Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
9 | * |
10 | * Authors: | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * Yaniv Kamay <yaniv@qumranet.com> | |
13 | * | |
14 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
15 | * the COPYING file in the top-level directory. | |
16 | * | |
17 | */ | |
18 | ||
af669ac6 | 19 | #include <kvm/iodev.h> |
6aa8b732 | 20 | |
edf88417 | 21 | #include <linux/kvm_host.h> |
6aa8b732 AK |
22 | #include <linux/kvm.h> |
23 | #include <linux/module.h> | |
24 | #include <linux/errno.h> | |
6aa8b732 | 25 | #include <linux/percpu.h> |
6aa8b732 AK |
26 | #include <linux/mm.h> |
27 | #include <linux/miscdevice.h> | |
28 | #include <linux/vmalloc.h> | |
6aa8b732 | 29 | #include <linux/reboot.h> |
6aa8b732 AK |
30 | #include <linux/debugfs.h> |
31 | #include <linux/highmem.h> | |
32 | #include <linux/file.h> | |
fb3600cc | 33 | #include <linux/syscore_ops.h> |
774c47f1 | 34 | #include <linux/cpu.h> |
174cd4b1 | 35 | #include <linux/sched/signal.h> |
6e84f315 | 36 | #include <linux/sched/mm.h> |
03441a34 | 37 | #include <linux/sched/stat.h> |
d9e368d6 AK |
38 | #include <linux/cpumask.h> |
39 | #include <linux/smp.h> | |
d6d28168 | 40 | #include <linux/anon_inodes.h> |
04d2cc77 | 41 | #include <linux/profile.h> |
7aa81cc0 | 42 | #include <linux/kvm_para.h> |
6fc138d2 | 43 | #include <linux/pagemap.h> |
8d4e1288 | 44 | #include <linux/mman.h> |
35149e21 | 45 | #include <linux/swap.h> |
e56d532f | 46 | #include <linux/bitops.h> |
547de29e | 47 | #include <linux/spinlock.h> |
6ff5894c | 48 | #include <linux/compat.h> |
bc6678a3 | 49 | #include <linux/srcu.h> |
8f0b1ab6 | 50 | #include <linux/hugetlb.h> |
5a0e3ad6 | 51 | #include <linux/slab.h> |
743eeb0b SL |
52 | #include <linux/sort.h> |
53 | #include <linux/bsearch.h> | |
6aa8b732 | 54 | |
e495606d | 55 | #include <asm/processor.h> |
e495606d | 56 | #include <asm/io.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
3e021bf5 | 59 | #include <asm/pgtable.h> |
6aa8b732 | 60 | |
5f94c174 | 61 | #include "coalesced_mmio.h" |
af585b92 | 62 | #include "async_pf.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
229456fc MT |
65 | #define CREATE_TRACE_POINTS |
66 | #include <trace/events/kvm.h> | |
67 | ||
536a6f88 JF |
68 | /* Worst case buffer size needed for holding an integer. */ |
69 | #define ITOA_MAX_LEN 12 | |
70 | ||
6aa8b732 AK |
71 | MODULE_AUTHOR("Qumranet"); |
72 | MODULE_LICENSE("GPL"); | |
73 | ||
920552b2 | 74 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 75 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 76 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 77 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 78 | |
aca6ff29 | 79 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 80 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 81 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 82 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 WL |
83 | |
84 | /* Default resets per-vcpu halt_poll_ns . */ | |
ec76d819 | 85 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 86 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 87 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 88 | |
fa40a821 MT |
89 | /* |
90 | * Ordering of locks: | |
91 | * | |
b7d409de | 92 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
93 | */ |
94 | ||
2f303b74 | 95 | DEFINE_SPINLOCK(kvm_lock); |
4a937f96 | 96 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 97 | LIST_HEAD(vm_list); |
133de902 | 98 | |
7f59f492 | 99 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 100 | static int kvm_usage_count; |
10474ae8 | 101 | static atomic_t hardware_enable_failed; |
1b6c0168 | 102 | |
c16f862d RR |
103 | struct kmem_cache *kvm_vcpu_cache; |
104 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 105 | |
15ad7146 AK |
106 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
107 | ||
76f7c879 | 108 | struct dentry *kvm_debugfs_dir; |
e23a808b | 109 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 110 | |
536a6f88 JF |
111 | static int kvm_debugfs_num_entries; |
112 | static const struct file_operations *stat_fops_per_vm[]; | |
113 | ||
bccf2150 AK |
114 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
115 | unsigned long arg); | |
de8e5d74 | 116 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
117 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
118 | unsigned long arg); | |
119 | #endif | |
10474ae8 AG |
120 | static int hardware_enable_all(void); |
121 | static void hardware_disable_all(void); | |
bccf2150 | 122 | |
e93f8a0f | 123 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 124 | |
bc009e43 | 125 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 126 | |
52480137 | 127 | __visible bool kvm_rebooting; |
b7c4145b | 128 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 129 | |
54dee993 MT |
130 | static bool largepages_enabled = true; |
131 | ||
286de8f6 CI |
132 | #define KVM_EVENT_CREATE_VM 0 |
133 | #define KVM_EVENT_DESTROY_VM 1 | |
134 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
135 | static unsigned long long kvm_createvm_count; | |
136 | static unsigned long long kvm_active_vms; | |
137 | ||
b1394e74 RK |
138 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
139 | unsigned long start, unsigned long end) | |
140 | { | |
141 | } | |
142 | ||
ba049e93 | 143 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 144 | { |
11feeb49 | 145 | if (pfn_valid(pfn)) |
bf4bea8e | 146 | return PageReserved(pfn_to_page(pfn)); |
cbff90a7 BAY |
147 | |
148 | return true; | |
149 | } | |
150 | ||
bccf2150 AK |
151 | /* |
152 | * Switches to specified vcpu, until a matching vcpu_put() | |
153 | */ | |
ec7660cc | 154 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 155 | { |
ec7660cc | 156 | int cpu = get_cpu(); |
15ad7146 | 157 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 158 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 159 | put_cpu(); |
6aa8b732 | 160 | } |
2f1fe811 | 161 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 162 | |
313a3dc7 | 163 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 164 | { |
15ad7146 | 165 | preempt_disable(); |
313a3dc7 | 166 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
167 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
168 | preempt_enable(); | |
6aa8b732 | 169 | } |
2f1fe811 | 170 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 171 | |
7a97cec2 PB |
172 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
173 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
174 | { | |
175 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
176 | ||
177 | /* | |
178 | * We need to wait for the VCPU to reenable interrupts and get out of | |
179 | * READING_SHADOW_PAGE_TABLES mode. | |
180 | */ | |
181 | if (req & KVM_REQUEST_WAIT) | |
182 | return mode != OUTSIDE_GUEST_MODE; | |
183 | ||
184 | /* | |
185 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
186 | */ | |
187 | return mode == IN_GUEST_MODE; | |
188 | } | |
189 | ||
d9e368d6 AK |
190 | static void ack_flush(void *_completed) |
191 | { | |
d9e368d6 AK |
192 | } |
193 | ||
b49defe8 PB |
194 | static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait) |
195 | { | |
196 | if (unlikely(!cpus)) | |
197 | cpus = cpu_online_mask; | |
198 | ||
199 | if (cpumask_empty(cpus)) | |
200 | return false; | |
201 | ||
202 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
203 | return true; | |
204 | } | |
205 | ||
445b8236 | 206 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
d9e368d6 | 207 | { |
597a5f55 | 208 | int i, cpu, me; |
6ef7a1bc | 209 | cpumask_var_t cpus; |
b49defe8 | 210 | bool called; |
d9e368d6 | 211 | struct kvm_vcpu *vcpu; |
d9e368d6 | 212 | |
79f55997 | 213 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
6ef7a1bc | 214 | |
3cba4130 | 215 | me = get_cpu(); |
988a2cae | 216 | kvm_for_each_vcpu(i, vcpu, kvm) { |
3cba4130 | 217 | kvm_make_request(req, vcpu); |
d9e368d6 | 218 | cpu = vcpu->cpu; |
6b7e2d09 | 219 | |
178f02ff RK |
220 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
221 | continue; | |
6c6e8360 | 222 | |
6b7e2d09 | 223 | if (cpus != NULL && cpu != -1 && cpu != me && |
7a97cec2 | 224 | kvm_request_needs_ipi(vcpu, req)) |
b49defe8 | 225 | __cpumask_set_cpu(cpu, cpus); |
49846896 | 226 | } |
b49defe8 | 227 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 228 | put_cpu(); |
6ef7a1bc | 229 | free_cpumask_var(cpus); |
49846896 | 230 | return called; |
d9e368d6 AK |
231 | } |
232 | ||
a6d51016 | 233 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 234 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 235 | { |
4ae3cb3a LT |
236 | /* |
237 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
238 | * kvm_make_all_cpus_request. | |
239 | */ | |
240 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
241 | ||
242 | /* | |
243 | * We want to publish modifications to the page tables before reading | |
244 | * mode. Pairs with a memory barrier in arch-specific code. | |
245 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
246 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
247 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
248 | * | |
249 | * There is already an smp_mb__after_atomic() before | |
250 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
251 | * barrier here. | |
252 | */ | |
445b8236 | 253 | if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
49846896 | 254 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 255 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 256 | } |
2ba9f0d8 | 257 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 258 | #endif |
2e53d63a | 259 | |
49846896 RR |
260 | void kvm_reload_remote_mmus(struct kvm *kvm) |
261 | { | |
445b8236 | 262 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 263 | } |
2e53d63a | 264 | |
fb3f0f51 RR |
265 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
266 | { | |
267 | struct page *page; | |
268 | int r; | |
269 | ||
270 | mutex_init(&vcpu->mutex); | |
271 | vcpu->cpu = -1; | |
fb3f0f51 RR |
272 | vcpu->kvm = kvm; |
273 | vcpu->vcpu_id = id; | |
34bb10b7 | 274 | vcpu->pid = NULL; |
8577370f | 275 | init_swait_queue_head(&vcpu->wq); |
af585b92 | 276 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 277 | |
bf9f6ac8 FW |
278 | vcpu->pre_pcpu = -1; |
279 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
280 | ||
fb3f0f51 RR |
281 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
282 | if (!page) { | |
283 | r = -ENOMEM; | |
284 | goto fail; | |
285 | } | |
286 | vcpu->run = page_address(page); | |
287 | ||
4c088493 R |
288 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
289 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 290 | vcpu->preempted = false; |
4c088493 | 291 | |
e9b11c17 | 292 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 293 | if (r < 0) |
e9b11c17 | 294 | goto fail_free_run; |
fb3f0f51 RR |
295 | return 0; |
296 | ||
fb3f0f51 RR |
297 | fail_free_run: |
298 | free_page((unsigned long)vcpu->run); | |
299 | fail: | |
76fafa5e | 300 | return r; |
fb3f0f51 RR |
301 | } |
302 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
303 | ||
304 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
305 | { | |
0e4524a5 CB |
306 | /* |
307 | * no need for rcu_read_lock as VCPU_RUN is the only place that | |
308 | * will change the vcpu->pid pointer and on uninit all file | |
309 | * descriptors are already gone. | |
310 | */ | |
311 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
e9b11c17 | 312 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
313 | free_page((unsigned long)vcpu->run); |
314 | } | |
315 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
316 | ||
e930bffe AA |
317 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
318 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
319 | { | |
320 | return container_of(mn, struct kvm, mmu_notifier); | |
321 | } | |
322 | ||
3da0dd43 IE |
323 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
324 | struct mm_struct *mm, | |
325 | unsigned long address, | |
326 | pte_t pte) | |
327 | { | |
328 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 329 | int idx; |
3da0dd43 | 330 | |
bc6678a3 | 331 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
332 | spin_lock(&kvm->mmu_lock); |
333 | kvm->mmu_notifier_seq++; | |
334 | kvm_set_spte_hva(kvm, address, pte); | |
335 | spin_unlock(&kvm->mmu_lock); | |
bc6678a3 | 336 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
337 | } |
338 | ||
e930bffe AA |
339 | static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
340 | struct mm_struct *mm, | |
341 | unsigned long start, | |
342 | unsigned long end) | |
343 | { | |
344 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 345 | int need_tlb_flush = 0, idx; |
e930bffe | 346 | |
bc6678a3 | 347 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
348 | spin_lock(&kvm->mmu_lock); |
349 | /* | |
350 | * The count increase must become visible at unlock time as no | |
351 | * spte can be established without taking the mmu_lock and | |
352 | * count is also read inside the mmu_lock critical section. | |
353 | */ | |
354 | kvm->mmu_notifier_count++; | |
b3ae2096 | 355 | need_tlb_flush = kvm_unmap_hva_range(kvm, start, end); |
a4ee1ca4 | 356 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
357 | /* we've to flush the tlb before the pages can be freed */ |
358 | if (need_tlb_flush) | |
359 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
360 | |
361 | spin_unlock(&kvm->mmu_lock); | |
b1394e74 RK |
362 | |
363 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
364 | ||
565f3be2 | 365 | srcu_read_unlock(&kvm->srcu, idx); |
e930bffe AA |
366 | } |
367 | ||
368 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
369 | struct mm_struct *mm, | |
370 | unsigned long start, | |
371 | unsigned long end) | |
372 | { | |
373 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
374 | ||
375 | spin_lock(&kvm->mmu_lock); | |
376 | /* | |
377 | * This sequence increase will notify the kvm page fault that | |
378 | * the page that is going to be mapped in the spte could have | |
379 | * been freed. | |
380 | */ | |
381 | kvm->mmu_notifier_seq++; | |
a355aa54 | 382 | smp_wmb(); |
e930bffe AA |
383 | /* |
384 | * The above sequence increase must be visible before the | |
a355aa54 PM |
385 | * below count decrease, which is ensured by the smp_wmb above |
386 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
387 | */ |
388 | kvm->mmu_notifier_count--; | |
389 | spin_unlock(&kvm->mmu_lock); | |
390 | ||
391 | BUG_ON(kvm->mmu_notifier_count < 0); | |
392 | } | |
393 | ||
394 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
395 | struct mm_struct *mm, | |
57128468 ALC |
396 | unsigned long start, |
397 | unsigned long end) | |
e930bffe AA |
398 | { |
399 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 400 | int young, idx; |
e930bffe | 401 | |
bc6678a3 | 402 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 403 | spin_lock(&kvm->mmu_lock); |
e930bffe | 404 | |
57128468 | 405 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
406 | if (young) |
407 | kvm_flush_remote_tlbs(kvm); | |
408 | ||
565f3be2 TY |
409 | spin_unlock(&kvm->mmu_lock); |
410 | srcu_read_unlock(&kvm->srcu, idx); | |
411 | ||
e930bffe AA |
412 | return young; |
413 | } | |
414 | ||
1d7715c6 VD |
