex_init(&kvm->irq_lock); | |
621 | mutex_init(&kvm->slots_lock); | |
e3736c3e | 622 | refcount_set(&kvm->users_count, 1); |
e9ad4ec8 PB |
623 | INIT_LIST_HEAD(&kvm->devices); |
624 | ||
e08b9637 | 625 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 626 | if (r) |
719d93cd | 627 | goto out_err_no_disable; |
10474ae8 AG |
628 | |
629 | r = hardware_enable_all(); | |
630 | if (r) | |
719d93cd | 631 | goto out_err_no_disable; |
10474ae8 | 632 | |
c77dcacb | 633 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 634 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 635 | #endif |
6aa8b732 | 636 | |
1e702d9a AW |
637 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
638 | ||
46a26bf5 | 639 | r = -ENOMEM; |
f481b069 | 640 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 PB |
641 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
642 | if (!slots) | |
f481b069 | 643 | goto out_err_no_srcu; |
4bd518f1 PB |
644 | /* |
645 | * Generations must be different for each address space. | |
646 | * Init kvm generation close to the maximum to easily test the | |
647 | * code of handling generation number wrap-around. | |
648 | */ | |
649 | slots->generation = i * 2 - 150; | |
650 | rcu_assign_pointer(kvm->memslots[i], slots); | |
f481b069 | 651 | } |
00f034a1 | 652 | |
bc6678a3 | 653 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
654 | goto out_err_no_srcu; |
655 | if (init_srcu_struct(&kvm->irq_srcu)) | |
656 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
657 | for (i = 0; i < KVM_NR_BUSES; i++) { |
658 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
659 | GFP_KERNEL); | |
57e7fbee | 660 | if (!kvm->buses[i]) |
e93f8a0f | 661 | goto out_err; |
e93f8a0f | 662 | } |
e930bffe | 663 | |
74b5c5bf MW |
664 | r = kvm_init_mmu_notifier(kvm); |
665 | if (r) | |
666 | goto out_err; | |
667 | ||
2f303b74 | 668 | spin_lock(&kvm_lock); |
5e58cfe4 | 669 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 670 | spin_unlock(&kvm_lock); |
d89f5eff | 671 | |
2ecd9d29 PZ |
672 | preempt_notifier_inc(); |
673 | ||
f17abe9a | 674 | return kvm; |
10474ae8 AG |
675 | |
676 | out_err: | |
719d93cd CB |
677 | cleanup_srcu_struct(&kvm->irq_srcu); |
678 | out_err_no_irq_srcu: | |
57e7fbee | 679 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 680 | out_err_no_srcu: |
10474ae8 | 681 | hardware_disable_all(); |
719d93cd | 682 | out_err_no_disable: |
e93f8a0f MT |
683 | for (i = 0; i < KVM_NR_BUSES; i++) |
684 | kfree(kvm->buses[i]); | |
f481b069 PB |
685 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
686 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
d89f5eff | 687 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 688 | mmdrop(current->mm); |
10474ae8 | 689 | return ERR_PTR(r); |
f17abe9a AK |
690 | } |
691 | ||
92eca8fa TY |
692 | /* |
693 | * Avoid using vmalloc for a small buffer. | |
694 | * Should not be used when the size is statically known. | |
695 | */ | |
c1a7b32a | 696 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
697 | { |
698 | if (size > PAGE_SIZE) | |
699 | return vzalloc(size); | |
700 | else | |
701 | return kzalloc(size, GFP_KERNEL); | |
702 | } | |
703 | ||
07f0a7bd SW |
704 | static void kvm_destroy_devices(struct kvm *kvm) |
705 | { | |
e6e3b5a6 | 706 | struct kvm_device *dev, *tmp; |
07f0a7bd | 707 | |
a28ebea2 CD |
708 | /* |
709 | * We do not need to take the kvm->lock here, because nobody else | |
710 | * has a reference to the struct kvm at this point and therefore | |
711 | * cannot access the devices list anyhow. | |
712 | */ | |
e6e3b5a6 GT |
713 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
714 | list_del(&dev->vm_node); | |
07f0a7bd SW |
715 | dev->ops->destroy(dev); |
716 | } | |
717 | } | |
718 | ||
f17abe9a AK |
719 | static void kvm_destroy_vm(struct kvm *kvm) |
720 | { | |
e93f8a0f | 721 | int i; |
6d4e4c4f AK |
722 | struct mm_struct *mm = kvm->mm; |
723 | ||
536a6f88 | 724 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 725 | kvm_arch_sync_events(kvm); |
2f303b74 | 726 | spin_lock(&kvm_lock); |
133de902 | 727 | list_del(&kvm->vm_list); |
2f303b74 | 728 | spin_unlock(&kvm_lock); |
399ec807 | 729 | kvm_free_irq_routing(kvm); |
df630b8c | 730 | for (i = 0; i < KVM_NR_BUSES; i++) { |
e93f8a0f | 731 | kvm_io_bus_destroy(kvm->buses[i]); |
df630b8c PX |
732 | kvm->buses[i] = NULL; |
733 | } | |
980da6ce | 734 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
735 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
736 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 737 | #else |
2df72e9b | 738 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 739 | #endif |
d19a9cd2 | 740 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 741 | kvm_destroy_devices(kvm); |
f481b069 PB |
742 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
743 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
820b3fcd | 744 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
745 | cleanup_srcu_struct(&kvm->srcu); |
746 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 747 | preempt_notifier_dec(); |
10474ae8 | 748 | hardware_disable_all(); |
6d4e4c4f | 749 | mmdrop(mm); |
f17abe9a AK |
750 | } |
751 | ||
d39f13b0 IE |
752 | void kvm_get_kvm(struct kvm *kvm) |
753 | { | |
e3736c3e | 754 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
755 | } |
756 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
757 | ||
758 | void kvm_put_kvm(struct kvm *kvm) | |
759 | { | |
e3736c3e | 760 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
761 | kvm_destroy_vm(kvm); |
762 | } | |
763 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
764 | ||
765 | ||
f17abe9a AK |
766 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
767 | { | |
768 | struct kvm *kvm = filp->private_data; | |
769 | ||
721eecbf GH |
770 | kvm_irqfd_release(kvm); |
771 | ||
d39f13b0 | 772 | kvm_put_kvm(kvm); |
6aa8b732 AK |
773 | return 0; |
774 | } | |
775 | ||
515a0127 TY |
776 | /* |
777 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 778 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 779 | */ |
a36a57b1 TY |
780 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
781 | { | |
515a0127 | 782 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 783 | |
92eca8fa | 784 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
785 | if (!memslot->dirty_bitmap) |
786 | return -ENOMEM; | |
787 | ||
a36a57b1 TY |
788 | return 0; |
789 | } | |
790 | ||
bf3e05bc | 791 | /* |
0e60b079 IM |
792 | * Insert memslot and re-sort memslots based on their GFN, |
793 | * so binary search could be used to lookup GFN. | |
794 | * Sorting algorithm takes advantage of having initially | |
795 | * sorted array and known changed memslot position. | |
bf3e05bc | 796 | */ |
5cc15027 PB |
797 | static void update_memslots(struct kvm_memslots *slots, |
798 | struct kvm_memory_slot *new) | |
bf3e05bc | 799 | { |
8593176c PB |
800 | int id = new->id; |
801 | int i = slots->id_to_index[id]; | |
063584d4 | 802 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 803 | |
8593176c | 804 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 805 | if (!new->npages) { |
dbaff309 | 806 | WARN_ON(!mslots[i].npages); |
9c1a5d38 IM |
807 | if (mslots[i].npages) |
808 | slots->used_slots--; | |
809 | } else { | |
810 | if (!mslots[i].npages) | |
811 | slots->used_slots++; | |
812 | } | |
0e60b079 | 813 | |
7f379cff | 814 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
815 | new->base_gfn <= mslots[i + 1].base_gfn) { |
816 | if (!mslots[i + 1].npages) | |
817 | break; | |
7f379cff IM |
818 | mslots[i] = mslots[i + 1]; |
819 | slots->id_to_index[mslots[i].id] = i; | |
820 | i++; | |
821 | } | |
efbeec70 PB |
822 | |
823 | /* | |
824 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
825 | * so that it moves before all those with base_gfn == npages == 0. | |
826 | * | |
827 | * On the other hand, if new->npages is zero, the above loop has | |
828 | * already left i pointing to the beginning of the empty part of | |
829 | * mslots, and the ">=" would move the hole backwards in this | |
830 | * case---which is wrong. So skip the loop when deleting a slot. | |
831 | */ | |
832 | if (new->npages) { | |
833 | while (i > 0 && | |
834 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
835 | mslots[i] = mslots[i - 1]; | |
836 | slots->id_to_index[mslots[i].id] = i; | |
837 | i--; | |
838 | } | |
dbaff309 PB |
839 | } else |
840 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 841 | |
8593176c PB |
842 | mslots[i] = *new; |
843 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
844 | } |
845 | ||
09170a49 | 846 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 847 | { |
4d8b81ab XG |
848 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
849 | ||
0f8a4de3 | 850 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
851 | valid_flags |= KVM_MEM_READONLY; |
852 | #endif | |
853 | ||
854 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
855 | return -EINVAL; |
856 | ||
857 | return 0; | |
858 | } | |
859 | ||
7ec4fb44 | 860 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 861 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 862 | { |
f481b069 | 863 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
7ec4fb44 | 864 | |
ee3d1570 DM |
865 | /* |
866 | * Set the low bit in the generation, which disables SPTE caching | |
867 | * until the end of synchronize_srcu_expedited. | |
868 | */ | |
869 | WARN_ON(old_memslots->generation & 1); | |
870 | slots->generation = old_memslots->generation + 1; | |
871 | ||
f481b069 | 872 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 873 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 874 | |
ee3d1570 DM |
875 | /* |
876 | * Increment the new memslot generation a second time. This prevents | |
877 | * vm exits that race with memslot updates from caching a memslot | |
878 | * generation that will (potentially) be valid forever. | |
4bd518f1 PB |
879 | * |
880 | * Generations must be unique even across address spaces. We do not need | |
881 | * a global counter for that, instead the generation space is evenly split | |
882 | * across address spaces. For example, with two address spaces, address | |
883 | * space 0 will use generations 0, 4, 8, ... while * address space 1 will | |
884 | * use generations 2, 6, 10, 14, ... | |
ee3d1570 | 885 | */ |
4bd518f1 | 886 | slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1; |
ee3d1570 | 887 | |
15f46015 | 888 | kvm_arch_memslots_updated(kvm, slots); |
e59dbe09 TY |
889 | |
890 | return old_memslots; | |
7ec4fb44 GN |
891 | } |
892 | ||
6aa8b732 AK |
893 | /* |
894 | * Allocate some memory and give it an address in the guest physical address | |
895 | * space. | |
896 | * | |
897 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 898 | * |
02d5d55b | 899 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 900 | */ |
f78e0e2e | 901 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 902 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 903 | { |
8234b22e | 904 | int r; |
6aa8b732 | 905 | gfn_t base_gfn; |
28bcb112 | 906 | unsigned long npages; |
a843fac2 | 907 | struct kvm_memory_slot *slot; |
6aa8b732 | 908 | struct kvm_memory_slot old, new; |
b7f69c55 | 909 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 910 | int as_id, id; |
f64c0398 | 911 | enum kvm_mr_change change; |
6aa8b732 | 912 | |
a50d64d6 XG |
913 | r = check_memory_region_flags(mem); |
914 | if (r) | |
915 | goto out; | |
916 | ||
6aa8b732 | 917 | r = -EINVAL; |
f481b069 PB |
918 | as_id = mem->slot >> 16; |
919 | id = (u16)mem->slot; | |
920 | ||
6aa8b732 AK |
921 | /* General sanity checks */ |
922 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
923 | goto out; | |
924 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
925 | goto out; | |
fa3d315a | 926 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 927 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 928 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
929 | !access_ok(VERIFY_WRITE, |
930 | (void __user *)(unsigned long)mem->userspace_addr, | |
931 | mem->memory_size))) | |
78749809 | 932 | goto out; |
f481b069 | 933 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
934 | goto out; |
935 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
936 | goto out; | |
937 | ||
f481b069 | 938 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
939 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
940 | npages = mem->memory_size >> PAGE_SHIFT; | |
941 | ||
660c22c4 TY |
942 | if (npages > KVM_MEM_MAX_NR_PAGES) |
943 | goto out; | |
944 | ||
a843fac2 | 945 | new = old = *slot; |
6aa8b732 | 946 | |
f481b069 | 947 | new.id = id; |
6aa8b732 AK |
948 | new.base_gfn = base_gfn; |
949 | new.npages = npages; | |
950 | new.flags = mem->flags; | |
951 | ||
f64c0398 TY |
952 | if (npages) { |
953 | if (!old.npages) | |
954 | change = KVM_MR_CREATE; | |
955 | else { /* Modify an existing slot. */ | |
956 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
957 | (npages != old.npages) || |
958 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
959 | goto out; |
960 | ||
961 | if (base_gfn != old.base_gfn) | |
962 | change = KVM_MR_MOVE; | |
963 | else if (new.flags != old.flags) | |
964 | change = KVM_MR_FLAGS_ONLY; | |
965 | else { /* Nothing to change. */ | |
966 | r = 0; | |
967 | goto out; | |
968 | } | |
969 | } | |
09170a49 PB |
970 | } else { |
971 | if (!old.npages) | |
972 | goto out; | |
973 | ||
f64c0398 | 974 | change = KVM_MR_DELETE; |
09170a49 PB |
975 | new.base_gfn = 0; |
976 | new.flags = 0; | |
977 | } | |
6aa8b732 | 978 | |
f64c0398 | 979 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
980 | /* Check for overlaps */ |
981 | r = -EEXIST; | |
f481b069 | 982 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
a843fac2 | 983 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
f481b069 | 984 | (slot->id == id)) |
0a706bee TY |
985 | continue; |
986 | if (!((base_gfn + npages <= slot->base_gfn) || | |
987 | (base_gfn >= slot->base_gfn + slot->npages))) | |
988 | goto out; | |
989 | } | |
6aa8b732 | 990 | } |
6aa8b732 | 991 | |
6aa8b732 AK |
992 | /* Free page dirty bitmap if unneeded */ |
993 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 994 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
995 | |
996 | r = -ENOMEM; | |
f64c0398 | 997 | if (change == KVM_MR_CREATE) { |
189a2f7b | 998 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 999 | |
5587027c | 1000 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 1001 | goto out_free; |
6aa8b732 | 1002 | } |
ec04b260 | 1003 | |
6aa8b732 AK |
1004 | /* Allocate page dirty bitmap if needed */ |
1005 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 1006 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 1007 | goto out_free; |
6aa8b732 AK |
1008 | } |
1009 | ||
74496134 | 1010 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
f2a81036 PB |
1011 | if (!slots) |
1012 | goto out_free; | |
f481b069 | 1013 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 1014 | |
f64c0398 | 1015 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 1016 | slot = id_to_memslot(slots, id); |
28a37544 XG |
1017 | slot->flags |= KVM_MEMSLOT_INVALID; |
1018 | ||
f481b069 | 1019 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 1020 | |
