]>
Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
9 | * |
10 | * Authors: | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * Yaniv Kamay <yaniv@qumranet.com> | |
13 | * | |
14 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
15 | * the COPYING file in the top-level directory. | |
16 | * | |
17 | */ | |
18 | ||
af669ac6 | 19 | #include <kvm/iodev.h> |
6aa8b732 | 20 | |
edf88417 | 21 | #include <linux/kvm_host.h> |
6aa8b732 AK |
22 | #include <linux/kvm.h> |
23 | #include <linux/module.h> | |
24 | #include <linux/errno.h> | |
6aa8b732 | 25 | #include <linux/percpu.h> |
6aa8b732 AK |
26 | #include <linux/mm.h> |
27 | #include <linux/miscdevice.h> | |
28 | #include <linux/vmalloc.h> | |
6aa8b732 | 29 | #include <linux/reboot.h> |
6aa8b732 AK |
30 | #include <linux/debugfs.h> |
31 | #include <linux/highmem.h> | |
32 | #include <linux/file.h> | |
fb3600cc | 33 | #include <linux/syscore_ops.h> |
774c47f1 | 34 | #include <linux/cpu.h> |
e8edc6e0 | 35 | #include <linux/sched.h> |
d9e368d6 AK |
36 | #include <linux/cpumask.h> |
37 | #include <linux/smp.h> | |
d6d28168 | 38 | #include <linux/anon_inodes.h> |
04d2cc77 | 39 | #include <linux/profile.h> |
7aa81cc0 | 40 | #include <linux/kvm_para.h> |
6fc138d2 | 41 | #include <linux/pagemap.h> |
8d4e1288 | 42 | #include <linux/mman.h> |
35149e21 | 43 | #include <linux/swap.h> |
e56d532f | 44 | #include <linux/bitops.h> |
547de29e | 45 | #include <linux/spinlock.h> |
6ff5894c | 46 | #include <linux/compat.h> |
bc6678a3 | 47 | #include <linux/srcu.h> |
8f0b1ab6 | 48 | #include <linux/hugetlb.h> |
5a0e3ad6 | 49 | #include <linux/slab.h> |
743eeb0b SL |
50 | #include <linux/sort.h> |
51 | #include <linux/bsearch.h> | |
6aa8b732 | 52 | |
e495606d | 53 | #include <asm/processor.h> |
e495606d | 54 | #include <asm/io.h> |
2ea75be3 | 55 | #include <asm/ioctl.h> |
7c0f6ba6 | 56 | #include <linux/uaccess.h> |
3e021bf5 | 57 | #include <asm/pgtable.h> |
6aa8b732 | 58 | |
5f94c174 | 59 | #include "coalesced_mmio.h" |
af585b92 | 60 | #include "async_pf.h" |
3c3c29fd | 61 | #include "vfio.h" |
5f94c174 | 62 | |
229456fc MT |
63 | #define CREATE_TRACE_POINTS |
64 | #include <trace/events/kvm.h> | |
65 | ||
536a6f88 JF |
66 | /* Worst case buffer size needed for holding an integer. */ |
67 | #define ITOA_MAX_LEN 12 | |
68 | ||
6aa8b732 AK |
69 | MODULE_AUTHOR("Qumranet"); |
70 | MODULE_LICENSE("GPL"); | |
71 | ||
920552b2 | 72 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 73 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
f7819512 | 74 | module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR); |
ec76d819 | 75 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 76 | |
aca6ff29 | 77 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 78 | unsigned int halt_poll_ns_grow = 2; |
6b6de68c | 79 | module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR); |
ec76d819 | 80 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 WL |
81 | |
82 | /* Default resets per-vcpu halt_poll_ns . */ | |
ec76d819 | 83 | unsigned int halt_poll_ns_shrink; |
6b6de68c | 84 | module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR); |
ec76d819 | 85 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 86 | |
fa40a821 MT |
87 | /* |
88 | * Ordering of locks: | |
89 | * | |
b7d409de | 90 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
91 | */ |
92 | ||
2f303b74 | 93 | DEFINE_SPINLOCK(kvm_lock); |
4a937f96 | 94 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 95 | LIST_HEAD(vm_list); |
133de902 | 96 | |
7f59f492 | 97 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 98 | static int kvm_usage_count; |
10474ae8 | 99 | static atomic_t hardware_enable_failed; |
1b6c0168 | 100 | |
c16f862d RR |
101 | struct kmem_cache *kvm_vcpu_cache; |
102 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 103 | |
15ad7146 AK |
104 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
105 | ||
76f7c879 | 106 | struct dentry *kvm_debugfs_dir; |
e23a808b | 107 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 108 | |
536a6f88 JF |
109 | static int kvm_debugfs_num_entries; |
110 | static const struct file_operations *stat_fops_per_vm[]; | |
111 | ||
bccf2150 AK |
112 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
113 | unsigned long arg); | |
de8e5d74 | 114 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
115 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
116 | unsigned long arg); | |
117 | #endif | |
10474ae8 AG |
118 | static int hardware_enable_all(void); |
119 | static void hardware_disable_all(void); | |
bccf2150 | 120 | |
e93f8a0f | 121 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 122 | |
ba049e93 | 123 | static void kvm_release_pfn_dirty(kvm_pfn_t pfn); |
bc009e43 | 124 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 125 | |
52480137 | 126 | __visible bool kvm_rebooting; |
b7c4145b | 127 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 128 | |
54dee993 MT |
129 | static bool largepages_enabled = true; |
130 | ||
ba049e93 | 131 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 132 | { |
11feeb49 | 133 | if (pfn_valid(pfn)) |
bf4bea8e | 134 | return PageReserved(pfn_to_page(pfn)); |
cbff90a7 BAY |
135 | |
136 | return true; | |
137 | } | |
138 | ||
bccf2150 AK |
139 | /* |
140 | * Switches to specified vcpu, until a matching vcpu_put() | |
141 | */ | |
9fc77441 | 142 | int vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 143 | { |
15ad7146 AK |
144 | int cpu; |
145 | ||
9fc77441 MT |
146 | if (mutex_lock_killable(&vcpu->mutex)) |
147 | return -EINTR; | |
15ad7146 AK |
148 | cpu = get_cpu(); |
149 | preempt_notifier_register(&vcpu->preempt_notifier); | |
313a3dc7 | 150 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 151 | put_cpu(); |
9fc77441 | 152 | return 0; |
6aa8b732 | 153 | } |
2f1fe811 | 154 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 155 | |
313a3dc7 | 156 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 157 | { |
15ad7146 | 158 | preempt_disable(); |
313a3dc7 | 159 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
160 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
161 | preempt_enable(); | |
6aa8b732 AK |
162 | mutex_unlock(&vcpu->mutex); |
163 | } | |
2f1fe811 | 164 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 165 | |
d9e368d6 AK |
166 | static void ack_flush(void *_completed) |
167 | { | |
d9e368d6 AK |
168 | } |
169 | ||
445b8236 | 170 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 171 | { |
597a5f55 | 172 | int i, cpu, me; |
6ef7a1bc RR |
173 | cpumask_var_t cpus; |
174 | bool called = true; | |
d9e368d6 | 175 | struct kvm_vcpu *vcpu; |
d9e368d6 | 176 | |
79f55997 | 177 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 178 | |
3cba4130 | 179 | me = get_cpu(); |
988a2cae | 180 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 181 | kvm_make_request(req, vcpu); |
d9e368d6 | 182 | cpu = vcpu->cpu; |
6b7e2d09 | 183 | |
a30a0509 LT |
184 | /* Set ->requests bit before we read ->mode. */ |
185 | smp_mb__after_atomic(); | |
6b7e2d09 XG |
186 | |
187 | if (cpus != NULL && cpu != -1 && cpu != me && | |
188 | kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE) | |
6ef7a1bc | 189 | cpumask_set_cpu(cpu, cpus); |
49846896 | 190 | } |
6ef7a1bc RR |
191 | if (unlikely(cpus == NULL)) |
192 | smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1); | |
193 | else if (!cpumask_empty(cpus)) | |
194 | smp_call_function_many(cpus, ack_flush, NULL, 1); | |
195 | else | |
196 | called = false; | |
3cba4130 | 197 | put_cpu(); |
6ef7a1bc | 198 | free_cpumask_var(cpus); |
49846896 | 199 | return called; |
d9e368d6 AK |
200 | } |
201 | ||
a6d51016 | 202 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 203 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 204 | { |
4ae3cb3a LT |
205 | /* |
206 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
207 | * kvm_make_all_cpus_request. | |
208 | */ | |
209 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
210 | ||
211 | /* | |
212 | * We want to publish modifications to the page tables before reading | |
213 | * mode. Pairs with a memory barrier in arch-specific code. | |
214 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
215 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
216 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
217 | * | |
218 | * There is already an smp_mb__after_atomic() before | |
219 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
220 | * barrier here. | |
221 | */ | |
445b8236 | 222 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
49846896 | 223 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 224 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 225 | } |
2ba9f0d8 | 226 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 227 | #endif |
2e53d63a | 228 | |
49846896 RR |
229 | void kvm_reload_remote_mmus(struct kvm *kvm) |
230 | { | |
445b8236 | 231 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 232 | } |
2e53d63a | 233 | |
fb3f0f51 RR |
234 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
235 | { | |
236 | struct page *page; | |
237 | int r; | |
238 | ||
239 | mutex_init(&vcpu->mutex); | |
240 | vcpu->cpu = -1; | |
fb3f0f51 RR |
241 | vcpu->kvm = kvm; |
242 | vcpu->vcpu_id = id; | |
34bb10b7 | 243 | vcpu->pid = NULL; |
8577370f | 244 | init_swait_queue_head(&vcpu->wq); |
af585b92 | 245 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 246 | |
bf9f6ac8 FW |
247 | vcpu->pre_pcpu = -1; |
248 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
249 | ||
fb3f0f51 RR |
250 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
251 | if (!page) { | |
252 | r = -ENOMEM; | |
253 | goto fail; | |
254 | } | |
255 | vcpu->run = page_address(page); | |
256 | ||
4c088493 R |
257 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
258 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 259 | vcpu->preempted = false; |
4c088493 | 260 | |
e9b11c17 | 261 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 262 | if (r < 0) |
e9b11c17 | 263 | goto fail_free_run; |
fb3f0f51 RR |
264 | return 0; |
265 | ||
fb3f0f51 RR |
266 | fail_free_run: |
267 | free_page((unsigned long)vcpu->run); | |
268 | fail: | |
76fafa5e | 269 | return r; |
fb3f0f51 RR |
270 | } |
271 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
272 | ||
273 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
274 | { | |
34bb10b7 | 275 | put_pid(vcpu->pid); |
e9b11c17 | 276 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
277 | free_page((unsigned long)vcpu->run); |
278 | } | |
279 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
280 | ||
e930bffe AA |
281 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
282 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
283 | { | |
284 | return container_of(mn, struct kvm, mmu_notifier); | |
285 | } | |
286 | ||
287 | static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn, | |
288 | struct mm_struct *mm, | |
289 | unsigned long address) | |
290 | { | |
291 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 292 | int need_tlb_flush, idx; |
e930bffe AA |
293 | |
294 | /* | |
295 | * When ->invalidate_page runs, the linux pte has been zapped | |
296 | * already but the page is still allocated until | |
297 | * ->invalidate_page returns. So if we increase the sequence | |
298 | * here the kvm page fault will notice if the spte can't be | |
299 | * established because the page is going to be freed. If | |
300 | * instead the kvm page fault establishes the spte before | |
301 | * ->invalidate_page runs, kvm_unmap_hva will release it | |
302 | * before returning. | |
303 | * | |
304 | * The sequence increase only need to be seen at spin_unlock | |
305 | * time, and not at spin_lock time. | |
306 | * | |
307 | * Increasing the sequence after the spin_unlock would be | |
308 | * unsafe because the kvm page fault could then establish the | |
309 | * pte after kvm_unmap_hva returned, without noticing the page | |
310 | * is going to be freed. | |
311 | */ | |
bc6678a3 | 312 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 313 | spin_lock(&kvm->mmu_lock); |
565f3be2 | 314 | |
e930bffe | 315 | kvm->mmu_notifier_seq++; |
a4ee1ca4 | 316 | need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty; |
e930bffe AA |
317 | /* we've to flush the tlb before the pages can be freed */ |
318 | if (need_tlb_flush) | |
319 | kvm_flush_remote_tlbs(kvm); | |
320 | ||
565f3be2 | 321 | spin_unlock(&kvm->mmu_lock); |
fe71557a TC |
322 | |
323 | kvm_arch_mmu_notifier_invalidate_page(kvm, address); | |
324 | ||
565f3be2 | 325 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
326 | } |
327 | ||
3da0dd43 IE |
328 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
329 | struct mm_struct *mm, | |
330 | unsigned long address, | |
331 | pte_t pte) | |
332 | { | |
333 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 334 | int idx; |
3da0dd43 | 335 | |
bc6678a3 | 336 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
337 | spin_lock(&kvm->mmu_lock); |
338 | kvm->mmu_notifier_seq++; | |
339 | kvm_set_spte_hva(kvm, address, pte); | |
340 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 341 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
342 | } |
343 | ||
e930bffe AA |
344 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
345 | struct mm_struct *mm, | |
346 | unsigned long start, | |
347 | unsigned long end) | |
348 | { | |
349 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 350 | int need_tlb_flush = 0, idx; |
e930bffe | 351 | |
bc6678a3 | 352 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
353 | spin_lock(&kvm->mmu_lock); |
354 | /* | |
355 | * The count increase must become visible at unlock time as no | |
356 | * spte can be established without taking the mmu_lock and | |
357 | * count is also read inside the mmu_lock critical section. | |
358 | */ | |
359 | kvm->mmu_notifier_count++; | |
b3ae2096 | 360 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 361 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
362 | /* we've to flush the tlb before the pages can be freed */ |
363 | if (need_tlb_flush) | |
364 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
365 | |
366 | spin_unlock(&kvm->mmu_lock); | |
367 | srcu_read_unlock(&kvm->srcu, idx); | |
e930bffe AA |
368 | } |
369 | ||
370 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
371 | struct mm_struct *mm, | |
372 | unsigned long start, | |
373 | unsigned long end) | |
374 | { | |
375 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
376 | ||
377 | spin_lock(&kvm->mmu_lock); | |
378 | /* | |
379 | * This sequence increase will notify the kvm page fault that | |
380 | * the page that is going to be mapped in the spte could have | |
381 | * been freed. | |
382 | */ | |
383 | kvm->mmu_notifier_seq++; | |
a355aa54 | 384 | smp_wmb(); |
e930bffe AA |
385 | /* |
386 | * The above sequence increase must be visible before the | |
a355aa54 PM |
387 | * below count decrease, which is ensured by the smp_wmb above |
388 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
389 | */ |
390 | kvm->mmu_notifier_count--; | |
391 | spin_unlock(&kvm->mmu_lock); | |
392 | ||
393 | BUG_ON(kvm->mmu_notifier_count < 0); | |
394 | } | |
395 | ||
396 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
397 | struct mm_struct *mm, | |
57128468 ALC |
398 | unsigned long start, |
399 | unsigned long end) | |
e930bffe AA |
400 | { |
401 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 402 | int young, idx; |
e930bffe | 403 | |
bc6678a3 | 404 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 405 | spin_lock(&kvm->mmu_lock); |
e930bffe | 406 | |
57128468 | 407 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
408 | if (young) |
409 | kvm_flush_remote_tlbs(kvm); | |
410 | ||
565f3be2 TY |
411 | spin_unlock(&kvm->mmu_lock); |
412 | srcu_read_unlock(&kvm->srcu, idx); | |
413 | ||
e930bffe AA |
414 | return young; |
415 | } | |
416 | ||
1d7715c6 VD |
417 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
418 | struct mm_struct *mm, | |
419 | unsigned long start, | |
420 | unsigned long end) | |
421 | { | |
422 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
423 | int young, idx; | |
424 | ||
425 | idx = srcu_read_lock(&kvm->srcu); | |
426 | spin_lock(&kvm->mmu_lock); | |
427 | /* | |
428 | * Even though we do not flush TLB, this will still adversely | |
429 | * affect performance on pre-Haswell Intel EPT, where there is | |
430 | * no EPT Access Bit to clear so that we have to tear down EPT | |
431 | * tables instead. If we find this unacceptable, we can always | |
432 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
433 | * do anything on clear_young. | |
434 | * | |
435 | * Also note that currently we never issue secondary TLB flushes | |
436 | * from clear_young, leaving this job up to the regular system | |
437 | * cadence. If we find this inaccurate, we might come up with a | |
438 | * more sophisticated heuristic later. | |
439 | */ | |
440 | young = kvm_age_hva(kvm, start, end); | |
441 | spin_unlock(&kvm->mmu_lock); | |
442 | srcu_read_unlock(&kvm->srcu, idx); | |
443 | ||
444 | return young; | |
445 | } | |
446 | ||
8ee53820 AA |
447 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
448 | struct mm_struct *mm, | |
449 | unsigned long address) | |
450 | { | |
451 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
452 | int young, idx; | |
453 | ||
454 | idx = srcu_read_lock(&kvm->srcu); | |
455 | spin_lock(&kvm->mmu_lock); | |
456 | young = kvm_test_age_hva(kvm, address); | |
457 | spin_unlock(&kvm->mmu_lock); | |
458 | srcu_read_unlock(&kvm->srcu, idx); | |
459 | ||
460 | return young; | |
461 | } | |
462 | ||
85db06e5 MT |
463 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
464 | struct mm_struct *mm) | |
465 | { | |
466 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
467 | int idx; |
468 | ||
469 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 470 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 471 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