415 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
416 | struct mm_struct *mm, | |
417 | unsigned long start, | |
418 | unsigned long end) | |
419 | { | |
420 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
421 | int young, idx; | |
422 | ||
423 | idx = srcu_read_lock(&kvm->srcu); | |
424 | spin_lock(&kvm->mmu_lock); | |
425 | /* | |
426 | * Even though we do not flush TLB, this will still adversely | |
427 | * affect performance on pre-Haswell Intel EPT, where there is | |
428 | * no EPT Access Bit to clear so that we have to tear down EPT | |
429 | * tables instead. If we find this unacceptable, we can always | |
430 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
431 | * do anything on clear_young. | |
432 | * | |
433 | * Also note that currently we never issue secondary TLB flushes | |
434 | * from clear_young, leaving this job up to the regular system | |
435 | * cadence. If we find this inaccurate, we might come up with a | |
436 | * more sophisticated heuristic later. | |
437 | */ | |
438 | young = kvm_age_hva(kvm, start, end); | |
439 | spin_unlock(&kvm->mmu_lock); | |
440 | srcu_read_unlock(&kvm->srcu, idx); | |
441 | ||
442 | return young; | |
443 | } | |
444 | ||
8ee53820 AA |
445 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
446 | struct mm_struct *mm, | |
447 | unsigned long address) | |
448 | { | |
449 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
450 | int young, idx; | |
451 | ||
452 | idx = srcu_read_lock(&kvm->srcu); | |
453 | spin_lock(&kvm->mmu_lock); | |
454 | young = kvm_test_age_hva(kvm, address); | |
455 | spin_unlock(&kvm->mmu_lock); | |
456 | srcu_read_unlock(&kvm->srcu, idx); | |
457 | ||
458 | return young; | |
459 | } | |
460 | ||
85db06e5 MT |
461 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
462 | struct mm_struct *mm) | |
463 | { | |
464 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
465 | int idx; |
466 | ||
467 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 468 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 469 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
470 | } |
471 | ||
e930bffe | 472 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e930bffe AA |
473 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
474 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
475 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 476 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 477 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 478 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 479 | .release = kvm_mmu_notifier_release, |
e930bffe | 480 | }; |
4c07b0a4 AK |
481 | |
482 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
483 | { | |
484 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
485 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
486 | } | |
487 | ||
488 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
489 | ||
490 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
491 | { | |
492 | return 0; | |
493 | } | |
494 | ||
e930bffe AA |
495 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
496 | ||
a47d2b07 | 497 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
498 | { |
499 | int i; | |
a47d2b07 | 500 | struct kvm_memslots *slots; |
bf3e05bc | 501 | |
a7c3e901 | 502 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); |
a47d2b07 PB |
503 | if (!slots) |
504 | return NULL; | |
505 | ||
bf3e05bc | 506 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 507 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
508 | |
509 | return slots; | |
510 | } | |
511 | ||
512 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
513 | { | |
514 | if (!memslot->dirty_bitmap) | |
515 | return; | |
516 | ||
517 | kvfree(memslot->dirty_bitmap); | |
518 | memslot->dirty_bitmap = NULL; | |
519 | } | |
520 | ||
521 | /* | |
522 | * Free any memory in @free but not in @dont. | |
523 | */ | |
524 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
525 | struct kvm_memory_slot *dont) | |
526 | { | |
527 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
528 | kvm_destroy_dirty_bitmap(free); | |
529 | ||
530 | kvm_arch_free_memslot(kvm, free, dont); | |
531 | ||
532 | free->npages = 0; | |
533 | } | |
534 | ||
535 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
536 | { | |
537 | struct kvm_memory_slot *memslot; | |
538 | ||
539 | if (!slots) | |
540 | return; | |
541 | ||
542 | kvm_for_each_memslot(memslot, slots) | |
543 | kvm_free_memslot(kvm, memslot, NULL); | |
544 | ||
545 | kvfree(slots); | |
bf3e05bc XG |
546 | } |
547 | ||
536a6f88 JF |
548 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
549 | { | |
550 | int i; | |
551 | ||
552 | if (!kvm->debugfs_dentry) | |
553 | return; | |
554 | ||
555 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
556 | ||
9d5a1dce LC |
557 | if (kvm->debugfs_stat_data) { |
558 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
559 | kfree(kvm->debugfs_stat_data[i]); | |
560 | kfree(kvm->debugfs_stat_data); | |
561 | } | |
536a6f88 JF |
562 | } |
563 | ||
564 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
565 | { | |
566 | char dir_name[ITOA_MAX_LEN * 2]; | |
567 | struct kvm_stat_data *stat_data; | |
568 | struct kvm_stats_debugfs_item *p; | |
569 | ||
570 | if (!debugfs_initialized()) | |
571 | return 0; | |
572 | ||
573 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
574 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, | |
575 | kvm_debugfs_dir); | |
576 | if (!kvm->debugfs_dentry) | |
577 | return -ENOMEM; | |
578 | ||
579 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
580 | sizeof(*kvm->debugfs_stat_data), | |
581 | GFP_KERNEL); | |
582 | if (!kvm->debugfs_stat_data) | |
583 | return -ENOMEM; | |
584 | ||
585 | for (p = debugfs_entries; p->name; p++) { | |
586 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL); | |
587 | if (!stat_data) | |
588 | return -ENOMEM; | |
589 | ||
590 | stat_data->kvm = kvm; | |
591 | stat_data->offset = p->offset; | |
592 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; | |
ce35ef27 | 593 | if (!debugfs_create_file(p->name, 0644, |
536a6f88 JF |
594 | kvm->debugfs_dentry, |
595 | stat_data, | |
596 | stat_fops_per_vm[p->kind])) | |
597 | return -ENOMEM; | |
598 | } | |
599 | return 0; | |
600 | } | |
601 | ||
e08b9637 | 602 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 603 | { |
d89f5eff JK |
604 | int r, i; |
605 | struct kvm *kvm = kvm_arch_alloc_vm(); | |
6aa8b732 | 606 | |
d89f5eff JK |
607 | if (!kvm) |
608 | return ERR_PTR(-ENOMEM); | |
609 | ||
e9ad4ec8 | 610 | spin_lock_init(&kvm->mmu_lock); |
f1f10076 | 611 | mmgrab(current->mm); |
e9ad4ec8 PB |
612 | kvm->mm = current->mm; |
613 | kvm_eventfd_init(kvm); | |
614 | mutex_init(&kvm->lock); | |
615 | mutex_init(&kvm->irq_lock); | |
616 | mutex_init(&kvm->slots_lock); | |
e3736c3e | 617 | refcount_set(&kvm->users_count, 1); |
e9ad4ec8 PB |
618 | INIT_LIST_HEAD(&kvm->devices); |
619 | ||
e08b9637 | 620 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 621 | if (r) |
719d93cd | 622 | goto out_err_no_disable; |
10474ae8 AG |
623 | |
624 | r = hardware_enable_all(); | |
625 | if (r) | |
719d93cd | 626 | goto out_err_no_disable; |
10474ae8 | 627 | |
c77dcacb | 628 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 629 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 630 | #endif |
6aa8b732 | 631 | |
1e702d9a AW |
632 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
633 | ||
46a26bf5 | 634 | r = -ENOMEM; |
f481b069 | 635 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 PB |
636 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
637 | if (!slots) | |
f481b069 | 638 | goto out_err_no_srcu; |
4bd518f1 PB |
639 | /* |
640 | * Generations must be different for each address space. | |
641 | * Init kvm generation close to the maximum to easily test the | |
642 | * code of handling generation number wrap-around. | |
643 | */ | |
644 | slots->generation = i * 2 - 150; | |
645 | rcu_assign_pointer(kvm->memslots[i], slots); | |
f481b069 | 646 | } |
00f034a1 | 647 | |
bc6678a3 | 648 | if (init_srcu_struct(&kvm->srcu)) |
719d93cd CB |
649 | goto out_err_no_srcu; |
650 | if (init_srcu_struct(&kvm->irq_srcu)) | |
651 | goto out_err_no_irq_srcu; | |
e93f8a0f | 652 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 CB |
653 | rcu_assign_pointer(kvm->buses[i], |
654 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL)); | |
57e7fbee | 655 | if (!kvm->buses[i]) |
e93f8a0f | 656 | goto out_err; |
e93f8a0f | 657 | } |
e930bffe | 658 | |
74b5c5bf MW |
659 | r = kvm_init_mmu_notifier(kvm); |
660 | if (r) | |
661 | goto out_err; | |
662 | ||
2f303b74 | 663 | spin_lock(&kvm_lock); |
5e58cfe4 | 664 | list_add(&kvm->vm_list, &vm_list); |
2f303b74 | 665 | spin_unlock(&kvm_lock); |
d89f5eff | 666 | |
2ecd9d29 PZ |
667 | preempt_notifier_inc(); |
668 | ||
f17abe9a | 669 | return kvm; |
10474ae8 AG |
670 | |
671 | out_err: | |
719d93cd CB |
672 | cleanup_srcu_struct(&kvm->irq_srcu); |
673 | out_err_no_irq_srcu: | |
57e7fbee | 674 | cleanup_srcu_struct(&kvm->srcu); |
719d93cd | 675 | out_err_no_srcu: |
10474ae8 | 676 | hardware_disable_all(); |
719d93cd | 677 | out_err_no_disable: |
021086e3 | 678 | refcount_set(&kvm->users_count, 0); |
e93f8a0f | 679 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 680 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 681 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 682 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
d89f5eff | 683 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 684 | mmdrop(current->mm); |
10474ae8 | 685 | return ERR_PTR(r); |
f17abe9a AK |
686 | } |
687 | ||
07f0a7bd SW |
688 | static void kvm_destroy_devices(struct kvm *kvm) |
689 | { | |
e6e3b5a6 | 690 | struct kvm_device *dev, *tmp; |
07f0a7bd | 691 | |
a28ebea2 CD |
692 | /* |
693 | * We do not need to take the kvm->lock here, because nobody else | |
694 | * has a reference to the struct kvm at this point and therefore | |
695 | * cannot access the devices list anyhow. | |
696 | */ | |
e6e3b5a6 GT |
697 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
698 | list_del(&dev->vm_node); | |
07f0a7bd SW |
699 | dev->ops->destroy(dev); |
700 | } | |
701 | } | |
702 | ||
f17abe9a AK |
703 | static void kvm_destroy_vm(struct kvm *kvm) |
704 | { | |
e93f8a0f | 705 | int i; |
6d4e4c4f AK |
706 | struct mm_struct *mm = kvm->mm; |
707 | ||
286de8f6 | 708 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 709 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 710 | kvm_arch_sync_events(kvm); |
2f303b74 | 711 | spin_lock(&kvm_lock); |
133de902 | 712 | list_del(&kvm->vm_list); |
2f303b74 | 713 | spin_unlock(&kvm_lock); |
399ec807 | 714 | kvm_free_irq_routing(kvm); |
df630b8c | 715 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 716 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 717 | |
4a12f951 CB |
718 | if (bus) |
719 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
720 | kvm->buses[i] = NULL; |
721 | } | |
980da6ce | 722 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
723 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
724 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 725 | #else |
2df72e9b | 726 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 727 | #endif |
d19a9cd2 | 728 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 729 | kvm_destroy_devices(kvm); |
f481b069 | 730 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 731 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 732 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
733 | cleanup_srcu_struct(&kvm->srcu); |
734 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 735 | preempt_notifier_dec(); |
10474ae8 | 736 | hardware_disable_all(); |
6d4e4c4f | 737 | mmdrop(mm); |
f17abe9a AK |
738 | } |
739 | ||
d39f13b0 IE |
740 | void kvm_get_kvm(struct kvm *kvm) |
741 | { | |
e3736c3e | 742 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
743 | } |
744 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
745 | ||
746 | void kvm_put_kvm(struct kvm *kvm) | |
747 | { | |
e3736c3e | 748 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
749 | kvm_destroy_vm(kvm); |
750 | } | |
751 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
752 | ||
753 | ||
f17abe9a AK |
754 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
755 | { | |
756 | struct kvm *kvm = filp->private_data; | |
757 | ||
721eecbf GH |
758 | kvm_irqfd_release(kvm); |
759 | ||
d39f13b0 | 760 | kvm_put_kvm(kvm); |
6aa8b732 AK |
761 | return 0; |
762 | } | |
763 | ||
515a0127 TY |
764 | /* |
765 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 766 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 767 | */ |
a36a57b1 TY |
768 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
769 | { | |
515a0127 | 770 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 771 | |
a7c3e901 | 772 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL); |
a36a57b1 TY |
773 | if (!memslot->dirty_bitmap) |
774 | return -ENOMEM; | |
775 | ||
a36a57b1 TY |
776 | return 0; |
777 | } | |
778 | ||
bf3e05bc | 779 | /* |
0e60b079 IM |
780 | * Insert memslot and re-sort memslots based on their GFN, |
781 | * so binary search could be used to lookup GFN. | |
782 | * Sorting algorithm takes advantage of having initially | |
783 | * sorted array and known changed memslot position. | |
bf3e05bc | 784 | */ |
5cc15027 PB |
785 | static void update_memslots(struct kvm_memslots *slots, |
786 | struct kvm_memory_slot *new) | |
bf3e05bc | 787 | { |
8593176c PB |
788 | int id = new->id; |
789 | int i = slots->id_to_index[id]; | |
063584d4 | 790 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 791 | |
8593176c | 792 | WARN_ON(mslots[i].id != id); |
9c1a5d38 | 793 | if (!new->npages) { |
dbaff309 | 794 | WARN_ON(!mslots[i].npages); |
9c1a5d38 IM |
795 | if (mslots[i].npages) |
796 | slots->used_slots--; | |
797 | } else { | |
798 | if (!mslots[i].npages) | |
799 | slots->used_slots++; | |
800 | } | |
0e60b079 | 801 | |
7f379cff | 802 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
803 | new->base_gfn <= mslots[i + 1].base_gfn) { |
804 | if (!mslots[i + 1].npages) | |
805 | break; | |
7f379cff IM |
806 | mslots[i] = mslots[i + 1]; |
807 | slots->id_to_index[mslots[i].id] = i; | |
808 | i++; | |
809 | } | |
efbeec70 PB |
810 | |
811 | /* | |
812 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
813 | * so that it moves before all those with base_gfn == npages == 0. | |
814 | * | |
815 | * On the other hand, if new->npages is zero, the above loop has | |
816 | * already left i pointing to the beginning of the empty part of | |
817 | * mslots, and the ">=" would move the hole backwards in this | |
818 | * case---which is wrong. So skip the loop when deleting a slot. | |
819 | */ | |
820 | if (new->npages) { | |
821 | while (i > 0 && | |
822 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
823 | mslots[i] = mslots[i - 1]; | |
824 | slots->id_to_index[mslots[i].id] = i; | |
825 | i--; | |
826 | } | |
dbaff309 PB |
827 | } else |
828 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 829 | |
8593176c PB |
830 | mslots[i] = *new; |
831 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
832 | } |
833 | ||
09170a49 | 834 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 835 | { |
4d8b81ab XG |