e40f193f AW |
1021 | /* slot was deleted or moved, clear iommu mapping */ |
1022 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
1023 | /* From this point no new shadow pages pointing to a deleted, |
1024 | * or moved, memslot will be created. | |
bc6678a3 MT |
1025 | * |
1026 | * validation of sp->gfn happens in: | |
b7d409de XL |
1027 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
1028 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 1029 | */ |
2df72e9b | 1030 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
1031 | |
1032 | /* | |
1033 | * We can re-use the old_memslots from above, the only difference | |
1034 | * from the currently installed memslots is the invalid flag. This | |
1035 | * will get overwritten by update_memslots anyway. | |
1036 | */ | |
b7f69c55 | 1037 | slots = old_memslots; |
bc6678a3 | 1038 | } |
34d4cb8f | 1039 | |
7b6195a9 | 1040 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 1041 | if (r) |
b7f69c55 | 1042 | goto out_slots; |
f7784b8e | 1043 | |
a47d2b07 | 1044 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 1045 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 1046 | new.dirty_bitmap = NULL; |
db3fe4eb | 1047 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
1048 | } |
1049 | ||
5cc15027 | 1050 | update_memslots(slots, &new); |
f481b069 | 1051 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 1052 | |
f36f3f28 | 1053 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 1054 | |
a47d2b07 | 1055 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 1056 | kvfree(old_memslots); |
bc6678a3 | 1057 | |
261874b0 AW |
1058 | /* |
1059 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
1060 | * un-mapped and re-mapped if their base changes. Since base change |
1061 | * unmapping is handled above with slot deletion, mapping alone is | |
1062 | * needed here. Anything else the iommu might care about for existing | |
1063 | * slots (size changes, userspace addr changes and read-only flag | |
1064 | * changes) is disallowed above, so any other attribute changes getting | |
1065 | * here can be skipped. | |
261874b0 | 1066 | */ |
75d61fbc TY |
1067 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
1068 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 1069 | return r; |
bc6678a3 MT |
1070 | } |
1071 | ||
6aa8b732 AK |
1072 | return 0; |
1073 | ||
e40f193f | 1074 | out_slots: |
74496134 | 1075 | kvfree(slots); |
f78e0e2e | 1076 | out_free: |
a47d2b07 | 1077 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
1078 | out: |
1079 | return r; | |
210c7c4d | 1080 | } |
f78e0e2e SY |
1081 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1082 | ||
1083 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1084 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1085 | { |
1086 | int r; | |
1087 | ||
79fac95e | 1088 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1089 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1090 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1091 | return r; |
1092 | } | |
210c7c4d IE |
1093 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1094 | ||
7940876e SH |
1095 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1096 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1097 | { |
f481b069 | 1098 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1099 | return -EINVAL; |
09170a49 | 1100 | |
47ae31e2 | 1101 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1102 | } |
1103 | ||
5bb064dc ZX |
1104 | int kvm_get_dirty_log(struct kvm *kvm, |
1105 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 1106 | { |
9f6b8029 | 1107 | struct kvm_memslots *slots; |
6aa8b732 | 1108 | struct kvm_memory_slot *memslot; |
843574a3 | 1109 | int i, as_id, id; |
87bf6e7d | 1110 | unsigned long n; |
6aa8b732 AK |
1111 | unsigned long any = 0; |
1112 | ||
f481b069 PB |
1113 | as_id = log->slot >> 16; |
1114 | id = (u16)log->slot; | |
1115 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1116 | return -EINVAL; |
6aa8b732 | 1117 | |
f481b069 PB |
1118 | slots = __kvm_memslots(kvm, as_id); |
1119 | memslot = id_to_memslot(slots, id); | |
6aa8b732 | 1120 | if (!memslot->dirty_bitmap) |
843574a3 | 1121 | return -ENOENT; |
6aa8b732 | 1122 | |
87bf6e7d | 1123 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 1124 | |
cd1a4a98 | 1125 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1126 | any = memslot->dirty_bitmap[i]; |
1127 | ||
6aa8b732 | 1128 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
843574a3 | 1129 | return -EFAULT; |
6aa8b732 | 1130 | |
5bb064dc ZX |
1131 | if (any) |
1132 | *is_dirty = 1; | |
843574a3 | 1133 | return 0; |
6aa8b732 | 1134 | } |
2ba9f0d8 | 1135 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1136 | |
ba0513b5 MS |
1137 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1138 | /** | |
1139 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1140 | * are dirty write protect them for next write. | |
1141 | * @kvm: pointer to kvm instance | |
1142 | * @log: slot id and address to which we copy the log | |
1143 | * @is_dirty: flag set if any page is dirty | |
1144 | * | |
1145 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1146 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1147 | * following order: | |
1148 | * | |
1149 | * 1. Take a snapshot of the bit and clear it if needed. | |
1150 | * 2. Write protect the corresponding page. | |
1151 | * 3. Copy the snapshot to the userspace. | |
1152 | * 4. Upon return caller flushes TLB's if needed. | |
1153 | * | |
1154 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1155 | * entry. This is not a problem because the page is reported dirty using | |
1156 | * the snapshot taken before and step 4 ensures that writes done after | |
1157 | * exiting to userspace will be logged for the next call. | |
1158 | * | |
1159 | */ | |
1160 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1161 | struct kvm_dirty_log *log, bool *is_dirty) | |
1162 | { | |
9f6b8029 | 1163 | struct kvm_memslots *slots; |
ba0513b5 | 1164 | struct kvm_memory_slot *memslot; |
58d6db34 | 1165 | int i, as_id, id; |
ba0513b5 MS |
1166 | unsigned long n; |
1167 | unsigned long *dirty_bitmap; | |
1168 | unsigned long *dirty_bitmap_buffer; | |
1169 | ||
f481b069 PB |
1170 | as_id = log->slot >> 16; |
1171 | id = (u16)log->slot; | |
1172 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1173 | return -EINVAL; |
ba0513b5 | 1174 | |
f481b069 PB |
1175 | slots = __kvm_memslots(kvm, as_id); |
1176 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1177 | |
1178 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1179 | if (!dirty_bitmap) |
58d6db34 | 1180 | return -ENOENT; |
ba0513b5 MS |
1181 | |
1182 | n = kvm_dirty_bitmap_bytes(memslot); | |
1183 | ||
1184 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1185 | memset(dirty_bitmap_buffer, 0, n); | |
1186 | ||
1187 | spin_lock(&kvm->mmu_lock); | |
1188 | *is_dirty = false; | |
1189 | for (i = 0; i < n / sizeof(long); i++) { | |
1190 | unsigned long mask; | |
1191 | gfn_t offset; | |
1192 | ||
1193 | if (!dirty_bitmap[i]) | |
1194 | continue; | |
1195 | ||
1196 | *is_dirty = true; | |
1197 | ||
1198 | mask = xchg(&dirty_bitmap[i], 0); | |
1199 | dirty_bitmap_buffer[i] = mask; | |
1200 | ||
58d2930f TY |
1201 | if (mask) { |
1202 | offset = i * BITS_PER_LONG; | |
1203 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1204 | offset, mask); | |
1205 | } | |
ba0513b5 MS |
1206 | } |
1207 | ||
1208 | spin_unlock(&kvm->mmu_lock); | |
ba0513b5 | 1209 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
58d6db34 ME |
1210 | return -EFAULT; |
1211 | return 0; | |
ba0513b5 MS |
1212 | } |
1213 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1214 | #endif | |
1215 | ||
db3fe4eb TY |
1216 | bool kvm_largepages_enabled(void) |
1217 | { | |
1218 | return largepages_enabled; | |
1219 | } | |
1220 | ||
54dee993 MT |
1221 | void kvm_disable_largepages(void) |
1222 | { | |
1223 | largepages_enabled = false; | |
1224 | } | |
1225 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1226 | ||
49c7754c GN |
1227 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1228 | { | |
1229 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1230 | } | |
a1f4d395 | 1231 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1232 | |
8e73485c PB |
1233 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1234 | { | |
1235 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1236 | } | |
1237 | ||
33e94154 | 1238 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1239 | { |
bf3e05bc | 1240 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1241 | |
bbacc0c1 | 1242 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc | 1243 | memslot->flags & KVM_MEMSLOT_INVALID) |
33e94154 | 1244 | return false; |
e0d62c7f | 1245 | |
33e94154 | 1246 | return true; |
e0d62c7f IE |
1247 | } |
1248 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1249 | ||
8f0b1ab6 JR |
1250 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1251 | { | |
1252 | struct vm_area_struct *vma; | |
1253 | unsigned long addr, size; | |
1254 | ||
1255 | size = PAGE_SIZE; | |
1256 | ||
1257 | addr = gfn_to_hva(kvm, gfn); | |
1258 | if (kvm_is_error_hva(addr)) | |
1259 | return PAGE_SIZE; | |
1260 | ||
1261 | down_read(¤t->mm->mmap_sem); | |
1262 | vma = find_vma(current->mm, addr); | |
1263 | if (!vma) | |
1264 | goto out; | |
1265 | ||
1266 | size = vma_kernel_pagesize(vma); | |
1267 | ||
1268 | out: | |
1269 | up_read(¤t->mm->mmap_sem); | |
1270 | ||
1271 | return size; | |
1272 | } | |
1273 | ||
4d8b81ab XG |
1274 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1275 | { | |
1276 | return slot->flags & KVM_MEM_READONLY; | |
1277 | } | |
1278 | ||
4d8b81ab XG |
1279 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1280 | gfn_t *nr_pages, bool write) | |
539cb660 | 1281 | { |
bc6678a3 | 1282 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1283 | return KVM_HVA_ERR_BAD; |
48987781 | 1284 | |
4d8b81ab XG |
1285 | if (memslot_is_readonly(slot) && write) |
1286 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1287 | |
1288 | if (nr_pages) | |
1289 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1290 | ||
4d8b81ab | 1291 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1292 | } |
48987781 | 1293 | |
4d8b81ab XG |
1294 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1295 | gfn_t *nr_pages) | |
1296 | { | |
1297 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1298 | } |
48987781 | 1299 | |
4d8b81ab | 1300 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1301 | gfn_t gfn) |
4d8b81ab XG |
1302 | { |
1303 | return gfn_to_hva_many(slot, gfn, NULL); | |
1304 | } | |
1305 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1306 | ||
48987781 XG |
1307 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1308 | { | |
49c7754c | 1309 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1310 | } |
0d150298 | 1311 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1312 | |
8e73485c PB |
1313 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1314 | { | |
1315 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1316 | } | |
1317 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1318 | ||
86ab8cff | 1319 | /* |
ba6a3541 PB |
1320 | * If writable is set to false, the hva returned by this function is only |
1321 | * allowed to be read. | |
86ab8cff | 1322 | */ |
64d83126 CD |
1323 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1324 | gfn_t gfn, bool *writable) | |
86ab8cff | 1325 | { |
a2ac07fe GN |
1326 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1327 | ||
1328 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1329 | *writable = !memslot_is_readonly(slot); |
1330 | ||
a2ac07fe | 1331 | return hva; |
86ab8cff XG |
1332 | } |
1333 | ||
64d83126 CD |
1334 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1335 | { | |
1336 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1337 | ||
1338 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1339 | } | |
1340 | ||
8e73485c PB |
1341 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1342 | { | |
1343 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1344 | ||
1345 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1346 | } | |
1347 | ||
d4edcf0d DH |
1348 | static int get_user_page_nowait(unsigned long start, int write, |
1349 | struct page **page) | |
0857b9e9 | 1350 | { |
0d731759 | 1351 | int flags = FOLL_NOWAIT | FOLL_HWPOISON; |
0857b9e9 GN |
1352 | |
1353 | if (write) | |
1354 | flags |= FOLL_WRITE; | |
1355 | ||
0d731759 | 1356 | return get_user_pages(start, 1, flags, page, NULL); |
0857b9e9 GN |
1357 | } |
1358 | ||
fafc3dba HY |
1359 | static inline int check_user_page_hwpoison(unsigned long addr) |
1360 | { | |
0d731759 | 1361 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1362 | |
0d731759 | 1363 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1364 | return rc == -EHWPOISON; |
1365 | } | |
1366 | ||
2fc84311 XG |
1367 | /* |
1368 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1369 | * true indicates success, otherwise false is returned. | |
1370 | */ | |
1371 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
ba049e93 | 1372 | bool write_fault, bool *writable, kvm_pfn_t *pfn) |
954bbbc2 | 1373 | { |
8d4e1288 | 1374 | struct page *page[1]; |
2fc84311 | 1375 | int npages; |
954bbbc2 | 1376 | |
2fc84311 XG |
1377 | if (!(async || atomic)) |
1378 | return false; | |
af585b92 | 1379 | |
12ce13fe XG |
1380 | /* |
1381 | * Fast pin a writable pfn only if it is a write fault request | |
1382 | * or the caller allows to map a writable pfn for a read fault | |
1383 | * request. | |
1384 | */ | |
1385 | if (!(write_fault || writable)) | |
1386 | return false; | |
612819c3 | 1387 | |
2fc84311 XG |
1388 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1389 | if (npages == 1) { | |
1390 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1391 | |
2fc84311 XG |
1392 | if (writable) |
1393 | *writable = true; | |
1394 | return true; | |
1395 | } | |
af585b92 | 1396 | |
2fc84311 XG |
1397 | return false; |
1398 | } | |
612819c3 | 1399 | |
2fc84311 XG |
1400 | /* |
1401 | * The slow path to get the pfn of the specified host virtual address, | |
1402 | * 1 indicates success, -errno is returned if error is detected. | |
1403 | */ | |
1404 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1405 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 XG |
1406 | { |
1407 | struct page *page[1]; | |
1408 | int npages = 0; | |
612819c3 | 1409 | |
2fc84311 XG |
1410 | might_sleep(); |
1411 | ||
1412 | if (writable) | |
1413 | *writable = write_fault; | |
1414 | ||
1415 | if (async) { | |
1416 | down_read(¤t->mm->mmap_sem); | |
d4edcf0d | 1417 | npages = get_user_page_nowait(addr, write_fault, page); |
2fc84311 | 1418 | up_read(¤t->mm->mmap_sem); |
d4944b0e | 1419 | } else { |
8b7457ef | 1420 | unsigned int flags = FOLL_HWPOISON; |
d4944b0e LS |
1421 | |
1422 | if (write_fault) | |
1423 | flags |= FOLL_WRITE; | |
1424 | ||
8b7457ef | 1425 | npages = get_user_pages_unlocked(addr, 1, page, flags); |
d4944b0e | 1426 | } |
2fc84311 XG |
1427 | if (npages != 1) |
1428 | return npages; | |
1429 | ||
1430 | /* map read fault as writable if possible */ | |