472 | } |
473 | ||
e930bffe AA |
474 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
475 | .invalidate_page = kvm_mmu_notifier_invalidate_page, | |
476 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, | |
477 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
478 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 479 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 480 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 481 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 482 | .release = kvm_mmu_notifier_release, |
e930bffe | 483 | }; |
4c07b0a4 AK |
484 | |
485 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
486 | { | |
487 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
488 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
489 | } | |
490 | ||
491 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
492 | ||
493 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
494 | { | |
495 | return 0; | |
496 | } | |
497 | ||
e930bffe AA |
498 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
499 | ||
a47d2b07 | 500 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
501 | { |
502 | int i; | |
a47d2b07 | 503 | struct kvm_memslots *slots; |
bf3e05bc | 504 | |
a47d2b07 PB |
505 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
506 | if (!slots) | |
507 | return NULL; | |
508 | ||
509 | /* | |
510 | * Init kvm generation close to the maximum to easily test the | |
511 | * code of handling generation number wrap-around. | |
512 | */ | |
513 | slots->generation = -150; | |
bf3e05bc | 514 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 515 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
516 | |
517 | return slots; | |
518 | } | |
519 | ||
520 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
521 | { | |
522 | if (!memslot->dirty_bitmap) | |
523 | return; | |
524 | ||
525 | kvfree(memslot->dirty_bitmap); | |
526 | memslot->dirty_bitmap = NULL; | |
527 | } | |
528 | ||
529 | /* | |
530 | * Free any memory in @free but not in @dont. | |
531 | */ | |
532 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
533 | struct kvm_memory_slot *dont) | |
534 | { | |
535 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
536 | kvm_destroy_dirty_bitmap(free); | |
537 | ||
538 | kvm_arch_free_memslot(kvm, free, dont); | |
539 | ||
540 | free->npages = 0; | |
541 | } | |
542 | ||
543 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
544 | { | |
545 | struct kvm_memory_slot *memslot; | |
546 | ||
547 | if (!slots) | |
548 | return; | |
549 | ||
550 | kvm_for_each_memslot(memslot, slots) | |
551 | kvm_free_memslot(kvm, memslot, NULL); | |
552 | ||
553 | kvfree(slots); | |
bf3e05bc XG |
554 | } |
555 | ||
536a6f88 JF |
556 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
557 | { | |
558 | int i; | |
559 | ||
560 | if (!kvm->debugfs_dentry) | |
561 | return; | |
562 | ||
563 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
564 | ||
9d5a1dce LC |
565 | if (kvm->debugfs_stat_data) { |
566 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
567 | kfree(kvm->debugfs_stat_data[i]); | |
568 | kfree(kvm->debugfs_stat_data); | |
569 | } | |
536a6f88 JF |
570 | } |
571 | ||
572 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
573 | { | |
574 | char dir_name[ITOA_MAX_LEN * 2]; | |
575 | struct kvm_stat_data *stat_data; | |
576 | struct kvm_stats_debugfs_item *p; | |
577 | ||
578 | if (!debugfs_initialized()) | |
579 | return 0; | |
580 | ||
581 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
582 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, | |
583 | kvm_debugfs_dir); | |
584 | if (!kvm->debugfs_dentry) | |
585 | return -ENOMEM; | |
586 | ||
587 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
588 | sizeof(*kvm->debugfs_stat_data), | |
589 | GFP_KERNEL); | |
590 | if (!kvm->debugfs_stat_data) | |
591 | return -ENOMEM; | |
592 | ||
593 | for (p = debugfs_entries; p->name; p++) { | |
594 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL); | |
595 | if (!stat_data) | |
596 | return -ENOMEM; | |
597 | ||
598 | stat_data->kvm = kvm; | |
599 | stat_data->offset = p->offset; | |
600 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; | |
ce35ef27 | 601 | if (!debugfs_create_file(p->name, 0644, |
536a6f88 JF |
602 | kvm->debugfs_dentry, |
603 | stat_data, | |
604 | stat_fops_per_vm[p->kind])) | |
605 | return -ENOMEM; | |
606 | } | |
607 | return 0; | |
608 | } | |
609 | ||
e08b9637 | 610 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 611 | { |
d89f5eff JK |
612 | int r, i; |
613 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 614 | |
d89f5eff JK |
615 | if (!kvm) |
616 | return ERR_PTR(-ENOMEM); | |
617 | ||
e9ad4ec8 PB |
618 | spin_lock_init(&kvm->mmu_lock); |
619 | atomic_inc(¤t->mm->mm_count); | |
620 | kvm->mm = current->mm; | |
621 | kvm_eventfd_init(kvm); | |
622 | mutex_init(&kvm->lock); | |
623 | mutex_init(&kvm->irq_lock); | |
624 | mutex_init(&kvm->slots_lock); | |
625 | atomic_set(&kvm->users_count, 1); | |
626 | INIT_LIST_HEAD(&kvm->devices); | |
627 | ||
e08b9637 | 628 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 629 | if (r) |
719d93cd | 630 | goto out_err_no_disable; |
10474ae8 AG |
631 | |
632 | r = hardware_enable_all(); | |
633 | if (r) | |
719d93cd | 634 | goto out_err_no_disable; |
10474ae8 | 635 | |
c77dcacb | 636 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 637 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 638 | #endif |
6aa8b732 | 639 | |
1e702d9a AW |
640 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
641 | ||
46a26bf5 | 642 | r = -ENOMEM; |
f481b069 PB |
643 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
644 | kvm->memslots[i] = kvm_alloc_memslots(); | |
645 | if (!kvm->memslots[i]) | |
646 | goto out_err_no_srcu; | |
647 | } | |
00f034a1 | 648 | |
bc6678a3 | 649 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
650 | goto out_err_no_srcu; |
651 | if (init_srcu_struct(&kvm->irq_srcu)) | |
652 | goto out_err_no_irq_srcu; | |
e93f8a0f MT |
653 | for (i = 0; i < KVM_NR_BUSES; i++) { |
654 | kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus), | |
655 | GFP_KERNEL); | |
57e7fbee | 656 | if (!kvm->buses[i]) |
e93f8a0f | 657 | goto out_err; |
e93f8a0f | 658 | } |
e930bffe | 659 | |
74b5c5bf MW |
660 | r = kvm_init_mmu_notifier(kvm); |
661 | if (r) | |
662 | goto out_err; | |
663 | ||
2f303b74 | 664 | spin_lock(&kvm_lock); |
5e58cfe4 | 665 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 666 | spin_unlock(&kvm_lock); |
d89f5eff | 667 | |
2ecd9d29 PZ |
668 | preempt_notifier_inc(); |
669 | ||
f17abe9a | 670 | return kvm; |
10474ae8 AG |
671 | |
672 | out_err: | |
719d93cd CB |
673 | cleanup_srcu_struct(&kvm->irq_srcu); |
674 | out_err_no_irq_srcu: | |
57e7fbee | 675 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 676 | out_err_no_srcu: |
10474ae8 | 677 | hardware_disable_all(); |
719d93cd | 678 | out_err_no_disable: |
e93f8a0f MT |
679 | for (i = 0; i < KVM_NR_BUSES; i++) |
680 | kfree(kvm->buses[i]); | |
f481b069 PB |
681 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
682 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
d89f5eff | 683 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 684 | mmdrop(current->mm); |
10474ae8 | 685 | return ERR_PTR(r); |
f17abe9a AK |
686 | } |
687 | ||
92eca8fa TY |
688 | /* |
689 | * Avoid using vmalloc for a small buffer. | |
690 | * Should not be used when the size is statically known. | |
691 | */ | |
c1a7b32a | 692 | void *kvm_kvzalloc(unsigned long size) |
92eca8fa TY |
693 | { |
694 | if (size > PAGE_SIZE) | |
695 | return vzalloc(size); | |
696 | else | |
697 | return kzalloc(size, GFP_KERNEL); | |
698 | } | |
699 | ||
07f0a7bd SW |
700 | static void kvm_destroy_devices(struct kvm *kvm) |
701 | { | |
e6e3b5a6 | 702 | struct kvm_device *dev, *tmp; |
07f0a7bd | 703 | |
a28ebea2 CD |
704 | /* |
705 | * We do not need to take the kvm->lock here, because nobody else | |
706 | * has a reference to the struct kvm at this point and therefore | |
707 | * cannot access the devices list anyhow. | |
708 | */ | |
e6e3b5a6 GT |
709 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
710 | list_del(&dev->vm_node); | |
07f0a7bd SW |
711 | dev->ops->destroy(dev); |
712 | } | |
713 | } | |
714 | ||
f17abe9a AK |
715 | static void kvm_destroy_vm(struct kvm *kvm) |
716 | { | |
e93f8a0f | 717 | int i; |
6d4e4c4f AK |
718 | struct mm_struct *mm = kvm->mm; |
719 | ||
536a6f88 | 720 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 721 | kvm_arch_sync_events(kvm); |
2f303b74 | 722 | spin_lock(&kvm_lock); |
133de902 | 723 | list_del(&kvm->vm_list); |
2f303b74 | 724 | spin_unlock(&kvm_lock); |
399ec807 | 725 | kvm_free_irq_routing(kvm); |
46a92167 | 726 | for (i = 0; i < KVM_NR_BUSES; i++) { |
e93f8a0f | 727 | kvm_io_bus_destroy(kvm->buses[i]); |
46a92167 PX |
728 | kvm->buses[i] = NULL; |
729 | } | |
980da6ce | 730 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
731 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
732 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 733 | #else |
2df72e9b | 734 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 735 | #endif |
d19a9cd2 | 736 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 737 | kvm_destroy_devices(kvm); |
f481b069 PB |
738 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
739 | kvm_free_memslots(kvm, kvm->memslots[i]); | |
820b3fcd | 740 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
741 | cleanup_srcu_struct(&kvm->srcu); |
742 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 743 | preempt_notifier_dec(); |
10474ae8 | 744 | hardware_disable_all(); |
6d4e4c4f | 745 | mmdrop(mm); |
f17abe9a AK |
746 | } |
747 | ||
d39f13b0 IE |
748 | void kvm_get_kvm(struct kvm *kvm) |
749 | { | |
750 | atomic_inc(&kvm->users_count); | |
751 | } | |
752 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
753 | ||
754 | void kvm_put_kvm(struct kvm *kvm) | |
755 | { | |
756 | if (atomic_dec_and_test(&kvm->users_count)) | |
757 | kvm_destroy_vm(kvm); | |
758 | } | |
759 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
760 | ||
761 | ||
f17abe9a AK |
762 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
763 | { | |
764 | struct kvm *kvm = filp->private_data; | |
765 | ||
721eecbf GH |
766 | kvm_irqfd_release(kvm); |
767 | ||
d39f13b0 | 768 | kvm_put_kvm(kvm); |
6aa8b732 AK |
769 | return 0; |
770 | } | |
771 | ||
515a0127 TY |
772 | /* |
773 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 774 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 775 | */ |
a36a57b1 TY |
776 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
777 | { | |
515a0127 | 778 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 779 | |
92eca8fa | 780 | memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes); |
a36a57b1 TY |
781 | if (!memslot->dirty_bitmap) |
782 | return -ENOMEM; | |
783 | ||
a36a57b1 TY |
784 | return 0; |
785 | } | |
786 | ||
bf3e05bc | 787 | /* |
0e60b079 IM |
788 | * Insert memslot and re-sort memslots based on their GFN, |
789 | * so binary search could be used to lookup GFN. | |
790 | * Sorting algorithm takes advantage of having initially | |
791 | * sorted array and known changed memslot position. | |
bf3e05bc | 792 | */ |
5cc15027 PB |
793 | static void update_memslots(struct kvm_memslots *slots, |
794 | struct kvm_memory_slot *new) | |
bf3e05bc | 795 | { |
8593176c PB |
796 | int id = new->id; |
797 | int i = slots->id_to_index[id]; | |
063584d4 | 798 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 799 | |
8593176c | 800 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 801 | if (!new->npages) { |
dbaff309 | 802 | WARN_ON(!mslots[i].npages); |
9c1a5d38 IM |
803 | if (mslots[i].npages) |
804 | slots->used_slots--; | |
805 | } else { | |
806 | if (!mslots[i].npages) | |
807 | slots->used_slots++; | |
808 | } | |
0e60b079 | 809 | |
7f379cff | 810 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
811 | new->base_gfn <= mslots[i + 1].base_gfn) { |
812 | if (!mslots[i + 1].npages) | |
813 | break; | |
7f379cff IM |
814 | mslots[i] = mslots[i + 1]; |
815 | slots->id_to_index[mslots[i].id] = i; | |
816 | i++; | |
817 | } | |
efbeec70 PB |
818 | |
819 | /* | |
820 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
821 | * so that it moves before all those with base_gfn == npages == 0. | |
822 | * | |
823 | * On the other hand, if new->npages is zero, the above loop has | |
824 | * already left i pointing to the beginning of the empty part of | |
825 | * mslots, and the ">=" would move the hole backwards in this | |
826 | * case---which is wrong. So skip the loop when deleting a slot. | |
827 | */ | |
828 | if (new->npages) { | |
829 | while (i > 0 && | |
830 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
831 | mslots[i] = mslots[i - 1]; | |
832 | slots->id_to_index[mslots[i].id] = i; | |
833 | i--; | |
834 | } | |
dbaff309 PB |
835 | } else |
836 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 837 | |
8593176c PB |
838 | mslots[i] = *new; |
839 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
840 | } |
841 | ||
09170a49 | 842 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 843 | { |
4d8b81ab XG |
844 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
845 | ||
0f8a4de3 | 846 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
847 | valid_flags |= KVM_MEM_READONLY; |
848 | #endif | |
849 | ||
850 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
851 | return -EINVAL; |
852 | ||
853 | return 0; | |
854 | } | |
855 | ||
7ec4fb44 | 856 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 857 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 858 | { |
f481b069 | 859 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
7ec4fb44 | 860 | |
ee3d1570 DM |
861 | /* |
862 | * Set the low bit in the generation, which disables SPTE caching | |
863 | * until the end of synchronize_srcu_expedited. | |
864 | */ | |
865 | WARN_ON(old_memslots->generation & 1); | |
866 | slots->generation = old_memslots->generation + 1; | |
867 | ||
f481b069 | 868 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 869 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 870 | |
ee3d1570 DM |
871 | /* |
872 | * Increment the new memslot generation a second time. This prevents | |
873 | * vm exits that race with memslot updates from caching a memslot | |
874 | * generation that will (potentially) be valid forever. | |
875 | */ | |
876 | slots->generation++; | |
877 | ||
15f46015 | 878 | kvm_arch_memslots_updated(kvm, slots); |
e59dbe09 TY |
879 | |
880 | return old_memslots; | |
7ec4fb44 GN |
881 | } |
882 | ||
6aa8b732 AK |
883 | /* |
884 | * Allocate some memory and give it an address in the guest physical address | |
885 | * space. | |
886 | * | |
887 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 888 | * |
02d5d55b | 889 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 890 | */ |
f78e0e2e | 891 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 892 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 893 | { |
8234b22e | 894 | int r; |
6aa8b732 | 895 | gfn_t base_gfn; |
28bcb112 | 896 | unsigned long npages; |
a843fac2 | 897 | struct kvm_memory_slot *slot; |
6aa8b732 | 898 | struct kvm_memory_slot old, new; |
b7f69c55 | 899 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 900 | int as_id, id; |
f64c0398 | 901 | enum kvm_mr_change change; |
6aa8b732 | 902 | |
a50d64d6 XG |
903 | r = check_memory_region_flags(mem); |
904 | if (r) | |
905 | goto out; | |
906 | ||
6aa8b732 | 907 | r = -EINVAL; |
f481b069 PB |
908 | as_id = mem->slot >> 16; |
909 | id = (u16)mem->slot; | |
910 | ||
6aa8b732 AK |
911 | /* General sanity checks */ |
912 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
913 | goto out; | |
914 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
915 | goto out; | |
fa3d315a | 916 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 917 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 918 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
919 | !access_ok(VERIFY_WRITE, |
920 | (void __user *)(unsigned long)mem->userspace_addr, | |
921 | mem->memory_size))) | |
78749809 | 922 | goto out; |
f481b069 | 923 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
924 | goto out; |
925 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
926 | goto out; | |
927 | ||
f481b069 | 928 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
929 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
930 | npages = mem->memory_size >> PAGE_SHIFT; | |
931 | ||
660c22c4 TY |
932 | if (npages > KVM_MEM_MAX_NR_PAGES) |
933 | goto out; | |
934 | ||
a843fac2 | 935 | new = old = *slot; |
6aa8b732 | 936 | |
f481b069 | 937 | new.id = id; |
6aa8b732 AK |
938 | new.base_gfn = base_gfn; |
939 | new.npages = npages; | |
940 | new.flags = mem->flags; | |
941 | ||
f64c0398 TY |
942 | if (npages) { |
943 | if (!old.npages) | |
944 | change = KVM_MR_CREATE; | |
945 | else { /* Modify an existing slot. */ | |
946 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
947 | (npages != old.npages) || |
948 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
949 | goto out; |
950 | ||
951 | if (base_gfn != old.base_gfn) | |
952 | change = KVM_MR_MOVE; | |
953 | else if (new.flags != old.flags) | |
954 | change = KVM_MR_FLAGS_ONLY; | |
955 | else { /* Nothing to change. */ | |
956 | r = 0; | |
957 | goto out; | |
958 | } | |
959 | } | |
09170a49 PB |
960 | } else { |
961 | if (!old.npages) | |
962 | goto out; | |
963 | ||
f64c0398 | 964 | change = KVM_MR_DELETE; |
09170a49 PB |
965 | new.base_gfn = 0; |
966 | new.flags = 0; | |
967 | } | |
6aa8b732 | 968 | |
f64c0398 | 969 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
970 | /* Check for overlaps */ |
971 | r = -EEXIST; | |
f481b069 | 972 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
a843fac2 | 973 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
f481b069 | 974 | (slot->id == id)) |
0a706bee TY |
975 | continue; |
976 | if (!((base_gfn + npages <= slot->base_gfn) || | |