836 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
837 | ||
0f8a4de3 | 838 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
839 | valid_flags |= KVM_MEM_READONLY; |
840 | #endif | |
841 | ||
842 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
843 | return -EINVAL; |
844 | ||
845 | return 0; | |
846 | } | |
847 | ||
7ec4fb44 | 848 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 849 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 850 | { |
f481b069 | 851 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
7ec4fb44 | 852 | |
ee3d1570 DM |
853 | /* |
854 | * Set the low bit in the generation, which disables SPTE caching | |
855 | * until the end of synchronize_srcu_expedited. | |
856 | */ | |
857 | WARN_ON(old_memslots->generation & 1); | |
858 | slots->generation = old_memslots->generation + 1; | |
859 | ||
f481b069 | 860 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 861 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 862 | |
ee3d1570 DM |
863 | /* |
864 | * Increment the new memslot generation a second time. This prevents | |
865 | * vm exits that race with memslot updates from caching a memslot | |
866 | * generation that will (potentially) be valid forever. | |
4bd518f1 PB |
867 | * |
868 | * Generations must be unique even across address spaces. We do not need | |
869 | * a global counter for that, instead the generation space is evenly split | |
870 | * across address spaces. For example, with two address spaces, address | |
871 | * space 0 will use generations 0, 4, 8, ... while * address space 1 will | |
872 | * use generations 2, 6, 10, 14, ... | |
ee3d1570 | 873 | */ |
4bd518f1 | 874 | slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1; |
ee3d1570 | 875 | |
15f46015 | 876 | kvm_arch_memslots_updated(kvm, slots); |
e59dbe09 TY |
877 | |
878 | return old_memslots; | |
7ec4fb44 GN |
879 | } |
880 | ||
6aa8b732 AK |
881 | /* |
882 | * Allocate some memory and give it an address in the guest physical address | |
883 | * space. | |
884 | * | |
885 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 886 | * |
02d5d55b | 887 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 888 | */ |
f78e0e2e | 889 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 890 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 891 | { |
8234b22e | 892 | int r; |
6aa8b732 | 893 | gfn_t base_gfn; |
28bcb112 | 894 | unsigned long npages; |
a843fac2 | 895 | struct kvm_memory_slot *slot; |
6aa8b732 | 896 | struct kvm_memory_slot old, new; |
b7f69c55 | 897 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 898 | int as_id, id; |
f64c0398 | 899 | enum kvm_mr_change change; |
6aa8b732 | 900 | |
a50d64d6 XG |
901 | r = check_memory_region_flags(mem); |
902 | if (r) | |
903 | goto out; | |
904 | ||
6aa8b732 | 905 | r = -EINVAL; |
f481b069 PB |
906 | as_id = mem->slot >> 16; |
907 | id = (u16)mem->slot; | |
908 | ||
6aa8b732 AK |
909 | /* General sanity checks */ |
910 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
911 | goto out; | |
912 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
913 | goto out; | |
fa3d315a | 914 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 915 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 916 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
9e3bb6b6 HC |
917 | !access_ok(VERIFY_WRITE, |
918 | (void __user *)(unsigned long)mem->userspace_addr, | |
919 | mem->memory_size))) | |
78749809 | 920 | goto out; |
f481b069 | 921 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
922 | goto out; |
923 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
924 | goto out; | |
925 | ||
f481b069 | 926 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
927 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
928 | npages = mem->memory_size >> PAGE_SHIFT; | |
929 | ||
660c22c4 TY |
930 | if (npages > KVM_MEM_MAX_NR_PAGES) |
931 | goto out; | |
932 | ||
a843fac2 | 933 | new = old = *slot; |
6aa8b732 | 934 | |
f481b069 | 935 | new.id = id; |
6aa8b732 AK |
936 | new.base_gfn = base_gfn; |
937 | new.npages = npages; | |
938 | new.flags = mem->flags; | |
939 | ||
f64c0398 TY |
940 | if (npages) { |
941 | if (!old.npages) | |
942 | change = KVM_MR_CREATE; | |
943 | else { /* Modify an existing slot. */ | |
944 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
945 | (npages != old.npages) || |
946 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
947 | goto out; |
948 | ||
949 | if (base_gfn != old.base_gfn) | |
950 | change = KVM_MR_MOVE; | |
951 | else if (new.flags != old.flags) | |
952 | change = KVM_MR_FLAGS_ONLY; | |
953 | else { /* Nothing to change. */ | |
954 | r = 0; | |
955 | goto out; | |
956 | } | |
957 | } | |
09170a49 PB |
958 | } else { |
959 | if (!old.npages) | |
960 | goto out; | |
961 | ||
f64c0398 | 962 | change = KVM_MR_DELETE; |
09170a49 PB |
963 | new.base_gfn = 0; |
964 | new.flags = 0; | |
965 | } | |
6aa8b732 | 966 | |
f64c0398 | 967 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
968 | /* Check for overlaps */ |
969 | r = -EEXIST; | |
f481b069 | 970 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
a843fac2 | 971 | if ((slot->id >= KVM_USER_MEM_SLOTS) || |
f481b069 | 972 | (slot->id == id)) |
0a706bee TY |
973 | continue; |
974 | if (!((base_gfn + npages <= slot->base_gfn) || | |
975 | (base_gfn >= slot->base_gfn + slot->npages))) | |
976 | goto out; | |
977 | } | |
6aa8b732 | 978 | } |
6aa8b732 | 979 | |
6aa8b732 AK |
980 | /* Free page dirty bitmap if unneeded */ |
981 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 982 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
983 | |
984 | r = -ENOMEM; | |
f64c0398 | 985 | if (change == KVM_MR_CREATE) { |
189a2f7b | 986 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 987 | |
5587027c | 988 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 989 | goto out_free; |
6aa8b732 | 990 | } |
ec04b260 | 991 | |
6aa8b732 AK |
992 | /* Allocate page dirty bitmap if needed */ |
993 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 994 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 995 | goto out_free; |
6aa8b732 AK |
996 | } |
997 | ||
a7c3e901 | 998 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); |
f2a81036 PB |
999 | if (!slots) |
1000 | goto out_free; | |
f481b069 | 1001 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 1002 | |
f64c0398 | 1003 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 1004 | slot = id_to_memslot(slots, id); |
28a37544 XG |
1005 | slot->flags |= KVM_MEMSLOT_INVALID; |
1006 | ||
f481b069 | 1007 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 1008 | |
12d6e753 MT |
1009 | /* From this point no new shadow pages pointing to a deleted, |
1010 | * or moved, memslot will be created. | |
bc6678a3 MT |
1011 | * |
1012 | * validation of sp->gfn happens in: | |
b7d409de XL |
1013 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
1014 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 1015 | */ |
2df72e9b | 1016 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
1017 | |
1018 | /* | |
1019 | * We can re-use the old_memslots from above, the only difference | |
1020 | * from the currently installed memslots is the invalid flag. This | |
1021 | * will get overwritten by update_memslots anyway. | |
1022 | */ | |
b7f69c55 | 1023 | slots = old_memslots; |
bc6678a3 | 1024 | } |
34d4cb8f | 1025 | |
7b6195a9 | 1026 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 1027 | if (r) |
b7f69c55 | 1028 | goto out_slots; |
f7784b8e | 1029 | |
a47d2b07 | 1030 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 1031 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 1032 | new.dirty_bitmap = NULL; |
db3fe4eb | 1033 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
1034 | } |
1035 | ||
5cc15027 | 1036 | update_memslots(slots, &new); |
f481b069 | 1037 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 1038 | |
f36f3f28 | 1039 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 1040 | |
a47d2b07 | 1041 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 1042 | kvfree(old_memslots); |
6aa8b732 AK |
1043 | return 0; |
1044 | ||
e40f193f | 1045 | out_slots: |
74496134 | 1046 | kvfree(slots); |
f78e0e2e | 1047 | out_free: |
a47d2b07 | 1048 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
1049 | out: |
1050 | return r; | |
210c7c4d | 1051 | } |
f78e0e2e SY |
1052 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1053 | ||
1054 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1055 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1056 | { |
1057 | int r; | |
1058 | ||
79fac95e | 1059 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1060 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1061 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1062 | return r; |
1063 | } | |
210c7c4d IE |
1064 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1065 | ||
7940876e SH |
1066 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1067 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1068 | { |
f481b069 | 1069 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1070 | return -EINVAL; |
09170a49 | 1071 | |
47ae31e2 | 1072 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1073 | } |
1074 | ||
5bb064dc ZX |
1075 | int kvm_get_dirty_log(struct kvm *kvm, |
1076 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 1077 | { |
9f6b8029 | 1078 | struct kvm_memslots *slots; |
6aa8b732 | 1079 | struct kvm_memory_slot *memslot; |
843574a3 | 1080 | int i, as_id, id; |
87bf6e7d | 1081 | unsigned long n; |
6aa8b732 AK |
1082 | unsigned long any = 0; |
1083 | ||
f481b069 PB |
1084 | as_id = log->slot >> 16; |
1085 | id = (u16)log->slot; | |
1086 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1087 | return -EINVAL; |
6aa8b732 | 1088 | |
f481b069 PB |
1089 | slots = __kvm_memslots(kvm, as_id); |
1090 | memslot = id_to_memslot(slots, id); | |
6aa8b732 | 1091 | if (!memslot->dirty_bitmap) |
843574a3 | 1092 | return -ENOENT; |
6aa8b732 | 1093 | |
87bf6e7d | 1094 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 1095 | |
cd1a4a98 | 1096 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1097 | any = memslot->dirty_bitmap[i]; |
1098 | ||
6aa8b732 | 1099 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
843574a3 | 1100 | return -EFAULT; |
6aa8b732 | 1101 | |
5bb064dc ZX |
1102 | if (any) |
1103 | *is_dirty = 1; | |
843574a3 | 1104 | return 0; |
6aa8b732 | 1105 | } |
2ba9f0d8 | 1106 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1107 | |
ba0513b5 MS |
1108 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1109 | /** | |
1110 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages | |
1111 | * are dirty write protect them for next write. | |
1112 | * @kvm: pointer to kvm instance | |
1113 | * @log: slot id and address to which we copy the log | |
1114 | * @is_dirty: flag set if any page is dirty | |
1115 | * | |
1116 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1117 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1118 | * following order: | |
1119 | * | |
1120 | * 1. Take a snapshot of the bit and clear it if needed. | |
1121 | * 2. Write protect the corresponding page. | |
1122 | * 3. Copy the snapshot to the userspace. | |
1123 | * 4. Upon return caller flushes TLB's if needed. | |
1124 | * | |
1125 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1126 | * entry. This is not a problem because the page is reported dirty using | |
1127 | * the snapshot taken before and step 4 ensures that writes done after | |
1128 | * exiting to userspace will be logged for the next call. | |
1129 | * | |
1130 | */ | |
1131 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
1132 | struct kvm_dirty_log *log, bool *is_dirty) | |
1133 | { | |
9f6b8029 | 1134 | struct kvm_memslots *slots; |
ba0513b5 | 1135 | struct kvm_memory_slot *memslot; |
58d6db34 | 1136 | int i, as_id, id; |
ba0513b5 MS |
1137 | unsigned long n; |
1138 | unsigned long *dirty_bitmap; | |
1139 | unsigned long *dirty_bitmap_buffer; | |
1140 | ||
f481b069 PB |
1141 | as_id = log->slot >> 16; |
1142 | id = (u16)log->slot; | |
1143 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1144 | return -EINVAL; |
ba0513b5 | 1145 | |
f481b069 PB |
1146 | slots = __kvm_memslots(kvm, as_id); |
1147 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1148 | |
1149 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1150 | if (!dirty_bitmap) |
58d6db34 | 1151 | return -ENOENT; |
ba0513b5 MS |
1152 | |
1153 | n = kvm_dirty_bitmap_bytes(memslot); | |
1154 | ||
1155 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); | |
1156 | memset(dirty_bitmap_buffer, 0, n); | |
1157 | ||
1158 | spin_lock(&kvm->mmu_lock); | |
1159 | *is_dirty = false; | |
1160 | for (i = 0; i < n / sizeof(long); i++) { | |
1161 | unsigned long mask; | |
1162 | gfn_t offset; | |
1163 | ||
1164 | if (!dirty_bitmap[i]) | |
1165 | continue; | |
1166 | ||
1167 | *is_dirty = true; | |
1168 | ||
1169 | mask = xchg(&dirty_bitmap[i], 0); | |
1170 | dirty_bitmap_buffer[i] = mask; | |
1171 | ||
58d2930f TY |
1172 | if (mask) { |
1173 | offset = i * BITS_PER_LONG; | |
1174 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1175 | offset, mask); | |
1176 | } | |
ba0513b5 MS |
1177 | } |
1178 | ||
1179 | spin_unlock(&kvm->mmu_lock); | |
ba0513b5 | 1180 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
58d6db34 ME |
1181 | return -EFAULT; |
1182 | return 0; | |
ba0513b5 MS |
1183 | } |
1184 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1185 | #endif | |
1186 | ||
db3fe4eb TY |
1187 | bool kvm_largepages_enabled(void) |
1188 | { | |
1189 | return largepages_enabled; | |
1190 | } | |
1191 | ||
54dee993 MT |
1192 | void kvm_disable_largepages(void) |
1193 | { | |
1194 | largepages_enabled = false; | |
1195 | } | |
1196 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1197 | ||
49c7754c GN |
1198 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1199 | { | |
1200 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1201 | } | |
a1f4d395 | 1202 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1203 | |
8e73485c PB |
1204 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1205 | { | |
1206 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1207 | } | |
1208 | ||
33e94154 | 1209 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1210 | { |
bf3e05bc | 1211 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1212 | |
bbacc0c1 | 1213 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc | 1214 | memslot->flags & KVM_MEMSLOT_INVALID) |
33e94154 | 1215 | return false; |
e0d62c7f | 1216 | |
33e94154 | 1217 | return true; |
e0d62c7f IE |
1218 | } |
1219 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1220 | ||
8f0b1ab6 JR |
1221 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1222 | { | |
1223 | struct vm_area_struct *vma; | |
1224 | unsigned long addr, size; | |
1225 | ||
1226 | size = PAGE_SIZE; | |
1227 | ||
1228 | addr = gfn_to_hva(kvm, gfn); | |
1229 | if (kvm_is_error_hva(addr)) | |
1230 | return PAGE_SIZE; | |
1231 | ||
1232 | down_read(¤t->mm->mmap_sem); | |
1233 | vma = find_vma(current->mm, addr); | |
1234 | if (!vma) | |
1235 | goto out; | |
1236 | ||
1237 | size = vma_kernel_pagesize(vma); | |
1238 | ||
1239 | out: | |
1240 | up_read(¤t->mm->mmap_sem); | |
1241 | ||
1242 | return size; | |
1243 | } | |
1244 | ||
4d8b81ab XG |