12ce13fe | 1431 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1432 | struct page *wpage[1]; |
1433 | ||
1434 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1435 | if (npages == 1) { | |
1436 | *writable = true; | |
1437 | put_page(page[0]); | |
1438 | page[0] = wpage[0]; | |
612819c3 | 1439 | } |
2fc84311 XG |
1440 | |
1441 | npages = 1; | |
887c08ac | 1442 | } |
2fc84311 XG |
1443 | *pfn = page_to_pfn(page[0]); |
1444 | return npages; | |
1445 | } | |
539cb660 | 1446 | |
4d8b81ab XG |
1447 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1448 | { | |
1449 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1450 | return false; | |
2e2e3738 | 1451 | |
4d8b81ab XG |
1452 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1453 | return false; | |
887c08ac | 1454 | |
4d8b81ab XG |
1455 | return true; |
1456 | } | |
bf998156 | 1457 | |
92176a8e PB |
1458 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1459 | unsigned long addr, bool *async, | |
1460 | bool write_fault, kvm_pfn_t *p_pfn) | |
1461 | { | |
add6a0cd PB |
1462 | unsigned long pfn; |
1463 | int r; | |
1464 | ||
1465 | r = follow_pfn(vma, addr, &pfn); | |
1466 | if (r) { | |
1467 | /* | |
1468 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1469 | * not call the fault handler, so do it here. | |
1470 | */ | |
1471 | bool unlocked = false; | |
1472 | r = fixup_user_fault(current, current->mm, addr, | |
1473 | (write_fault ? FAULT_FLAG_WRITE : 0), | |
1474 | &unlocked); | |
1475 | if (unlocked) | |
1476 | return -EAGAIN; | |
1477 | if (r) | |
1478 | return r; | |
1479 | ||
1480 | r = follow_pfn(vma, addr, &pfn); | |
1481 | if (r) | |
1482 | return r; | |
1483 | ||
1484 | } | |
1485 | ||
1486 | ||
1487 | /* | |
1488 | * Get a reference here because callers of *hva_to_pfn* and | |
1489 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1490 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1491 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1492 | * simply do nothing for reserved pfns. | |
1493 | * | |
1494 | * Whoever called remap_pfn_range is also going to call e.g. | |
1495 | * unmap_mapping_range before the underlying pages are freed, | |
1496 | * causing a call to our MMU notifier. | |
1497 | */ | |
1498 | kvm_get_pfn(pfn); | |
1499 | ||
1500 | *p_pfn = pfn; | |
92176a8e PB |
1501 | return 0; |
1502 | } | |
1503 | ||
12ce13fe XG |
1504 | /* |
1505 | * Pin guest page in memory and return its pfn. | |
1506 | * @addr: host virtual address which maps memory to the guest | |
1507 | * @atomic: whether this function can sleep | |
1508 | * @async: whether this function need to wait IO complete if the | |
1509 | * host page is not in the memory | |
1510 | * @write_fault: whether we should get a writable host page | |
1511 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1512 | * | |
1513 | * The function will map a writable host page for these two cases: | |
1514 | * 1): @write_fault = true | |
1515 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1516 | * whether the mapping is writable. | |
1517 | */ | |
ba049e93 | 1518 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1519 | bool write_fault, bool *writable) |
1520 | { | |
1521 | struct vm_area_struct *vma; | |
ba049e93 | 1522 | kvm_pfn_t pfn = 0; |
92176a8e | 1523 | int npages, r; |
2e2e3738 | 1524 | |
2fc84311 XG |
1525 | /* we can do it either atomically or asynchronously, not both */ |
1526 | BUG_ON(atomic && async); | |
8d4e1288 | 1527 | |
2fc84311 XG |
1528 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1529 | return pfn; | |
1530 | ||
1531 | if (atomic) | |
1532 | return KVM_PFN_ERR_FAULT; | |
1533 | ||
1534 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1535 | if (npages == 1) | |
1536 | return pfn; | |
8d4e1288 | 1537 | |
2fc84311 XG |
1538 | down_read(¤t->mm->mmap_sem); |
1539 | if (npages == -EHWPOISON || | |
1540 | (!async && check_user_page_hwpoison(addr))) { | |
1541 | pfn = KVM_PFN_ERR_HWPOISON; | |
1542 | goto exit; | |
1543 | } | |
1544 | ||
add6a0cd | 1545 | retry: |
2fc84311 XG |
1546 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
1547 | ||
1548 | if (vma == NULL) | |
1549 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e PB |
1550 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1551 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn); | |
add6a0cd PB |
1552 | if (r == -EAGAIN) |
1553 | goto retry; | |
92176a8e PB |
1554 | if (r < 0) |
1555 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 1556 | } else { |
4d8b81ab | 1557 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1558 | *async = true; |
1559 | pfn = KVM_PFN_ERR_FAULT; | |
1560 | } | |
1561 | exit: | |
1562 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1563 | return pfn; |
35149e21 AL |
1564 | } |
1565 | ||
ba049e93 DW |
1566 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
1567 | bool atomic, bool *async, bool write_fault, | |
1568 | bool *writable) | |
887c08ac | 1569 | { |
4d8b81ab XG |
1570 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1571 | ||
b2740d35 PB |
1572 | if (addr == KVM_HVA_ERR_RO_BAD) { |
1573 | if (writable) | |
1574 | *writable = false; | |
4d8b81ab | 1575 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 1576 | } |
4d8b81ab | 1577 | |
b2740d35 PB |
1578 | if (kvm_is_error_hva(addr)) { |
1579 | if (writable) | |
1580 | *writable = false; | |
81c52c56 | 1581 | return KVM_PFN_NOSLOT; |
b2740d35 | 1582 | } |
4d8b81ab XG |
1583 | |
1584 | /* Do not map writable pfn in the readonly memslot. */ | |
1585 | if (writable && memslot_is_readonly(slot)) { | |
1586 | *writable = false; | |
1587 | writable = NULL; | |
1588 | } | |
1589 | ||
1590 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1591 | writable); | |
887c08ac | 1592 | } |
3520469d | 1593 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1594 | |
ba049e93 | 1595 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
1596 | bool *writable) |
1597 | { | |
e37afc6e PB |
1598 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1599 | write_fault, writable); | |
612819c3 MT |
1600 | } |
1601 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1602 | ||
ba049e93 | 1603 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1604 | { |
4d8b81ab | 1605 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1606 | } |
e37afc6e | 1607 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1608 | |
ba049e93 | 1609 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1610 | { |
4d8b81ab | 1611 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1612 | } |
037d92dc | 1613 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1614 | |
ba049e93 | 1615 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1616 | { |
1617 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1618 | } | |
1619 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1620 | ||
ba049e93 | 1621 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1622 | { |
1623 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1624 | } | |
1625 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1626 | ||
ba049e93 | 1627 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1628 | { |
1629 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1630 | } | |
1631 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1632 | ||
ba049e93 | 1633 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1634 | { |
1635 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1636 | } | |
1637 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1638 | ||
d9ef13c2 PB |
1639 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1640 | struct page **pages, int nr_pages) | |
48987781 XG |
1641 | { |
1642 | unsigned long addr; | |
1643 | gfn_t entry; | |
1644 | ||
d9ef13c2 | 1645 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1646 | if (kvm_is_error_hva(addr)) |
1647 | return -1; | |
1648 | ||
1649 | if (entry < nr_pages) | |
1650 | return 0; | |
1651 | ||
1652 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1653 | } | |
1654 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1655 | ||
ba049e93 | 1656 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 1657 | { |
81c52c56 | 1658 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1659 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1660 | |
bf4bea8e | 1661 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1662 | WARN_ON(1); |
6cede2e6 | 1663 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1664 | } |
a2766325 XG |
1665 | |
1666 | return pfn_to_page(pfn); | |
1667 | } | |
1668 | ||
35149e21 AL |
1669 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1670 | { | |
ba049e93 | 1671 | kvm_pfn_t pfn; |
2e2e3738 AL |
1672 | |
1673 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1674 | |
a2766325 | 1675 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1676 | } |
1677 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1678 | ||
8e73485c PB |
1679 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1680 | { | |
ba049e93 | 1681 | kvm_pfn_t pfn; |
8e73485c PB |
1682 | |
1683 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1684 | ||
1685 | return kvm_pfn_to_page(pfn); | |
1686 | } | |
1687 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1688 | ||
b4231d61 IE |
1689 | void kvm_release_page_clean(struct page *page) |
1690 | { | |
32cad84f XG |
1691 | WARN_ON(is_error_page(page)); |
1692 | ||
35149e21 | 1693 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1694 | } |
1695 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1696 | ||
ba049e93 | 1697 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 1698 | { |
bf4bea8e | 1699 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1700 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1701 | } |
1702 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1703 | ||
b4231d61 | 1704 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1705 | { |
a2766325 XG |
1706 | WARN_ON(is_error_page(page)); |
1707 | ||
35149e21 AL |
1708 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1709 | } | |
1710 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1711 | ||
ba049e93 | 1712 | static void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
1713 | { |
1714 | kvm_set_pfn_dirty(pfn); | |
1715 | kvm_release_pfn_clean(pfn); | |
1716 | } | |
35149e21 | 1717 | |
ba049e93 | 1718 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 1719 | { |
bf4bea8e | 1720 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 | 1721 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1722 | |
2e2e3738 AL |
1723 | if (!PageReserved(page)) |
1724 | SetPageDirty(page); | |
1725 | } | |
8a7ae055 | 1726 | } |
35149e21 AL |
1727 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1728 | ||
ba049e93 | 1729 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 1730 | { |
bf4bea8e | 1731 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1732 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1733 | } |
1734 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1735 | ||
ba049e93 | 1736 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 1737 | { |
bf4bea8e | 1738 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1739 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1740 | } |
1741 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1742 | |
195aefde IE |
1743 | static int next_segment(unsigned long len, int offset) |
1744 | { | |
1745 | if (len > PAGE_SIZE - offset) | |
1746 | return PAGE_SIZE - offset; | |
1747 | else | |
1748 | return len; | |
1749 | } | |
1750 | ||
8e73485c PB |
1751 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1752 | void *data, int offset, int len) | |
195aefde | 1753 | { |
e0506bcb IE |
1754 | int r; |
1755 | unsigned long addr; | |
195aefde | 1756 | |
8e73485c | 1757 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1758 | if (kvm_is_error_hva(addr)) |
1759 | return -EFAULT; | |
3180a7fc | 1760 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1761 | if (r) |
195aefde | 1762 | return -EFAULT; |
195aefde IE |
1763 | return 0; |
1764 | } | |
8e73485c PB |
1765 | |
1766 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1767 | int len) | |
1768 | { | |
1769 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1770 | ||
1771 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1772 | } | |
195aefde IE |
1773 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1774 | ||
8e73485c PB |
1775 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1776 | int offset, int len) | |
1777 | { | |
1778 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1779 | ||
1780 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1781 | } | |
1782 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1783 | ||
195aefde IE |
1784 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1785 | { | |
1786 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1787 | int seg; | |
1788 | int offset = offset_in_page(gpa); | |
1789 | int ret; | |
1790 | ||
1791 | while ((seg = next_segment(len, offset)) != 0) { | |
1792 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1793 | if (ret < 0) | |
1794 | return ret; | |
1795 | offset = 0; | |
1796 | len -= seg; | |
1797 | data += seg; | |
1798 | ++gfn; | |
1799 | } | |
1800 | return 0; | |
1801 | } | |
1802 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1803 | ||
8e73485c | 1804 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 1805 | { |
7ec54588 | 1806 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 1807 | int seg; |
7ec54588 | 1808 | int offset = offset_in_page(gpa); |
8e73485c PB |
1809 | int ret; |
1810 | ||
1811 | while ((seg = next_segment(len, offset)) != 0) { | |
1812 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
1813 | if (ret < 0) | |
1814 | return ret; | |
1815 | offset = 0; | |
1816 | len -= seg; | |
1817 | data += seg; | |
1818 | ++gfn; | |
1819 | } | |
1820 | return 0; | |
1821 | } | |
1822 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 1823 | |
8e73485c PB |
1824 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1825 | void *data, int offset, unsigned long len) | |
1826 | { | |
1827 | int r; | |
1828 | unsigned long addr; | |
1829 | ||
1830 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
1831 | if (kvm_is_error_hva(addr)) |
1832 | return -EFAULT; | |
0aac03f0 | 1833 | pagefault_disable(); |
3180a7fc | 1834 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1835 | pagefault_enable(); |
7ec54588 MT |
1836 | if (r) |
1837 | return -EFAULT; | |
1838 | return 0; | |
1839 | } | |
7ec54588 | 1840 | |
8e73485c PB |
1841 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1842 | unsigned long len) | |
1843 | { | |
1844 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1845 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1846 | int offset = offset_in_page(gpa); | |
1847 | ||
1848 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1849 | } | |
1850 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
1851 | ||
1852 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
1853 | void *data, unsigned long len) | |
1854 | { | |
1855 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1856 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1857 | int offset = offset_in_page(gpa); | |
1858 | ||
1859 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1860 | } | |
1861 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