977 | (base_gfn >= slot->base_gfn + slot->npages))) | |
978 | goto out; | |
979 | } | |
6aa8b732 | 980 | } |
6aa8b732 | 981 | |
6aa8b732 AK |
982 | /* Free page dirty bitmap if unneeded */ |
983 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 984 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
985 | |
986 | r = -ENOMEM; | |
f64c0398 | 987 | if (change == KVM_MR_CREATE) { |
189a2f7b | 988 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 989 | |
5587027c | 990 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 991 | goto out_free; |
6aa8b732 | 992 | } |
ec04b260 | 993 | |
6aa8b732 AK |
994 | /* Allocate page dirty bitmap if needed */ |
995 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 996 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 997 | goto out_free; |
6aa8b732 AK |
998 | } |
999 | ||
74496134 | 1000 | slots = kvm_kvzalloc(sizeof(struct kvm_memslots)); |
f2a81036 PB |
1001 | if (!slots) |
1002 | goto out_free; | |
f481b069 | 1003 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 1004 | |
f64c0398 | 1005 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 1006 | slot = id_to_memslot(slots, id); |
28a37544 XG |
1007 | slot->flags |= KVM_MEMSLOT_INVALID; |
1008 | ||
f481b069 | 1009 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 1010 | |
e40f193f AW |
1011 | /* slot was deleted or moved, clear iommu mapping */ |
1012 | kvm_iommu_unmap_pages(kvm, &old); | |
12d6e753 MT |
1013 | /* From this point no new shadow pages pointing to a deleted, |
1014 | * or moved, memslot will be created. | |
bc6678a3 MT |
1015 | * |
1016 | * validation of sp->gfn happens in: | |
b7d409de XL |
1017 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
1018 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 1019 | */ |
2df72e9b | 1020 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
1021 | |
1022 | /* | |
1023 | * We can re-use the old_memslots from above, the only difference | |
1024 | * from the currently installed memslots is the invalid flag. This | |
1025 | * will get overwritten by update_memslots anyway. | |
1026 | */ | |
b7f69c55 | 1027 | slots = old_memslots; |
bc6678a3 | 1028 | } |
34d4cb8f | 1029 | |
7b6195a9 | 1030 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 1031 | if (r) |
b7f69c55 | 1032 | goto out_slots; |
f7784b8e | 1033 | |
a47d2b07 | 1034 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 1035 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 1036 | new.dirty_bitmap = NULL; |
db3fe4eb | 1037 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
1038 | } |
1039 | ||
5cc15027 | 1040 | update_memslots(slots, &new); |
f481b069 | 1041 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 1042 | |
f36f3f28 | 1043 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 1044 | |
a47d2b07 | 1045 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 1046 | kvfree(old_memslots); |
bc6678a3 | 1047 | |
261874b0 AW |
1048 | /* |
1049 | * IOMMU mapping: New slots need to be mapped. Old slots need to be | |
75d61fbc TY |
1050 | * un-mapped and re-mapped if their base changes. Since base change |
1051 | * unmapping is handled above with slot deletion, mapping alone is | |
1052 | * needed here. Anything else the iommu might care about for existing | |
1053 | * slots (size changes, userspace addr changes and read-only flag | |
1054 | * changes) is disallowed above, so any other attribute changes getting | |
1055 | * here can be skipped. | |
261874b0 | 1056 | */ |
75d61fbc TY |
1057 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
1058 | r = kvm_iommu_map_pages(kvm, &new); | |
e0230e13 | 1059 | return r; |
bc6678a3 MT |
1060 | } |
1061 | ||
6aa8b732 AK |
1062 | return 0; |
1063 | ||
e40f193f | 1064 | out_slots: |
74496134 | 1065 | kvfree(slots); |
f78e0e2e | 1066 | out_free: |
a47d2b07 | 1067 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
1068 | out: |
1069 | return r; | |
210c7c4d | 1070 | } |
f78e0e2e SY |
1071 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1072 | ||
1073 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1074 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1075 | { |
1076 | int r; | |
1077 | ||
79fac95e | 1078 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1079 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1080 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1081 | return r; |
1082 | } | |
210c7c4d IE |
1083 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1084 | ||
7940876e SH |
1085 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1086 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1087 | { |
f481b069 | 1088 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1089 | return -EINVAL; |
09170a49 | 1090 | |
47ae31e2 | 1091 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1092 | } |
1093 | ||
5bb064dc ZX |
1094 | int kvm_get_dirty_log(struct kvm *kvm, |
1095 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 1096 | { |
9f6b8029 | 1097 | struct kvm_memslots *slots; |
6aa8b732 | 1098 | struct kvm_memory_slot *memslot; |
f481b069 | 1099 | int r, i, as_id, id; |
87bf6e7d | 1100 | unsigned long n; |
6aa8b732 AK |
1101 | unsigned long any = 0; |
1102 | ||
6aa8b732 | 1103 | r = -EINVAL; |
f481b069 PB |
1104 | as_id = log->slot >> 16; |
1105 | id = (u16)log->slot; | |
1106 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
6aa8b732 AK |
1107 | goto out; |
1108 | ||
f481b069 PB |
1109 | slots = __kvm_memslots(kvm, as_id); |
1110 | memslot = id_to_memslot(slots, id); | |
6aa8b732 AK |
1111 | r = -ENOENT; |
1112 | if (!memslot->dirty_bitmap) | |
1113 | goto out; | |
1114 | ||
87bf6e7d | 1115 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 1116 | |
cd1a4a98 | 1117 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1118 | any = memslot->dirty_bitmap[i]; |
1119 | ||
1120 | r = -EFAULT; | |
1121 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) | |
1122 | goto out; | |
1123 | ||
5bb064dc ZX |
1124 | if (any) |
1125 | *is_dirty = 1; | |
6aa8b732 AK |
1126 | |
1127 | r = 0; | |
6aa8b732 | 1128 | out: |
6aa8b732 AK |
1129 | return r; |
1130 | } | |
2ba9f0d8 | 1131 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1132 | |
ba0513b5 MS |
1133 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1134 | /** | |
1135 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1136 | * are dirty write protect them for next write. | |
1137 | * @kvm: pointer to kvm instance | |
1138 | * @log: slot id and address to which we copy the log | |
1139 | * @is_dirty: flag set if any page is dirty | |
1140 | * | |
1141 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1142 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1143 | * following order: | |
1144 | * | |
1145 | * 1. Take a snapshot of the bit and clear it if needed. | |
1146 | * 2. Write protect the corresponding page. | |
1147 | * 3. Copy the snapshot to the userspace. | |
1148 | * 4. Upon return caller flushes TLB's if needed. | |
1149 | * | |
1150 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1151 | * entry. This is not a problem because the page is reported dirty using | |
1152 | * the snapshot taken before and step 4 ensures that writes done after | |
1153 | * exiting to userspace will be logged for the next call. | |
1154 | * | |
1155 | */ | |
1156 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1157 | struct kvm_dirty_log *log, bool *is_dirty) | |
1158 | { | |
9f6b8029 | 1159 | struct kvm_memslots *slots; |
ba0513b5 | 1160 | struct kvm_memory_slot *memslot; |
f481b069 | 1161 | int r, i, as_id, id; |
ba0513b5 MS |
1162 | unsigned long n; |
1163 | unsigned long *dirty_bitmap; | |
1164 | unsigned long *dirty_bitmap_buffer; | |
1165 | ||
1166 | r = -EINVAL; | |
f481b069 PB |
1167 | as_id = log->slot >> 16; |
1168 | id = (u16)log->slot; | |
1169 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
ba0513b5 MS |
1170 | goto out; |
1171 | ||
f481b069 PB |
1172 | slots = __kvm_memslots(kvm, as_id); |
1173 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1174 | |
1175 | dirty_bitmap = memslot->dirty_bitmap; | |
1176 | r = -ENOENT; | |
1177 | if (!dirty_bitmap) | |
1178 | goto out; | |
1179 | ||
1180 | n = kvm_dirty_bitmap_bytes(memslot); | |
1181 | ||
1182 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1183 | memset(dirty_bitmap_buffer, 0, n); | |
1184 | ||
1185 | spin_lock(&kvm->mmu_lock); | |
1186 | *is_dirty = false; | |
1187 | for (i = 0; i < n / sizeof(long); i++) { | |
1188 | unsigned long mask; | |
1189 | gfn_t offset; | |
1190 | ||
1191 | if (!dirty_bitmap[i]) | |
1192 | continue; | |
1193 | ||
1194 | *is_dirty = true; | |
1195 | ||
1196 | mask = xchg(&dirty_bitmap[i], 0); | |
1197 | dirty_bitmap_buffer[i] = mask; | |
1198 | ||
58d2930f TY |
1199 | if (mask) { |
1200 | offset = i * BITS_PER_LONG; | |
1201 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1202 | offset, mask); | |
1203 | } | |
ba0513b5 MS |
1204 | } |
1205 | ||
1206 | spin_unlock(&kvm->mmu_lock); | |
1207 | ||
1208 | r = -EFAULT; | |
1209 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1210 | goto out; | |
1211 | ||
1212 | r = 0; | |
1213 | out: | |
1214 | return r; | |
1215 | } | |
1216 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1217 | #endif | |
1218 | ||
db3fe4eb TY |
1219 | bool kvm_largepages_enabled(void) |
1220 | { | |
1221 | return largepages_enabled; | |
1222 | } | |
1223 | ||
54dee993 MT |
1224 | void kvm_disable_largepages(void) |
1225 | { | |
1226 | largepages_enabled = false; | |
1227 | } | |
1228 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1229 | ||
49c7754c GN |
1230 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1231 | { | |
1232 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1233 | } | |
a1f4d395 | 1234 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1235 | |
8e73485c PB |
1236 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1237 | { | |
1238 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1239 | } | |
1240 | ||
33e94154 | 1241 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1242 | { |
bf3e05bc | 1243 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1244 | |
bbacc0c1 | 1245 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc | 1246 | memslot->flags & KVM_MEMSLOT_INVALID) |
33e94154 | 1247 | return false; |
e0d62c7f | 1248 | |
33e94154 | 1249 | return true; |
e0d62c7f IE |
1250 | } |
1251 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1252 | ||
8f0b1ab6 JR |
1253 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1254 | { | |
1255 | struct vm_area_struct *vma; | |
1256 | unsigned long addr, size; | |
1257 | ||
1258 | size = PAGE_SIZE; | |
1259 | ||
1260 | addr = gfn_to_hva(kvm, gfn); | |
1261 | if (kvm_is_error_hva(addr)) | |
1262 | return PAGE_SIZE; | |
1263 | ||
1264 | down_read(¤t->mm->mmap_sem); | |
1265 | vma = find_vma(current->mm, addr); | |
1266 | if (!vma) | |
1267 | goto out; | |
1268 | ||
1269 | size = vma_kernel_pagesize(vma); | |
1270 | ||
1271 | out: | |
1272 | up_read(¤t->mm->mmap_sem); | |
1273 | ||
1274 | return size; | |
1275 | } | |
1276 | ||
4d8b81ab XG |
1277 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1278 | { | |
1279 | return slot->flags & KVM_MEM_READONLY; | |
1280 | } | |
1281 | ||
4d8b81ab XG |
1282 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1283 | gfn_t *nr_pages, bool write) | |
539cb660 | 1284 | { |
bc6678a3 | 1285 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1286 | return KVM_HVA_ERR_BAD; |
48987781 | 1287 | |
4d8b81ab XG |
1288 | if (memslot_is_readonly(slot) && write) |
1289 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1290 | |
1291 | if (nr_pages) | |
1292 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1293 | ||
4d8b81ab | 1294 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1295 | } |
48987781 | 1296 | |
4d8b81ab XG |
1297 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1298 | gfn_t *nr_pages) | |
1299 | { | |
1300 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1301 | } |
48987781 | 1302 | |
4d8b81ab | 1303 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1304 | gfn_t gfn) |
4d8b81ab XG |
1305 | { |
1306 | return gfn_to_hva_many(slot, gfn, NULL); | |
1307 | } | |
1308 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1309 | ||
48987781 XG |
1310 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1311 | { | |
49c7754c | 1312 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1313 | } |
0d150298 | 1314 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1315 | |
8e73485c PB |
1316 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1317 | { | |
1318 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1319 | } | |
1320 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1321 | ||
86ab8cff | 1322 | /* |
ba6a3541 PB |
1323 | * If writable is set to false, the hva returned by this function is only |
1324 | * allowed to be read. | |
86ab8cff | 1325 | */ |
64d83126 CD |
1326 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1327 | gfn_t gfn, bool *writable) | |
86ab8cff | 1328 | { |
a2ac07fe GN |
1329 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1330 | ||
1331 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1332 | *writable = !memslot_is_readonly(slot); |
1333 | ||
a2ac07fe | 1334 | return hva; |
86ab8cff XG |
1335 | } |
1336 | ||
64d83126 CD |
1337 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1338 | { | |
1339 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1340 | ||
1341 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1342 | } | |
1343 | ||
8e73485c PB |
1344 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1345 | { | |
1346 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1347 | ||
1348 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1349 | } | |
1350 | ||
d4edcf0d DH |
1351 | static int get_user_page_nowait(unsigned long start, int write, |
1352 | struct page **page) | |
0857b9e9 | 1353 | { |
0d731759 | 1354 | int flags = FOLL_NOWAIT | FOLL_HWPOISON; |
0857b9e9 GN |
1355 | |
1356 | if (write) | |
1357 | flags |= FOLL_WRITE; | |
1358 | ||
0d731759 | 1359 | return get_user_pages(start, 1, flags, page, NULL); |
0857b9e9 GN |
1360 | } |
1361 | ||
fafc3dba HY |
1362 | static inline int check_user_page_hwpoison(unsigned long addr) |
1363 | { | |
0d731759 | 1364 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1365 | |
0d731759 | 1366 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1367 | return rc == -EHWPOISON; |
1368 | } | |
1369 | ||
2fc84311 XG |
1370 | /* |
1371 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1372 | * true indicates success, otherwise false is returned. | |
1373 | */ | |
1374 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
ba049e93 | 1375 | bool write_fault, bool *writable, kvm_pfn_t *pfn) |
954bbbc2 | 1376 | { |
8d4e1288 | 1377 | struct page *page[1]; |
2fc84311 | 1378 | int npages; |
954bbbc2 | 1379 | |
2fc84311 XG |
1380 | if (!(async || atomic)) |
1381 | return false; | |
af585b92 | 1382 | |
12ce13fe XG |
1383 | /* |
1384 | * Fast pin a writable pfn only if it is a write fault request | |
1385 | * or the caller allows to map a writable pfn for a read fault | |
1386 | * request. | |
1387 | */ | |
1388 | if (!(write_fault || writable)) | |
1389 | return false; | |
612819c3 | 1390 | |
2fc84311 XG |
1391 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1392 | if (npages == 1) { | |
1393 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1394 | |
2fc84311 XG |
1395 | if (writable) |
1396 | *writable = true; | |
1397 | return true; | |
1398 | } | |
af585b92 | 1399 | |
2fc84311 XG |
1400 | return false; |
1401 | } | |
612819c3 | 1402 | |
2fc84311 XG |
1403 | /* |
1404 | * The slow path to get the pfn of the specified host virtual address, | |
1405 | * 1 indicates success, -errno is returned if error is detected. | |
1406 | */ | |
1407 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1408 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 XG |
1409 | { |
1410 | struct page *page[1]; | |
1411 | int npages = 0; | |
612819c3 | 1412 | |
2fc84311 XG |
1413 | might_sleep(); |
1414 | ||
1415 | if (writable) | |
1416 | *writable = write_fault; | |
1417 | ||
1418 | if (async) { | |
1419 | down_read(¤t->mm->mmap_sem); | |
d4edcf0d | 1420 | npages = get_user_page_nowait(addr, write_fault, page); |
2fc84311 | 1421 | up_read(¤t->mm->mmap_sem); |
d4944b0e | 1422 | } else { |
8b7457ef | 1423 | unsigned int flags = FOLL_HWPOISON; |
d4944b0e LS |
1424 | |
1425 | if (write_fault) | |
1426 | flags |= FOLL_WRITE; | |
1427 | ||
8b7457ef | 1428 | npages = get_user_pages_unlocked(addr, 1, page, flags); |
d4944b0e | 1429 | } |
2fc84311 XG |
1430 | if (npages != 1) |
1431 | return npages; | |
1432 | ||
1433 | /* map read fault as writable if possible */ | |
12ce13fe | 1434 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1435 | struct page *wpage[1]; |
1436 | ||
1437 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1438 | if (npages == 1) { | |
1439 | *writable = true; | |
1440 | put_page(page[0]); | |
1441 | page[0] = wpage[0]; | |
612819c3 | 1442 | } |
2fc84311 XG |
1443 | |
1444 | npages = 1; | |
887c08ac | 1445 | } |
2fc84311 XG |
1446 | *pfn = page_to_pfn(page[0]); |
1447 | return npages; | |
1448 | } | |
539cb660 | 1449 | |
4d8b81ab XG |
1450 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1451 | { | |
1452 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1453 | return false; | |
2e2e3738 | 1454 | |
4d8b81ab XG |
1455 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1456 | return false; | |
887c08ac | 1457 | |
4d8b81ab XG |
1458 | return true; |
1459 | } | |
bf998156 | 1460 | |
92176a8e PB |
1461 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1462 | unsigned long addr, bool *async, | |
1463 | bool write_fault, kvm_pfn_t *p_pfn) | |
1464 | { | |
add6a0cd PB |
1465 | unsigned long pfn; |
1466 | int r; | |
1467 | ||
1468 | r = follow_pfn(vma, addr, &pfn); | |
1469 | if (r) { | |
1470 | /* | |
1471 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1472 | * not call the fault handler, so do it here. | |