1245 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1246 | { | |
1247 | return slot->flags & KVM_MEM_READONLY; | |
1248 | } | |
1249 | ||
4d8b81ab XG |
1250 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1251 | gfn_t *nr_pages, bool write) | |
539cb660 | 1252 | { |
bc6678a3 | 1253 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1254 | return KVM_HVA_ERR_BAD; |
48987781 | 1255 | |
4d8b81ab XG |
1256 | if (memslot_is_readonly(slot) && write) |
1257 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1258 | |
1259 | if (nr_pages) | |
1260 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1261 | ||
4d8b81ab | 1262 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1263 | } |
48987781 | 1264 | |
4d8b81ab XG |
1265 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1266 | gfn_t *nr_pages) | |
1267 | { | |
1268 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1269 | } |
48987781 | 1270 | |
4d8b81ab | 1271 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1272 | gfn_t gfn) |
4d8b81ab XG |
1273 | { |
1274 | return gfn_to_hva_many(slot, gfn, NULL); | |
1275 | } | |
1276 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1277 | ||
48987781 XG |
1278 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1279 | { | |
49c7754c | 1280 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1281 | } |
0d150298 | 1282 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1283 | |
8e73485c PB |
1284 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1285 | { | |
1286 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1287 | } | |
1288 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1289 | ||
86ab8cff | 1290 | /* |
ba6a3541 PB |
1291 | * If writable is set to false, the hva returned by this function is only |
1292 | * allowed to be read. | |
86ab8cff | 1293 | */ |
64d83126 CD |
1294 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1295 | gfn_t gfn, bool *writable) | |
86ab8cff | 1296 | { |
a2ac07fe GN |
1297 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1298 | ||
1299 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1300 | *writable = !memslot_is_readonly(slot); |
1301 | ||
a2ac07fe | 1302 | return hva; |
86ab8cff XG |
1303 | } |
1304 | ||
64d83126 CD |
1305 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1306 | { | |
1307 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1308 | ||
1309 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1310 | } | |
1311 | ||
8e73485c PB |
1312 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1313 | { | |
1314 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1315 | ||
1316 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1317 | } | |
1318 | ||
d4edcf0d DH |
1319 | static int get_user_page_nowait(unsigned long start, int write, |
1320 | struct page **page) | |
0857b9e9 | 1321 | { |
0d731759 | 1322 | int flags = FOLL_NOWAIT | FOLL_HWPOISON; |
0857b9e9 GN |
1323 | |
1324 | if (write) | |
1325 | flags |= FOLL_WRITE; | |
1326 | ||
0d731759 | 1327 | return get_user_pages(start, 1, flags, page, NULL); |
0857b9e9 GN |
1328 | } |
1329 | ||
fafc3dba HY |
1330 | static inline int check_user_page_hwpoison(unsigned long addr) |
1331 | { | |
0d731759 | 1332 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1333 | |
0d731759 | 1334 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1335 | return rc == -EHWPOISON; |
1336 | } | |
1337 | ||
2fc84311 XG |
1338 | /* |
1339 | * The atomic path to get the writable pfn which will be stored in @pfn, | |
1340 | * true indicates success, otherwise false is returned. | |
1341 | */ | |
1342 | static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async, | |
ba049e93 | 1343 | bool write_fault, bool *writable, kvm_pfn_t *pfn) |
954bbbc2 | 1344 | { |
8d4e1288 | 1345 | struct page *page[1]; |
2fc84311 | 1346 | int npages; |
954bbbc2 | 1347 | |
2fc84311 XG |
1348 | if (!(async || atomic)) |
1349 | return false; | |
af585b92 | 1350 | |
12ce13fe XG |
1351 | /* |
1352 | * Fast pin a writable pfn only if it is a write fault request | |
1353 | * or the caller allows to map a writable pfn for a read fault | |
1354 | * request. | |
1355 | */ | |
1356 | if (!(write_fault || writable)) | |
1357 | return false; | |
612819c3 | 1358 | |
2fc84311 XG |
1359 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1360 | if (npages == 1) { | |
1361 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1362 | |
2fc84311 XG |
1363 | if (writable) |
1364 | *writable = true; | |
1365 | return true; | |
1366 | } | |
af585b92 | 1367 | |
2fc84311 XG |
1368 | return false; |
1369 | } | |
612819c3 | 1370 | |
2fc84311 XG |
1371 | /* |
1372 | * The slow path to get the pfn of the specified host virtual address, | |
1373 | * 1 indicates success, -errno is returned if error is detected. | |
1374 | */ | |
1375 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1376 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 XG |
1377 | { |
1378 | struct page *page[1]; | |
1379 | int npages = 0; | |
612819c3 | 1380 | |
2fc84311 XG |
1381 | might_sleep(); |
1382 | ||
1383 | if (writable) | |
1384 | *writable = write_fault; | |
1385 | ||
1386 | if (async) { | |
1387 | down_read(¤t->mm->mmap_sem); | |
d4edcf0d | 1388 | npages = get_user_page_nowait(addr, write_fault, page); |
2fc84311 | 1389 | up_read(¤t->mm->mmap_sem); |
d4944b0e | 1390 | } else { |
8b7457ef | 1391 | unsigned int flags = FOLL_HWPOISON; |
d4944b0e LS |
1392 | |
1393 | if (write_fault) | |
1394 | flags |= FOLL_WRITE; | |
1395 | ||
8b7457ef | 1396 | npages = get_user_pages_unlocked(addr, 1, page, flags); |
d4944b0e | 1397 | } |
2fc84311 XG |
1398 | if (npages != 1) |
1399 | return npages; | |
1400 | ||
1401 | /* map read fault as writable if possible */ | |
12ce13fe | 1402 | if (unlikely(!write_fault) && writable) { |
2fc84311 XG |
1403 | struct page *wpage[1]; |
1404 | ||
1405 | npages = __get_user_pages_fast(addr, 1, 1, wpage); | |
1406 | if (npages == 1) { | |
1407 | *writable = true; | |
1408 | put_page(page[0]); | |
1409 | page[0] = wpage[0]; | |
612819c3 | 1410 | } |
2fc84311 XG |
1411 | |
1412 | npages = 1; | |
887c08ac | 1413 | } |
2fc84311 XG |
1414 | *pfn = page_to_pfn(page[0]); |
1415 | return npages; | |
1416 | } | |
539cb660 | 1417 | |
4d8b81ab XG |
1418 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1419 | { | |
1420 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1421 | return false; | |
2e2e3738 | 1422 | |
4d8b81ab XG |
1423 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1424 | return false; | |
887c08ac | 1425 | |
4d8b81ab XG |
1426 | return true; |
1427 | } | |
bf998156 | 1428 | |
92176a8e PB |
1429 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1430 | unsigned long addr, bool *async, | |
1431 | bool write_fault, kvm_pfn_t *p_pfn) | |
1432 | { | |
add6a0cd PB |
1433 | unsigned long pfn; |
1434 | int r; | |
1435 | ||
1436 | r = follow_pfn(vma, addr, &pfn); | |
1437 | if (r) { | |
1438 | /* | |
1439 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1440 | * not call the fault handler, so do it here. | |
1441 | */ | |
1442 | bool unlocked = false; | |
1443 | r = fixup_user_fault(current, current->mm, addr, | |
1444 | (write_fault ? FAULT_FLAG_WRITE : 0), | |
1445 | &unlocked); | |
1446 | if (unlocked) | |
1447 | return -EAGAIN; | |
1448 | if (r) | |
1449 | return r; | |
1450 | ||
1451 | r = follow_pfn(vma, addr, &pfn); | |
1452 | if (r) | |
1453 | return r; | |
1454 | ||
1455 | } | |
1456 | ||
1457 | ||
1458 | /* | |
1459 | * Get a reference here because callers of *hva_to_pfn* and | |
1460 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1461 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1462 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1463 | * simply do nothing for reserved pfns. | |
1464 | * | |
1465 | * Whoever called remap_pfn_range is also going to call e.g. | |
1466 | * unmap_mapping_range before the underlying pages are freed, | |
1467 | * causing a call to our MMU notifier. | |
1468 | */ | |
1469 | kvm_get_pfn(pfn); | |
1470 | ||
1471 | *p_pfn = pfn; | |
92176a8e PB |
1472 | return 0; |
1473 | } | |
1474 | ||
12ce13fe XG |
1475 | /* |
1476 | * Pin guest page in memory and return its pfn. | |
1477 | * @addr: host virtual address which maps memory to the guest | |
1478 | * @atomic: whether this function can sleep | |
1479 | * @async: whether this function need to wait IO complete if the | |
1480 | * host page is not in the memory | |
1481 | * @write_fault: whether we should get a writable host page | |
1482 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1483 | * | |
1484 | * The function will map a writable host page for these two cases: | |
1485 | * 1): @write_fault = true | |
1486 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1487 | * whether the mapping is writable. | |
1488 | */ | |
ba049e93 | 1489 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1490 | bool write_fault, bool *writable) |
1491 | { | |
1492 | struct vm_area_struct *vma; | |
ba049e93 | 1493 | kvm_pfn_t pfn = 0; |
92176a8e | 1494 | int npages, r; |
2e2e3738 | 1495 | |
2fc84311 XG |
1496 | /* we can do it either atomically or asynchronously, not both */ |
1497 | BUG_ON(atomic && async); | |
8d4e1288 | 1498 | |
2fc84311 XG |
1499 | if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn)) |
1500 | return pfn; | |
1501 | ||
1502 | if (atomic) | |
1503 | return KVM_PFN_ERR_FAULT; | |
1504 | ||
1505 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1506 | if (npages == 1) | |
1507 | return pfn; | |
8d4e1288 | 1508 | |
2fc84311 XG |
1509 | down_read(¤t->mm->mmap_sem); |
1510 | if (npages == -EHWPOISON || | |
1511 | (!async && check_user_page_hwpoison(addr))) { | |
1512 | pfn = KVM_PFN_ERR_HWPOISON; | |
1513 | goto exit; | |
1514 | } | |
1515 | ||
add6a0cd | 1516 | retry: |
2fc84311 XG |
1517 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
1518 | ||
1519 | if (vma == NULL) | |
1520 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e PB |
1521 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1522 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn); | |
add6a0cd PB |
1523 | if (r == -EAGAIN) |
1524 | goto retry; | |
92176a8e PB |
1525 | if (r < 0) |
1526 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 1527 | } else { |
4d8b81ab | 1528 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1529 | *async = true; |
1530 | pfn = KVM_PFN_ERR_FAULT; | |
1531 | } | |
1532 | exit: | |
1533 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1534 | return pfn; |
35149e21 AL |
1535 | } |
1536 | ||
ba049e93 DW |
1537 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
1538 | bool atomic, bool *async, bool write_fault, | |
1539 | bool *writable) | |
887c08ac | 1540 | { |
4d8b81ab XG |
1541 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1542 | ||
b2740d35 PB |
1543 | if (addr == KVM_HVA_ERR_RO_BAD) { |
1544 | if (writable) | |
1545 | *writable = false; | |
4d8b81ab | 1546 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 1547 | } |
4d8b81ab | 1548 | |
b2740d35 PB |
1549 | if (kvm_is_error_hva(addr)) { |
1550 | if (writable) | |
1551 | *writable = false; | |
81c52c56 | 1552 | return KVM_PFN_NOSLOT; |
b2740d35 | 1553 | } |
4d8b81ab XG |
1554 | |
1555 | /* Do not map writable pfn in the readonly memslot. */ | |
1556 | if (writable && memslot_is_readonly(slot)) { | |
1557 | *writable = false; | |
1558 | writable = NULL; | |
1559 | } | |
1560 | ||
1561 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1562 | writable); | |
887c08ac | 1563 | } |
3520469d | 1564 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1565 | |
ba049e93 | 1566 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
1567 | bool *writable) |
1568 | { | |
e37afc6e PB |
1569 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1570 | write_fault, writable); | |
612819c3 MT |
1571 | } |
1572 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1573 | ||
ba049e93 | 1574 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1575 | { |
4d8b81ab | 1576 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1577 | } |
e37afc6e | 1578 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1579 | |
ba049e93 | 1580 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1581 | { |
4d8b81ab | 1582 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1583 | } |
037d92dc | 1584 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1585 | |
ba049e93 | 1586 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1587 | { |
1588 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1589 | } | |
1590 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1591 | ||
ba049e93 | 1592 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1593 | { |
1594 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1595 | } | |
1596 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1597 | ||
ba049e93 | 1598 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1599 | { |
1600 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1601 | } | |
1602 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1603 | ||
ba049e93 | 1604 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1605 | { |
1606 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1607 | } | |
1608 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1609 | ||
d9ef13c2 PB |
1610 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1611 | struct page **pages, int nr_pages) | |
48987781 XG |
1612 | { |
1613 | unsigned long addr; | |
076b925d | 1614 | gfn_t entry = 0; |
48987781 | 1615 | |
d9ef13c2 | 1616 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1617 | if (kvm_is_error_hva(addr)) |
1618 | return -1; | |
1619 | ||
1620 | if (entry < nr_pages) | |
1621 | return 0; | |
1622 | ||
1623 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1624 | } | |
1625 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1626 | ||
ba049e93 | 1627 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 1628 | { |
81c52c56 | 1629 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1630 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1631 | |
bf4bea8e | 1632 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1633 | WARN_ON(1); |
6cede2e6 | 1634 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1635 | } |
a2766325 XG |
1636 | |
1637 | return pfn_to_page(pfn); | |
1638 | } | |
1639 | ||
35149e21 AL |
1640 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1641 | { | |
ba049e93 | 1642 | kvm_pfn_t pfn; |
2e2e3738 AL |
1643 | |
1644 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1645 | |
a2766325 | 1646 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1647 | } |
1648 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1649 | ||
8e73485c PB |
1650 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1651 | { | |
ba049e93 | 1652 | kvm_pfn_t pfn; |
8e73485c PB |
1653 | |
1654 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1655 | ||
1656 | return kvm_pfn_to_page(pfn); | |
1657 | } | |
1658 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1659 | ||
b4231d61 IE |
1660 | void kvm_release_page_clean(struct page *page) |
1661 | { | |
32cad84f XG |
1662 | WARN_ON(is_error_page(page)); |
1663 | ||
35149e21 | 1664 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1665 | } |
1666 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1667 | ||
ba049e93 | 1668 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 1669 | { |