1862 | ||
1863 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
1864 | const void *data, int offset, int len) | |
195aefde | 1865 | { |
e0506bcb IE |
1866 | int r; |
1867 | unsigned long addr; | |
195aefde | 1868 | |
251eb841 | 1869 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
1870 | if (kvm_is_error_hva(addr)) |
1871 | return -EFAULT; | |
8b0cedff | 1872 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1873 | if (r) |
195aefde | 1874 | return -EFAULT; |
bc009e43 | 1875 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
1876 | return 0; |
1877 | } | |
8e73485c PB |
1878 | |
1879 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
1880 | const void *data, int offset, int len) | |
1881 | { | |
1882 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1883 | ||
1884 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1885 | } | |
195aefde IE |
1886 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
1887 | ||
8e73485c PB |
1888 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
1889 | const void *data, int offset, int len) | |
1890 | { | |
1891 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1892 | ||
1893 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1894 | } | |
1895 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
1896 | ||
195aefde IE |
1897 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
1898 | unsigned long len) | |
1899 | { | |
1900 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1901 | int seg; | |
1902 | int offset = offset_in_page(gpa); | |
1903 | int ret; | |
1904 | ||
1905 | while ((seg = next_segment(len, offset)) != 0) { | |
1906 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1907 | if (ret < 0) | |
1908 | return ret; | |
1909 | offset = 0; | |
1910 | len -= seg; | |
1911 | data += seg; | |
1912 | ++gfn; | |
1913 | } | |
1914 | return 0; | |
1915 | } | |
ff651cb6 | 1916 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1917 | |
8e73485c PB |
1918 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
1919 | unsigned long len) | |
1920 | { | |
1921 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1922 | int seg; | |
1923 | int offset = offset_in_page(gpa); | |
1924 | int ret; | |
1925 | ||
1926 | while ((seg = next_segment(len, offset)) != 0) { | |
1927 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
1928 | if (ret < 0) | |
1929 | return ret; | |
1930 | offset = 0; | |
1931 | len -= seg; | |
1932 | data += seg; | |
1933 | ++gfn; | |
1934 | } | |
1935 | return 0; | |
1936 | } | |
1937 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
1938 | ||
5a2d4365 PB |
1939 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
1940 | struct gfn_to_hva_cache *ghc, | |
1941 | gpa_t gpa, unsigned long len) | |
49c7754c | 1942 | { |
49c7754c | 1943 | int offset = offset_in_page(gpa); |
8f964525 AH |
1944 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1945 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1946 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1947 | gfn_t nr_pages_avail; | |
49c7754c GN |
1948 | |
1949 | ghc->gpa = gpa; | |
1950 | ghc->generation = slots->generation; | |
8f964525 | 1951 | ghc->len = len; |
5a2d4365 | 1952 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
ca3f0874 RK |
1953 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
1954 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { | |
49c7754c | 1955 | ghc->hva += offset; |
8f964525 AH |
1956 | } else { |
1957 | /* | |
1958 | * If the requested region crosses two memslots, we still | |
1959 | * verify that the entire region is valid here. | |
1960 | */ | |
1961 | while (start_gfn <= end_gfn) { | |
5a2d4365 | 1962 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
8f964525 AH |
1963 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, |
1964 | &nr_pages_avail); | |
1965 | if (kvm_is_error_hva(ghc->hva)) | |
1966 | return -EFAULT; | |
1967 | start_gfn += nr_pages_avail; | |
1968 | } | |
1969 | /* Use the slow path for cross page reads and writes. */ | |
1970 | ghc->memslot = NULL; | |
1971 | } | |
49c7754c GN |
1972 | return 0; |
1973 | } | |
5a2d4365 | 1974 | |
bbd64115 | 1975 | int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
1976 | gpa_t gpa, unsigned long len) |
1977 | { | |
bbd64115 | 1978 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
5a2d4365 PB |
1979 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
1980 | } | |
bbd64115 | 1981 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init); |
49c7754c | 1982 | |
bbd64115 CL |
1983 | int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc, |
1984 | void *data, int offset, unsigned long len) | |
49c7754c | 1985 | { |
bbd64115 | 1986 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
49c7754c | 1987 | int r; |
4ec6e863 | 1988 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 1989 | |
4ec6e863 | 1990 | BUG_ON(len + offset > ghc->len); |
8f964525 | 1991 | |
49c7754c | 1992 | if (slots->generation != ghc->generation) |
5a2d4365 | 1993 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
1994 | |
1995 | if (unlikely(!ghc->memslot)) | |
bbd64115 | 1996 | return kvm_vcpu_write_guest(vcpu, gpa, data, len); |
49c7754c GN |
1997 | |
1998 | if (kvm_is_error_hva(ghc->hva)) | |
1999 | return -EFAULT; | |
2000 | ||
4ec6e863 | 2001 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2002 | if (r) |
2003 | return -EFAULT; | |
4ec6e863 | 2004 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2005 | |
2006 | return 0; | |
2007 | } | |
bbd64115 | 2008 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached); |
4ec6e863 | 2009 | |
bbd64115 CL |
2010 | int kvm_vcpu_write_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc, |
2011 | void *data, unsigned long len) | |
4ec6e863 | 2012 | { |
bbd64115 | 2013 | return kvm_vcpu_write_guest_offset_cached(vcpu, ghc, data, 0, len); |
4ec6e863 | 2014 | } |
bbd64115 | 2015 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached); |
49c7754c | 2016 | |
bbd64115 CL |
2017 | int kvm_vcpu_read_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc, |
2018 | void *data, unsigned long len) | |
e03b644f | 2019 | { |
bbd64115 | 2020 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
e03b644f GN |
2021 | int r; |
2022 | ||
8f964525 AH |
2023 | BUG_ON(len > ghc->len); |
2024 | ||
e03b644f | 2025 | if (slots->generation != ghc->generation) |
5a2d4365 | 2026 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
2027 | |
2028 | if (unlikely(!ghc->memslot)) | |
bbd64115 | 2029 | return kvm_vcpu_read_guest(vcpu, ghc->gpa, data, len); |
e03b644f GN |
2030 | |
2031 | if (kvm_is_error_hva(ghc->hva)) | |
2032 | return -EFAULT; | |
2033 | ||
2034 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
2035 | if (r) | |
2036 | return -EFAULT; | |
2037 | ||
2038 | return 0; | |
2039 | } | |
bbd64115 | 2040 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached); |
e03b644f | 2041 | |
195aefde IE |
2042 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
2043 | { | |
8a3caa6d HC |
2044 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
2045 | ||
2046 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
2047 | } |
2048 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
2049 | ||
2050 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
2051 | { | |
2052 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2053 | int seg; | |
2054 | int offset = offset_in_page(gpa); | |
2055 | int ret; | |
2056 | ||
bfda0e84 | 2057 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
2058 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
2059 | if (ret < 0) | |
2060 | return ret; | |
2061 | offset = 0; | |
2062 | len -= seg; | |
2063 | ++gfn; | |
2064 | } | |
2065 | return 0; | |
2066 | } | |
2067 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2068 | ||
bc009e43 | 2069 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 2070 | gfn_t gfn) |
6aa8b732 | 2071 | { |
7e9d619d RR |
2072 | if (memslot && memslot->dirty_bitmap) { |
2073 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 2074 | |
b74ca3b3 | 2075 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
2076 | } |
2077 | } | |
2078 | ||
49c7754c GN |
2079 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2080 | { | |
2081 | struct kvm_memory_slot *memslot; | |
2082 | ||
2083 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 2084 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 2085 | } |
2ba9f0d8 | 2086 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2087 | |
8e73485c PB |
2088 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2089 | { | |
2090 | struct kvm_memory_slot *memslot; | |
2091 | ||
2092 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2093 | mark_page_dirty_in_slot(memslot, gfn); | |
2094 | } | |
2095 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2096 | ||
aca6ff29 WL |
2097 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
2098 | { | |
6b6de68c | 2099 | unsigned int old, val, grow; |
aca6ff29 | 2100 | |
2cbd7824 | 2101 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 2102 | grow = READ_ONCE(halt_poll_ns_grow); |
aca6ff29 | 2103 | /* 10us base */ |
6b6de68c | 2104 | if (val == 0 && grow) |
aca6ff29 WL |
2105 | val = 10000; |
2106 | else | |
6b6de68c | 2107 | val *= grow; |
aca6ff29 | 2108 | |
313f636d DM |
2109 | if (val > halt_poll_ns) |
2110 | val = halt_poll_ns; | |
2111 | ||
aca6ff29 | 2112 | vcpu->halt_poll_ns = val; |
2cbd7824 | 2113 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2114 | } |
2115 | ||
2116 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
2117 | { | |
6b6de68c | 2118 | unsigned int old, val, shrink; |
aca6ff29 | 2119 | |
2cbd7824 | 2120 | old = val = vcpu->halt_poll_ns; |
6b6de68c CB |
2121 | shrink = READ_ONCE(halt_poll_ns_shrink); |
2122 | if (shrink == 0) | |
aca6ff29 WL |
2123 | val = 0; |
2124 | else | |
6b6de68c | 2125 | val /= shrink; |
aca6ff29 WL |
2126 | |
2127 | vcpu->halt_poll_ns = val; | |
2cbd7824 | 2128 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2129 | } |
2130 | ||
f7819512 PB |
2131 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
2132 | { | |
2133 | if (kvm_arch_vcpu_runnable(vcpu)) { | |
2134 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
2135 | return -EINTR; | |
2136 | } | |
2137 | if (kvm_cpu_has_pending_timer(vcpu)) | |
2138 | return -EINTR; | |
2139 | if (signal_pending(current)) | |
2140 | return -EINTR; | |
2141 | ||
2142 | return 0; | |
2143 | } | |
2144 | ||
b6958ce4 ED |
2145 | /* |
2146 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
2147 | */ | |
8776e519 | 2148 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 2149 | { |
f7819512 | 2150 | ktime_t start, cur; |
8577370f | 2151 | DECLARE_SWAITQUEUE(wait); |
f7819512 | 2152 | bool waited = false; |
aca6ff29 | 2153 | u64 block_ns; |
f7819512 PB |
2154 | |
2155 | start = cur = ktime_get(); | |
19020f8a WL |
2156 | if (vcpu->halt_poll_ns) { |
2157 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); | |
f95ef0cd | 2158 | |
62bea5bf | 2159 | ++vcpu->stat.halt_attempted_poll; |
f7819512 PB |
2160 | do { |
2161 | /* | |
2162 | * This sets KVM_REQ_UNHALT if an interrupt | |
2163 | * arrives. | |
2164 | */ | |
2165 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
2166 | ++vcpu->stat.halt_successful_poll; | |
3491caf2 CB |
2167 | if (!vcpu_valid_wakeup(vcpu)) |
2168 | ++vcpu->stat.halt_poll_invalid; | |
f7819512 PB |
2169 | goto out; |
2170 | } | |
2171 | cur = ktime_get(); | |
2172 | } while (single_task_running() && ktime_before(cur, stop)); | |
2173 | } | |
e5c239cf | 2174 | |
3217f7c2 CD |
2175 | kvm_arch_vcpu_blocking(vcpu); |
2176 | ||
e5c239cf | 2177 | for (;;) { |
8577370f | 2178 | prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
e5c239cf | 2179 | |
f7819512 | 2180 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
2181 | break; |
2182 | ||
f7819512 | 2183 | waited = true; |
b6958ce4 | 2184 | schedule(); |
b6958ce4 | 2185 | } |
d3bef15f | 2186 | |
8577370f | 2187 | finish_swait(&vcpu->wq, &wait); |
f7819512 PB |
2188 | cur = ktime_get(); |
2189 | ||
3217f7c2 | 2190 | kvm_arch_vcpu_unblocking(vcpu); |
f7819512 | 2191 | out: |
aca6ff29 WL |
2192 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
2193 | ||
2086d320 CB |
2194 | if (!vcpu_valid_wakeup(vcpu)) |
2195 | shrink_halt_poll_ns(vcpu); | |
2196 | else if (halt_poll_ns) { | |
aca6ff29 WL |
2197 | if (block_ns <= vcpu->halt_poll_ns) |
2198 | ; | |
2199 | /* we had a long block, shrink polling */ | |
2086d320 | 2200 | else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns) |
aca6ff29 WL |
2201 | shrink_halt_poll_ns(vcpu); |
2202 | /* we had a short halt and our poll time is too small */ | |
2203 | else if (vcpu->halt_poll_ns < halt_poll_ns && | |
2204 | block_ns < halt_poll_ns) | |
2205 | grow_halt_poll_ns(vcpu); | |
edb9272f WL |
2206 | } else |
2207 | vcpu->halt_poll_ns = 0; | |
aca6ff29 | 2208 | |
3491caf2 CB |
2209 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
2210 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 2211 | } |
2ba9f0d8 | 2212 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 2213 | |
8c84780d | 2214 | #ifndef CONFIG_S390 |
dd1a4cc1 | 2215 | void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2216 | { |
8577370f | 2217 | struct swait_queue_head *wqp; |
b6d33834 CD |
2218 | |
2219 | wqp = kvm_arch_vcpu_wq(vcpu); | |
8577370f MT |
2220 | if (swait_active(wqp)) { |
2221 | swake_up(wqp); | |
b6d33834 CD |
2222 | ++vcpu->stat.halt_wakeup; |
2223 | } | |
2224 | ||
dd1a4cc1 RK |
2225 | } |
2226 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2227 | ||
2228 | /* | |
2229 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2230 | */ | |
2231 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2232 | { | |
2233 | int me; | |
2234 | int cpu = vcpu->cpu; | |
2235 | ||
2236 | kvm_vcpu_wake_up(vcpu); | |
b6d33834 CD |
2237 | me = get_cpu(); |
2238 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2239 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2240 | smp_send_reschedule(cpu); | |
2241 | put_cpu(); | |
2242 | } | |
a20ed54d | 2243 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 2244 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2245 | |
fa93384f | 2246 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2247 | { |
2248 | struct pid *pid; | |
2249 | struct task_struct *task = NULL; | |
fa93384f | 2250 | int ret = 0; |
41628d33 KW |
2251 | |
2252 | rcu_read_lock(); | |
2253 | pid = rcu_dereference(target->pid); | |
2254 | if (pid) | |
27fbe64b | 2255 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2256 | rcu_read_unlock(); |
2257 | if (!task) | |
c45c528e | 2258 | return ret; |
c45c528e | 2259 | ret = yield_to(task, 1); |
41628d33 | 2260 | put_task_struct(task); |
c45c528e R |
2261 | |
2262 | return ret; | |
41628d33 KW |
2263 | } |
2264 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2265 | ||
06e48c51 R |
2266 | /* |
2267 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2268 | * Most eligible candidate to yield is decided by following heuristics: | |
2269 | * | |
2270 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2271 | * (preempted lock holder), indicated by @in_spin_loop. | |
2272 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2273 | * | |