1473 | */ | |
1474 | bool unlocked = false; | |
1475 | r = fixup_user_fault(current, current->mm, addr, | |
1476 | (write_fault ? FAULT_FLAG_WRITE : 0), | |
1477 | &unlocked); | |
1478 | if (unlocked) | |
1479 | return -EAGAIN; | |
1480 | if (r) | |
1481 | return r; | |
1482 | ||
1483 | r = follow_pfn(vma, addr, &pfn); | |
1484 | if (r) | |
1485 | return r; | |
1486 | ||
1487 | } | |
1488 | ||
1489 | ||
1490 | /* | |
1491 | * Get a reference here because callers of *hva_to_pfn* and | |
1492 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1493 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1494 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1495 | * simply do nothing for reserved pfns. | |
1496 | * | |
1497 | * Whoever called remap_pfn_range is also going to call e.g. | |
1498 | * unmap_mapping_range before the underlying pages are freed, | |
1499 | * causing a call to our MMU notifier. | |
1500 | */ | |
1501 | kvm_get_pfn(pfn); | |
1502 | ||
1503 | *p_pfn = pfn; | |
92176a8e PB |
1504 | return 0; |
1505 | } | |
1506 | ||
12ce13fe XG |
1507 | /* |
1508 | * Pin guest page in memory and return its pfn. | |
1509 | * @addr: host virtual address which maps memory to the guest | |
1510 | * @atomic: whether this function can sleep | |
1511 | * @async: whether this function need to wait IO complete if the | |
1512 | * host page is not in the memory | |
1513 | * @write_fault: whether we should get a writable host page | |
1514 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1515 | * | |
1516 | * The function will map a writable host page for these two cases: | |
1517 | * 1): @write_fault = true | |
1518 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1519 | * whether the mapping is writable. | |
1520 | */ | |
ba049e93 | 1521 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1522 | bool write_fault, bool *writable) |
1523 | { | |
1524 | struct vm_area_struct *vma; | |
ba049e93 | 1525 | kvm_pfn_t pfn = 0; |
92176a8e | 1526 | int npages, r; |
2e2e3738 | 1527 | |
2fc84311 XG |
1528 | /* we can do it either atomically or asynchronously, not both */ |
1529 | BUG_ON(atomic && async); | |
8d4e1288 | 1530 | |
2fc84311 XG |
1531 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1532 | return pfn; | |
1533 | ||
1534 | if (atomic) | |
1535 | return KVM_PFN_ERR_FAULT; | |
1536 | ||
1537 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1538 | if (npages == 1) | |
1539 | return pfn; | |
8d4e1288 | 1540 | |
2fc84311 XG |
1541 | down_read(¤t->mm->mmap_sem); |
1542 | if (npages == -EHWPOISON || | |
1543 | (!async && check_user_page_hwpoison(addr))) { | |
1544 | pfn = KVM_PFN_ERR_HWPOISON; | |
1545 | goto exit; | |
1546 | } | |
1547 | ||
add6a0cd | 1548 | retry: |
2fc84311 XG |
1549 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
1550 | ||
1551 | if (vma == NULL) | |
1552 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e PB |
1553 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1554 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn); | |
add6a0cd PB |
1555 | if (r == -EAGAIN) |
1556 | goto retry; | |
92176a8e PB |
1557 | if (r < 0) |
1558 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 1559 | } else { |
4d8b81ab | 1560 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1561 | *async = true; |
1562 | pfn = KVM_PFN_ERR_FAULT; | |
1563 | } | |
1564 | exit: | |
1565 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1566 | return pfn; |
35149e21 AL |
1567 | } |
1568 | ||
ba049e93 DW |
1569 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
1570 | bool atomic, bool *async, bool write_fault, | |
1571 | bool *writable) | |
887c08ac | 1572 | { |
4d8b81ab XG |
1573 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1574 | ||
b2740d35 PB |
1575 | if (addr == KVM_HVA_ERR_RO_BAD) { |
1576 | if (writable) | |
1577 | *writable = false; | |
4d8b81ab | 1578 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 1579 | } |
4d8b81ab | 1580 | |
b2740d35 PB |
1581 | if (kvm_is_error_hva(addr)) { |
1582 | if (writable) | |
1583 | *writable = false; | |
81c52c56 | 1584 | return KVM_PFN_NOSLOT; |
b2740d35 | 1585 | } |
4d8b81ab XG |
1586 | |
1587 | /* Do not map writable pfn in the readonly memslot. */ | |
1588 | if (writable && memslot_is_readonly(slot)) { | |
1589 | *writable = false; | |
1590 | writable = NULL; | |
1591 | } | |
1592 | ||
1593 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1594 | writable); | |
887c08ac | 1595 | } |
3520469d | 1596 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1597 | |
ba049e93 | 1598 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
1599 | bool *writable) |
1600 | { | |
e37afc6e PB |
1601 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1602 | write_fault, writable); | |
612819c3 MT |
1603 | } |
1604 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1605 | ||
ba049e93 | 1606 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1607 | { |
4d8b81ab | 1608 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1609 | } |
e37afc6e | 1610 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1611 | |
ba049e93 | 1612 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1613 | { |
4d8b81ab | 1614 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1615 | } |
037d92dc | 1616 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1617 | |
ba049e93 | 1618 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1619 | { |
1620 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1621 | } | |
1622 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1623 | ||
ba049e93 | 1624 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1625 | { |
1626 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1627 | } | |
1628 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1629 | ||
ba049e93 | 1630 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1631 | { |
1632 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1633 | } | |
1634 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1635 | ||
ba049e93 | 1636 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1637 | { |
1638 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1639 | } | |
1640 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1641 | ||
d9ef13c2 PB |
1642 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1643 | struct page **pages, int nr_pages) | |
48987781 XG |
1644 | { |
1645 | unsigned long addr; | |
1646 | gfn_t entry; | |
1647 | ||
d9ef13c2 | 1648 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1649 | if (kvm_is_error_hva(addr)) |
1650 | return -1; | |
1651 | ||
1652 | if (entry < nr_pages) | |
1653 | return 0; | |
1654 | ||
1655 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1656 | } | |
1657 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1658 | ||
ba049e93 | 1659 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 1660 | { |
81c52c56 | 1661 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1662 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1663 | |
bf4bea8e | 1664 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1665 | WARN_ON(1); |
6cede2e6 | 1666 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1667 | } |
a2766325 XG |
1668 | |
1669 | return pfn_to_page(pfn); | |
1670 | } | |
1671 | ||
35149e21 AL |
1672 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1673 | { | |
ba049e93 | 1674 | kvm_pfn_t pfn; |
2e2e3738 AL |
1675 | |
1676 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1677 | |
a2766325 | 1678 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1679 | } |
1680 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1681 | ||
8e73485c PB |
1682 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1683 | { | |
ba049e93 | 1684 | kvm_pfn_t pfn; |
8e73485c PB |
1685 | |
1686 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1687 | ||
1688 | return kvm_pfn_to_page(pfn); | |
1689 | } | |
1690 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1691 | ||
b4231d61 IE |
1692 | void kvm_release_page_clean(struct page *page) |
1693 | { | |
32cad84f XG |
1694 | WARN_ON(is_error_page(page)); |
1695 | ||
35149e21 | 1696 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1697 | } |
1698 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1699 | ||
ba049e93 | 1700 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 1701 | { |
bf4bea8e | 1702 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1703 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1704 | } |
1705 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1706 | ||
b4231d61 | 1707 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1708 | { |
a2766325 XG |
1709 | WARN_ON(is_error_page(page)); |
1710 | ||
35149e21 AL |
1711 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1712 | } | |
1713 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1714 | ||
ba049e93 | 1715 | static void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
1716 | { |
1717 | kvm_set_pfn_dirty(pfn); | |
1718 | kvm_release_pfn_clean(pfn); | |
1719 | } | |
35149e21 | 1720 | |
ba049e93 | 1721 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 1722 | { |
bf4bea8e | 1723 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 | 1724 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1725 | |
2e2e3738 AL |
1726 | if (!PageReserved(page)) |
1727 | SetPageDirty(page); | |
1728 | } | |
8a7ae055 | 1729 | } |
35149e21 AL |
1730 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1731 | ||
ba049e93 | 1732 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 1733 | { |
bf4bea8e | 1734 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1735 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1736 | } |
1737 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1738 | ||
ba049e93 | 1739 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 1740 | { |
bf4bea8e | 1741 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1742 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1743 | } |
1744 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1745 | |
195aefde IE |
1746 | static int next_segment(unsigned long len, int offset) |
1747 | { | |
1748 | if (len > PAGE_SIZE - offset) | |
1749 | return PAGE_SIZE - offset; | |
1750 | else | |
1751 | return len; | |
1752 | } | |
1753 | ||
8e73485c PB |
1754 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1755 | void *data, int offset, int len) | |
195aefde | 1756 | { |
e0506bcb IE |
1757 | int r; |
1758 | unsigned long addr; | |
195aefde | 1759 | |
8e73485c | 1760 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1761 | if (kvm_is_error_hva(addr)) |
1762 | return -EFAULT; | |
3180a7fc | 1763 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1764 | if (r) |
195aefde | 1765 | return -EFAULT; |
195aefde IE |
1766 | return 0; |
1767 | } | |
8e73485c PB |
1768 | |
1769 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1770 | int len) | |
1771 | { | |
1772 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1773 | ||
1774 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1775 | } | |
195aefde IE |
1776 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1777 | ||
8e73485c PB |
1778 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1779 | int offset, int len) | |
1780 | { | |
1781 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1782 | ||
1783 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1784 | } | |
1785 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1786 | ||
195aefde IE |
1787 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1788 | { | |
1789 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1790 | int seg; | |
1791 | int offset = offset_in_page(gpa); | |
1792 | int ret; | |
1793 | ||
1794 | while ((seg = next_segment(len, offset)) != 0) { | |
1795 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1796 | if (ret < 0) | |
1797 | return ret; | |
1798 | offset = 0; | |
1799 | len -= seg; | |
1800 | data += seg; | |
1801 | ++gfn; | |
1802 | } | |
1803 | return 0; | |
1804 | } | |
1805 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1806 | ||
8e73485c | 1807 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 1808 | { |
7ec54588 | 1809 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 1810 | int seg; |
7ec54588 | 1811 | int offset = offset_in_page(gpa); |
8e73485c PB |
1812 | int ret; |
1813 | ||
1814 | while ((seg = next_segment(len, offset)) != 0) { | |
1815 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
1816 | if (ret < 0) | |
1817 | return ret; | |
1818 | offset = 0; | |
1819 | len -= seg; | |
1820 | data += seg; | |
1821 | ++gfn; | |
1822 | } | |
1823 | return 0; | |
1824 | } | |
1825 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 1826 | |
8e73485c PB |
1827 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1828 | void *data, int offset, unsigned long len) | |
1829 | { | |
1830 | int r; | |
1831 | unsigned long addr; | |
1832 | ||
1833 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
1834 | if (kvm_is_error_hva(addr)) |
1835 | return -EFAULT; | |
0aac03f0 | 1836 | pagefault_disable(); |
3180a7fc | 1837 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1838 | pagefault_enable(); |
7ec54588 MT |
1839 | if (r) |
1840 | return -EFAULT; | |
1841 | return 0; | |
1842 | } | |
7ec54588 | 1843 | |
8e73485c PB |
1844 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1845 | unsigned long len) | |
1846 | { | |
1847 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1848 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1849 | int offset = offset_in_page(gpa); | |
1850 | ||
1851 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1852 | } | |
1853 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
1854 | ||
1855 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
1856 | void *data, unsigned long len) | |
1857 | { | |
1858 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1859 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1860 | int offset = offset_in_page(gpa); | |
1861 | ||
1862 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1863 | } | |
1864 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
1865 | ||
1866 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
1867 | const void *data, int offset, int len) | |
195aefde | 1868 | { |
e0506bcb IE |
1869 | int r; |
1870 | unsigned long addr; | |
195aefde | 1871 | |
251eb841 | 1872 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
1873 | if (kvm_is_error_hva(addr)) |
1874 | return -EFAULT; | |
8b0cedff | 1875 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1876 | if (r) |
195aefde | 1877 | return -EFAULT; |
bc009e43 | 1878 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
1879 | return 0; |
1880 | } | |
8e73485c PB |
1881 | |
1882 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
1883 | const void *data, int offset, int len) | |
1884 | { | |
1885 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1886 | ||
1887 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1888 | } | |
195aefde IE |
1889 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
1890 | ||
8e73485c PB |
1891 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
1892 | const void *data, int offset, int len) | |
1893 | { | |
1894 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1895 | ||
1896 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1897 | } | |
1898 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
1899 | ||
195aefde IE |
1900 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
1901 | unsigned long len) | |
1902 | { | |
1903 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1904 | int seg; | |
1905 | int offset = offset_in_page(gpa); | |
1906 | int ret; | |
1907 | ||
1908 | while ((seg = next_segment(len, offset)) != 0) { | |
1909 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1910 | if (ret < 0) | |
1911 | return ret; | |
1912 | offset = 0; | |
1913 | len -= seg; | |
1914 | data += seg; | |
1915 | ++gfn; | |
1916 | } | |
1917 | return 0; | |
1918 | } | |
ff651cb6 | 1919 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1920 | |
8e73485c PB |
1921 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
1922 | unsigned long len) | |
1923 | { | |
1924 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1925 | int seg; | |
1926 | int offset = offset_in_page(gpa); | |
1927 | int ret; | |
1928 | ||
1929 | while ((seg = next_segment(len, offset)) != 0) { | |
1930 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
1931 | if (ret < 0) | |
1932 | return ret; | |
1933 | offset = 0; | |
1934 | len -= seg; | |
1935 | data += seg; | |
1936 | ++gfn; | |
1937 | } | |
1938 | return 0; | |
1939 | } | |
1940 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
1941 | ||
49c7754c | 1942 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
8f964525 | 1943 | gpa_t gpa, unsigned long len) |
49c7754c GN |
1944 | { |
1945 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1946 | int offset = offset_in_page(gpa); | |
8f964525 AH |
1947 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1948 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1949 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1950 | gfn_t nr_pages_avail; | |
49c7754c GN |
1951 | |
1952 | ghc->gpa = gpa; | |
1953 | ghc->generation = slots->generation; | |
8f964525 AH |
1954 | ghc->len = len; |
1955 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
ca3f0874 RK |
1956 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
1957 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { | |
49c7754c | 1958 | ghc->hva += offset; |
8f964525 AH |
1959 | } else { |
1960 | /* | |
1961 | * If the requested region crosses two memslots, we still | |
1962 | * verify that the entire region is valid here. | |
1963 | */ | |
1964 | while (start_gfn <= end_gfn) { | |
1965 | ghc->memslot = gfn_to_memslot(kvm, start_gfn); | |
1966 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
1967 | &nr_pages_avail); | |
1968 | if (kvm_is_error_hva(ghc->hva)) | |
1969 | return -EFAULT; | |
1970 | start_gfn += nr_pages_avail; | |
1971 | } | |