bf4bea8e | 1670 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1671 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1672 | } |
1673 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1674 | ||
b4231d61 | 1675 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1676 | { |
a2766325 XG |
1677 | WARN_ON(is_error_page(page)); |
1678 | ||
35149e21 AL |
1679 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1680 | } | |
1681 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1682 | ||
f7a6509f | 1683 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
1684 | { |
1685 | kvm_set_pfn_dirty(pfn); | |
1686 | kvm_release_pfn_clean(pfn); | |
1687 | } | |
f7a6509f | 1688 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 1689 | |
ba049e93 | 1690 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 1691 | { |
bf4bea8e | 1692 | if (!kvm_is_reserved_pfn(pfn)) { |
2e2e3738 | 1693 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1694 | |
2e2e3738 AL |
1695 | if (!PageReserved(page)) |
1696 | SetPageDirty(page); | |
1697 | } | |
8a7ae055 | 1698 | } |
35149e21 AL |
1699 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1700 | ||
ba049e93 | 1701 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 1702 | { |
bf4bea8e | 1703 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1704 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1705 | } |
1706 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1707 | ||
ba049e93 | 1708 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 1709 | { |
bf4bea8e | 1710 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1711 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1712 | } |
1713 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1714 | |
195aefde IE |
1715 | static int next_segment(unsigned long len, int offset) |
1716 | { | |
1717 | if (len > PAGE_SIZE - offset) | |
1718 | return PAGE_SIZE - offset; | |
1719 | else | |
1720 | return len; | |
1721 | } | |
1722 | ||
8e73485c PB |
1723 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1724 | void *data, int offset, int len) | |
195aefde | 1725 | { |
e0506bcb IE |
1726 | int r; |
1727 | unsigned long addr; | |
195aefde | 1728 | |
8e73485c | 1729 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1730 | if (kvm_is_error_hva(addr)) |
1731 | return -EFAULT; | |
3180a7fc | 1732 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1733 | if (r) |
195aefde | 1734 | return -EFAULT; |
195aefde IE |
1735 | return 0; |
1736 | } | |
8e73485c PB |
1737 | |
1738 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1739 | int len) | |
1740 | { | |
1741 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1742 | ||
1743 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1744 | } | |
195aefde IE |
1745 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1746 | ||
8e73485c PB |
1747 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1748 | int offset, int len) | |
1749 | { | |
1750 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1751 | ||
1752 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1753 | } | |
1754 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1755 | ||
195aefde IE |
1756 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1757 | { | |
1758 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1759 | int seg; | |
1760 | int offset = offset_in_page(gpa); | |
1761 | int ret; | |
1762 | ||
1763 | while ((seg = next_segment(len, offset)) != 0) { | |
1764 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1765 | if (ret < 0) | |
1766 | return ret; | |
1767 | offset = 0; | |
1768 | len -= seg; | |
1769 | data += seg; | |
1770 | ++gfn; | |
1771 | } | |
1772 | return 0; | |
1773 | } | |
1774 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1775 | ||
8e73485c | 1776 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 1777 | { |
7ec54588 | 1778 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 1779 | int seg; |
7ec54588 | 1780 | int offset = offset_in_page(gpa); |
8e73485c PB |
1781 | int ret; |
1782 | ||
1783 | while ((seg = next_segment(len, offset)) != 0) { | |
1784 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
1785 | if (ret < 0) | |
1786 | return ret; | |
1787 | offset = 0; | |
1788 | len -= seg; | |
1789 | data += seg; | |
1790 | ++gfn; | |
1791 | } | |
1792 | return 0; | |
1793 | } | |
1794 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 1795 | |
8e73485c PB |
1796 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1797 | void *data, int offset, unsigned long len) | |
1798 | { | |
1799 | int r; | |
1800 | unsigned long addr; | |
1801 | ||
1802 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
1803 | if (kvm_is_error_hva(addr)) |
1804 | return -EFAULT; | |
0aac03f0 | 1805 | pagefault_disable(); |
3180a7fc | 1806 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 1807 | pagefault_enable(); |
7ec54588 MT |
1808 | if (r) |
1809 | return -EFAULT; | |
1810 | return 0; | |
1811 | } | |
7ec54588 | 1812 | |
8e73485c PB |
1813 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
1814 | unsigned long len) | |
1815 | { | |
1816 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1817 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1818 | int offset = offset_in_page(gpa); | |
1819 | ||
1820 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1821 | } | |
1822 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
1823 | ||
1824 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
1825 | void *data, unsigned long len) | |
1826 | { | |
1827 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1828 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1829 | int offset = offset_in_page(gpa); | |
1830 | ||
1831 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
1832 | } | |
1833 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
1834 | ||
1835 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
1836 | const void *data, int offset, int len) | |
195aefde | 1837 | { |
e0506bcb IE |
1838 | int r; |
1839 | unsigned long addr; | |
195aefde | 1840 | |
251eb841 | 1841 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
1842 | if (kvm_is_error_hva(addr)) |
1843 | return -EFAULT; | |
8b0cedff | 1844 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 1845 | if (r) |
195aefde | 1846 | return -EFAULT; |
bc009e43 | 1847 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
1848 | return 0; |
1849 | } | |
8e73485c PB |
1850 | |
1851 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
1852 | const void *data, int offset, int len) | |
1853 | { | |
1854 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1855 | ||
1856 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1857 | } | |
195aefde IE |
1858 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
1859 | ||
8e73485c PB |
1860 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
1861 | const void *data, int offset, int len) | |
1862 | { | |
1863 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1864 | ||
1865 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
1866 | } | |
1867 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
1868 | ||
195aefde IE |
1869 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
1870 | unsigned long len) | |
1871 | { | |
1872 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1873 | int seg; | |
1874 | int offset = offset_in_page(gpa); | |
1875 | int ret; | |
1876 | ||
1877 | while ((seg = next_segment(len, offset)) != 0) { | |
1878 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
1879 | if (ret < 0) | |
1880 | return ret; | |
1881 | offset = 0; | |
1882 | len -= seg; | |
1883 | data += seg; | |
1884 | ++gfn; | |
1885 | } | |
1886 | return 0; | |
1887 | } | |
ff651cb6 | 1888 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 1889 | |
8e73485c PB |
1890 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
1891 | unsigned long len) | |
1892 | { | |
1893 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1894 | int seg; | |
1895 | int offset = offset_in_page(gpa); | |
1896 | int ret; | |
1897 | ||
1898 | while ((seg = next_segment(len, offset)) != 0) { | |
1899 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
1900 | if (ret < 0) | |
1901 | return ret; | |
1902 | offset = 0; | |
1903 | len -= seg; | |
1904 | data += seg; | |
1905 | ++gfn; | |
1906 | } | |
1907 | return 0; | |
1908 | } | |
1909 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
1910 | ||
5a2d4365 PB |
1911 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
1912 | struct gfn_to_hva_cache *ghc, | |
1913 | gpa_t gpa, unsigned long len) | |
49c7754c | 1914 | { |
49c7754c | 1915 | int offset = offset_in_page(gpa); |
8f964525 AH |
1916 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
1917 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
1918 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
1919 | gfn_t nr_pages_avail; | |
49c7754c GN |
1920 | |
1921 | ghc->gpa = gpa; | |
1922 | ghc->generation = slots->generation; | |
8f964525 | 1923 | ghc->len = len; |
5a2d4365 | 1924 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
ca3f0874 RK |
1925 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL); |
1926 | if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) { | |
49c7754c | 1927 | ghc->hva += offset; |
8f964525 AH |
1928 | } else { |
1929 | /* | |
1930 | * If the requested region crosses two memslots, we still | |
1931 | * verify that the entire region is valid here. | |
1932 | */ | |
1933 | while (start_gfn <= end_gfn) { | |
076b925d | 1934 | nr_pages_avail = 0; |
5a2d4365 | 1935 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
8f964525 AH |
1936 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, |
1937 | &nr_pages_avail); | |
1938 | if (kvm_is_error_hva(ghc->hva)) | |
1939 | return -EFAULT; | |
1940 | start_gfn += nr_pages_avail; | |
1941 | } | |
1942 | /* Use the slow path for cross page reads and writes. */ | |
1943 | ghc->memslot = NULL; | |
1944 | } | |
49c7754c GN |
1945 | return 0; |
1946 | } | |
5a2d4365 | 1947 | |
4e335d9e | 1948 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
1949 | gpa_t gpa, unsigned long len) |
1950 | { | |
4e335d9e | 1951 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
1952 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
1953 | } | |
4e335d9e | 1954 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 1955 | |
4e335d9e PB |
1956 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1957 | void *data, int offset, unsigned long len) | |
49c7754c | 1958 | { |
4e335d9e | 1959 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 1960 | int r; |
4ec6e863 | 1961 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 1962 | |
4ec6e863 | 1963 | BUG_ON(len + offset > ghc->len); |
8f964525 | 1964 | |
49c7754c | 1965 | if (slots->generation != ghc->generation) |
5a2d4365 | 1966 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
1967 | |
1968 | if (unlikely(!ghc->memslot)) | |
4e335d9e | 1969 | return kvm_write_guest(kvm, gpa, data, len); |
49c7754c GN |
1970 | |
1971 | if (kvm_is_error_hva(ghc->hva)) | |
1972 | return -EFAULT; | |
1973 | ||
4ec6e863 | 1974 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
1975 | if (r) |
1976 | return -EFAULT; | |
4ec6e863 | 1977 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
1978 | |
1979 | return 0; | |
1980 | } | |
4e335d9e | 1981 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 1982 | |
4e335d9e PB |
1983 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1984 | void *data, unsigned long len) | |
4ec6e863 | 1985 | { |
4e335d9e | 1986 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 1987 | } |
4e335d9e | 1988 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 1989 | |
4e335d9e PB |
1990 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
1991 | void *data, unsigned long len) | |
e03b644f | 1992 | { |
4e335d9e | 1993 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f GN |
1994 | int r; |
1995 | ||
8f964525 AH |
1996 | BUG_ON(len > ghc->len); |
1997 | ||
e03b644f | 1998 | if (slots->generation != ghc->generation) |
5a2d4365 | 1999 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
2000 | |
2001 | if (unlikely(!ghc->memslot)) | |
4e335d9e | 2002 | return kvm_read_guest(kvm, ghc->gpa, data, len); |
e03b644f GN |
2003 | |
2004 | if (kvm_is_error_hva(ghc->hva)) | |
2005 | return -EFAULT; | |
2006 | ||
2007 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
2008 | if (r) | |
2009 | return -EFAULT; | |
2010 | ||
2011 | return 0; | |
2012 | } | |
4e335d9e | 2013 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 2014 | |
195aefde IE |
2015 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
2016 | { | |
8a3caa6d HC |
2017 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
2018 | ||
2019 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
2020 | } |
2021 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
2022 | ||
2023 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
2024 | { | |
2025 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2026 | int seg; | |
2027 | int offset = offset_in_page(gpa); | |
2028 | int ret; | |
2029 | ||
bfda0e84 | 2030 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
2031 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
2032 | if (ret < 0) | |
2033 | return ret; | |
2034 | offset = 0; | |
2035 | len -= seg; | |
2036 | ++gfn; | |
2037 | } | |
2038 | return 0; | |
2039 | } | |
2040 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2041 | ||
bc009e43 | 2042 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 2043 | gfn_t gfn) |
6aa8b732 | 2044 | { |
7e9d619d RR |
2045 | if (memslot && memslot->dirty_bitmap) { |
2046 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 2047 | |
b74ca3b3 | 2048 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
2049 | } |
2050 | } | |
2051 | ||
49c7754c GN |
2052 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2053 | { | |
2054 | struct kvm_memory_slot *memslot; | |
2055 | ||
2056 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 2057 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 2058 | } |
2ba9f0d8 | 2059 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2060 | |
8e73485c PB |
2061 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2062 | { | |
2063 | struct kvm_memory_slot *memslot; | |
2064 | ||
2065 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2066 | mark_page_dirty_in_slot(memslot, gfn); | |
2067 | } | |
2068 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2069 | ||
20b7035c JS |
2070 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
2071 | { | |
2072 | if (!vcpu->sigset_active) | |
2073 | return; | |
2074 | ||
2075 | /* | |
2076 | * This does a lockless modification of ->real_blocked, which is fine | |
2077 | * because, only current can change ->real_blocked and all readers of | |
2078 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
2079 | * of ->blocked. | |
2080 | */ | |
2081 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
2082 | } | |
2083 | ||
2084 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
2085 | { | |
2086 | if (!vcpu->sigset_active) | |
2087 | return; | |
2088 | ||
2089 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
2090 | sigemptyset(¤t->real_blocked); | |
2091 | } | |
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 | |
178f02ff | 2210 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2211 | { |