2274 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2275 | * chance last time (mostly it has become eligible now since we have probably | |
2276 | * yielded to lockholder in last iteration. This is done by toggling | |
2277 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2278 | * | |
2279 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2280 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2281 | * burning. Giving priority for a potential lock-holder increases lock | |
2282 | * progress. | |
2283 | * | |
2284 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2285 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2286 | * and continue with next VCPU and so on. | |
2287 | */ | |
7940876e | 2288 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2289 | { |
4a55dd72 | 2290 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2291 | bool eligible; |
2292 | ||
2293 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2294 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2295 | |
2296 | if (vcpu->spin_loop.in_spin_loop) | |
2297 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2298 | ||
2299 | return eligible; | |
4a55dd72 SW |
2300 | #else |
2301 | return true; | |
06e48c51 | 2302 | #endif |
4a55dd72 | 2303 | } |
c45c528e | 2304 | |
217ece61 | 2305 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 2306 | { |
217ece61 RR |
2307 | struct kvm *kvm = me->kvm; |
2308 | struct kvm_vcpu *vcpu; | |
2309 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2310 | int yielded = 0; | |
c45c528e | 2311 | int try = 3; |
217ece61 RR |
2312 | int pass; |
2313 | int i; | |
d255f4f2 | 2314 | |
4c088493 | 2315 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2316 | /* |
2317 | * We boost the priority of a VCPU that is runnable but not | |
2318 | * currently running, because it got preempted by something | |
2319 | * else and called schedule in __vcpu_run. Hopefully that | |
2320 | * VCPU is holding the lock that we need and will release it. | |
2321 | * We approximate round-robin by starting at the last boosted VCPU. | |
2322 | */ | |
c45c528e | 2323 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2324 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2325 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2326 | i = last_boosted_vcpu; |
2327 | continue; | |
2328 | } else if (pass && i > last_boosted_vcpu) | |
2329 | break; | |
7bc7ae25 R |
2330 | if (!ACCESS_ONCE(vcpu->preempted)) |
2331 | continue; | |
217ece61 RR |
2332 | if (vcpu == me) |
2333 | continue; | |
8577370f | 2334 | if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 2335 | continue; |
06e48c51 R |
2336 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2337 | continue; | |
c45c528e R |
2338 | |
2339 | yielded = kvm_vcpu_yield_to(vcpu); | |
2340 | if (yielded > 0) { | |
217ece61 | 2341 | kvm->last_boosted_vcpu = i; |
217ece61 | 2342 | break; |
c45c528e R |
2343 | } else if (yielded < 0) { |
2344 | try--; | |
2345 | if (!try) | |
2346 | break; | |
217ece61 | 2347 | } |
217ece61 RR |
2348 | } |
2349 | } | |
4c088493 | 2350 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2351 | |
2352 | /* Ensure vcpu is not eligible during next spinloop */ | |
2353 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2354 | } |
2355 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2356 | ||
11bac800 | 2357 | static int kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 2358 | { |
11bac800 | 2359 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
2360 | struct page *page; |
2361 | ||
e4a533a4 | 2362 | if (vmf->pgoff == 0) |
039576c0 | 2363 | page = virt_to_page(vcpu->run); |
09566765 | 2364 | #ifdef CONFIG_X86 |
e4a533a4 | 2365 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2366 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
2367 | #endif |
2368 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2369 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
2370 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2371 | #endif |
039576c0 | 2372 | else |
5b1c1493 | 2373 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2374 | get_page(page); |
e4a533a4 | 2375 | vmf->page = page; |
2376 | return 0; | |
9a2bb7f4 AK |
2377 | } |
2378 | ||
f0f37e2f | 2379 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2380 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2381 | }; |
2382 | ||
2383 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2384 | { | |
2385 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2386 | return 0; | |
2387 | } | |
2388 | ||
bccf2150 AK |
2389 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2390 | { | |
2391 | struct kvm_vcpu *vcpu = filp->private_data; | |
2392 | ||
45b5939e | 2393 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
66c0b394 | 2394 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2395 | return 0; |
2396 | } | |
2397 | ||
3d3aab1b | 2398 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2399 | .release = kvm_vcpu_release, |
2400 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2401 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2402 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2403 | #endif | |
9a2bb7f4 | 2404 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2405 | .llseek = noop_llseek, |
bccf2150 AK |
2406 | }; |
2407 | ||
2408 | /* | |
2409 | * Allocates an inode for the vcpu. | |
2410 | */ | |
2411 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2412 | { | |
24009b05 | 2413 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2414 | } |
2415 | ||
45b5939e LC |
2416 | static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
2417 | { | |
2418 | char dir_name[ITOA_MAX_LEN * 2]; | |
2419 | int ret; | |
2420 | ||
2421 | if (!kvm_arch_has_vcpu_debugfs()) | |
2422 | return 0; | |
2423 | ||
2424 | if (!debugfs_initialized()) | |
2425 | return 0; | |
2426 | ||
2427 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
2428 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, | |
2429 | vcpu->kvm->debugfs_dentry); | |
2430 | if (!vcpu->debugfs_dentry) | |
2431 | return -ENOMEM; | |
2432 | ||
2433 | ret = kvm_arch_create_vcpu_debugfs(vcpu); | |
2434 | if (ret < 0) { | |
2435 | debugfs_remove_recursive(vcpu->debugfs_dentry); | |
2436 | return ret; | |
2437 | } | |
2438 | ||
2439 | return 0; | |
2440 | } | |
2441 | ||
c5ea7660 AK |
2442 | /* |
2443 | * Creates some virtual cpus. Good luck creating more than one. | |
2444 | */ | |
73880c80 | 2445 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2446 | { |
2447 | int r; | |
e09fefde | 2448 | struct kvm_vcpu *vcpu; |
c5ea7660 | 2449 | |
0b1b1dfd | 2450 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
2451 | return -EINVAL; |
2452 | ||
6c7caebc PB |
2453 | mutex_lock(&kvm->lock); |
2454 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
2455 | mutex_unlock(&kvm->lock); | |
2456 | return -EINVAL; | |
2457 | } | |
2458 | ||
2459 | kvm->created_vcpus++; | |
2460 | mutex_unlock(&kvm->lock); | |
2461 | ||
73880c80 | 2462 | vcpu = kvm_arch_vcpu_create(kvm, id); |
6c7caebc PB |
2463 | if (IS_ERR(vcpu)) { |
2464 | r = PTR_ERR(vcpu); | |
2465 | goto vcpu_decrement; | |
2466 | } | |
c5ea7660 | 2467 | |
15ad7146 AK |
2468 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2469 | ||
26e5215f AK |
2470 | r = kvm_arch_vcpu_setup(vcpu); |
2471 | if (r) | |
d780592b | 2472 | goto vcpu_destroy; |
26e5215f | 2473 | |
45b5939e LC |
2474 | r = kvm_create_vcpu_debugfs(vcpu); |
2475 | if (r) | |
2476 | goto vcpu_destroy; | |
2477 | ||
11ec2804 | 2478 | mutex_lock(&kvm->lock); |
e09fefde DH |
2479 | if (kvm_get_vcpu_by_id(kvm, id)) { |
2480 | r = -EEXIST; | |
2481 | goto unlock_vcpu_destroy; | |
2482 | } | |
73880c80 GN |
2483 | |
2484 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2485 | |
fb3f0f51 | 2486 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2487 | kvm_get_kvm(kvm); |
bccf2150 | 2488 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2489 | if (r < 0) { |
2490 | kvm_put_kvm(kvm); | |
d780592b | 2491 | goto unlock_vcpu_destroy; |
73880c80 GN |
2492 | } |
2493 | ||
2494 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
dd489240 PB |
2495 | |
2496 | /* | |
2497 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2498 | * before kvm->online_vcpu's incremented value. | |
2499 | */ | |
73880c80 GN |
2500 | smp_wmb(); |
2501 | atomic_inc(&kvm->online_vcpus); | |
2502 | ||
73880c80 | 2503 | mutex_unlock(&kvm->lock); |
42897d86 | 2504 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2505 | return r; |
39c3b86e | 2506 | |
d780592b | 2507 | unlock_vcpu_destroy: |
7d8fece6 | 2508 | mutex_unlock(&kvm->lock); |
45b5939e | 2509 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
d780592b | 2510 | vcpu_destroy: |
d40ccc62 | 2511 | kvm_arch_vcpu_destroy(vcpu); |
6c7caebc PB |
2512 | vcpu_decrement: |
2513 | mutex_lock(&kvm->lock); | |
2514 | kvm->created_vcpus--; | |
2515 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
2516 | return r; |
2517 | } | |
2518 | ||
1961d276 AK |
2519 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2520 | { | |
2521 | if (sigset) { | |
2522 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2523 | vcpu->sigset_active = 1; | |
2524 | vcpu->sigset = *sigset; | |
2525 | } else | |
2526 | vcpu->sigset_active = 0; | |
2527 | return 0; | |
2528 | } | |
2529 | ||
bccf2150 AK |
2530 | static long kvm_vcpu_ioctl(struct file *filp, |
2531 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2532 | { |
bccf2150 | 2533 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2534 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2535 | int r; |
fa3795a7 DH |
2536 | struct kvm_fpu *fpu = NULL; |
2537 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2538 | |
6d4e4c4f AK |
2539 | if (vcpu->kvm->mm != current->mm) |
2540 | return -EIO; | |
2122ff5e | 2541 | |
2ea75be3 DM |
2542 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2543 | return -EINVAL; | |
2544 | ||
2f4d9b54 | 2545 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2546 | /* |
2547 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2548 | * so vcpu_load() would break it. | |
2549 | */ | |
47b43c52 | 2550 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
2122ff5e AK |
2551 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
2552 | #endif | |
2553 | ||
2554 | ||
9fc77441 MT |
2555 | r = vcpu_load(vcpu); |
2556 | if (r) | |
2557 | return r; | |
6aa8b732 | 2558 | switch (ioctl) { |
9a2bb7f4 | 2559 | case KVM_RUN: |
f0fe5108 AK |
2560 | r = -EINVAL; |
2561 | if (arg) | |
2562 | goto out; | |
7a72f7a1 CB |
2563 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
2564 | /* The thread running this VCPU changed. */ | |
2565 | struct pid *oldpid = vcpu->pid; | |
2566 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
f95ef0cd | 2567 | |
7a72f7a1 CB |
2568 | rcu_assign_pointer(vcpu->pid, newpid); |
2569 | if (oldpid) | |
2570 | synchronize_rcu(); | |
2571 | put_pid(oldpid); | |
2572 | } | |
b6c7a5dc | 2573 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2574 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2575 | break; |
6aa8b732 | 2576 | case KVM_GET_REGS: { |
3e4bb3ac | 2577 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2578 | |
3e4bb3ac XZ |
2579 | r = -ENOMEM; |
2580 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2581 | if (!kvm_regs) | |
6aa8b732 | 2582 | goto out; |
3e4bb3ac XZ |
2583 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2584 | if (r) | |
2585 | goto out_free1; | |
6aa8b732 | 2586 | r = -EFAULT; |
3e4bb3ac XZ |
2587 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2588 | goto out_free1; | |
6aa8b732 | 2589 | r = 0; |
3e4bb3ac XZ |
2590 | out_free1: |
2591 | kfree(kvm_regs); | |
6aa8b732 AK |
2592 | break; |
2593 | } | |
2594 | case KVM_SET_REGS: { | |
3e4bb3ac | 2595 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2596 | |
3e4bb3ac | 2597 | r = -ENOMEM; |
ff5c2c03 SL |
2598 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2599 | if (IS_ERR(kvm_regs)) { | |
2600 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2601 | goto out; |
ff5c2c03 | 2602 | } |
3e4bb3ac | 2603 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2604 | kfree(kvm_regs); |
6aa8b732 AK |
2605 | break; |
2606 | } | |
2607 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2608 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2609 | r = -ENOMEM; | |
2610 | if (!kvm_sregs) | |
2611 | goto out; | |
2612 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2613 | if (r) |
2614 | goto out; | |
2615 | r = -EFAULT; | |
fa3795a7 | 2616 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2617 | goto out; |
2618 | r = 0; | |
2619 | break; | |
2620 | } | |
2621 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2622 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2623 | if (IS_ERR(kvm_sregs)) { | |
2624 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2625 | kvm_sregs = NULL; |
6aa8b732 | 2626 | goto out; |
ff5c2c03 | 2627 | } |
fa3795a7 | 2628 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2629 | break; |
2630 | } | |
62d9f0db MT |
2631 | case KVM_GET_MP_STATE: { |
2632 | struct kvm_mp_state mp_state; | |
2633 | ||
2634 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2635 | if (r) | |
2636 | goto out; | |
2637 | r = -EFAULT; | |
893bdbf1 | 2638 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2639 | goto out; |
2640 | r = 0; | |
2641 | break; | |
2642 | } | |
2643 | case KVM_SET_MP_STATE: { | |
2644 | struct kvm_mp_state mp_state; | |
2645 | ||
2646 | r = -EFAULT; | |
893bdbf1 | 2647 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2648 | goto out; |
2649 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2650 | break; |
2651 | } | |
6aa8b732 AK |
2652 | case KVM_TRANSLATE: { |
2653 | struct kvm_translation tr; | |
2654 | ||
2655 | r = -EFAULT; | |
893bdbf1 | 2656 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2657 | goto out; |
8b006791 | 2658 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2659 | if (r) |
2660 | goto out; | |
2661 | r = -EFAULT; | |
893bdbf1 | 2662 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2663 | goto out; |
2664 | r = 0; | |
2665 | break; | |
2666 | } | |
d0bfb940 JK |
2667 | case KVM_SET_GUEST_DEBUG: { |
2668 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2669 | |
2670 | r = -EFAULT; | |
893bdbf1 | 2671 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2672 | goto out; |
d0bfb940 | 2673 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2674 | break; |
2675 | } | |
1961d276 AK |
2676 | case KVM_SET_SIGNAL_MASK: { |
2677 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2678 | struct kvm_signal_mask kvm_sigmask; | |
2679 | sigset_t sigset, *p; | |
2680 | ||
2681 | p = NULL; | |
2682 | if (argp) { | |
2683 | r = -EFAULT; | |
2684 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2685 | sizeof(kvm_sigmask))) |
1961d276 AK |
2686 | goto out; |
2687 | r = -EINVAL; | |
893bdbf1 | 2688 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2689 | goto out; |
2690 | r = -EFAULT; | |
2691 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2692 | sizeof(sigset))) |
1961d276 AK |
2693 | goto out; |
2694 | p = &sigset; | |
2695 | } | |
376d41ff | 2696 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2697 | break; |
2698 | } | |
b8836737 | 2699 | case KVM_GET_FPU: { |
fa3795a7 DH |