1972 | /* Use the slow path for cross page reads and writes. */ | |
1973 | ghc->memslot = NULL; | |
1974 | } | |
49c7754c GN |
1975 | return 0; |
1976 | } | |
1977 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); | |
1978 | ||
4ec6e863 PX |
1979 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1980 | void *data, int offset, unsigned long len) | |
49c7754c GN |
1981 | { |
1982 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1983 | int r; | |
4ec6e863 | 1984 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 1985 | |
4ec6e863 | 1986 | BUG_ON(len + offset > ghc->len); |
8f964525 | 1987 | |
49c7754c | 1988 | if (slots->generation != ghc->generation) |
8f964525 AH |
1989 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
1990 | ||
1991 | if (unlikely(!ghc->memslot)) | |
4ec6e863 | 1992 | return kvm_write_guest(kvm, gpa, data, len); |
49c7754c GN |
1993 | |
1994 | if (kvm_is_error_hva(ghc->hva)) | |
1995 | return -EFAULT; | |
1996 | ||
4ec6e863 | 1997 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
1998 | if (r) |
1999 | return -EFAULT; | |
4ec6e863 | 2000 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2001 | |
2002 | return 0; | |
2003 | } | |
4ec6e863 PX |
2004 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
2005 | ||
2006 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
2007 | void *data, unsigned long len) | |
2008 | { | |
2009 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); | |
2010 | } | |
49c7754c GN |
2011 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
2012 | ||
e03b644f GN |
2013 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2014 | void *data, unsigned long len) | |
2015 | { | |
2016 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
2017 | int r; | |
2018 | ||
8f964525 AH |
2019 | BUG_ON(len > ghc->len); |
2020 | ||
e03b644f | 2021 | if (slots->generation != ghc->generation) |
8f964525 AH |
2022 | kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len); |
2023 | ||
2024 | if (unlikely(!ghc->memslot)) | |
2025 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
e03b644f GN |
2026 | |
2027 | if (kvm_is_error_hva(ghc->hva)) | |
2028 | return -EFAULT; | |
2029 | ||
2030 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
2031 | if (r) | |
2032 | return -EFAULT; | |
2033 | ||
2034 | return 0; | |
2035 | } | |
2036 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); | |
2037 | ||
195aefde IE |
2038 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
2039 | { | |
8a3caa6d HC |
2040 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
2041 | ||
2042 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
2043 | } |
2044 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
2045 | ||
2046 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
2047 | { | |
2048 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2049 | int seg; | |
2050 | int offset = offset_in_page(gpa); | |
2051 | int ret; | |
2052 | ||
bfda0e84 | 2053 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
2054 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
2055 | if (ret < 0) | |
2056 | return ret; | |
2057 | offset = 0; | |
2058 | len -= seg; | |
2059 | ++gfn; | |
2060 | } | |
2061 | return 0; | |
2062 | } | |
2063 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2064 | ||
bc009e43 | 2065 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 2066 | gfn_t gfn) |
6aa8b732 | 2067 | { |
7e9d619d RR |
2068 | if (memslot && memslot->dirty_bitmap) { |
2069 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 2070 | |
b74ca3b3 | 2071 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
2072 | } |
2073 | } | |
2074 | ||
49c7754c GN |
2075 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2076 | { | |
2077 | struct kvm_memory_slot *memslot; | |
2078 | ||
2079 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 2080 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 2081 | } |
2ba9f0d8 | 2082 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2083 | |
8e73485c PB |
2084 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2085 | { | |
2086 | struct kvm_memory_slot *memslot; | |
2087 | ||
2088 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2089 | mark_page_dirty_in_slot(memslot, gfn); | |
2090 | } | |
2091 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2092 | ||
aca6ff29 WL |
2093 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
2094 | { | |
6b6de68c | 2095 | unsigned int old, val, grow; |
aca6ff29 | 2096 | |
2cbd7824 | 2097 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 2098 | grow = READ_ONCE(halt_poll_ns_grow); |
aca6ff29 | 2099 | /* 10us base */ |
6b6de68c | 2100 | if (val == 0 && grow) |
aca6ff29 WL |
2101 | val = 10000; |
2102 | else | |
6b6de68c | 2103 | val *= grow; |
aca6ff29 | 2104 | |
313f636d DM |
2105 | if (val > halt_poll_ns) |
2106 | val = halt_poll_ns; | |
2107 | ||
aca6ff29 | 2108 | vcpu->halt_poll_ns = val; |
2cbd7824 | 2109 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2110 | } |
2111 | ||
2112 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
2113 | { | |
6b6de68c | 2114 | unsigned int old, val, shrink; |
aca6ff29 | 2115 | |
2cbd7824 | 2116 | old = val = vcpu->halt_poll_ns; |
6b6de68c CB |
2117 | shrink = READ_ONCE(halt_poll_ns_shrink); |
2118 | if (shrink == 0) | |
aca6ff29 WL |
2119 | val = 0; |
2120 | else | |
6b6de68c | 2121 | val /= shrink; |
aca6ff29 WL |
2122 | |
2123 | vcpu->halt_poll_ns = val; | |
2cbd7824 | 2124 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2125 | } |
2126 | ||
f7819512 PB |
2127 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
2128 | { | |
2129 | if (kvm_arch_vcpu_runnable(vcpu)) { | |
2130 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
2131 | return -EINTR; | |
2132 | } | |
2133 | if (kvm_cpu_has_pending_timer(vcpu)) | |
2134 | return -EINTR; | |
2135 | if (signal_pending(current)) | |
2136 | return -EINTR; | |
2137 | ||
2138 | return 0; | |
2139 | } | |
2140 | ||
b6958ce4 ED |
2141 | /* |
2142 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
2143 | */ | |
8776e519 | 2144 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 2145 | { |
f7819512 | 2146 | ktime_t start, cur; |
8577370f | 2147 | DECLARE_SWAITQUEUE(wait); |
f7819512 | 2148 | bool waited = false; |
aca6ff29 | 2149 | u64 block_ns; |
f7819512 PB |
2150 | |
2151 | start = cur = ktime_get(); | |
19020f8a WL |
2152 | if (vcpu->halt_poll_ns) { |
2153 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); | |
f95ef0cd | 2154 | |
62bea5bf | 2155 | ++vcpu->stat.halt_attempted_poll; |
f7819512 PB |
2156 | do { |
2157 | /* | |
2158 | * This sets KVM_REQ_UNHALT if an interrupt | |
2159 | * arrives. | |
2160 | */ | |
2161 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
2162 | ++vcpu->stat.halt_successful_poll; | |
3491caf2 CB |
2163 | if (!vcpu_valid_wakeup(vcpu)) |
2164 | ++vcpu->stat.halt_poll_invalid; | |
f7819512 PB |
2165 | goto out; |
2166 | } | |
2167 | cur = ktime_get(); | |
2168 | } while (single_task_running() && ktime_before(cur, stop)); | |
2169 | } | |
e5c239cf | 2170 | |
3217f7c2 CD |
2171 | kvm_arch_vcpu_blocking(vcpu); |
2172 | ||
e5c239cf | 2173 | for (;;) { |
8577370f | 2174 | prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
e5c239cf | 2175 | |
f7819512 | 2176 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
2177 | break; |
2178 | ||
f7819512 | 2179 | waited = true; |
b6958ce4 | 2180 | schedule(); |
b6958ce4 | 2181 | } |
d3bef15f | 2182 | |
8577370f | 2183 | finish_swait(&vcpu->wq, &wait); |
f7819512 PB |
2184 | cur = ktime_get(); |
2185 | ||
3217f7c2 | 2186 | kvm_arch_vcpu_unblocking(vcpu); |
f7819512 | 2187 | out: |
aca6ff29 WL |
2188 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
2189 | ||
2086d320 CB |
2190 | if (!vcpu_valid_wakeup(vcpu)) |
2191 | shrink_halt_poll_ns(vcpu); | |
2192 | else if (halt_poll_ns) { | |
aca6ff29 WL |
2193 | if (block_ns <= vcpu->halt_poll_ns) |
2194 | ; | |
2195 | /* we had a long block, shrink polling */ | |
2086d320 | 2196 | else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns) |
aca6ff29 WL |
2197 | shrink_halt_poll_ns(vcpu); |
2198 | /* we had a short halt and our poll time is too small */ | |
2199 | else if (vcpu->halt_poll_ns < halt_poll_ns && | |
2200 | block_ns < halt_poll_ns) | |
2201 | grow_halt_poll_ns(vcpu); | |
edb9272f WL |
2202 | } else |
2203 | vcpu->halt_poll_ns = 0; | |
aca6ff29 | 2204 | |
3491caf2 CB |
2205 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
2206 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 2207 | } |
2ba9f0d8 | 2208 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 2209 | |
8c84780d | 2210 | #ifndef CONFIG_S390 |
dd1a4cc1 | 2211 | void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2212 | { |
8577370f | 2213 | struct swait_queue_head *wqp; |
b6d33834 CD |
2214 | |
2215 | wqp = kvm_arch_vcpu_wq(vcpu); | |
8577370f MT |
2216 | if (swait_active(wqp)) { |
2217 | swake_up(wqp); | |
b6d33834 CD |
2218 | ++vcpu->stat.halt_wakeup; |
2219 | } | |
2220 | ||
dd1a4cc1 RK |
2221 | } |
2222 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2223 | ||
2224 | /* | |
2225 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2226 | */ | |
2227 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2228 | { | |
2229 | int me; | |
2230 | int cpu = vcpu->cpu; | |
2231 | ||
2232 | kvm_vcpu_wake_up(vcpu); | |
b6d33834 CD |
2233 | me = get_cpu(); |
2234 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2235 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2236 | smp_send_reschedule(cpu); | |
2237 | put_cpu(); | |
2238 | } | |
a20ed54d | 2239 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
8c84780d | 2240 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2241 | |
fa93384f | 2242 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2243 | { |
2244 | struct pid *pid; | |
2245 | struct task_struct *task = NULL; | |
fa93384f | 2246 | int ret = 0; |
41628d33 KW |
2247 | |
2248 | rcu_read_lock(); | |
2249 | pid = rcu_dereference(target->pid); | |
2250 | if (pid) | |
27fbe64b | 2251 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2252 | rcu_read_unlock(); |
2253 | if (!task) | |
c45c528e | 2254 | return ret; |
c45c528e | 2255 | ret = yield_to(task, 1); |
41628d33 | 2256 | put_task_struct(task); |
c45c528e R |
2257 | |
2258 | return ret; | |
41628d33 KW |
2259 | } |
2260 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2261 | ||
06e48c51 R |
2262 | /* |
2263 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2264 | * Most eligible candidate to yield is decided by following heuristics: | |
2265 | * | |
2266 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2267 | * (preempted lock holder), indicated by @in_spin_loop. | |
2268 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2269 | * | |
2270 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2271 | * chance last time (mostly it has become eligible now since we have probably | |
2272 | * yielded to lockholder in last iteration. This is done by toggling | |
2273 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2274 | * | |
2275 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2276 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2277 | * burning. Giving priority for a potential lock-holder increases lock | |
2278 | * progress. | |
2279 | * | |
2280 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2281 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2282 | * and continue with next VCPU and so on. | |
2283 | */ | |
7940876e | 2284 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2285 | { |
4a55dd72 | 2286 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2287 | bool eligible; |
2288 | ||
2289 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2290 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2291 | |
2292 | if (vcpu->spin_loop.in_spin_loop) | |
2293 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2294 | ||
2295 | return eligible; | |
4a55dd72 SW |
2296 | #else |
2297 | return true; | |
06e48c51 | 2298 | #endif |
4a55dd72 | 2299 | } |
c45c528e | 2300 | |
217ece61 | 2301 | void kvm_vcpu_on_spin(struct kvm_vcpu *me) |
d255f4f2 | 2302 | { |
217ece61 RR |
2303 | struct kvm *kvm = me->kvm; |
2304 | struct kvm_vcpu *vcpu; | |
2305 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2306 | int yielded = 0; | |
c45c528e | 2307 | int try = 3; |
217ece61 RR |
2308 | int pass; |
2309 | int i; | |
d255f4f2 | 2310 | |
4c088493 | 2311 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2312 | /* |
2313 | * We boost the priority of a VCPU that is runnable but not | |
2314 | * currently running, because it got preempted by something | |
2315 | * else and called schedule in __vcpu_run. Hopefully that | |
2316 | * VCPU is holding the lock that we need and will release it. | |
2317 | * We approximate round-robin by starting at the last boosted VCPU. | |
2318 | */ | |
c45c528e | 2319 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2320 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2321 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2322 | i = last_boosted_vcpu; |
2323 | continue; | |
2324 | } else if (pass && i > last_boosted_vcpu) | |
2325 | break; | |
7bc7ae25 R |
2326 | if (!ACCESS_ONCE(vcpu->preempted)) |
2327 | continue; | |
217ece61 RR |
2328 | if (vcpu == me) |
2329 | continue; | |
8577370f | 2330 | if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 2331 | continue; |
06e48c51 R |
2332 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2333 | continue; | |
c45c528e R |
2334 | |
2335 | yielded = kvm_vcpu_yield_to(vcpu); | |
2336 | if (yielded > 0) { | |
217ece61 | 2337 | kvm->last_boosted_vcpu = i; |
217ece61 | 2338 | break; |
c45c528e R |
2339 | } else if (yielded < 0) { |
2340 | try--; | |
2341 | if (!try) | |
2342 | break; | |
217ece61 | 2343 | } |
217ece61 RR |
2344 | } |
2345 | } | |
4c088493 | 2346 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2347 | |
2348 | /* Ensure vcpu is not eligible during next spinloop */ | |
2349 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2350 | } |
2351 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2352 | ||
e4a533a4 | 2353 | static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
9a2bb7f4 AK |
2354 | { |
2355 | struct kvm_vcpu *vcpu = vma->vm_file->private_data; | |
9a2bb7f4 AK |
2356 | struct page *page; |
2357 | ||
e4a533a4 | 2358 | if (vmf->pgoff == 0) |
039576c0 | 2359 | page = virt_to_page(vcpu->run); |
09566765 | 2360 | #ifdef CONFIG_X86 |
e4a533a4 | 2361 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2362 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 LV |
2363 | #endif |
2364 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
2365 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) | |
2366 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2367 | #endif |
039576c0 | 2368 | else |
5b1c1493 | 2369 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2370 | get_page(page); |
e4a533a4 | 2371 | vmf->page = page; |
2372 | return 0; | |
9a2bb7f4 AK |
2373 | } |
2374 | ||
f0f37e2f | 2375 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2376 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2377 | }; |
2378 | ||
2379 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2380 | { | |
2381 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2382 | return 0; | |
2383 | } | |
2384 | ||
bccf2150 AK |
2385 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2386 | { | |
2387 | struct kvm_vcpu *vcpu = filp->private_data; | |
2388 | ||
45b5939e | 2389 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
66c0b394 | 2390 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2391 | return 0; |
2392 | } | |
2393 | ||
3d3aab1b | 2394 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2395 | .release = kvm_vcpu_release, |
2396 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2397 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2398 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2399 | #endif | |
9a2bb7f4 | 2400 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2401 | .llseek = noop_llseek, |
bccf2150 AK |
2402 | }; |
2403 | ||
2404 | /* | |
2405 | * Allocates an inode for the vcpu. | |
2406 | */ | |
2407 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2408 | { | |
24009b05 | 2409 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2410 | } |
2411 | ||
45b5939e LC |
2412 | static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
2413 | { | |
2414 | char dir_name[ITOA_MAX_LEN * 2]; | |
2415 | int ret; | |
2416 | ||
2417 | if (!kvm_arch_has_vcpu_debugfs()) | |
2418 | return 0; | |
2419 | ||
2420 | if (!debugfs_initialized()) | |
2421 | return 0; | |
2422 | ||
2423 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
2424 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, | |
2425 | vcpu->kvm->debugfs_dentry); | |
2426 | if (!vcpu->debugfs_dentry) | |
2427 | return -ENOMEM; | |
2428 | ||
2429 | ret = kvm_arch_create_vcpu_debugfs(vcpu); | |
2430 | if (ret < 0) { | |
2431 | debugfs_remove_recursive(vcpu->debugfs_dentry); | |
2432 | return ret; | |
2433 | } | |
2434 | ||
2435 | return 0; | |
2436 | } | |
2437 | ||
c5ea7660 AK |
2438 | /* |
2439 | * Creates some virtual cpus. Good luck creating more than one. | |
2440 | */ | |
73880c80 | 2441 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2442 | { |
2443 | int r; | |
e09fefde | 2444 | struct kvm_vcpu *vcpu; |
c5ea7660 | 2445 | |
0b1b1dfd | 2446 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
2447 | return -EINVAL; |
2448 | ||
6c7caebc PB |
2449 | mutex_lock(&kvm->lock); |
2450 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
2451 | mutex_unlock(&kvm->lock); | |
2452 | return -EINVAL; | |
2453 | } | |
2454 | ||
2455 | kvm->created_vcpus++; | |
2456 | mutex_unlock(&kvm->lock); | |
2457 | ||
73880c80 | 2458 | vcpu = kvm_arch_vcpu_create(kvm, id); |
6c7caebc PB |
2459 | if (IS_ERR(vcpu)) { |
2460 | r = PTR_ERR(vcpu); | |
2461 | goto vcpu_decrement; | |
2462 | } | |
c5ea7660 | 2463 | |
15ad7146 AK |
2464 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2465 | ||