8577370f | 2212 | struct swait_queue_head *wqp; |
b6d33834 CD |
2213 | |
2214 | wqp = kvm_arch_vcpu_wq(vcpu); | |
5e0018b3 | 2215 | if (swq_has_sleeper(wqp)) { |
8577370f | 2216 | swake_up(wqp); |
b6d33834 | 2217 | ++vcpu->stat.halt_wakeup; |
178f02ff | 2218 | return true; |
b6d33834 CD |
2219 | } |
2220 | ||
178f02ff | 2221 | return false; |
dd1a4cc1 RK |
2222 | } |
2223 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2224 | ||
0266c894 | 2225 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
2226 | /* |
2227 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2228 | */ | |
2229 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2230 | { | |
2231 | int me; | |
2232 | int cpu = vcpu->cpu; | |
2233 | ||
178f02ff RK |
2234 | if (kvm_vcpu_wake_up(vcpu)) |
2235 | return; | |
2236 | ||
b6d33834 CD |
2237 | me = get_cpu(); |
2238 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2239 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2240 | smp_send_reschedule(cpu); | |
2241 | put_cpu(); | |
2242 | } | |
a20ed54d | 2243 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 2244 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2245 | |
fa93384f | 2246 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2247 | { |
2248 | struct pid *pid; | |
2249 | struct task_struct *task = NULL; | |
fa93384f | 2250 | int ret = 0; |
41628d33 KW |
2251 | |
2252 | rcu_read_lock(); | |
2253 | pid = rcu_dereference(target->pid); | |
2254 | if (pid) | |
27fbe64b | 2255 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2256 | rcu_read_unlock(); |
2257 | if (!task) | |
c45c528e | 2258 | return ret; |
c45c528e | 2259 | ret = yield_to(task, 1); |
41628d33 | 2260 | put_task_struct(task); |
c45c528e R |
2261 | |
2262 | return ret; | |
41628d33 KW |
2263 | } |
2264 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2265 | ||
06e48c51 R |
2266 | /* |
2267 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2268 | * Most eligible candidate to yield is decided by following heuristics: | |
2269 | * | |
2270 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2271 | * (preempted lock holder), indicated by @in_spin_loop. | |
2272 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2273 | * | |
2274 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2275 | * chance last time (mostly it has become eligible now since we have probably | |
2276 | * yielded to lockholder in last iteration. This is done by toggling | |
2277 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2278 | * | |
2279 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2280 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2281 | * burning. Giving priority for a potential lock-holder increases lock | |
2282 | * progress. | |
2283 | * | |
2284 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2285 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2286 | * and continue with next VCPU and so on. | |
2287 | */ | |
7940876e | 2288 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2289 | { |
4a55dd72 | 2290 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2291 | bool eligible; |
2292 | ||
2293 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2294 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2295 | |
2296 | if (vcpu->spin_loop.in_spin_loop) | |
2297 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2298 | ||
2299 | return eligible; | |
4a55dd72 SW |
2300 | #else |
2301 | return true; | |
06e48c51 | 2302 | #endif |
4a55dd72 | 2303 | } |
c45c528e | 2304 | |
199b5763 | 2305 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 2306 | { |
217ece61 RR |
2307 | struct kvm *kvm = me->kvm; |
2308 | struct kvm_vcpu *vcpu; | |
2309 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2310 | int yielded = 0; | |
c45c528e | 2311 | int try = 3; |
217ece61 RR |
2312 | int pass; |
2313 | int i; | |
d255f4f2 | 2314 | |
4c088493 | 2315 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2316 | /* |
2317 | * We boost the priority of a VCPU that is runnable but not | |
2318 | * currently running, because it got preempted by something | |
2319 | * else and called schedule in __vcpu_run. Hopefully that | |
2320 | * VCPU is holding the lock that we need and will release it. | |
2321 | * We approximate round-robin by starting at the last boosted VCPU. | |
2322 | */ | |
c45c528e | 2323 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2324 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2325 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2326 | i = last_boosted_vcpu; |
2327 | continue; | |
2328 | } else if (pass && i > last_boosted_vcpu) | |
2329 | break; | |
6aa7de05 | 2330 | if (!READ_ONCE(vcpu->preempted)) |
7bc7ae25 | 2331 | continue; |
217ece61 RR |
2332 | if (vcpu == me) |
2333 | continue; | |
8577370f | 2334 | if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu)) |
217ece61 | 2335 | continue; |
199b5763 LM |
2336 | if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu)) |
2337 | continue; | |
06e48c51 R |
2338 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2339 | continue; | |
c45c528e R |
2340 | |
2341 | yielded = kvm_vcpu_yield_to(vcpu); | |
2342 | if (yielded > 0) { | |
217ece61 | 2343 | kvm->last_boosted_vcpu = i; |
217ece61 | 2344 | break; |
c45c528e R |
2345 | } else if (yielded < 0) { |
2346 | try--; | |
2347 | if (!try) | |
2348 | break; | |
217ece61 | 2349 | } |
217ece61 RR |
2350 | } |
2351 | } | |
4c088493 | 2352 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2353 | |
2354 | /* Ensure vcpu is not eligible during next spinloop */ | |
2355 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2356 | } |
2357 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2358 | ||
11bac800 | 2359 | static int kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 2360 | { |
11bac800 | 2361 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
2362 | struct page *page; |
2363 | ||
e4a533a4 | 2364 | if (vmf->pgoff == 0) |
039576c0 | 2365 | page = virt_to_page(vcpu->run); |
09566765 | 2366 | #ifdef CONFIG_X86 |
e4a533a4 | 2367 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2368 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 2369 | #endif |
4b4357e0 | 2370 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
2371 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
2372 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2373 | #endif |
039576c0 | 2374 | else |
5b1c1493 | 2375 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2376 | get_page(page); |
e4a533a4 | 2377 | vmf->page = page; |
2378 | return 0; | |
9a2bb7f4 AK |
2379 | } |
2380 | ||
f0f37e2f | 2381 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2382 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2383 | }; |
2384 | ||
2385 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2386 | { | |
2387 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2388 | return 0; | |
2389 | } | |
2390 | ||
bccf2150 AK |
2391 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2392 | { | |
2393 | struct kvm_vcpu *vcpu = filp->private_data; | |
2394 | ||
45b5939e | 2395 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
66c0b394 | 2396 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2397 | return 0; |
2398 | } | |
2399 | ||
3d3aab1b | 2400 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2401 | .release = kvm_vcpu_release, |
2402 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
de8e5d74 | 2403 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2404 | .compat_ioctl = kvm_vcpu_compat_ioctl, |
2405 | #endif | |
9a2bb7f4 | 2406 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2407 | .llseek = noop_llseek, |
bccf2150 AK |
2408 | }; |
2409 | ||
2410 | /* | |
2411 | * Allocates an inode for the vcpu. | |
2412 | */ | |
2413 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2414 | { | |
24009b05 | 2415 | return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
bccf2150 AK |
2416 | } |
2417 | ||
45b5939e LC |
2418 | static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
2419 | { | |
2420 | char dir_name[ITOA_MAX_LEN * 2]; | |
2421 | int ret; | |
2422 | ||
2423 | if (!kvm_arch_has_vcpu_debugfs()) | |
2424 | return 0; | |
2425 | ||
2426 | if (!debugfs_initialized()) | |
2427 | return 0; | |
2428 | ||
2429 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
2430 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, | |
2431 | vcpu->kvm->debugfs_dentry); | |
2432 | if (!vcpu->debugfs_dentry) | |
2433 | return -ENOMEM; | |
2434 | ||
2435 | ret = kvm_arch_create_vcpu_debugfs(vcpu); | |
2436 | if (ret < 0) { | |
2437 | debugfs_remove_recursive(vcpu->debugfs_dentry); | |
2438 | return ret; | |
2439 | } | |
2440 | ||
2441 | return 0; | |
2442 | } | |
2443 | ||
c5ea7660 AK |
2444 | /* |
2445 | * Creates some virtual cpus. Good luck creating more than one. | |
2446 | */ | |
73880c80 | 2447 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2448 | { |
2449 | int r; | |
e09fefde | 2450 | struct kvm_vcpu *vcpu; |
c5ea7660 | 2451 | |
0b1b1dfd | 2452 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
2453 | return -EINVAL; |
2454 | ||
6c7caebc PB |
2455 | mutex_lock(&kvm->lock); |
2456 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
2457 | mutex_unlock(&kvm->lock); | |
2458 | return -EINVAL; | |
2459 | } | |
2460 | ||
2461 | kvm->created_vcpus++; | |
2462 | mutex_unlock(&kvm->lock); | |
2463 | ||
73880c80 | 2464 | vcpu = kvm_arch_vcpu_create(kvm, id); |
6c7caebc PB |
2465 | if (IS_ERR(vcpu)) { |
2466 | r = PTR_ERR(vcpu); | |
2467 | goto vcpu_decrement; | |
2468 | } | |
c5ea7660 | 2469 | |
15ad7146 AK |
2470 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2471 | ||
26e5215f AK |
2472 | r = kvm_arch_vcpu_setup(vcpu); |
2473 | if (r) | |
d780592b | 2474 | goto vcpu_destroy; |
26e5215f | 2475 | |
45b5939e LC |
2476 | r = kvm_create_vcpu_debugfs(vcpu); |
2477 | if (r) | |
2478 | goto vcpu_destroy; | |
2479 | ||
11ec2804 | 2480 | mutex_lock(&kvm->lock); |
e09fefde DH |
2481 | if (kvm_get_vcpu_by_id(kvm, id)) { |
2482 | r = -EEXIST; | |
2483 | goto unlock_vcpu_destroy; | |
2484 | } | |
73880c80 GN |
2485 | |
2486 | BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]); | |
c5ea7660 | 2487 | |
fb3f0f51 | 2488 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2489 | kvm_get_kvm(kvm); |
bccf2150 | 2490 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2491 | if (r < 0) { |
2492 | kvm_put_kvm(kvm); | |
d780592b | 2493 | goto unlock_vcpu_destroy; |
73880c80 GN |
2494 | } |
2495 | ||
2496 | kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu; | |
dd489240 PB |
2497 | |
2498 | /* | |
2499 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2500 | * before kvm->online_vcpu's incremented value. | |
2501 | */ | |
73880c80 GN |
2502 | smp_wmb(); |
2503 | atomic_inc(&kvm->online_vcpus); | |
2504 | ||
73880c80 | 2505 | mutex_unlock(&kvm->lock); |
42897d86 | 2506 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2507 | return r; |
39c3b86e | 2508 | |
d780592b | 2509 | unlock_vcpu_destroy: |
7d8fece6 | 2510 | mutex_unlock(&kvm->lock); |
45b5939e | 2511 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
d780592b | 2512 | vcpu_destroy: |
d40ccc62 | 2513 | kvm_arch_vcpu_destroy(vcpu); |
6c7caebc PB |
2514 | vcpu_decrement: |
2515 | mutex_lock(&kvm->lock); | |
2516 | kvm->created_vcpus--; | |
2517 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
2518 | return r; |
2519 | } | |
2520 | ||
1961d276 AK |
2521 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2522 | { | |
2523 | if (sigset) { | |
2524 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2525 | vcpu->sigset_active = 1; | |
2526 | vcpu->sigset = *sigset; | |
2527 | } else | |
2528 | vcpu->sigset_active = 0; | |
2529 | return 0; | |
2530 | } | |
2531 | ||
bccf2150 AK |
2532 | static long kvm_vcpu_ioctl(struct file *filp, |
2533 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2534 | { |
bccf2150 | 2535 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2536 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2537 | int r; |
fa3795a7 DH |
2538 | struct kvm_fpu *fpu = NULL; |
2539 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2540 | |
6d4e4c4f AK |
2541 | if (vcpu->kvm->mm != current->mm) |
2542 | return -EIO; | |
2122ff5e | 2543 | |
2ea75be3 DM |
2544 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2545 | return -EINVAL; | |
2546 | ||
2f4d9b54 | 2547 | #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS) |
2122ff5e AK |
2548 | /* |
2549 | * Special cases: vcpu ioctls that are asynchronous to vcpu execution, | |
2550 | * so vcpu_load() would break it. | |
2551 | */ | |
47b43c52 | 2552 | if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT) |
2122ff5e AK |
2553 | return kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
2554 | #endif | |
2555 | ||
2556 | ||
ec7660cc CD |
2557 | if (mutex_lock_killable(&vcpu->mutex)) |
2558 | return -EINTR; | |
6aa8b732 | 2559 | switch (ioctl) { |
0e4524a5 CB |
2560 | case KVM_RUN: { |
2561 | struct pid *oldpid; | |
f0fe5108 AK |
2562 | r = -EINVAL; |
2563 | if (arg) | |
2564 | goto out; | |
0e4524a5 CB |
2565 | oldpid = rcu_access_pointer(vcpu->pid); |
2566 | if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) { | |
7a72f7a1 | 2567 | /* The thread running this VCPU changed. */ |
7a72f7a1 | 2568 | struct pid *newpid = get_task_pid(current, PIDTYPE_PID); |
f95ef0cd | 2569 | |
7a72f7a1 CB |
2570 | rcu_assign_pointer(vcpu->pid, newpid); |
2571 | if (oldpid) | |
2572 | synchronize_rcu(); | |
2573 | put_pid(oldpid); | |
2574 | } | |
b6c7a5dc | 2575 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2576 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2577 | break; |
0e4524a5 | 2578 | } |
6aa8b732 | 2579 | case KVM_GET_REGS: { |
3e4bb3ac | 2580 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2581 | |
3e4bb3ac XZ |
2582 | r = -ENOMEM; |
2583 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL); | |
2584 | if (!kvm_regs) | |
6aa8b732 | 2585 | goto out; |
3e4bb3ac XZ |
2586 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2587 | if (r) | |
2588 | goto out_free1; | |
6aa8b732 | 2589 | r = -EFAULT; |
3e4bb3ac XZ |
2590 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2591 | goto out_free1; | |
6aa8b732 | 2592 | r = 0; |
3e4bb3ac XZ |
2593 | out_free1: |
2594 | kfree(kvm_regs); | |
6aa8b732 AK |
2595 | break; |
2596 | } | |
2597 | case KVM_SET_REGS: { | |
3e4bb3ac | 2598 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2599 | |
3e4bb3ac | 2600 | r = -ENOMEM; |
ff5c2c03 SL |
2601 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2602 | if (IS_ERR(kvm_regs)) { | |
2603 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2604 | goto out; |
ff5c2c03 | 2605 | } |
3e4bb3ac | 2606 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2607 | kfree(kvm_regs); |
6aa8b732 AK |
2608 | break; |
2609 | } | |
2610 | case KVM_GET_SREGS: { | |
fa3795a7 DH |
2611 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL); |
2612 | r = -ENOMEM; | |
2613 | if (!kvm_sregs) | |
2614 | goto out; | |
2615 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2616 | if (r) |
2617 | goto out; | |
2618 | r = -EFAULT; | |
fa3795a7 | 2619 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2620 | goto out; |
2621 | r = 0; | |
2622 | break; | |
2623 | } | |