2700 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2701 | r = -ENOMEM; | |
2702 | if (!fpu) | |
2703 | goto out; | |
2704 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2705 | if (r) |
2706 | goto out; | |
2707 | r = -EFAULT; | |
fa3795a7 | 2708 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2709 | goto out; |
2710 | r = 0; | |
2711 | break; | |
2712 | } | |
2713 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2714 | fpu = memdup_user(argp, sizeof(*fpu)); |
2715 | if (IS_ERR(fpu)) { | |
2716 | r = PTR_ERR(fpu); | |
18595411 | 2717 | fpu = NULL; |
b8836737 | 2718 | goto out; |
ff5c2c03 | 2719 | } |
fa3795a7 | 2720 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2721 | break; |
2722 | } | |
bccf2150 | 2723 | default: |
313a3dc7 | 2724 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2725 | } |
2726 | out: | |
2122ff5e | 2727 | vcpu_put(vcpu); |
fa3795a7 DH |
2728 | kfree(fpu); |
2729 | kfree(kvm_sregs); | |
bccf2150 AK |
2730 | return r; |
2731 | } | |
2732 | ||
de8e5d74 | 2733 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2734 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2735 | unsigned int ioctl, unsigned long arg) | |
2736 | { | |
2737 | struct kvm_vcpu *vcpu = filp->private_data; | |
2738 | void __user *argp = compat_ptr(arg); | |
2739 | int r; | |
2740 | ||
2741 | if (vcpu->kvm->mm != current->mm) | |
2742 | return -EIO; | |
2743 | ||
2744 | switch (ioctl) { | |
2745 | case KVM_SET_SIGNAL_MASK: { | |
2746 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2747 | struct kvm_signal_mask kvm_sigmask; | |
2748 | compat_sigset_t csigset; | |
2749 | sigset_t sigset; | |
2750 | ||
2751 | if (argp) { | |
2752 | r = -EFAULT; | |
2753 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2754 | sizeof(kvm_sigmask))) |
1dda606c AG |
2755 | goto out; |
2756 | r = -EINVAL; | |
893bdbf1 | 2757 | if (kvm_sigmask.len != sizeof(csigset)) |
1dda606c AG |
2758 | goto out; |
2759 | r = -EFAULT; | |
2760 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
893bdbf1 | 2761 | sizeof(csigset))) |
1dda606c | 2762 | goto out; |
760a9a30 AC |
2763 | sigset_from_compat(&sigset, &csigset); |
2764 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2765 | } else | |
2766 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2767 | break; |
2768 | } | |
2769 | default: | |
2770 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2771 | } | |
2772 | ||
2773 | out: | |
2774 | return r; | |
2775 | } | |
2776 | #endif | |
2777 | ||
852b6d57 SW |
2778 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2779 | int (*accessor)(struct kvm_device *dev, | |
2780 | struct kvm_device_attr *attr), | |
2781 | unsigned long arg) | |
2782 | { | |
2783 | struct kvm_device_attr attr; | |
2784 | ||
2785 | if (!accessor) | |
2786 | return -EPERM; | |
2787 | ||
2788 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2789 | return -EFAULT; | |
2790 | ||
2791 | return accessor(dev, &attr); | |
2792 | } | |
2793 | ||
2794 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2795 | unsigned long arg) | |
2796 | { | |
2797 | struct kvm_device *dev = filp->private_data; | |
2798 | ||
2799 | switch (ioctl) { | |
2800 | case KVM_SET_DEVICE_ATTR: | |
2801 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2802 | case KVM_GET_DEVICE_ATTR: | |
2803 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2804 | case KVM_HAS_DEVICE_ATTR: | |
2805 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2806 | default: | |
2807 | if (dev->ops->ioctl) | |
2808 | return dev->ops->ioctl(dev, ioctl, arg); | |
2809 | ||
2810 | return -ENOTTY; | |
2811 | } | |
2812 | } | |
2813 | ||
852b6d57 SW |
2814 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2815 | { | |
2816 | struct kvm_device *dev = filp->private_data; | |
2817 | struct kvm *kvm = dev->kvm; | |
2818 | ||
852b6d57 SW |
2819 | kvm_put_kvm(kvm); |
2820 | return 0; | |
2821 | } | |
2822 | ||
2823 | static const struct file_operations kvm_device_fops = { | |
2824 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2825 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2826 | .compat_ioctl = kvm_device_ioctl, |
2827 | #endif | |
852b6d57 SW |
2828 | .release = kvm_device_release, |
2829 | }; | |
2830 | ||
2831 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2832 | { | |
2833 | if (filp->f_op != &kvm_device_fops) | |
2834 | return NULL; | |
2835 | ||
2836 | return filp->private_data; | |
2837 | } | |
2838 | ||
d60eacb0 | 2839 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2840 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2841 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2842 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2843 | #endif |
d60eacb0 | 2844 | |
5975a2e0 | 2845 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2846 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2847 | #endif |
d60eacb0 WD |
2848 | }; |
2849 | ||
2850 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2851 | { | |
2852 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2853 | return -ENOSPC; | |
2854 | ||
2855 | if (kvm_device_ops_table[type] != NULL) | |
2856 | return -EEXIST; | |
2857 | ||
2858 | kvm_device_ops_table[type] = ops; | |
2859 | return 0; | |
2860 | } | |
2861 | ||
571ee1b6 WL |
2862 | void kvm_unregister_device_ops(u32 type) |
2863 | { | |
2864 | if (kvm_device_ops_table[type] != NULL) | |
2865 | kvm_device_ops_table[type] = NULL; | |
2866 | } | |
2867 | ||
852b6d57 SW |
2868 | static int kvm_ioctl_create_device(struct kvm *kvm, |
2869 | struct kvm_create_device *cd) | |
2870 | { | |
2871 | struct kvm_device_ops *ops = NULL; | |
2872 | struct kvm_device *dev; | |
2873 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2874 | int ret; | |
2875 | ||
d60eacb0 WD |
2876 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
2877 | return -ENODEV; | |
2878 | ||
2879 | ops = kvm_device_ops_table[cd->type]; | |
2880 | if (ops == NULL) | |
852b6d57 | 2881 | return -ENODEV; |
852b6d57 SW |
2882 | |
2883 | if (test) | |
2884 | return 0; | |
2885 | ||
2886 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2887 | if (!dev) | |
2888 | return -ENOMEM; | |
2889 | ||
2890 | dev->ops = ops; | |
2891 | dev->kvm = kvm; | |
852b6d57 | 2892 | |
a28ebea2 | 2893 | mutex_lock(&kvm->lock); |
852b6d57 SW |
2894 | ret = ops->create(dev, cd->type); |
2895 | if (ret < 0) { | |
a28ebea2 | 2896 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
2897 | kfree(dev); |
2898 | return ret; | |
2899 | } | |
a28ebea2 CD |
2900 | list_add(&dev->vm_node, &kvm->devices); |
2901 | mutex_unlock(&kvm->lock); | |
852b6d57 | 2902 | |
023e9fdd CD |
2903 | if (ops->init) |
2904 | ops->init(dev); | |
2905 | ||
24009b05 | 2906 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 2907 | if (ret < 0) { |
a28ebea2 CD |
2908 | mutex_lock(&kvm->lock); |
2909 | list_del(&dev->vm_node); | |
2910 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 2911 | ops->destroy(dev); |
852b6d57 SW |
2912 | return ret; |
2913 | } | |
2914 | ||
2915 | kvm_get_kvm(kvm); | |
2916 | cd->fd = ret; | |
2917 | return 0; | |
2918 | } | |
2919 | ||
92b591a4 AG |
2920 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2921 | { | |
2922 | switch (arg) { | |
2923 | case KVM_CAP_USER_MEMORY: | |
2924 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2925 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
2926 | case KVM_CAP_INTERNAL_ERROR_DATA: |
2927 | #ifdef CONFIG_HAVE_KVM_MSI | |
2928 | case KVM_CAP_SIGNAL_MSI: | |
2929 | #endif | |
297e2105 | 2930 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 2931 | case KVM_CAP_IRQFD: |
92b591a4 AG |
2932 | case KVM_CAP_IRQFD_RESAMPLE: |
2933 | #endif | |
e9ea5069 | 2934 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 AG |
2935 | case KVM_CAP_CHECK_EXTENSION_VM: |
2936 | return 1; | |
2937 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2938 | case KVM_CAP_IRQ_ROUTING: | |
2939 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
2940 | #endif |
2941 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
2942 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
2943 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 2944 | #endif |
0b1b1dfd GK |
2945 | case KVM_CAP_MAX_VCPU_ID: |
2946 | return KVM_MAX_VCPU_ID; | |
92b591a4 AG |
2947 | default: |
2948 | break; | |
2949 | } | |
2950 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2951 | } | |
2952 | ||
bccf2150 AK |
2953 | static long kvm_vm_ioctl(struct file *filp, |
2954 | unsigned int ioctl, unsigned long arg) | |
2955 | { | |
2956 | struct kvm *kvm = filp->private_data; | |
2957 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2958 | int r; |
bccf2150 | 2959 | |
6d4e4c4f AK |
2960 | if (kvm->mm != current->mm) |
2961 | return -EIO; | |
bccf2150 AK |
2962 | switch (ioctl) { |
2963 | case KVM_CREATE_VCPU: | |
2964 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2965 | break; |
6fc138d2 IE |
2966 | case KVM_SET_USER_MEMORY_REGION: { |
2967 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2968 | ||
2969 | r = -EFAULT; | |
2970 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 2971 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
2972 | goto out; |
2973 | ||
47ae31e2 | 2974 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2975 | break; |
2976 | } | |
2977 | case KVM_GET_DIRTY_LOG: { | |
2978 | struct kvm_dirty_log log; | |
2979 | ||
2980 | r = -EFAULT; | |
893bdbf1 | 2981 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 2982 | goto out; |
2c6f5df9 | 2983 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2984 | break; |
2985 | } | |
5f94c174 LV |
2986 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2987 | case KVM_REGISTER_COALESCED_MMIO: { | |
2988 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2989 | |
5f94c174 | 2990 | r = -EFAULT; |
893bdbf1 | 2991 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2992 | goto out; |
5f94c174 | 2993 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2994 | break; |
2995 | } | |
2996 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2997 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2998 | |
5f94c174 | 2999 | r = -EFAULT; |
893bdbf1 | 3000 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3001 | goto out; |
5f94c174 | 3002 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3003 | break; |
3004 | } | |
3005 | #endif | |
721eecbf GH |
3006 | case KVM_IRQFD: { |
3007 | struct kvm_irqfd data; | |
3008 | ||
3009 | r = -EFAULT; | |
893bdbf1 | 3010 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3011 | goto out; |
d4db2935 | 3012 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3013 | break; |
3014 | } | |
d34e6b17 GH |
3015 | case KVM_IOEVENTFD: { |
3016 | struct kvm_ioeventfd data; | |
3017 | ||
3018 | r = -EFAULT; | |
893bdbf1 | 3019 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3020 | goto out; |
3021 | r = kvm_ioeventfd(kvm, &data); | |
3022 | break; | |
3023 | } | |
07975ad3 JK |
3024 | #ifdef CONFIG_HAVE_KVM_MSI |
3025 | case KVM_SIGNAL_MSI: { | |
3026 | struct kvm_msi msi; | |
3027 | ||
3028 | r = -EFAULT; | |
893bdbf1 | 3029 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3030 | goto out; |
3031 | r = kvm_send_userspace_msi(kvm, &msi); | |
3032 | break; | |
3033 | } | |
23d43cf9 CD |
3034 | #endif |
3035 | #ifdef __KVM_HAVE_IRQ_LINE | |
3036 | case KVM_IRQ_LINE_STATUS: | |
3037 | case KVM_IRQ_LINE: { | |
3038 | struct kvm_irq_level irq_event; | |
3039 | ||
3040 | r = -EFAULT; | |
893bdbf1 | 3041 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3042 | goto out; |
3043 | ||
aa2fbe6d YZ |
3044 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3045 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3046 | if (r) |
3047 | goto out; | |
3048 | ||
3049 | r = -EFAULT; | |
3050 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3051 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3052 | goto out; |
3053 | } | |
3054 | ||
3055 | r = 0; | |
3056 | break; | |
3057 | } | |
73880c80 | 3058 | #endif |
aa8d5944 AG |
3059 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3060 | case KVM_SET_GSI_ROUTING: { | |
3061 | struct kvm_irq_routing routing; | |
3062 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3063 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3064 | |
3065 | r = -EFAULT; | |
3066 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3067 | goto out; | |
3068 | r = -EINVAL; | |
caf1ff26 | 3069 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3070 | goto out; |
3071 | if (routing.flags) | |
3072 | goto out; | |
f8c1b85b PB |
3073 | if (routing.nr) { |
3074 | r = -ENOMEM; | |
3075 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
3076 | if (!entries) | |
3077 | goto out; | |
3078 | r = -EFAULT; | |
3079 | urouting = argp; | |
3080 | if (copy_from_user(entries, urouting->entries, | |
3081 | routing.nr * sizeof(*entries))) | |
3082 | goto out_free_irq_routing; | |
3083 | } | |
aa8d5944 AG |
3084 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3085 | routing.flags); | |
a642a175 | 3086 | out_free_irq_routing: |
aa8d5944 AG |
3087 | vfree(entries); |
3088 | break; | |
3089 | } | |
3090 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3091 | case KVM_CREATE_DEVICE: { |
3092 | struct kvm_create_device cd; | |
3093 | ||
3094 | r = -EFAULT; | |
3095 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3096 | goto out; | |
3097 | ||
3098 | r = kvm_ioctl_create_device(kvm, &cd); | |
3099 | if (r) | |
3100 | goto out; | |
3101 | ||
3102 | r = -EFAULT; | |
3103 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3104 | goto out; | |
3105 | ||
3106 | r = 0; | |
3107 | break; | |
3108 | } | |
92b591a4 AG |
3109 | case KVM_CHECK_EXTENSION: |
3110 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3111 | break; | |
f17abe9a | 3112 | default: |
1fe779f8 | 3113 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3114 | } |
3115 | out: | |
3116 | return r; | |
3117 | } | |
3118 | ||
de8e5d74 | 3119 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3120 | struct compat_kvm_dirty_log { |
3121 | __u32 slot; | |
3122 | __u32 padding1; | |
3123 | union { | |
3124 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3125 | __u64 padding2; | |
3126 | }; | |
3127 | }; | |
3128 | ||
3129 | static long kvm_vm_compat_ioctl(struct file *filp, | |
3130 | unsigned int ioctl, unsigned long arg) | |
3131 | { | |
3132 | struct kvm *kvm = filp->private_data; | |
3133 | int r; | |
3134 | ||
3135 | if (kvm->mm != current->mm) | |
3136 | return -EIO; | |
3137 | switch (ioctl) { | |
3138 | case KVM_GET_DIRTY_LOG: { | |
3139 | struct compat_kvm_dirty_log compat_log; | |
3140 | struct kvm_dirty_log log; | |
3141 | ||
6ff5894c AB |
3142 | if (copy_from_user(&compat_log, (void __user *)arg, |
3143 | sizeof(compat_log))) | |
f6a3b168 | 3144 | return -EFAULT; |
6ff5894c AB |
3145 | log.slot = compat_log.slot; |
3146 | log.padding1 = compat_log.padding1; | |
3147 | log.padding2 = compat_log.padding2; | |
3148 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3149 | ||
3150 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
3151 | break; |
3152 | } | |
3153 | default: | |
3154 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
3155 | } | |
6ff5894c AB |
3156 | return r; |
3157 | } | |
3158 | #endif | |
3159 | ||