26e5215f AK |
2466 | r = kvm_arch_vcpu_setup(vcpu); |
2467 | if (r) | |
d780592b | 2468 | goto vcpu_destroy; |
26e5215f | 2469 | |
45b5939e LC |
2470 | r = kvm_create_vcpu_debugfs(vcpu); |
2471 | if (r) | |
2472 | goto vcpu_destroy; | |
2473 | ||
11ec2804 | 2474 | mutex_lock(&kvm->lock); |
e09fefde DH |
2475 | if (kvm_get_vcpu_by_id(kvm, id)) { |
2476 | r = -EEXIST; | |
2477 | goto unlock_vcpu_destroy; | |
2478 | } | |
73880c80 GN |
2479 | |
2480 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2481 | |
fb3f0f51 | 2482 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2483 | kvm_get_kvm(kvm); |
bccf2150 | 2484 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2485 | if (r < 0) { |
2486 | kvm_put_kvm(kvm); | |
d780592b | 2487 | goto unlock_vcpu_destroy; |
73880c80 GN |
2488 | } |
2489 | ||
2490 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
dd489240 PB |
2491 | |
2492 | /* | |
2493 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2494 | * before kvm->online_vcpu's incremented value. | |
2495 | */ | |
73880c80 GN |
2496 | smp_wmb(); |
2497 | atomic_inc(&kvm->online_vcpus); | |
2498 | ||
73880c80 | 2499 | mutex_unlock(&kvm->lock); |
42897d86 | 2500 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2501 | return r; |
39c3b86e | 2502 | |
d780592b | 2503 | unlock_vcpu_destroy: |
7d8fece6 | 2504 | mutex_unlock(&kvm->lock); |
45b5939e | 2505 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
d780592b | 2506 | vcpu_destroy: |
d40ccc62 | 2507 | kvm_arch_vcpu_destroy(vcpu); |
6c7caebc PB |
2508 | vcpu_decrement: |
2509 | mutex_lock(&kvm->lock); | |
2510 | kvm->created_vcpus--; | |
2511 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
2512 | return r; |
2513 | } | |
2514 | ||
1961d276 AK |
2515 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2516 | { | |
2517 | if (sigset) { | |
2518 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2519 | vcpu->sigset_active = 1; | |
2520 | vcpu->sigset = *sigset; | |
2521 | } else | |
2522 | vcpu->sigset_active = 0; | |
2523 | return 0; | |
2524 | } | |
2525 | ||
bccf2150 AK |
2526 | static long kvm_vcpu_ioctl(struct file *filp, |
2527 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2528 | { |
bccf2150 | 2529 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2530 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2531 | int r; |
fa3795a7 DH |
2532 | struct kvm_fpu *fpu = NULL; |
2533 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2534 | |
6d4e4c4f AK |
2535 | if (vcpu->kvm->mm != current->mm) |
2536 | return -EIO; | |
2122ff5e | 2537 | |
2ea75be3 DM |
2538 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2539 | return -EINVAL; | |
2540 | ||
2f4d9b54 | 2541 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2542 | /* |
2543 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2544 | * so vcpu_load() would break it. | |
2545 | */ | |
47b43c52 | 2546 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
2122ff5e AK |
2547 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
2548 | #endif | |
2549 | ||
2550 | ||
9fc77441 MT |
2551 | r = vcpu_load(vcpu); |
2552 | if (r) | |
2553 | return r; | |
6aa8b732 | 2554 | switch (ioctl) { |
9a2bb7f4 | 2555 | case KVM_RUN: |
f0fe5108 AK |
2556 | r = -EINVAL; |
2557 | if (arg) | |
2558 | goto out; | |
7a72f7a1 CB |
2559 | if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) { |
2560 | /* The thread running this VCPU changed. */ | |
2561 | struct pid *oldpid = vcpu->pid; | |
2562 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); | |
f95ef0cd | 2563 | |
7a72f7a1 CB |
2564 | rcu_assign_pointer(vcpu->pid, newpid); |
2565 | if (oldpid) | |
2566 | synchronize_rcu(); | |
2567 | put_pid(oldpid); | |
2568 | } | |
b6c7a5dc | 2569 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2570 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2571 | break; |
6aa8b732 | 2572 | case KVM_GET_REGS: { |
3e4bb3ac | 2573 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2574 | |
3e4bb3ac XZ |
2575 | r = -ENOMEM; |
2576 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2577 | if (!kvm_regs) | |
6aa8b732 | 2578 | goto out; |
3e4bb3ac XZ |
2579 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2580 | if (r) | |
2581 | goto out_free1; | |
6aa8b732 | 2582 | r = -EFAULT; |
3e4bb3ac XZ |
2583 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2584 | goto out_free1; | |
6aa8b732 | 2585 | r = 0; |
3e4bb3ac XZ |
2586 | out_free1: |
2587 | kfree(kvm_regs); | |
6aa8b732 AK |
2588 | break; |
2589 | } | |
2590 | case KVM_SET_REGS: { | |
3e4bb3ac | 2591 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2592 | |
3e4bb3ac | 2593 | r = -ENOMEM; |
ff5c2c03 SL |
2594 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2595 | if (IS_ERR(kvm_regs)) { | |
2596 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2597 | goto out; |
ff5c2c03 | 2598 | } |
3e4bb3ac | 2599 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2600 | kfree(kvm_regs); |
6aa8b732 AK |
2601 | break; |
2602 | } | |
2603 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2604 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2605 | r = -ENOMEM; | |
2606 | if (!kvm_sregs) | |
2607 | goto out; | |
2608 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2609 | if (r) |
2610 | goto out; | |
2611 | r = -EFAULT; | |
fa3795a7 | 2612 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2613 | goto out; |
2614 | r = 0; | |
2615 | break; | |
2616 | } | |
2617 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2618 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2619 | if (IS_ERR(kvm_sregs)) { | |
2620 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2621 | kvm_sregs = NULL; |
6aa8b732 | 2622 | goto out; |
ff5c2c03 | 2623 | } |
fa3795a7 | 2624 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2625 | break; |
2626 | } | |
62d9f0db MT |
2627 | case KVM_GET_MP_STATE: { |
2628 | struct kvm_mp_state mp_state; | |
2629 | ||
2630 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2631 | if (r) | |
2632 | goto out; | |
2633 | r = -EFAULT; | |
893bdbf1 | 2634 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2635 | goto out; |
2636 | r = 0; | |
2637 | break; | |
2638 | } | |
2639 | case KVM_SET_MP_STATE: { | |
2640 | struct kvm_mp_state mp_state; | |
2641 | ||
2642 | r = -EFAULT; | |
893bdbf1 | 2643 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2644 | goto out; |
2645 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2646 | break; |
2647 | } | |
6aa8b732 AK |
2648 | case KVM_TRANSLATE: { |
2649 | struct kvm_translation tr; | |
2650 | ||
2651 | r = -EFAULT; | |
893bdbf1 | 2652 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2653 | goto out; |
8b006791 | 2654 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2655 | if (r) |
2656 | goto out; | |
2657 | r = -EFAULT; | |
893bdbf1 | 2658 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2659 | goto out; |
2660 | r = 0; | |
2661 | break; | |
2662 | } | |
d0bfb940 JK |
2663 | case KVM_SET_GUEST_DEBUG: { |
2664 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2665 | |
2666 | r = -EFAULT; | |
893bdbf1 | 2667 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2668 | goto out; |
d0bfb940 | 2669 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2670 | break; |
2671 | } | |
1961d276 AK |
2672 | case KVM_SET_SIGNAL_MASK: { |
2673 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2674 | struct kvm_signal_mask kvm_sigmask; | |
2675 | sigset_t sigset, *p; | |
2676 | ||
2677 | p = NULL; | |
2678 | if (argp) { | |
2679 | r = -EFAULT; | |
2680 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2681 | sizeof(kvm_sigmask))) |
1961d276 AK |
2682 | goto out; |
2683 | r = -EINVAL; | |
893bdbf1 | 2684 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2685 | goto out; |
2686 | r = -EFAULT; | |
2687 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2688 | sizeof(sigset))) |
1961d276 AK |
2689 | goto out; |
2690 | p = &sigset; | |
2691 | } | |
376d41ff | 2692 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2693 | break; |
2694 | } | |
b8836737 | 2695 | case KVM_GET_FPU: { |
fa3795a7 DH |
2696 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2697 | r = -ENOMEM; | |
2698 | if (!fpu) | |
2699 | goto out; | |
2700 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2701 | if (r) |
2702 | goto out; | |
2703 | r = -EFAULT; | |
fa3795a7 | 2704 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2705 | goto out; |
2706 | r = 0; | |
2707 | break; | |
2708 | } | |
2709 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2710 | fpu = memdup_user(argp, sizeof(*fpu)); |
2711 | if (IS_ERR(fpu)) { | |
2712 | r = PTR_ERR(fpu); | |
18595411 | 2713 | fpu = NULL; |
b8836737 | 2714 | goto out; |
ff5c2c03 | 2715 | } |
fa3795a7 | 2716 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2717 | break; |
2718 | } | |
bccf2150 | 2719 | default: |
313a3dc7 | 2720 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2721 | } |
2722 | out: | |
2122ff5e | 2723 | vcpu_put(vcpu); |
fa3795a7 DH |
2724 | kfree(fpu); |
2725 | kfree(kvm_sregs); | |
bccf2150 AK |
2726 | return r; |
2727 | } | |
2728 | ||
de8e5d74 | 2729 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2730 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2731 | unsigned int ioctl, unsigned long arg) | |
2732 | { | |
2733 | struct kvm_vcpu *vcpu = filp->private_data; | |
2734 | void __user *argp = compat_ptr(arg); | |
2735 | int r; | |
2736 | ||
2737 | if (vcpu->kvm->mm != current->mm) | |
2738 | return -EIO; | |
2739 | ||
2740 | switch (ioctl) { | |
2741 | case KVM_SET_SIGNAL_MASK: { | |
2742 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2743 | struct kvm_signal_mask kvm_sigmask; | |
2744 | compat_sigset_t csigset; | |
2745 | sigset_t sigset; | |
2746 | ||
2747 | if (argp) { | |
2748 | r = -EFAULT; | |
2749 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2750 | sizeof(kvm_sigmask))) |
1dda606c AG |
2751 | goto out; |
2752 | r = -EINVAL; | |
893bdbf1 | 2753 | if (kvm_sigmask.len != sizeof(csigset)) |
1dda606c AG |
2754 | goto out; |
2755 | r = -EFAULT; | |
2756 | if (copy_from_user(&csigset, sigmask_arg->sigset, | |
893bdbf1 | 2757 | sizeof(csigset))) |
1dda606c | 2758 | goto out; |
760a9a30 AC |
2759 | sigset_from_compat(&sigset, &csigset); |
2760 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); | |
2761 | } else | |
2762 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2763 | break; |
2764 | } | |
2765 | default: | |
2766 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2767 | } | |
2768 | ||
2769 | out: | |
2770 | return r; | |
2771 | } | |
2772 | #endif | |
2773 | ||
852b6d57 SW |
2774 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2775 | int (*accessor)(struct kvm_device *dev, | |
2776 | struct kvm_device_attr *attr), | |
2777 | unsigned long arg) | |
2778 | { | |
2779 | struct kvm_device_attr attr; | |
2780 | ||
2781 | if (!accessor) | |
2782 | return -EPERM; | |
2783 | ||
2784 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2785 | return -EFAULT; | |
2786 | ||
2787 | return accessor(dev, &attr); | |
2788 | } | |
2789 | ||
2790 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2791 | unsigned long arg) | |
2792 | { | |
2793 | struct kvm_device *dev = filp->private_data; | |
2794 | ||
2795 | switch (ioctl) { | |
2796 | case KVM_SET_DEVICE_ATTR: | |
2797 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2798 | case KVM_GET_DEVICE_ATTR: | |
2799 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2800 | case KVM_HAS_DEVICE_ATTR: | |
2801 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2802 | default: | |
2803 | if (dev->ops->ioctl) | |
2804 | return dev->ops->ioctl(dev, ioctl, arg); | |
2805 | ||
2806 | return -ENOTTY; | |
2807 | } | |
2808 | } | |
2809 | ||
852b6d57 SW |
2810 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2811 | { | |
2812 | struct kvm_device *dev = filp->private_data; | |
2813 | struct kvm *kvm = dev->kvm; | |
2814 | ||
852b6d57 SW |
2815 | kvm_put_kvm(kvm); |
2816 | return 0; | |
2817 | } | |
2818 | ||
2819 | static const struct file_operations kvm_device_fops = { | |
2820 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2821 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2822 | .compat_ioctl = kvm_device_ioctl, |
2823 | #endif | |
852b6d57 SW |
2824 | .release = kvm_device_release, |
2825 | }; | |
2826 | ||
2827 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2828 | { | |
2829 | if (filp->f_op != &kvm_device_fops) | |
2830 | return NULL; | |
2831 | ||
2832 | return filp->private_data; | |
2833 | } | |
2834 | ||
d60eacb0 | 2835 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2836 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2837 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2838 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2839 | #endif |
d60eacb0 | 2840 | |
5975a2e0 | 2841 | #ifdef CONFIG_KVM_XICS |
d60eacb0 | 2842 | [KVM_DEV_TYPE_XICS] = &kvm_xics_ops, |
ec53500f | 2843 | #endif |
d60eacb0 WD |
2844 | }; |
2845 | ||
2846 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
2847 | { | |
2848 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
2849 | return -ENOSPC; | |
2850 | ||
2851 | if (kvm_device_ops_table[type] != NULL) | |
2852 | return -EEXIST; | |
2853 | ||
2854 | kvm_device_ops_table[type] = ops; | |
2855 | return 0; | |
2856 | } | |
2857 | ||
571ee1b6 WL |
2858 | void kvm_unregister_device_ops(u32 type) |
2859 | { | |
2860 | if (kvm_device_ops_table[type] != NULL) | |
2861 | kvm_device_ops_table[type] = NULL; | |
2862 | } | |
2863 | ||
852b6d57 SW |
2864 | static int kvm_ioctl_create_device(struct kvm *kvm, |
2865 | struct kvm_create_device *cd) | |
2866 | { | |
2867 | struct kvm_device_ops *ops = NULL; | |
2868 | struct kvm_device *dev; | |
2869 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
2870 | int ret; | |
2871 | ||
d60eacb0 WD |
2872 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
2873 | return -ENODEV; | |
2874 | ||
2875 | ops = kvm_device_ops_table[cd->type]; | |
2876 | if (ops == NULL) | |
852b6d57 | 2877 | return -ENODEV; |
852b6d57 SW |
2878 | |
2879 | if (test) | |
2880 | return 0; | |
2881 | ||
2882 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
2883 | if (!dev) | |
2884 | return -ENOMEM; | |
2885 | ||
2886 | dev->ops = ops; | |
2887 | dev->kvm = kvm; | |
852b6d57 | 2888 | |
a28ebea2 | 2889 | mutex_lock(&kvm->lock); |
852b6d57 SW |
2890 | ret = ops->create(dev, cd->type); |
2891 | if (ret < 0) { | |
a28ebea2 | 2892 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
2893 | kfree(dev); |
2894 | return ret; | |
2895 | } | |
a28ebea2 CD |
2896 | list_add(&dev->vm_node, &kvm->devices); |
2897 | mutex_unlock(&kvm->lock); | |
852b6d57 | 2898 | |
023e9fdd CD |
2899 | if (ops->init) |
2900 | ops->init(dev); | |
2901 | ||
24009b05 | 2902 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 2903 | if (ret < 0) { |
a28ebea2 CD |
2904 | mutex_lock(&kvm->lock); |
2905 | list_del(&dev->vm_node); | |
2906 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 2907 | ops->destroy(dev); |
852b6d57 SW |
2908 | return ret; |
2909 | } | |
2910 | ||
2911 | kvm_get_kvm(kvm); | |
2912 | cd->fd = ret; | |
2913 | return 0; | |
2914 | } | |
2915 | ||
92b591a4 AG |
2916 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
2917 | { | |
2918 | switch (arg) { | |
2919 | case KVM_CAP_USER_MEMORY: | |
2920 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
2921 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
2922 | case KVM_CAP_INTERNAL_ERROR_DATA: |
2923 | #ifdef CONFIG_HAVE_KVM_MSI | |
2924 | case KVM_CAP_SIGNAL_MSI: | |
2925 | #endif | |
297e2105 | 2926 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 2927 | case KVM_CAP_IRQFD: |
92b591a4 AG |
2928 | case KVM_CAP_IRQFD_RESAMPLE: |
2929 | #endif | |
e9ea5069 | 2930 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 AG |
2931 | case KVM_CAP_CHECK_EXTENSION_VM: |
2932 | return 1; | |
2933 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING | |
2934 | case KVM_CAP_IRQ_ROUTING: | |
2935 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
2936 | #endif |
2937 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
2938 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
2939 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 2940 | #endif |
0b1b1dfd GK |
2941 | case KVM_CAP_MAX_VCPU_ID: |
2942 | return KVM_MAX_VCPU_ID; | |
92b591a4 AG |
2943 | default: |
2944 | break; | |
2945 | } | |
2946 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2947 | } | |
2948 | ||
bccf2150 AK |
2949 | static long kvm_vm_ioctl(struct file *filp, |
2950 | unsigned int ioctl, unsigned long arg) | |
2951 | { | |
2952 | struct kvm *kvm = filp->private_data; | |
2953 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2954 | int r; |
bccf2150 | 2955 | |
6d4e4c4f AK |
2956 | if (kvm->mm != current->mm) |
2957 | return -EIO; | |
bccf2150 AK |
2958 | switch (ioctl) { |
2959 | case KVM_CREATE_VCPU: | |
2960 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2961 | break; |
6fc138d2 IE |
2962 | case KVM_SET_USER_MEMORY_REGION: { |
2963 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2964 | ||
2965 | r = -EFAULT; | |
2966 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 2967 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
2968 | goto out; |
2969 | ||
47ae31e2 | 2970 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2971 | break; |
2972 | } | |
2973 | case KVM_GET_DIRTY_LOG: { | |
2974 | struct kvm_dirty_log log; | |
2975 | ||
2976 | r = -EFAULT; | |
893bdbf1 | 2977 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 2978 | goto out; |
2c6f5df9 | 2979 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2980 | break; |
2981 | } | |
5f94c174 LV |
2982 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
2983 | case KVM_REGISTER_COALESCED_MMIO: { | |
2984 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2985 | |
5f94c174 | 2986 | r = -EFAULT; |
893bdbf1 | 2987 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2988 | goto out; |