2624 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2625 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2626 | if (IS_ERR(kvm_sregs)) { | |
2627 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2628 | kvm_sregs = NULL; |
6aa8b732 | 2629 | goto out; |
ff5c2c03 | 2630 | } |
fa3795a7 | 2631 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2632 | break; |
2633 | } | |
62d9f0db MT |
2634 | case KVM_GET_MP_STATE: { |
2635 | struct kvm_mp_state mp_state; | |
2636 | ||
2637 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2638 | if (r) | |
2639 | goto out; | |
2640 | r = -EFAULT; | |
893bdbf1 | 2641 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2642 | goto out; |
2643 | r = 0; | |
2644 | break; | |
2645 | } | |
2646 | case KVM_SET_MP_STATE: { | |
2647 | struct kvm_mp_state mp_state; | |
2648 | ||
2649 | r = -EFAULT; | |
893bdbf1 | 2650 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2651 | goto out; |
2652 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2653 | break; |
2654 | } | |
6aa8b732 AK |
2655 | case KVM_TRANSLATE: { |
2656 | struct kvm_translation tr; | |
2657 | ||
2658 | r = -EFAULT; | |
893bdbf1 | 2659 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2660 | goto out; |
8b006791 | 2661 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2662 | if (r) |
2663 | goto out; | |
2664 | r = -EFAULT; | |
893bdbf1 | 2665 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2666 | goto out; |
2667 | r = 0; | |
2668 | break; | |
2669 | } | |
d0bfb940 JK |
2670 | case KVM_SET_GUEST_DEBUG: { |
2671 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2672 | |
2673 | r = -EFAULT; | |
893bdbf1 | 2674 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2675 | goto out; |
d0bfb940 | 2676 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2677 | break; |
2678 | } | |
1961d276 AK |
2679 | case KVM_SET_SIGNAL_MASK: { |
2680 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2681 | struct kvm_signal_mask kvm_sigmask; | |
2682 | sigset_t sigset, *p; | |
2683 | ||
2684 | p = NULL; | |
2685 | if (argp) { | |
2686 | r = -EFAULT; | |
2687 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2688 | sizeof(kvm_sigmask))) |
1961d276 AK |
2689 | goto out; |
2690 | r = -EINVAL; | |
893bdbf1 | 2691 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2692 | goto out; |
2693 | r = -EFAULT; | |
2694 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2695 | sizeof(sigset))) |
1961d276 AK |
2696 | goto out; |
2697 | p = &sigset; | |
2698 | } | |
376d41ff | 2699 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2700 | break; |
2701 | } | |
b8836737 | 2702 | case KVM_GET_FPU: { |
fa3795a7 DH |
2703 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL); |
2704 | r = -ENOMEM; | |
2705 | if (!fpu) | |
2706 | goto out; | |
2707 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2708 | if (r) |
2709 | goto out; | |
2710 | r = -EFAULT; | |
fa3795a7 | 2711 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2712 | goto out; |
2713 | r = 0; | |
2714 | break; | |
2715 | } | |
2716 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2717 | fpu = memdup_user(argp, sizeof(*fpu)); |
2718 | if (IS_ERR(fpu)) { | |
2719 | r = PTR_ERR(fpu); | |
18595411 | 2720 | fpu = NULL; |
b8836737 | 2721 | goto out; |
ff5c2c03 | 2722 | } |
fa3795a7 | 2723 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2724 | break; |
2725 | } | |
bccf2150 | 2726 | default: |
313a3dc7 | 2727 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2728 | } |
2729 | out: | |
ec7660cc | 2730 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
2731 | kfree(fpu); |
2732 | kfree(kvm_sregs); | |
bccf2150 AK |
2733 | return r; |
2734 | } | |
2735 | ||
de8e5d74 | 2736 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2737 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2738 | unsigned int ioctl, unsigned long arg) | |
2739 | { | |
2740 | struct kvm_vcpu *vcpu = filp->private_data; | |
2741 | void __user *argp = compat_ptr(arg); | |
2742 | int r; | |
2743 | ||
2744 | if (vcpu->kvm->mm != current->mm) | |
2745 | return -EIO; | |
2746 | ||
2747 | switch (ioctl) { | |
2748 | case KVM_SET_SIGNAL_MASK: { | |
2749 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2750 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
2751 | sigset_t sigset; |
2752 | ||
2753 | if (argp) { | |
2754 | r = -EFAULT; | |
2755 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2756 | sizeof(kvm_sigmask))) |
1dda606c AG |
2757 | goto out; |
2758 | r = -EINVAL; | |
3968cf62 | 2759 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
2760 | goto out; |
2761 | r = -EFAULT; | |
3968cf62 | 2762 | if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset)) |
1dda606c | 2763 | goto out; |
760a9a30 AC |
2764 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
2765 | } else | |
2766 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
2767 | break; |
2768 | } | |
2769 | default: | |
2770 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
2771 | } | |
2772 | ||
2773 | out: | |
2774 | return r; | |
2775 | } | |
2776 | #endif | |
2777 | ||
852b6d57 SW |
2778 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
2779 | int (*accessor)(struct kvm_device *dev, | |
2780 | struct kvm_device_attr *attr), | |
2781 | unsigned long arg) | |
2782 | { | |
2783 | struct kvm_device_attr attr; | |
2784 | ||
2785 | if (!accessor) | |
2786 | return -EPERM; | |
2787 | ||
2788 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
2789 | return -EFAULT; | |
2790 | ||
2791 | return accessor(dev, &attr); | |
2792 | } | |
2793 | ||
2794 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
2795 | unsigned long arg) | |
2796 | { | |
2797 | struct kvm_device *dev = filp->private_data; | |
2798 | ||
2799 | switch (ioctl) { | |
2800 | case KVM_SET_DEVICE_ATTR: | |
2801 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
2802 | case KVM_GET_DEVICE_ATTR: | |
2803 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
2804 | case KVM_HAS_DEVICE_ATTR: | |
2805 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
2806 | default: | |
2807 | if (dev->ops->ioctl) | |
2808 | return dev->ops->ioctl(dev, ioctl, arg); | |
2809 | ||
2810 | return -ENOTTY; | |
2811 | } | |
2812 | } | |
2813 | ||
852b6d57 SW |
2814 | static int kvm_device_release(struct inode *inode, struct file *filp) |
2815 | { | |
2816 | struct kvm_device *dev = filp->private_data; | |
2817 | struct kvm *kvm = dev->kvm; | |
2818 | ||
852b6d57 SW |
2819 | kvm_put_kvm(kvm); |
2820 | return 0; | |
2821 | } | |
2822 | ||
2823 | static const struct file_operations kvm_device_fops = { | |
2824 | .unlocked_ioctl = kvm_device_ioctl, | |
de8e5d74 | 2825 | #ifdef CONFIG_KVM_COMPAT |
db6ae615 SW |
2826 | .compat_ioctl = kvm_device_ioctl, |
2827 | #endif | |
852b6d57 SW |
2828 | .release = kvm_device_release, |
2829 | }; | |
2830 | ||
2831 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
2832 | { | |
2833 | if (filp->f_op != &kvm_device_fops) | |
2834 | return NULL; | |
2835 | ||
2836 | return filp->private_data; | |
2837 | } | |
2838 | ||
d60eacb0 | 2839 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 2840 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
2841 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
2842 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 2843 | #endif |
d60eacb0 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; | |
4b4357e0 | 2933 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
2934 | case KVM_CAP_COALESCED_MMIO: |
2935 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
2936 | #endif | |
92b591a4 AG |
2937 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
2938 | case KVM_CAP_IRQ_ROUTING: | |
2939 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
2940 | #endif |
2941 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
2942 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
2943 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 2944 | #endif |
0b1b1dfd GK |
2945 | case KVM_CAP_MAX_VCPU_ID: |
2946 | return KVM_MAX_VCPU_ID; | |
92b591a4 AG |
2947 | default: |
2948 | break; | |
2949 | } | |
2950 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
2951 | } | |
2952 | ||
bccf2150 AK |
2953 | static long kvm_vm_ioctl(struct file *filp, |
2954 | unsigned int ioctl, unsigned long arg) | |
2955 | { | |
2956 | struct kvm *kvm = filp->private_data; | |
2957 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 2958 | int r; |
bccf2150 | 2959 | |
6d4e4c4f AK |
2960 | if (kvm->mm != current->mm) |
2961 | return -EIO; | |
bccf2150 AK |
2962 | switch (ioctl) { |
2963 | case KVM_CREATE_VCPU: | |
2964 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 2965 | break; |
6fc138d2 IE |
2966 | case KVM_SET_USER_MEMORY_REGION: { |
2967 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2968 | ||
2969 | r = -EFAULT; | |
2970 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 2971 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
2972 | goto out; |
2973 | ||
47ae31e2 | 2974 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
2975 | break; |
2976 | } | |
2977 | case KVM_GET_DIRTY_LOG: { | |
2978 | struct kvm_dirty_log log; | |
2979 | ||
2980 | r = -EFAULT; | |
893bdbf1 | 2981 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 2982 | goto out; |
2c6f5df9 | 2983 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
2984 | break; |
2985 | } | |
4b4357e0 | 2986 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
2987 | case KVM_REGISTER_COALESCED_MMIO: { |
2988 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2989 | |
5f94c174 | 2990 | r = -EFAULT; |
893bdbf1 | 2991 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 2992 | goto out; |
5f94c174 | 2993 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
2994 | break; |
2995 | } | |
2996 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
2997 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 2998 | |
5f94c174 | 2999 | r = -EFAULT; |
893bdbf1 | 3000 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3001 | goto out; |
5f94c174 | 3002 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3003 | break; |
3004 | } | |
3005 | #endif | |
721eecbf GH |
3006 | case KVM_IRQFD: { |
3007 | struct kvm_irqfd data; | |
3008 | ||
3009 | r = -EFAULT; | |
893bdbf1 | 3010 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3011 | goto out; |
d4db2935 | 3012 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3013 | break; |
3014 | } | |
d34e6b17 GH |
3015 | case KVM_IOEVENTFD: { |
3016 | struct kvm_ioeventfd data; | |
3017 | ||
3018 | r = -EFAULT; | |
893bdbf1 | 3019 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3020 | goto out; |
3021 | r = kvm_ioeventfd(kvm, &data); | |
3022 | break; | |
3023 | } | |
07975ad3 JK |
3024 | #ifdef CONFIG_HAVE_KVM_MSI |
3025 | case KVM_SIGNAL_MSI: { | |
3026 | struct kvm_msi msi; | |
3027 | ||
3028 | r = -EFAULT; | |
893bdbf1 | 3029 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3030 | goto out; |
3031 | r = kvm_send_userspace_msi(kvm, &msi); | |
3032 | break; | |
3033 | } | |
23d43cf9 CD |
3034 | #endif |
3035 | #ifdef __KVM_HAVE_IRQ_LINE | |
3036 | case KVM_IRQ_LINE_STATUS: | |
3037 | case KVM_IRQ_LINE: { | |
3038 | struct kvm_irq_level irq_event; | |
3039 | ||
3040 | r = -EFAULT; | |
893bdbf1 | 3041 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3042 | goto out; |
3043 | ||
aa2fbe6d YZ |
3044 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3045 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3046 | if (r) |
3047 | goto out; | |
3048 | ||
3049 | r = -EFAULT; | |
3050 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3051 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3052 | goto out; |
3053 | } | |
3054 | ||
3055 | r = 0; | |
3056 | break; | |
3057 | } | |
73880c80 | 3058 | #endif |
aa8d5944 AG |
3059 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3060 | case KVM_SET_GSI_ROUTING: { | |
3061 | struct kvm_irq_routing routing; | |
3062 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3063 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3064 | |
3065 | r = -EFAULT; | |
3066 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3067 | goto out; | |
3068 | r = -EINVAL; | |
5c0aea0e DH |
3069 | if (!kvm_arch_can_set_irq_routing(kvm)) |
3070 | goto out; | |
caf1ff26 | 3071 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3072 | goto out; |
3073 | if (routing.flags) | |
3074 | goto out; | |
f8c1b85b PB |
3075 | if (routing.nr) { |
3076 | r = -ENOMEM; | |
3077 | entries = vmalloc(routing.nr * sizeof(*entries)); | |
3078 | if (!entries) | |
3079 | goto out; | |
3080 | r = -EFAULT; | |
3081 | urouting = argp; | |
3082 | if (copy_from_user(entries, urouting->entries, | |
3083 | routing.nr * sizeof(*entries))) | |
3084 | goto out_free_irq_routing; | |
3085 | } | |
aa8d5944 AG |
3086 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3087 | routing.flags); | |
a642a175 | 3088 | out_free_irq_routing: |
aa8d5944 AG |
3089 | vfree(entries); |
3090 | break; | |
3091 | } | |
3092 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3093 | case KVM_CREATE_DEVICE: { |
3094 | struct kvm_create_device cd; | |
3095 | ||
3096 | r = -EFAULT; | |
3097 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3098 | goto out; | |
3099 | ||
3100 | r = kvm_ioctl_create_device(kvm, &cd); | |
3101 | if (r) | |
3102 | goto out; | |
3103 | ||
3104 | r = -EFAULT; | |
3105 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3106 | goto out; | |
3107 | ||
3108 | r = 0; | |
3109 | break; | |
3110 | } | |
92b591a4 AG |
3111 | case KVM_CHECK_EXTENSION: |
3112 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3113 | break; | |
f17abe9a | 3114 | default: |
1fe779f8 | 3115 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3116 | } |
3117 | out: | |
3118 | return r; | |
3119 | } | |
3120 | ||
de8e5d74 | 3121 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3122 | struct compat_kvm_dirty_log { |
3123 | __u32 slot; | |
3124 | __u32 padding1; | |
3125 | union { | |
3126 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3127 | __u64 padding2; | |
3128 | }; | |
3129 | }; | |
3130 | ||
3131 | static long kvm_vm_compat_ioctl(struct file *filp, | |
3132 | unsigned int ioctl, unsigned long arg) | |
3133 | { | |
3134 | struct kvm *kvm = filp->private_data; | |
3135 | int r; | |
3136 | ||
3137 | if (kvm->mm != current->mm) | |
3138 | return -EIO; | |
3139 | switch (ioctl) { | |
3140 | case KVM_GET_DIRTY_LOG: { | |
3141 | struct compat_kvm_dirty_log compat_log; | |
3142 | struct kvm_dirty_log log; | |
3143 | ||
6ff5894c AB |
3144 | if (copy_from_user(&compat_log, (void __user *)arg, |
3145 | sizeof(compat_log))) | |
f6a3b168 | 3146 | return -EFAULT; |
6ff5894c AB |
3147 | log.slot = compat_log.slot; |
3148 | log.padding1 = compat_log.padding1; | |
3149 | log.padding2 = compat_log.padding2; | |
3150 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3151 | ||
3152 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
3153 | break; |
3154 | } | |
3155 | default: | |
3156 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
3157 | } | |
6ff5894c AB |
3158 | return r; |
3159 | } | |
3160 | #endif | |
3161 | ||
3d3aab1b | 3162 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
3163 | .release = kvm_vm_release, |
3164 | .unlocked_ioctl = kvm_vm_ioctl, | |
de8e5d74 | 3165 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3166 | .compat_ioctl = kvm_vm_compat_ioctl, |