3d3aab1b | 3160 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
3161 | .release = kvm_vm_release, |
3162 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 3163 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3164 | .compat_ioctl = kvm_vm_compat_ioctl, |
3165 | #endif | |
6038f373 | 3166 | .llseek = noop_llseek, |
f17abe9a AK |
3167 | }; |
3168 | ||
e08b9637 | 3169 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 3170 | { |
aac87636 | 3171 | int r; |
f17abe9a | 3172 | struct kvm *kvm; |
506cfba9 | 3173 | struct file *file; |
f17abe9a | 3174 | |
e08b9637 | 3175 | kvm = kvm_create_vm(type); |
d6d28168 AK |
3176 | if (IS_ERR(kvm)) |
3177 | return PTR_ERR(kvm); | |
6ce5a090 TY |
3178 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
3179 | r = kvm_coalesced_mmio_init(kvm); | |
3180 | if (r < 0) { | |
3181 | kvm_put_kvm(kvm); | |
3182 | return r; | |
3183 | } | |
3184 | #endif | |
506cfba9 | 3185 | r = get_unused_fd_flags(O_CLOEXEC); |
536a6f88 | 3186 | if (r < 0) { |
66c0b394 | 3187 | kvm_put_kvm(kvm); |
536a6f88 JF |
3188 | return r; |
3189 | } | |
506cfba9 AV |
3190 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
3191 | if (IS_ERR(file)) { | |
3192 | put_unused_fd(r); | |
3193 | kvm_put_kvm(kvm); | |
3194 | return PTR_ERR(file); | |
3195 | } | |
536a6f88 JF |
3196 | |
3197 | if (kvm_create_vm_debugfs(kvm, r) < 0) { | |
506cfba9 AV |
3198 | put_unused_fd(r); |
3199 | fput(file); | |
536a6f88 JF |
3200 | return -ENOMEM; |
3201 | } | |
f17abe9a | 3202 | |
506cfba9 | 3203 | fd_install(r, file); |
aac87636 | 3204 | return r; |
f17abe9a AK |
3205 | } |
3206 | ||
3207 | static long kvm_dev_ioctl(struct file *filp, | |
3208 | unsigned int ioctl, unsigned long arg) | |
3209 | { | |
07c45a36 | 3210 | long r = -EINVAL; |
f17abe9a AK |
3211 | |
3212 | switch (ioctl) { | |
3213 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
3214 | if (arg) |
3215 | goto out; | |
f17abe9a AK |
3216 | r = KVM_API_VERSION; |
3217 | break; | |
3218 | case KVM_CREATE_VM: | |
e08b9637 | 3219 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 3220 | break; |
018d00d2 | 3221 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 3222 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 3223 | break; |
07c45a36 | 3224 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
3225 | if (arg) |
3226 | goto out; | |
adb1ff46 AK |
3227 | r = PAGE_SIZE; /* struct kvm_run */ |
3228 | #ifdef CONFIG_X86 | |
3229 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
3230 | #endif |
3231 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
3232 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 3233 | #endif |
07c45a36 | 3234 | break; |
d4c9ff2d FEL |
3235 | case KVM_TRACE_ENABLE: |
3236 | case KVM_TRACE_PAUSE: | |
3237 | case KVM_TRACE_DISABLE: | |
2023a29c | 3238 | r = -EOPNOTSUPP; |
d4c9ff2d | 3239 | break; |
6aa8b732 | 3240 | default: |
043405e1 | 3241 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
3242 | } |
3243 | out: | |
3244 | return r; | |
3245 | } | |
3246 | ||
6aa8b732 | 3247 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
3248 | .unlocked_ioctl = kvm_dev_ioctl, |
3249 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 3250 | .llseek = noop_llseek, |
6aa8b732 AK |
3251 | }; |
3252 | ||
3253 | static struct miscdevice kvm_dev = { | |
bbe4432e | 3254 | KVM_MINOR, |
6aa8b732 AK |
3255 | "kvm", |
3256 | &kvm_chardev_ops, | |
3257 | }; | |
3258 | ||
75b7127c | 3259 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
3260 | { |
3261 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 3262 | int r; |
1b6c0168 | 3263 | |
7f59f492 | 3264 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3265 | return; |
10474ae8 | 3266 | |
7f59f492 | 3267 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 3268 | |
13a34e06 | 3269 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
3270 | |
3271 | if (r) { | |
3272 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
3273 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 3274 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 3275 | } |
1b6c0168 AK |
3276 | } |
3277 | ||
8c18b2d2 | 3278 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 3279 | { |
4a937f96 | 3280 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3281 | if (kvm_usage_count) |
3282 | hardware_enable_nolock(NULL); | |
4a937f96 | 3283 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3284 | return 0; |
75b7127c TY |
3285 | } |
3286 | ||
3287 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
3288 | { |
3289 | int cpu = raw_smp_processor_id(); | |
3290 | ||
7f59f492 | 3291 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3292 | return; |
7f59f492 | 3293 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 3294 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
3295 | } |
3296 | ||
8c18b2d2 | 3297 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 3298 | { |
4a937f96 | 3299 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3300 | if (kvm_usage_count) |
3301 | hardware_disable_nolock(NULL); | |
4a937f96 | 3302 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3303 | return 0; |
75b7127c TY |
3304 | } |
3305 | ||
10474ae8 AG |
3306 | static void hardware_disable_all_nolock(void) |
3307 | { | |
3308 | BUG_ON(!kvm_usage_count); | |
3309 | ||
3310 | kvm_usage_count--; | |
3311 | if (!kvm_usage_count) | |
75b7127c | 3312 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
3313 | } |
3314 | ||
3315 | static void hardware_disable_all(void) | |
3316 | { | |
4a937f96 | 3317 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3318 | hardware_disable_all_nolock(); |
4a937f96 | 3319 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3320 | } |
3321 | ||
3322 | static int hardware_enable_all(void) | |
3323 | { | |
3324 | int r = 0; | |
3325 | ||
4a937f96 | 3326 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3327 | |
3328 | kvm_usage_count++; | |
3329 | if (kvm_usage_count == 1) { | |
3330 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3331 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3332 | |
3333 | if (atomic_read(&hardware_enable_failed)) { | |
3334 | hardware_disable_all_nolock(); | |
3335 | r = -EBUSY; | |
3336 | } | |
3337 | } | |
3338 | ||
4a937f96 | 3339 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3340 | |
3341 | return r; | |
3342 | } | |
3343 | ||
9a2b85c6 | 3344 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3345 | void *v) |
9a2b85c6 | 3346 | { |
8e1c1815 SY |
3347 | /* |
3348 | * Some (well, at least mine) BIOSes hang on reboot if | |
3349 | * in vmx root mode. | |
3350 | * | |
3351 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3352 | */ | |
1170adc6 | 3353 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3354 | kvm_rebooting = true; |
75b7127c | 3355 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3356 | return NOTIFY_OK; |
3357 | } | |
3358 | ||
3359 | static struct notifier_block kvm_reboot_notifier = { | |
3360 | .notifier_call = kvm_reboot, | |
3361 | .priority = 0, | |
3362 | }; | |
3363 | ||
e93f8a0f | 3364 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3365 | { |
3366 | int i; | |
3367 | ||
3368 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3369 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3370 | |
3371 | kvm_iodevice_destructor(pos); | |
3372 | } | |
e93f8a0f | 3373 | kfree(bus); |
2eeb2e94 GH |
3374 | } |
3375 | ||
c21fbff1 | 3376 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3377 | const struct kvm_io_range *r2) |
743eeb0b | 3378 | { |
8f4216c7 JW |
3379 | gpa_t addr1 = r1->addr; |
3380 | gpa_t addr2 = r2->addr; | |
3381 | ||
3382 | if (addr1 < addr2) | |
743eeb0b | 3383 | return -1; |
8f4216c7 JW |
3384 | |
3385 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
3386 | * accept any overlapping write. Any order is acceptable for | |
3387 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
3388 | * we process all of them. | |
3389 | */ | |
3390 | if (r2->len) { | |
3391 | addr1 += r1->len; | |
3392 | addr2 += r2->len; | |
3393 | } | |
3394 | ||
3395 | if (addr1 > addr2) | |
743eeb0b | 3396 | return 1; |
8f4216c7 | 3397 | |
743eeb0b SL |
3398 | return 0; |
3399 | } | |
3400 | ||
a343c9b7 PB |
3401 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3402 | { | |
c21fbff1 | 3403 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3404 | } |
3405 | ||
39369f7a | 3406 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
3407 | gpa_t addr, int len) |
3408 | { | |
743eeb0b SL |
3409 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
3410 | .addr = addr, | |
3411 | .len = len, | |
3412 | .dev = dev, | |
3413 | }; | |
3414 | ||
3415 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
3416 | kvm_io_bus_sort_cmp, NULL); | |
3417 | ||
3418 | return 0; | |
3419 | } | |
3420 | ||
39369f7a | 3421 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3422 | gpa_t addr, int len) |
3423 | { | |
3424 | struct kvm_io_range *range, key; | |
3425 | int off; | |
3426 | ||
3427 | key = (struct kvm_io_range) { | |
3428 | .addr = addr, | |
3429 | .len = len, | |
3430 | }; | |
3431 | ||
3432 | range = bsearch(&key, bus->range, bus->dev_count, | |
3433 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3434 | if (range == NULL) | |
3435 | return -ENOENT; | |
3436 | ||
3437 | off = range - bus->range; | |
3438 | ||
c21fbff1 | 3439 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3440 | off--; |
3441 | ||
3442 | return off; | |
3443 | } | |
3444 | ||
e32edf4f | 3445 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3446 | struct kvm_io_range *range, const void *val) |
3447 | { | |
3448 | int idx; | |
3449 | ||
3450 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3451 | if (idx < 0) | |
3452 | return -EOPNOTSUPP; | |
3453 | ||
3454 | while (idx < bus->dev_count && | |
c21fbff1 | 3455 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3456 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3457 | range->len, val)) |
3458 | return idx; | |
3459 | idx++; | |
3460 | } | |
3461 | ||
3462 | return -EOPNOTSUPP; | |
3463 | } | |
3464 | ||
bda9020e | 3465 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3466 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3467 | int len, const void *val) |
2eeb2e94 | 3468 | { |
90d83dc3 | 3469 | struct kvm_io_bus *bus; |
743eeb0b | 3470 | struct kvm_io_range range; |
126a5af5 | 3471 | int r; |
743eeb0b SL |
3472 | |
3473 | range = (struct kvm_io_range) { | |
3474 | .addr = addr, | |
3475 | .len = len, | |
3476 | }; | |
90d83dc3 | 3477 | |
e32edf4f NN |
3478 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3479 | r = __kvm_io_bus_write(vcpu, bus, &range, val); | |
126a5af5 CH |
3480 | return r < 0 ? r : 0; |
3481 | } | |
3482 | ||
3483 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3484 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3485 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3486 | { |
3487 | struct kvm_io_bus *bus; | |
3488 | struct kvm_io_range range; | |
3489 | ||
3490 | range = (struct kvm_io_range) { | |
3491 | .addr = addr, | |
3492 | .len = len, | |
3493 | }; | |
3494 | ||
e32edf4f | 3495 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
126a5af5 CH |
3496 | |
3497 | /* First try the device referenced by cookie. */ | |
3498 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3499 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3500 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3501 | val)) |
3502 | return cookie; | |
3503 | ||
3504 | /* | |
3505 | * cookie contained garbage; fall back to search and return the | |
3506 | * correct cookie value. | |
3507 | */ | |
e32edf4f | 3508 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3509 | } |
3510 | ||
e32edf4f NN |
3511 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3512 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3513 | { |
3514 | int idx; | |
3515 | ||
3516 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3517 | if (idx < 0) |
3518 | return -EOPNOTSUPP; | |
3519 | ||
3520 | while (idx < bus->dev_count && | |
c21fbff1 | 3521 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3522 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3523 | range->len, val)) |
3524 | return idx; | |
743eeb0b SL |
3525 | idx++; |
3526 | } | |
3527 | ||
bda9020e MT |
3528 | return -EOPNOTSUPP; |
3529 | } | |
68c3b4d1 | 3530 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3531 | |
bda9020e | 3532 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3533 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3534 | int len, void *val) |
bda9020e | 3535 | { |
90d83dc3 | 3536 | struct kvm_io_bus *bus; |
743eeb0b | 3537 | struct kvm_io_range range; |
126a5af5 | 3538 | int r; |
743eeb0b SL |
3539 | |
3540 | range = (struct kvm_io_range) { | |
3541 | .addr = addr, | |
3542 | .len = len, | |
3543 | }; | |
e93f8a0f | 3544 | |
e32edf4f NN |
3545 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3546 | r = __kvm_io_bus_read(vcpu, bus, &range, val); | |
126a5af5 CH |
3547 | return r < 0 ? r : 0; |
3548 | } | |
743eeb0b | 3549 | |
2eeb2e94 | 3550 | |
79fac95e | 3551 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3552 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3553 | int len, struct kvm_io_device *dev) | |
6c474694 | 3554 | { |
e93f8a0f | 3555 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3556 | |
e93f8a0f | 3557 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3558 | /* exclude ioeventfd which is limited by maximum fd */ |
3559 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3560 | return -ENOSPC; |
2eeb2e94 | 3561 | |
d3febddd | 3562 | new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
a1300716 | 3563 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
e93f8a0f MT |
3564 | if (!new_bus) |
3565 | return -ENOMEM; | |
a1300716 AK |
3566 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3567 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3568 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3569 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3570 | synchronize_srcu_expedited(&kvm->srcu); | |
3571 | kfree(bus); | |
090b7aff GH |
3572 | |
3573 | return 0; | |
3574 | } | |
3575 | ||
79fac95e | 3576 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3577 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3578 | struct kvm_io_device *dev) | |
090b7aff | 3579 | { |
e93f8a0f MT |
3580 | int i, r; |
3581 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3582 | |
cdfca7b3 | 3583 | bus = kvm->buses[bus_idx]; |
df630b8c PX |
3584 | |
3585 | /* | |
3586 | * It's possible the bus being released before hand. If so, | |
3587 | * we're done here. | |
3588 | */ | |
3589 | if (!bus) | |
3590 | return 0; | |
3591 | ||
e93f8a0f | 3592 | r = -ENOENT; |
a1300716 AK |
3593 | for (i = 0; i < bus->dev_count; i++) |
3594 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3595 | r = 0; |
090b7aff GH |
3596 | break; |
3597 | } | |
e93f8a0f | 3598 | |
a1300716 | 3599 | if (r) |
e93f8a0f | 3600 | return r; |
a1300716 | 3601 | |
d3febddd | 3602 | new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) * |
a1300716 AK |
3603 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
3604 | if (!new_bus) | |
3605 | return -ENOMEM; | |
3606 | ||
3607 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3608 | new_bus->dev_count--; | |