5f94c174 | 2989 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2990 | break; |
2991 | } | |
2992 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2993 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2994 | |
5f94c174 | 2995 | r = -EFAULT; |
893bdbf1 | 2996 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2997 | goto out; |
5f94c174 | 2998 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2999 | break; |
3000 | } | |
3001 | #endif | |
721eecbf GH |
3002 | case KVM_IRQFD: { |
3003 | struct kvm_irqfd data; | |
3004 | ||
3005 | r = -EFAULT; | |
893bdbf1 | 3006 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3007 | goto out; |
d4db2935 | 3008 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3009 | break; |
3010 | } | |
d34e6b17 GH |
3011 | case KVM_IOEVENTFD: { |
3012 | struct kvm_ioeventfd data; | |
3013 | ||
3014 | r = -EFAULT; | |
893bdbf1 | 3015 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3016 | goto out; |
3017 | r = kvm_ioeventfd(kvm, &data); | |
3018 | break; | |
3019 | } | |
07975ad3 JK |
3020 | #ifdef CONFIG_HAVE_KVM_MSI |
3021 | case KVM_SIGNAL_MSI: { | |
3022 | struct kvm_msi msi; | |
3023 | ||
3024 | r = -EFAULT; | |
893bdbf1 | 3025 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3026 | goto out; |
3027 | r = kvm_send_userspace_msi(kvm, &msi); | |
3028 | break; | |
3029 | } | |
23d43cf9 CD |
3030 | #endif |
3031 | #ifdef __KVM_HAVE_IRQ_LINE | |
3032 | case KVM_IRQ_LINE_STATUS: | |
3033 | case KVM_IRQ_LINE: { | |
3034 | struct kvm_irq_level irq_event; | |
3035 | ||
3036 | r = -EFAULT; | |
893bdbf1 | 3037 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3038 | goto out; |
3039 | ||
aa2fbe6d YZ |
3040 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3041 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3042 | if (r) |
3043 | goto out; | |
3044 | ||
3045 | r = -EFAULT; | |
3046 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3047 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3048 | goto out; |
3049 | } | |
3050 | ||
3051 | r = 0; | |
3052 | break; | |
3053 | } | |
73880c80 | 3054 | #endif |
aa8d5944 AG |
3055 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3056 | case KVM_SET_GSI_ROUTING: { | |
3057 | struct kvm_irq_routing routing; | |
3058 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3059 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3060 | |
3061 | r = -EFAULT; | |
3062 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3063 | goto out; | |
3064 | r = -EINVAL; | |
caf1ff26 | 3065 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3066 | goto out; |
3067 | if (routing.flags) | |
3068 | goto out; | |
f8c1b85b PB |
3069 | if (routing.nr) { |
3070 | r = -ENOMEM; | |
3071 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
3072 | if (!entries) | |
3073 | goto out; | |
3074 | r = -EFAULT; | |
3075 | urouting = argp; | |
3076 | if (copy_from_user(entries, urouting->entries, | |
3077 | routing.nr * sizeof(*entries))) | |
3078 | goto out_free_irq_routing; | |
3079 | } | |
aa8d5944 AG |
3080 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3081 | routing.flags); | |
a642a175 | 3082 | out_free_irq_routing: |
aa8d5944 AG |
3083 | vfree(entries); |
3084 | break; | |
3085 | } | |
3086 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3087 | case KVM_CREATE_DEVICE: { |
3088 | struct kvm_create_device cd; | |
3089 | ||
3090 | r = -EFAULT; | |
3091 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3092 | goto out; | |
3093 | ||
3094 | r = kvm_ioctl_create_device(kvm, &cd); | |
3095 | if (r) | |
3096 | goto out; | |
3097 | ||
3098 | r = -EFAULT; | |
3099 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3100 | goto out; | |
3101 | ||
3102 | r = 0; | |
3103 | break; | |
3104 | } | |
92b591a4 AG |
3105 | case KVM_CHECK_EXTENSION: |
3106 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3107 | break; | |
f17abe9a | 3108 | default: |
1fe779f8 | 3109 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3110 | } |
3111 | out: | |
3112 | return r; | |
3113 | } | |
3114 | ||
de8e5d74 | 3115 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3116 | struct compat_kvm_dirty_log { |
3117 | __u32 slot; | |
3118 | __u32 padding1; | |
3119 | union { | |
3120 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3121 | __u64 padding2; | |
3122 | }; | |
3123 | }; | |
3124 | ||
3125 | static long kvm_vm_compat_ioctl(struct file *filp, | |
3126 | unsigned int ioctl, unsigned long arg) | |
3127 | { | |
3128 | struct kvm *kvm = filp->private_data; | |
3129 | int r; | |
3130 | ||
3131 | if (kvm->mm != current->mm) | |
3132 | return -EIO; | |
3133 | switch (ioctl) { | |
3134 | case KVM_GET_DIRTY_LOG: { | |
3135 | struct compat_kvm_dirty_log compat_log; | |
3136 | struct kvm_dirty_log log; | |
3137 | ||
3138 | r = -EFAULT; | |
3139 | if (copy_from_user(&compat_log, (void __user *)arg, | |
3140 | sizeof(compat_log))) | |
3141 | goto out; | |
3142 | log.slot = compat_log.slot; | |
3143 | log.padding1 = compat_log.padding1; | |
3144 | log.padding2 = compat_log.padding2; | |
3145 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3146 | ||
3147 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
3148 | break; |
3149 | } | |
3150 | default: | |
3151 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
3152 | } | |
3153 | ||
3154 | out: | |
3155 | return r; | |
3156 | } | |
3157 | #endif | |
3158 | ||
3d3aab1b | 3159 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
3160 | .release = kvm_vm_release, |
3161 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 3162 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3163 | .compat_ioctl = kvm_vm_compat_ioctl, |
3164 | #endif | |
6038f373 | 3165 | .llseek = noop_llseek, |
f17abe9a AK |
3166 | }; |
3167 | ||
e08b9637 | 3168 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 3169 | { |
aac87636 | 3170 | int r; |
f17abe9a | 3171 | struct kvm *kvm; |
506cfba9 | 3172 | struct file *file; |
f17abe9a | 3173 | |
e08b9637 | 3174 | kvm = kvm_create_vm(type); |
d6d28168 AK |
3175 | if (IS_ERR(kvm)) |
3176 | return PTR_ERR(kvm); | |
6ce5a090 TY |
3177 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET |
3178 | r = kvm_coalesced_mmio_init(kvm); | |
3179 | if (r < 0) { | |
3180 | kvm_put_kvm(kvm); | |
3181 | return r; | |
3182 | } | |
3183 | #endif | |
506cfba9 | 3184 | r = get_unused_fd_flags(O_CLOEXEC); |
536a6f88 | 3185 | if (r < 0) { |
66c0b394 | 3186 | kvm_put_kvm(kvm); |
536a6f88 JF |
3187 | return r; |
3188 | } | |
506cfba9 AV |
3189 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
3190 | if (IS_ERR(file)) { | |
3191 | put_unused_fd(r); | |
3192 | kvm_put_kvm(kvm); | |
3193 | return PTR_ERR(file); | |
3194 | } | |
536a6f88 JF |
3195 | |
3196 | if (kvm_create_vm_debugfs(kvm, r) < 0) { | |
506cfba9 AV |
3197 | put_unused_fd(r); |
3198 | fput(file); | |
536a6f88 JF |
3199 | return -ENOMEM; |
3200 | } | |
f17abe9a | 3201 | |
506cfba9 | 3202 | fd_install(r, file); |
aac87636 | 3203 | return r; |
f17abe9a AK |
3204 | } |
3205 | ||
3206 | static long kvm_dev_ioctl(struct file *filp, | |
3207 | unsigned int ioctl, unsigned long arg) | |
3208 | { | |
07c45a36 | 3209 | long r = -EINVAL; |
f17abe9a AK |
3210 | |
3211 | switch (ioctl) { | |
3212 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
3213 | if (arg) |
3214 | goto out; | |
f17abe9a AK |
3215 | r = KVM_API_VERSION; |
3216 | break; | |
3217 | case KVM_CREATE_VM: | |
e08b9637 | 3218 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 3219 | break; |
018d00d2 | 3220 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 3221 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 3222 | break; |
07c45a36 | 3223 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
3224 | if (arg) |
3225 | goto out; | |
adb1ff46 AK |
3226 | r = PAGE_SIZE; /* struct kvm_run */ |
3227 | #ifdef CONFIG_X86 | |
3228 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 LV |
3229 | #endif |
3230 | #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET | |
3231 | r += PAGE_SIZE; /* coalesced mmio ring page */ | |
adb1ff46 | 3232 | #endif |
07c45a36 | 3233 | break; |
d4c9ff2d FEL |
3234 | case KVM_TRACE_ENABLE: |
3235 | case KVM_TRACE_PAUSE: | |
3236 | case KVM_TRACE_DISABLE: | |
2023a29c | 3237 | r = -EOPNOTSUPP; |
d4c9ff2d | 3238 | break; |
6aa8b732 | 3239 | default: |
043405e1 | 3240 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
3241 | } |
3242 | out: | |
3243 | return r; | |
3244 | } | |
3245 | ||
6aa8b732 | 3246 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
3247 | .unlocked_ioctl = kvm_dev_ioctl, |
3248 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 3249 | .llseek = noop_llseek, |
6aa8b732 AK |
3250 | }; |
3251 | ||
3252 | static struct miscdevice kvm_dev = { | |
bbe4432e | 3253 | KVM_MINOR, |
6aa8b732 AK |
3254 | "kvm", |
3255 | &kvm_chardev_ops, | |
3256 | }; | |
3257 | ||
75b7127c | 3258 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
3259 | { |
3260 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 3261 | int r; |
1b6c0168 | 3262 | |
7f59f492 | 3263 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3264 | return; |
10474ae8 | 3265 | |
7f59f492 | 3266 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 3267 | |
13a34e06 | 3268 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
3269 | |
3270 | if (r) { | |
3271 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
3272 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 3273 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 3274 | } |
1b6c0168 AK |
3275 | } |
3276 | ||
8c18b2d2 | 3277 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 3278 | { |
4a937f96 | 3279 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3280 | if (kvm_usage_count) |
3281 | hardware_enable_nolock(NULL); | |
4a937f96 | 3282 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3283 | return 0; |
75b7127c TY |
3284 | } |
3285 | ||
3286 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
3287 | { |
3288 | int cpu = raw_smp_processor_id(); | |
3289 | ||
7f59f492 | 3290 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3291 | return; |
7f59f492 | 3292 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 3293 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
3294 | } |
3295 | ||
8c18b2d2 | 3296 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 3297 | { |
4a937f96 | 3298 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3299 | if (kvm_usage_count) |
3300 | hardware_disable_nolock(NULL); | |
4a937f96 | 3301 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3302 | return 0; |
75b7127c TY |
3303 | } |
3304 | ||
10474ae8 AG |
3305 | static void hardware_disable_all_nolock(void) |
3306 | { | |
3307 | BUG_ON(!kvm_usage_count); | |
3308 | ||
3309 | kvm_usage_count--; | |
3310 | if (!kvm_usage_count) | |
75b7127c | 3311 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
3312 | } |
3313 | ||
3314 | static void hardware_disable_all(void) | |
3315 | { | |
4a937f96 | 3316 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3317 | hardware_disable_all_nolock(); |
4a937f96 | 3318 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3319 | } |
3320 | ||
3321 | static int hardware_enable_all(void) | |
3322 | { | |
3323 | int r = 0; | |
3324 | ||
4a937f96 | 3325 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3326 | |
3327 | kvm_usage_count++; | |
3328 | if (kvm_usage_count == 1) { | |
3329 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3330 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3331 | |
3332 | if (atomic_read(&hardware_enable_failed)) { | |
3333 | hardware_disable_all_nolock(); | |
3334 | r = -EBUSY; | |
3335 | } | |
3336 | } | |
3337 | ||
4a937f96 | 3338 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3339 | |
3340 | return r; | |
3341 | } | |
3342 | ||
9a2b85c6 | 3343 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3344 | void *v) |
9a2b85c6 | 3345 | { |
8e1c1815 SY |
3346 | /* |
3347 | * Some (well, at least mine) BIOSes hang on reboot if | |
3348 | * in vmx root mode. | |
3349 | * | |
3350 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3351 | */ | |
1170adc6 | 3352 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3353 | kvm_rebooting = true; |
75b7127c | 3354 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3355 | return NOTIFY_OK; |
3356 | } | |
3357 | ||
3358 | static struct notifier_block kvm_reboot_notifier = { | |
3359 | .notifier_call = kvm_reboot, | |
3360 | .priority = 0, | |
3361 | }; | |
3362 | ||
e93f8a0f | 3363 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3364 | { |
3365 | int i; | |
3366 | ||
3367 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3368 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3369 | |
3370 | kvm_iodevice_destructor(pos); | |
3371 | } | |
e93f8a0f | 3372 | kfree(bus); |
2eeb2e94 GH |
3373 | } |
3374 | ||
c21fbff1 | 3375 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3376 | const struct kvm_io_range *r2) |
743eeb0b | 3377 | { |
8f4216c7 JW |
3378 | gpa_t addr1 = r1->addr; |
3379 | gpa_t addr2 = r2->addr; | |
3380 | ||
3381 | if (addr1 < addr2) | |
743eeb0b | 3382 | return -1; |
8f4216c7 JW |
3383 | |
3384 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
3385 | * accept any overlapping write. Any order is acceptable for | |
3386 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
3387 | * we process all of them. | |
3388 | */ | |
3389 | if (r2->len) { | |
3390 | addr1 += r1->len; | |
3391 | addr2 += r2->len; | |
3392 | } | |
3393 | ||
3394 | if (addr1 > addr2) | |
743eeb0b | 3395 | return 1; |
8f4216c7 | 3396 | |
743eeb0b SL |
3397 | return 0; |
3398 | } | |
3399 | ||
a343c9b7 PB |
3400 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3401 | { | |
c21fbff1 | 3402 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3403 | } |
3404 | ||
39369f7a | 3405 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
3406 | gpa_t addr, int len) |
3407 | { | |
743eeb0b SL |
3408 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
3409 | .addr = addr, | |
3410 | .len = len, | |
3411 | .dev = dev, | |
3412 | }; | |
3413 | ||
3414 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
3415 | kvm_io_bus_sort_cmp, NULL); | |
3416 | ||
3417 | return 0; | |
3418 | } | |
3419 | ||
39369f7a | 3420 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3421 | gpa_t addr, int len) |
3422 | { | |
3423 | struct kvm_io_range *range, key; | |
3424 | int off; | |
3425 | ||
3426 | key = (struct kvm_io_range) { | |
3427 | .addr = addr, | |
3428 | .len = len, | |
3429 | }; | |
3430 | ||
3431 | range = bsearch(&key, bus->range, bus->dev_count, | |
3432 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3433 | if (range == NULL) | |
3434 | return -ENOENT; | |
3435 | ||
3436 | off = range - bus->range; | |
3437 | ||
c21fbff1 | 3438 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3439 | off--; |
3440 | ||
3441 | return off; | |
3442 | } | |
3443 | ||
e32edf4f | 3444 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3445 | struct kvm_io_range *range, const void *val) |
3446 | { | |
3447 | int idx; | |
3448 | ||
3449 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3450 | if (idx < 0) | |
3451 | return -EOPNOTSUPP; | |
3452 | ||
3453 | while (idx < bus->dev_count && | |
c21fbff1 | 3454 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3455 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3456 | range->len, val)) |
3457 | return idx; | |
3458 | idx++; | |
3459 | } | |
3460 | ||
3461 | return -EOPNOTSUPP; | |
3462 | } | |
3463 | ||
bda9020e | 3464 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3465 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3466 | int len, const void *val) |
2eeb2e94 | 3467 | { |
90d83dc3 | 3468 | struct kvm_io_bus *bus; |
743eeb0b | 3469 | struct kvm_io_range range; |
126a5af5 | 3470 | int r; |
743eeb0b SL |
3471 | |
3472 | range = (struct kvm_io_range) { | |
3473 | .addr = addr, | |
3474 | .len = len, | |
3475 | }; | |
90d83dc3 | 3476 | |
e32edf4f NN |
3477 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3478 | r = __kvm_io_bus_write(vcpu, bus, &range, val); | |
126a5af5 CH |
3479 | return r < 0 ? r : 0; |
3480 | } | |
3481 | ||
3482 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3483 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3484 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3485 | { |
3486 | struct kvm_io_bus *bus; | |
3487 | struct kvm_io_range range; | |
3488 | ||
3489 | range = (struct kvm_io_range) { | |
3490 | .addr = addr, | |
3491 | .len = len, | |
3492 | }; | |
3493 | ||
e32edf4f | 3494 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
126a5af5 CH |
3495 | |
3496 | /* First try the device referenced by cookie. */ | |
3497 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3498 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3499 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3500 | val)) |
3501 | return cookie; | |
3502 | ||
3503 | /* | |
3504 | * cookie contained garbage; fall back to search and return the | |
3505 | * correct cookie value. | |
3506 | */ | |
e32edf4f | 3507 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3508 | } |
3509 | ||
e32edf4f NN |
3510 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3511 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3512 | { |
3513 | int idx; | |
3514 | ||
3515 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3516 | if (idx < 0) |
3517 | return -EOPNOTSUPP; | |
3518 | ||
3519 | while (idx < bus->dev_count && | |
c21fbff1 | 3520 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3521 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3522 | range->len, val)) |
3523 | return idx; | |
743eeb0b SL |
3524 | idx++; |
3525 | } | |
3526 | ||
bda9020e MT |
3527 | return -EOPNOTSUPP; |
3528 | } | |
68c3b4d1 | 3529 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3530 | |
bda9020e | 3531 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3532 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3533 | int len, void *val) |
bda9020e | 3534 | { |
90d83dc3 | 3535 | struct kvm_io_bus *bus; |
743eeb0b | 3536 | struct kvm_io_range range; |
126a5af5 | 3537 | int r; |
743eeb0b SL |
3538 | |
3539 | range = (struct kvm_io_range) { | |
3540 | .addr = addr, | |
3541 | .len = len, | |
3542 | }; | |
e93f8a0f | 3543 | |
e32edf4f NN |
3544 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
3545 | r = __kvm_io_bus_read(vcpu, bus, &range, val); | |
126a5af5 CH |
3546 | return r < 0 ? r : 0; |
3547 | } | |
743eeb0b | 3548 | |
2eeb2e94 | 3549 | |
79fac95e | 3550 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3551 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3552 | int len, struct kvm_io_device *dev) | |
6c474694 | 3553 | { |
e93f8a0f | 3554 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3555 | |
e93f8a0f | 3556 | bus = kvm->buses[bus_idx]; |
6ea34c9b AK |
3557 | /* exclude ioeventfd which is limited by maximum fd */ |
3558 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3559 | return -ENOSPC; |
2eeb2e94 | 3560 | |
d3febddd | 3561 | new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
a1300716 | 3562 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
e93f8a0f MT |
3563 | if (!new_bus) |
3564 | return -ENOMEM; | |
a1300716 AK |
3565 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3566 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3567 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3568 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3569 | synchronize_srcu_expedited(&kvm->srcu); | |
3570 | kfree(bus); | |
090b7aff GH |
3571 | |
3572 | return 0; | |
3573 | } | |
3574 | ||
79fac95e | 3575 | /* Caller must hold slots_lock. */ |
e93f8a0f MT |
3576 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3577 | struct kvm_io_device *dev) | |
090b7aff | 3578 | { |
e93f8a0f MT |
3579 | int i, r; |
3580 | struct kvm_io_bus *new_bus, *bus; | |
090b7aff | 3581 | |
cdfca7b3 | 3582 | bus = kvm->buses[bus_idx]; |
46a92167 PX |
3583 | |
3584 | /* | |
3585 | * It's possible the bus being released before hand. If so, | |
3586 | * we're done here. | |
3587 | */ | |
3588 | if (!bus) | |
3589 | return 0; | |
3590 | ||
e93f8a0f | 3591 | r = -ENOENT; |
a1300716 AK |
3592 | for (i = 0; i < bus->dev_count; i++) |
3593 | if (bus->range[i].dev == dev) { | |
e93f8a0f | 3594 | r = 0; |
090b7aff GH |
3595 | break; |
3596 | } | |
e93f8a0f | 3597 | |
a1300716 | 3598 | if (r) |
e93f8a0f | 3599 | return r; |
a1300716 | 3600 | |
d3febddd | 3601 | new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) * |
a1300716 AK |
3602 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
3603 | if (!new_bus) | |
3604 | return -ENOMEM; | |
3605 | ||
3606 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3607 | new_bus->dev_count--; | |
3608 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3609 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3610 | |
3611 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
3612 | synchronize_srcu_expedited(&kvm->srcu); | |
3613 | kfree(bus); | |
3614 | return r; | |
2eeb2e94 GH |
3615 | } |
3616 | ||
8a39d006 AP |
3617 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3618 | gpa_t addr) | |
3619 | { | |
3620 | struct kvm_io_bus *bus; | |
3621 | int dev_idx, srcu_idx; | |
3622 | struct kvm_io_device *iodev = NULL; | |
3623 | ||
3624 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
3625 | ||
3626 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
3627 | ||
3628 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
3629 | if (dev_idx < 0) | |
3630 | goto out_unlock; | |
3631 | ||
3632 | iodev = bus->range[dev_idx].dev; | |
3633 | ||
3634 | out_unlock: | |
3635 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
3636 | ||
3637 | return iodev; | |
3638 | } | |
3639 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
3640 | ||
536a6f88 JF |
3641 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
3642 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
3643 | const char *fmt) | |
3644 | { | |
3645 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3646 | inode->i_private; | |
3647 | ||
3648 | /* The debugfs files are a reference to the kvm struct which | |
3649 | * is still valid when kvm_destroy_vm is called. | |
3650 | * To avoid the race between open and the removal of the debugfs | |
3651 | * directory we test against the users count. | |
3652 | */ | |
3653 | if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0)) | |
3654 | return -ENOENT; | |
3655 | ||
3656 | if (simple_attr_open(inode, file, get, set, fmt)) { | |
3657 | kvm_put_kvm(stat_data->kvm); | |
3658 | return -ENOMEM; | |
3659 | } | |
3660 | ||
3661 | return 0; | |
3662 | } | |
3663 | ||
3664 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
3665 | { | |
3666 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3667 | inode->i_private; | |
3668 | ||
3669 | simple_attr_release(inode, file); | |
3670 | kvm_put_kvm(stat_data->kvm); | |
3671 | ||
3672 | return 0; | |
3673 | } | |
3674 | ||
3675 | static int vm_stat_get_per_vm(void *data, u64 *val) | |
3676 | { | |
3677 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3678 | ||
8a7e75d4 | 3679 | *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset); |
536a6f88 JF |
3680 | |
3681 | return 0; | |
3682 | } | |
3683 | ||
ce35ef27 SJS |
3684 | static int vm_stat_clear_per_vm(void *data, u64 val) |
3685 | { | |
3686 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3687 | ||
3688 | if (val) | |
3689 | return -EINVAL; | |
3690 | ||
3691 | *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0; | |
3692 | ||
3693 | return 0; | |
3694 | } | |
3695 | ||
536a6f88 JF |
3696 | static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file) |
3697 | { | |
3698 | __simple_attr_check_format("%llu\n", 0ull); | |
3699 | return kvm_debugfs_open(inode, file, vm_stat_get_per_vm, | |
ce35ef27 | 3700 | vm_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
3701 | } |
3702 | ||
3703 | static const struct file_operations vm_stat_get_per_vm_fops = { | |
3704 | .owner = THIS_MODULE, | |
3705 | .open = vm_stat_get_per_vm_open, | |
3706 | .release = kvm_debugfs_release, | |
3707 | .read = simple_attr_read, | |
3708 | .write = simple_attr_write, | |
3709 | .llseek = generic_file_llseek, | |
3710 | }; | |
3711 | ||
3712 | static int vcpu_stat_get_per_vm(void *data, u64 *val) | |
3713 | { | |
3714 | int i; | |
3715 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3716 | struct kvm_vcpu *vcpu; | |
3717 | ||
3718 | *val = 0; | |
3719 | ||
3720 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
8a7e75d4 | 3721 | *val += *(u64 *)((void *)vcpu + stat_data->offset); |
536a6f88 JF |
3722 | |
3723 | return 0; | |
3724 | } | |
3725 | ||
ce35ef27 SJS |
3726 | static int vcpu_stat_clear_per_vm(void *data, u64 val) |
3727 | { | |
3728 | int i; | |
3729 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3730 | struct kvm_vcpu *vcpu; | |
3731 | ||
3732 | if (val) | |
3733 | return -EINVAL; | |
3734 | ||
3735 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
3736 | *(u64 *)((void *)vcpu + stat_data->offset) = 0; | |
3737 | ||
3738 | return 0; | |
3739 | } | |
3740 | ||
536a6f88 JF |
3741 | static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file) |
3742 | { | |
3743 | __simple_attr_check_format("%llu\n", 0ull); | |
3744 | return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm, | |
ce35ef27 | 3745 | vcpu_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
3746 | } |
3747 | ||
3748 | static const struct file_operations vcpu_stat_get_per_vm_fops = { | |
3749 | .owner = THIS_MODULE, | |
3750 | .open = vcpu_stat_get_per_vm_open, | |
3751 | .release = kvm_debugfs_release, | |
3752 | .read = simple_attr_read, | |
3753 | .write = simple_attr_write, | |
3754 | .llseek = generic_file_llseek, | |
3755 | }; | |
3756 | ||
3757 | static const struct file_operations *stat_fops_per_vm[] = { | |
3758 | [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops, | |
3759 | [KVM_STAT_VM] = &vm_stat_get_per_vm_fops, | |
3760 | }; | |
3761 | ||
8b88b099 | 3762 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3763 | { |
3764 | unsigned offset = (long)_offset; | |
ba1389b7 | 3765 | struct kvm *kvm; |
536a6f88 JF |
3766 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
3767 | u64 tmp_val; | |
ba1389b7 | 3768 | |
8b88b099 | 3769 | *val = 0; |
2f303b74 | 3770 | spin_lock(&kvm_lock); |
536a6f88 JF |
3771 | list_for_each_entry(kvm, &vm_list, vm_list) { |
3772 | stat_tmp.kvm = kvm; | |
3773 | vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
3774 | *val += tmp_val; | |
3775 | } | |
2f303b74 | 3776 | spin_unlock(&kvm_lock); |
8b88b099 | 3777 | return 0; |
ba1389b7 AK |
3778 | } |
3779 | ||
ce35ef27 SJS |
3780 | static int vm_stat_clear(void *_offset, u64 val) |
3781 | { | |
3782 | unsigned offset = (long)_offset; | |
3783 | struct kvm *kvm; | |
3784 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
3785 | ||
3786 | if (val) | |
3787 | return -EINVAL; | |
3788 | ||
3789 | spin_lock(&kvm_lock); | |
3790 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3791 | stat_tmp.kvm = kvm; | |
3792 | vm_stat_clear_per_vm((void *)&stat_tmp, 0); | |
3793 | } | |
3794 | spin_unlock(&kvm_lock); | |
3795 | ||
3796 | return 0; | |
3797 | } | |
3798 | ||
3799 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 3800 | |
8b88b099 | 3801 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3802 | { |
3803 | unsigned offset = (long)_offset; | |
1165f5fe | 3804 | struct kvm *kvm; |
536a6f88 JF |
3805 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
3806 | u64 tmp_val; | |
1165f5fe | 3807 | |
8b88b099 | 3808 | *val = 0; |
2f303b74 | 3809 | spin_lock(&kvm_lock); |
536a6f88 JF |
3810 | list_for_each_entry(kvm, &vm_list, vm_list) { |
3811 | stat_tmp.kvm = kvm; | |
3812 | vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
3813 | *val += tmp_val; | |
3814 | } | |
2f303b74 | 3815 | spin_unlock(&kvm_lock); |
8b88b099 | 3816 | return 0; |
1165f5fe AK |
3817 | } |
3818 | ||
ce35ef27 SJS |
3819 | static int vcpu_stat_clear(void *_offset, u64 val) |
3820 | { | |
3821 | unsigned offset = (long)_offset; | |
3822 | struct kvm *kvm; | |
3823 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
3824 | ||
3825 | if (val) | |
3826 | return -EINVAL; | |
3827 | ||
3828 | spin_lock(&kvm_lock); | |
3829 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3830 | stat_tmp.kvm = kvm; | |
3831 | vcpu_stat_clear_per_vm((void *)&stat_tmp, 0); | |
3832 | } | |
3833 | spin_unlock(&kvm_lock); | |
3834 | ||
3835 | return 0; | |
3836 | } | |
3837 | ||
3838 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
3839 | "%llu\n"); | |
ba1389b7 | 3840 | |
828c0950 | 3841 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3842 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3843 | [KVM_STAT_VM] = &vm_stat_fops, | |
3844 | }; | |
1165f5fe | 3845 | |
4f69b680 | 3846 | static int kvm_init_debug(void) |
6aa8b732 | 3847 | { |
0c8eb04a | 3848 | int r = -EEXIST; |
6aa8b732 AK |
3849 | struct kvm_stats_debugfs_item *p; |
3850 | ||
76f7c879 | 3851 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3852 | if (kvm_debugfs_dir == NULL) |
3853 | goto out; | |
3854 | ||
536a6f88 JF |
3855 | kvm_debugfs_num_entries = 0; |
3856 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
ce35ef27 | 3857 | if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir, |
4bd33b56 JF |
3858 | (void *)(long)p->offset, |
3859 | stat_fops[p->kind])) | |
4f69b680 H |
3860 | goto out_dir; |
3861 | } | |
3862 | ||
3863 | return 0; | |
3864 | ||
3865 | out_dir: | |
3866 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3867 | out: | |
3868 | return r; | |
6aa8b732 AK |
3869 | } |
3870 | ||
fb3600cc | 3871 | static int kvm_suspend(void) |
59ae6c6b | 3872 | { |
10474ae8 | 3873 | if (kvm_usage_count) |
75b7127c | 3874 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3875 | return 0; |
3876 | } | |
3877 | ||
fb3600cc | 3878 | static void kvm_resume(void) |
59ae6c6b | 3879 | { |
ca84d1a2 | 3880 | if (kvm_usage_count) { |
4a937f96 | 3881 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3882 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3883 | } |
59ae6c6b AK |
3884 | } |
3885 | ||
fb3600cc | 3886 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3887 | .suspend = kvm_suspend, |
3888 | .resume = kvm_resume, | |
3889 | }; | |
3890 | ||
15ad7146 AK |
3891 | static inline |
3892 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3893 | { | |
3894 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3895 | } | |
3896 | ||
3897 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3898 | { | |
3899 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 3900 | |
3a08a8f9 R |
3901 | if (vcpu->preempted) |
3902 | vcpu->preempted = false; | |
15ad7146 | 3903 | |
e790d9ef RK |
3904 | kvm_arch_sched_in(vcpu, cpu); |
3905 | ||
e9b11c17 | 3906 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3907 | } |
3908 | ||
3909 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3910 | struct task_struct *next) | |
3911 | { | |
3912 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3913 | ||
3a08a8f9 R |
3914 | if (current->state == TASK_RUNNING) |
3915 | vcpu->preempted = true; | |
e9b11c17 | 3916 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3917 | } |
3918 | ||
0ee75bea | 3919 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3920 | struct module *module) |
6aa8b732 AK |
3921 | { |
3922 | int r; | |
002c7f7c | 3923 | int cpu; |
6aa8b732 | 3924 | |
f8c16bba ZX |
3925 | r = kvm_arch_init(opaque); |
3926 | if (r) | |
d2308784 | 3927 | goto out_fail; |
cb498ea2 | 3928 | |
7dac16c3 AH |
3929 | /* |
3930 | * kvm_arch_init makes sure there's at most one caller | |
3931 | * for architectures that support multiple implementations, | |
3932 | * like intel and amd on x86. | |
36343f6e PB |
3933 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
3934 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 3935 | */ |
36343f6e PB |
3936 | r = kvm_irqfd_init(); |
3937 | if (r) | |
3938 | goto out_irqfd; | |
7dac16c3 | 3939 | |
8437a617 | 3940 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
3941 | r = -ENOMEM; |
3942 | goto out_free_0; | |
3943 | } | |
3944 | ||
e9b11c17 | 3945 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 3946 | if (r < 0) |
7f59f492 | 3947 | goto out_free_0a; |
6aa8b732 | 3948 | |
002c7f7c YS |
3949 | for_each_online_cpu(cpu) { |
3950 | smp_call_function_single(cpu, | |
e9b11c17 | 3951 | kvm_arch_check_processor_compat, |
8691e5a8 | 3952 | &r, 1); |
002c7f7c | 3953 | if (r < 0) |
d2308784 | 3954 | goto out_free_1; |
002c7f7c YS |
3955 | } |
3956 | ||
73c1b41e | 3957 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 3958 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 3959 | if (r) |
d2308784 | 3960 | goto out_free_2; |
6aa8b732 AK |
3961 | register_reboot_notifier(&kvm_reboot_notifier); |
3962 | ||
c16f862d | 3963 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
3964 | if (!vcpu_align) |
3965 | vcpu_align = __alignof__(struct kvm_vcpu); | |
3966 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
56919c5c | 3967 | 0, NULL); |
c16f862d RR |
3968 | if (!kvm_vcpu_cache) { |
3969 | r = -ENOMEM; | |
fb3600cc | 3970 | goto out_free_3; |
c16f862d RR |
3971 | } |
3972 | ||
af585b92 GN |
3973 | r = kvm_async_pf_init(); |
3974 | if (r) | |
3975 | goto out_free; | |
3976 | ||
6aa8b732 | 3977 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
3978 | kvm_vm_fops.owner = module; |
3979 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
3980 | |
3981 | r = misc_register(&kvm_dev); | |
3982 | if (r) { | |
1170adc6 | 3983 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 3984 | goto out_unreg; |
6aa8b732 AK |
3985 | } |
3986 | ||
fb3600cc RW |
3987 | register_syscore_ops(&kvm_syscore_ops); |
3988 | ||
15ad7146 AK |
3989 | kvm_preempt_ops.sched_in = kvm_sched_in; |
3990 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
3991 | ||
4f69b680 H |
3992 | r = kvm_init_debug(); |
3993 | if (r) { | |
1170adc6 | 3994 | pr_err("kvm: create debugfs files failed\n"); |
4f69b680 H |
3995 | goto out_undebugfs; |
3996 | } | |
0ea4ed8e | 3997 | |
3c3c29fd PB |
3998 | r = kvm_vfio_ops_init(); |
3999 | WARN_ON(r); | |
4000 | ||
c7addb90 | 4001 | return 0; |
6aa8b732 | 4002 | |
4f69b680 H |
4003 | out_undebugfs: |
4004 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 4005 | misc_deregister(&kvm_dev); |
af585b92 GN |
4006 | out_unreg: |
4007 | kvm_async_pf_deinit(); | |
6aa8b732 | 4008 | out_free: |
c16f862d | 4009 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4010 | out_free_3: |
6aa8b732 | 4011 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4012 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4013 | out_free_2: |
d2308784 | 4014 | out_free_1: |
e9b11c17 | 4015 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
4016 | out_free_0a: |
4017 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 4018 | out_free_0: |
a0f155e9 | 4019 | kvm_irqfd_exit(); |
36343f6e | 4020 | out_irqfd: |
7dac16c3 AH |
4021 | kvm_arch_exit(); |
4022 | out_fail: | |
6aa8b732 AK |
4023 | return r; |
4024 | } | |
cb498ea2 | 4025 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 4026 | |
cb498ea2 | 4027 | void kvm_exit(void) |
6aa8b732 | 4028 | { |
4bd33b56 | 4029 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 4030 | misc_deregister(&kvm_dev); |
c16f862d | 4031 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 4032 | kvm_async_pf_deinit(); |
fb3600cc | 4033 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 4034 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4035 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 4036 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 4037 | kvm_arch_hardware_unsetup(); |
f8c16bba | 4038 | kvm_arch_exit(); |
a0f155e9 | 4039 | kvm_irqfd_exit(); |
7f59f492 | 4040 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 4041 | kvm_vfio_ops_exit(); |
6aa8b732 | 4042 | } |
cb498ea2 | 4043 | EXPORT_SYMBOL_GPL(kvm_exit); |