3609 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3610 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3611 | |
3612 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3613 | synchronize_srcu_expedited(&kvm->srcu); | |
3614 | kfree(bus); | |
3615 | return r; | |
2eeb2e94 GH |
3616 | } |
3617 | ||
8a39d006 AP |
3618 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3619 | gpa_t addr) | |
3620 | { | |
3621 | struct kvm_io_bus *bus; | |
3622 | int dev_idx, srcu_idx; | |
3623 | struct kvm_io_device *iodev = NULL; | |
3624 | ||
3625 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
3626 | ||
3627 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
3628 | ||
3629 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
3630 | if (dev_idx < 0) | |
3631 | goto out_unlock; | |
3632 | ||
3633 | iodev = bus->range[dev_idx].dev; | |
3634 | ||
3635 | out_unlock: | |
3636 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
3637 | ||
3638 | return iodev; | |
3639 | } | |
3640 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
3641 | ||
536a6f88 JF |
3642 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
3643 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
3644 | const char *fmt) | |
3645 | { | |
3646 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3647 | inode->i_private; | |
3648 | ||
3649 | /* The debugfs files are a reference to the kvm struct which | |
3650 | * is still valid when kvm_destroy_vm is called. | |
3651 | * To avoid the race between open and the removal of the debugfs | |
3652 | * directory we test against the users count. | |
3653 | */ | |
e3736c3e | 3654 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
3655 | return -ENOENT; |
3656 | ||
3657 | if (simple_attr_open(inode, file, get, set, fmt)) { | |
3658 | kvm_put_kvm(stat_data->kvm); | |
3659 | return -ENOMEM; | |
3660 | } | |
3661 | ||
3662 | return 0; | |
3663 | } | |
3664 | ||
3665 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
3666 | { | |
3667 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3668 | inode->i_private; | |
3669 | ||
3670 | simple_attr_release(inode, file); | |
3671 | kvm_put_kvm(stat_data->kvm); | |
3672 | ||
3673 | return 0; | |
3674 | } | |
3675 | ||
3676 | static int vm_stat_get_per_vm(void *data, u64 *val) | |
3677 | { | |
3678 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3679 | ||
8a7e75d4 | 3680 | *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset); |
536a6f88 JF |
3681 | |
3682 | return 0; | |
3683 | } | |
3684 | ||
ce35ef27 SJS |
3685 | static int vm_stat_clear_per_vm(void *data, u64 val) |
3686 | { | |
3687 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3688 | ||
3689 | if (val) | |
3690 | return -EINVAL; | |
3691 | ||
3692 | *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0; | |
3693 | ||
3694 | return 0; | |
3695 | } | |
3696 | ||
536a6f88 JF |
3697 | static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file) |
3698 | { | |
3699 | __simple_attr_check_format("%llu\n", 0ull); | |
3700 | return kvm_debugfs_open(inode, file, vm_stat_get_per_vm, | |
ce35ef27 | 3701 | vm_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
3702 | } |
3703 | ||
3704 | static const struct file_operations vm_stat_get_per_vm_fops = { | |
3705 | .owner = THIS_MODULE, | |
3706 | .open = vm_stat_get_per_vm_open, | |
3707 | .release = kvm_debugfs_release, | |
3708 | .read = simple_attr_read, | |
3709 | .write = simple_attr_write, | |
3710 | .llseek = generic_file_llseek, | |
3711 | }; | |
3712 | ||
3713 | static int vcpu_stat_get_per_vm(void *data, u64 *val) | |
3714 | { | |
3715 | int i; | |
3716 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3717 | struct kvm_vcpu *vcpu; | |
3718 | ||
3719 | *val = 0; | |
3720 | ||
3721 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
8a7e75d4 | 3722 | *val += *(u64 *)((void *)vcpu + stat_data->offset); |
536a6f88 JF |
3723 | |
3724 | return 0; | |
3725 | } | |
3726 | ||
ce35ef27 SJS |
3727 | static int vcpu_stat_clear_per_vm(void *data, u64 val) |
3728 | { | |
3729 | int i; | |
3730 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3731 | struct kvm_vcpu *vcpu; | |
3732 | ||
3733 | if (val) | |
3734 | return -EINVAL; | |
3735 | ||
3736 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
3737 | *(u64 *)((void *)vcpu + stat_data->offset) = 0; | |
3738 | ||
3739 | return 0; | |
3740 | } | |
3741 | ||
536a6f88 JF |
3742 | static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file) |
3743 | { | |
3744 | __simple_attr_check_format("%llu\n", 0ull); | |
3745 | return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm, | |
ce35ef27 | 3746 | vcpu_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
3747 | } |
3748 | ||
3749 | static const struct file_operations vcpu_stat_get_per_vm_fops = { | |
3750 | .owner = THIS_MODULE, | |
3751 | .open = vcpu_stat_get_per_vm_open, | |
3752 | .release = kvm_debugfs_release, | |
3753 | .read = simple_attr_read, | |
3754 | .write = simple_attr_write, | |
3755 | .llseek = generic_file_llseek, | |
3756 | }; | |
3757 | ||
3758 | static const struct file_operations *stat_fops_per_vm[] = { | |
3759 | [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops, | |
3760 | [KVM_STAT_VM] = &vm_stat_get_per_vm_fops, | |
3761 | }; | |
3762 | ||
8b88b099 | 3763 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3764 | { |
3765 | unsigned offset = (long)_offset; | |
ba1389b7 | 3766 | struct kvm *kvm; |
536a6f88 JF |
3767 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
3768 | u64 tmp_val; | |
ba1389b7 | 3769 | |
8b88b099 | 3770 | *val = 0; |
2f303b74 | 3771 | spin_lock(&kvm_lock); |
536a6f88 JF |
3772 | list_for_each_entry(kvm, &vm_list, vm_list) { |
3773 | stat_tmp.kvm = kvm; | |
3774 | vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
3775 | *val += tmp_val; | |
3776 | } | |
2f303b74 | 3777 | spin_unlock(&kvm_lock); |
8b88b099 | 3778 | return 0; |
ba1389b7 AK |
3779 | } |
3780 | ||
ce35ef27 SJS |
3781 | static int vm_stat_clear(void *_offset, u64 val) |
3782 | { | |
3783 | unsigned offset = (long)_offset; | |
3784 | struct kvm *kvm; | |
3785 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
3786 | ||
3787 | if (val) | |
3788 | return -EINVAL; | |
3789 | ||
3790 | spin_lock(&kvm_lock); | |
3791 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3792 | stat_tmp.kvm = kvm; | |
3793 | vm_stat_clear_per_vm((void *)&stat_tmp, 0); | |
3794 | } | |
3795 | spin_unlock(&kvm_lock); | |
3796 | ||
3797 | return 0; | |
3798 | } | |
3799 | ||
3800 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 3801 | |
8b88b099 | 3802 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3803 | { |
3804 | unsigned offset = (long)_offset; | |
1165f5fe | 3805 | struct kvm *kvm; |
536a6f88 JF |
3806 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
3807 | u64 tmp_val; | |
1165f5fe | 3808 | |
8b88b099 | 3809 | *val = 0; |
2f303b74 | 3810 | spin_lock(&kvm_lock); |
536a6f88 JF |
3811 | list_for_each_entry(kvm, &vm_list, vm_list) { |
3812 | stat_tmp.kvm = kvm; | |
3813 | vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
3814 | *val += tmp_val; | |
3815 | } | |
2f303b74 | 3816 | spin_unlock(&kvm_lock); |
8b88b099 | 3817 | return 0; |
1165f5fe AK |
3818 | } |
3819 | ||
ce35ef27 SJS |
3820 | static int vcpu_stat_clear(void *_offset, u64 val) |
3821 | { | |
3822 | unsigned offset = (long)_offset; | |
3823 | struct kvm *kvm; | |
3824 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
3825 | ||
3826 | if (val) | |
3827 | return -EINVAL; | |
3828 | ||
3829 | spin_lock(&kvm_lock); | |
3830 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3831 | stat_tmp.kvm = kvm; | |
3832 | vcpu_stat_clear_per_vm((void *)&stat_tmp, 0); | |
3833 | } | |
3834 | spin_unlock(&kvm_lock); | |
3835 | ||
3836 | return 0; | |
3837 | } | |
3838 | ||
3839 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
3840 | "%llu\n"); | |
ba1389b7 | 3841 | |
828c0950 | 3842 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3843 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3844 | [KVM_STAT_VM] = &vm_stat_fops, | |
3845 | }; | |
1165f5fe | 3846 | |
4f69b680 | 3847 | static int kvm_init_debug(void) |
6aa8b732 | 3848 | { |
0c8eb04a | 3849 | int r = -EEXIST; |
6aa8b732 AK |
3850 | struct kvm_stats_debugfs_item *p; |
3851 | ||
76f7c879 | 3852 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3853 | if (kvm_debugfs_dir == NULL) |
3854 | goto out; | |
3855 | ||
536a6f88 JF |
3856 | kvm_debugfs_num_entries = 0; |
3857 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
ce35ef27 | 3858 | if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir, |
4bd33b56 JF |
3859 | (void *)(long)p->offset, |
3860 | stat_fops[p->kind])) | |
4f69b680 H |
3861 | goto out_dir; |
3862 | } | |
3863 | ||
3864 | return 0; | |
3865 | ||
3866 | out_dir: | |
3867 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3868 | out: | |
3869 | return r; | |
6aa8b732 AK |
3870 | } |
3871 | ||
fb3600cc | 3872 | static int kvm_suspend(void) |
59ae6c6b | 3873 | { |
10474ae8 | 3874 | if (kvm_usage_count) |
75b7127c | 3875 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3876 | return 0; |
3877 | } | |
3878 | ||
fb3600cc | 3879 | static void kvm_resume(void) |
59ae6c6b | 3880 | { |
ca84d1a2 | 3881 | if (kvm_usage_count) { |
4a937f96 | 3882 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3883 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3884 | } |
59ae6c6b AK |
3885 | } |
3886 | ||
fb3600cc | 3887 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3888 | .suspend = kvm_suspend, |
3889 | .resume = kvm_resume, | |
3890 | }; | |
3891 | ||
15ad7146 AK |
3892 | static inline |
3893 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3894 | { | |
3895 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3896 | } | |
3897 | ||
3898 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3899 | { | |
3900 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 3901 | |
3a08a8f9 R |
3902 | if (vcpu->preempted) |
3903 | vcpu->preempted = false; | |
15ad7146 | 3904 | |
e790d9ef RK |
3905 | kvm_arch_sched_in(vcpu, cpu); |
3906 | ||
e9b11c17 | 3907 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3908 | } |
3909 | ||
3910 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3911 | struct task_struct *next) | |
3912 | { | |
3913 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3914 | ||
3a08a8f9 R |
3915 | if (current->state == TASK_RUNNING) |
3916 | vcpu->preempted = true; | |
e9b11c17 | 3917 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3918 | } |
3919 | ||
0ee75bea | 3920 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3921 | struct module *module) |
6aa8b732 AK |
3922 | { |
3923 | int r; | |
002c7f7c | 3924 | int cpu; |
6aa8b732 | 3925 | |
f8c16bba ZX |
3926 | r = kvm_arch_init(opaque); |
3927 | if (r) | |
d2308784 | 3928 | goto out_fail; |
cb498ea2 | 3929 | |
7dac16c3 AH |
3930 | /* |
3931 | * kvm_arch_init makes sure there's at most one caller | |
3932 | * for architectures that support multiple implementations, | |
3933 | * like intel and amd on x86. | |
36343f6e PB |
3934 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
3935 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 3936 | */ |
36343f6e PB |
3937 | r = kvm_irqfd_init(); |
3938 | if (r) | |
3939 | goto out_irqfd; | |
7dac16c3 | 3940 | |
8437a617 | 3941 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3942 | r = -ENOMEM; |
3943 | goto out_free_0; | |
3944 | } | |
3945 | ||
e9b11c17 | 3946 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3947 | if (r < 0) |
7f59f492 | 3948 | goto out_free_0a; |
6aa8b732 | 3949 | |
002c7f7c YS |
3950 | for_each_online_cpu(cpu) { |
3951 | smp_call_function_single(cpu, | |
e9b11c17 | 3952 | kvm_arch_check_processor_compat, |
8691e5a8 | 3953 | &r, 1); |
002c7f7c | 3954 | if (r < 0) |
d2308784 | 3955 | goto out_free_1; |
002c7f7c YS |
3956 | } |
3957 | ||
73c1b41e | 3958 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 3959 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 3960 | if (r) |
d2308784 | 3961 | goto out_free_2; |
6aa8b732 AK |
3962 | register_reboot_notifier(&kvm_reboot_notifier); |
3963 | ||
c16f862d | 3964 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3965 | if (!vcpu_align) |
3966 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3967 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3968 | 0, NULL); |
c16f862d RR |
3969 | if (!kvm_vcpu_cache) { |
3970 | r = -ENOMEM; | |
fb3600cc | 3971 | goto out_free_3; |
c16f862d RR |
3972 | } |
3973 | ||
af585b92 GN |
3974 | r = kvm_async_pf_init(); |
3975 | if (r) | |
3976 | goto out_free; | |
3977 | ||
6aa8b732 | 3978 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3979 | kvm_vm_fops.owner = module; |
3980 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3981 | |
3982 | r = misc_register(&kvm_dev); | |
3983 | if (r) { | |
1170adc6 | 3984 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 3985 | goto out_unreg; |
6aa8b732 AK |
3986 | } |
3987 | ||
fb3600cc RW |
3988 | register_syscore_ops(&kvm_syscore_ops); |
3989 | ||
15ad7146 AK |
3990 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3991 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3992 | ||
4f69b680 H |
3993 | r = kvm_init_debug(); |
3994 | if (r) { | |
1170adc6 | 3995 | pr_err("kvm: create debugfs files failed\n"); |
4f69b680 H |
3996 | goto out_undebugfs; |
3997 | } | |
0ea4ed8e | 3998 | |
3c3c29fd PB |
3999 | r = kvm_vfio_ops_init(); |
4000 | WARN_ON(r); | |
4001 | ||
c7addb90 | 4002 | return 0; |
6aa8b732 | 4003 | |
4f69b680 H |
4004 | out_undebugfs: |
4005 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 4006 | misc_deregister(&kvm_dev); |
af585b92 GN |
4007 | out_unreg: |
4008 | kvm_async_pf_deinit(); | |
6aa8b732 | 4009 | out_free: |
c16f862d | 4010 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4011 | out_free_3: |
6aa8b732 | 4012 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4013 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4014 | out_free_2: |
d2308784 | 4015 | out_free_1: |
e9b11c17 | 4016 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
4017 | out_free_0a: |
4018 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 4019 | out_free_0: |
a0f155e9 | 4020 | kvm_irqfd_exit(); |
36343f6e | 4021 | out_irqfd: |
7dac16c3 AH |
4022 | kvm_arch_exit(); |
4023 | out_fail: | |
6aa8b732 AK |
4024 | return r; |
4025 | } | |
cb498ea2 | 4026 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 4027 | |
cb498ea2 | 4028 | void kvm_exit(void) |
6aa8b732 | 4029 | { |
4bd33b56 | 4030 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 4031 | misc_deregister(&kvm_dev); |
c16f862d | 4032 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 4033 | kvm_async_pf_deinit(); |
fb3600cc | 4034 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 4035 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4036 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 4037 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 4038 | kvm_arch_hardware_unsetup(); |
f8c16bba | 4039 | kvm_arch_exit(); |
a0f155e9 | 4040 | kvm_irqfd_exit(); |
7f59f492 | 4041 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 4042 | kvm_vfio_ops_exit(); |
6aa8b732 | 4043 | } |
cb498ea2 | 4044 | EXPORT_SYMBOL_GPL(kvm_exit); |