3167 | #endif | |
6038f373 | 3168 | .llseek = noop_llseek, |
f17abe9a AK |
3169 | }; |
3170 | ||
e08b9637 | 3171 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 3172 | { |
aac87636 | 3173 | int r; |
f17abe9a | 3174 | struct kvm *kvm; |
506cfba9 | 3175 | struct file *file; |
f17abe9a | 3176 | |
e08b9637 | 3177 | kvm = kvm_create_vm(type); |
d6d28168 AK |
3178 | if (IS_ERR(kvm)) |
3179 | return PTR_ERR(kvm); | |
4b4357e0 | 3180 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 3181 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
3182 | if (r < 0) |
3183 | goto put_kvm; | |
6ce5a090 | 3184 | #endif |
506cfba9 | 3185 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
3186 | if (r < 0) |
3187 | goto put_kvm; | |
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); | |
78588335 ME |
3192 | r = PTR_ERR(file); |
3193 | goto put_kvm; | |
506cfba9 | 3194 | } |
536a6f88 | 3195 | |
525df861 PB |
3196 | /* |
3197 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
3198 | * already set, with ->release() being kvm_vm_release(). In error | |
3199 | * cases it will be called by the final fput(file) and will take | |
3200 | * care of doing kvm_put_kvm(kvm). | |
3201 | */ | |
536a6f88 | 3202 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
3203 | put_unused_fd(r); |
3204 | fput(file); | |
536a6f88 JF |
3205 | return -ENOMEM; |
3206 | } | |
286de8f6 | 3207 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 3208 | |
506cfba9 | 3209 | fd_install(r, file); |
aac87636 | 3210 | return r; |
78588335 ME |
3211 | |
3212 | put_kvm: | |
3213 | kvm_put_kvm(kvm); | |
3214 | return r; | |
f17abe9a AK |
3215 | } |
3216 | ||
3217 | static long kvm_dev_ioctl(struct file *filp, | |
3218 | unsigned int ioctl, unsigned long arg) | |
3219 | { | |
07c45a36 | 3220 | long r = -EINVAL; |
f17abe9a AK |
3221 | |
3222 | switch (ioctl) { | |
3223 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
3224 | if (arg) |
3225 | goto out; | |
f17abe9a AK |
3226 | r = KVM_API_VERSION; |
3227 | break; | |
3228 | case KVM_CREATE_VM: | |
e08b9637 | 3229 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 3230 | break; |
018d00d2 | 3231 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 3232 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 3233 | break; |
07c45a36 | 3234 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
3235 | if (arg) |
3236 | goto out; | |
adb1ff46 AK |
3237 | r = PAGE_SIZE; /* struct kvm_run */ |
3238 | #ifdef CONFIG_X86 | |
3239 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 3240 | #endif |
4b4357e0 | 3241 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 3242 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 3243 | #endif |
07c45a36 | 3244 | break; |
d4c9ff2d FEL |
3245 | case KVM_TRACE_ENABLE: |
3246 | case KVM_TRACE_PAUSE: | |
3247 | case KVM_TRACE_DISABLE: | |
2023a29c | 3248 | r = -EOPNOTSUPP; |
d4c9ff2d | 3249 | break; |
6aa8b732 | 3250 | default: |
043405e1 | 3251 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
3252 | } |
3253 | out: | |
3254 | return r; | |
3255 | } | |
3256 | ||
6aa8b732 | 3257 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 AK |
3258 | .unlocked_ioctl = kvm_dev_ioctl, |
3259 | .compat_ioctl = kvm_dev_ioctl, | |
6038f373 | 3260 | .llseek = noop_llseek, |
6aa8b732 AK |
3261 | }; |
3262 | ||
3263 | static struct miscdevice kvm_dev = { | |
bbe4432e | 3264 | KVM_MINOR, |
6aa8b732 AK |
3265 | "kvm", |
3266 | &kvm_chardev_ops, | |
3267 | }; | |
3268 | ||
75b7127c | 3269 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
3270 | { |
3271 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 3272 | int r; |
1b6c0168 | 3273 | |
7f59f492 | 3274 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3275 | return; |
10474ae8 | 3276 | |
7f59f492 | 3277 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 3278 | |
13a34e06 | 3279 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
3280 | |
3281 | if (r) { | |
3282 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
3283 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 3284 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 3285 | } |
1b6c0168 AK |
3286 | } |
3287 | ||
8c18b2d2 | 3288 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 3289 | { |
4a937f96 | 3290 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3291 | if (kvm_usage_count) |
3292 | hardware_enable_nolock(NULL); | |
4a937f96 | 3293 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3294 | return 0; |
75b7127c TY |
3295 | } |
3296 | ||
3297 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
3298 | { |
3299 | int cpu = raw_smp_processor_id(); | |
3300 | ||
7f59f492 | 3301 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3302 | return; |
7f59f492 | 3303 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 3304 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
3305 | } |
3306 | ||
8c18b2d2 | 3307 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 3308 | { |
4a937f96 | 3309 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3310 | if (kvm_usage_count) |
3311 | hardware_disable_nolock(NULL); | |
4a937f96 | 3312 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3313 | return 0; |
75b7127c TY |
3314 | } |
3315 | ||
10474ae8 AG |
3316 | static void hardware_disable_all_nolock(void) |
3317 | { | |
3318 | BUG_ON(!kvm_usage_count); | |
3319 | ||
3320 | kvm_usage_count--; | |
3321 | if (!kvm_usage_count) | |
75b7127c | 3322 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
3323 | } |
3324 | ||
3325 | static void hardware_disable_all(void) | |
3326 | { | |
4a937f96 | 3327 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3328 | hardware_disable_all_nolock(); |
4a937f96 | 3329 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3330 | } |
3331 | ||
3332 | static int hardware_enable_all(void) | |
3333 | { | |
3334 | int r = 0; | |
3335 | ||
4a937f96 | 3336 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3337 | |
3338 | kvm_usage_count++; | |
3339 | if (kvm_usage_count == 1) { | |
3340 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3341 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3342 | |
3343 | if (atomic_read(&hardware_enable_failed)) { | |
3344 | hardware_disable_all_nolock(); | |
3345 | r = -EBUSY; | |
3346 | } | |
3347 | } | |
3348 | ||
4a937f96 | 3349 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3350 | |
3351 | return r; | |
3352 | } | |
3353 | ||
9a2b85c6 | 3354 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3355 | void *v) |
9a2b85c6 | 3356 | { |
8e1c1815 SY |
3357 | /* |
3358 | * Some (well, at least mine) BIOSes hang on reboot if | |
3359 | * in vmx root mode. | |
3360 | * | |
3361 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3362 | */ | |
1170adc6 | 3363 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3364 | kvm_rebooting = true; |
75b7127c | 3365 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3366 | return NOTIFY_OK; |
3367 | } | |
3368 | ||
3369 | static struct notifier_block kvm_reboot_notifier = { | |
3370 | .notifier_call = kvm_reboot, | |
3371 | .priority = 0, | |
3372 | }; | |
3373 | ||
e93f8a0f | 3374 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3375 | { |
3376 | int i; | |
3377 | ||
3378 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3379 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3380 | |
3381 | kvm_iodevice_destructor(pos); | |
3382 | } | |
e93f8a0f | 3383 | kfree(bus); |
2eeb2e94 GH |
3384 | } |
3385 | ||
c21fbff1 | 3386 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3387 | const struct kvm_io_range *r2) |
743eeb0b | 3388 | { |
8f4216c7 JW |
3389 | gpa_t addr1 = r1->addr; |
3390 | gpa_t addr2 = r2->addr; | |
3391 | ||
3392 | if (addr1 < addr2) | |
743eeb0b | 3393 | return -1; |
8f4216c7 JW |
3394 | |
3395 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
3396 | * accept any overlapping write. Any order is acceptable for | |
3397 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
3398 | * we process all of them. | |
3399 | */ | |
3400 | if (r2->len) { | |
3401 | addr1 += r1->len; | |
3402 | addr2 += r2->len; | |
3403 | } | |
3404 | ||
3405 | if (addr1 > addr2) | |
743eeb0b | 3406 | return 1; |
8f4216c7 | 3407 | |
743eeb0b SL |
3408 | return 0; |
3409 | } | |
3410 | ||
a343c9b7 PB |
3411 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3412 | { | |
c21fbff1 | 3413 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3414 | } |
3415 | ||
39369f7a | 3416 | static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev, |
743eeb0b SL |
3417 | gpa_t addr, int len) |
3418 | { | |
743eeb0b SL |
3419 | bus->range[bus->dev_count++] = (struct kvm_io_range) { |
3420 | .addr = addr, | |
3421 | .len = len, | |
3422 | .dev = dev, | |
3423 | }; | |
3424 | ||
3425 | sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range), | |
3426 | kvm_io_bus_sort_cmp, NULL); | |
3427 | ||
3428 | return 0; | |
3429 | } | |
3430 | ||
39369f7a | 3431 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3432 | gpa_t addr, int len) |
3433 | { | |
3434 | struct kvm_io_range *range, key; | |
3435 | int off; | |
3436 | ||
3437 | key = (struct kvm_io_range) { | |
3438 | .addr = addr, | |
3439 | .len = len, | |
3440 | }; | |
3441 | ||
3442 | range = bsearch(&key, bus->range, bus->dev_count, | |
3443 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3444 | if (range == NULL) | |
3445 | return -ENOENT; | |
3446 | ||
3447 | off = range - bus->range; | |
3448 | ||
c21fbff1 | 3449 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3450 | off--; |
3451 | ||
3452 | return off; | |
3453 | } | |
3454 | ||
e32edf4f | 3455 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3456 | struct kvm_io_range *range, const void *val) |
3457 | { | |
3458 | int idx; | |
3459 | ||
3460 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3461 | if (idx < 0) | |
3462 | return -EOPNOTSUPP; | |
3463 | ||
3464 | while (idx < bus->dev_count && | |
c21fbff1 | 3465 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3466 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3467 | range->len, val)) |
3468 | return idx; | |
3469 | idx++; | |
3470 | } | |
3471 | ||
3472 | return -EOPNOTSUPP; | |
3473 | } | |
3474 | ||
bda9020e | 3475 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3476 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3477 | int len, const void *val) |
2eeb2e94 | 3478 | { |
90d83dc3 | 3479 | struct kvm_io_bus *bus; |
743eeb0b | 3480 | struct kvm_io_range range; |
126a5af5 | 3481 | int r; |
743eeb0b SL |
3482 | |
3483 | range = (struct kvm_io_range) { | |
3484 | .addr = addr, | |
3485 | .len = len, | |
3486 | }; | |
90d83dc3 | 3487 | |
e32edf4f | 3488 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3489 | if (!bus) |
3490 | return -ENOMEM; | |
e32edf4f | 3491 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3492 | return r < 0 ? r : 0; |
3493 | } | |
3494 | ||
3495 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3496 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3497 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3498 | { |
3499 | struct kvm_io_bus *bus; | |
3500 | struct kvm_io_range range; | |
3501 | ||
3502 | range = (struct kvm_io_range) { | |
3503 | .addr = addr, | |
3504 | .len = len, | |
3505 | }; | |
3506 | ||
e32edf4f | 3507 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3508 | if (!bus) |
3509 | return -ENOMEM; | |
126a5af5 CH |
3510 | |
3511 | /* First try the device referenced by cookie. */ | |
3512 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3513 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3514 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3515 | val)) |
3516 | return cookie; | |
3517 | ||
3518 | /* | |
3519 | * cookie contained garbage; fall back to search and return the | |
3520 | * correct cookie value. | |
3521 | */ | |
e32edf4f | 3522 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3523 | } |
3524 | ||
e32edf4f NN |
3525 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3526 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3527 | { |
3528 | int idx; | |
3529 | ||
3530 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3531 | if (idx < 0) |
3532 | return -EOPNOTSUPP; | |
3533 | ||
3534 | while (idx < bus->dev_count && | |
c21fbff1 | 3535 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3536 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3537 | range->len, val)) |
3538 | return idx; | |
743eeb0b SL |
3539 | idx++; |
3540 | } | |
3541 | ||
bda9020e MT |
3542 | return -EOPNOTSUPP; |
3543 | } | |
68c3b4d1 | 3544 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
2eeb2e94 | 3545 | |
bda9020e | 3546 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3547 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3548 | int len, void *val) |
bda9020e | 3549 | { |
90d83dc3 | 3550 | struct kvm_io_bus *bus; |
743eeb0b | 3551 | struct kvm_io_range range; |
126a5af5 | 3552 | int r; |
743eeb0b SL |
3553 | |
3554 | range = (struct kvm_io_range) { | |
3555 | .addr = addr, | |
3556 | .len = len, | |
3557 | }; | |
e93f8a0f | 3558 | |
e32edf4f | 3559 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3560 | if (!bus) |
3561 | return -ENOMEM; | |
e32edf4f | 3562 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
3563 | return r < 0 ? r : 0; |
3564 | } | |
743eeb0b | 3565 | |
2eeb2e94 | 3566 | |
79fac95e | 3567 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3568 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3569 | int len, struct kvm_io_device *dev) | |
6c474694 | 3570 | { |
e93f8a0f | 3571 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3572 | |
4a12f951 | 3573 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
3574 | if (!bus) |
3575 | return -ENOMEM; | |
3576 | ||
6ea34c9b AK |
3577 | /* exclude ioeventfd which is limited by maximum fd */ |
3578 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3579 | return -ENOSPC; |
2eeb2e94 | 3580 | |
d3febddd | 3581 | new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) * |
a1300716 | 3582 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
e93f8a0f MT |
3583 | if (!new_bus) |
3584 | return -ENOMEM; | |
a1300716 AK |
3585 | memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count * |
3586 | sizeof(struct kvm_io_range))); | |
743eeb0b | 3587 | kvm_io_bus_insert_dev(new_bus, dev, addr, len); |
e93f8a0f MT |
3588 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3589 | synchronize_srcu_expedited(&kvm->srcu); | |
3590 | kfree(bus); | |
090b7aff GH |
3591 | |
3592 | return 0; | |
3593 | } | |
3594 | ||
79fac95e | 3595 | /* Caller must hold slots_lock. */ |
90db1043 DH |
3596 | void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3597 | struct kvm_io_device *dev) | |
090b7aff | 3598 | { |
90db1043 | 3599 | int i; |
e93f8a0f | 3600 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3601 | |
4a12f951 | 3602 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 3603 | if (!bus) |
90db1043 | 3604 | return; |
df630b8c | 3605 | |
a1300716 AK |
3606 | for (i = 0; i < bus->dev_count; i++) |
3607 | if (bus->range[i].dev == dev) { | |
090b7aff GH |
3608 | break; |
3609 | } | |
e93f8a0f | 3610 | |
90db1043 DH |
3611 | if (i == bus->dev_count) |
3612 | return; | |
a1300716 | 3613 | |
d3febddd | 3614 | new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) * |
a1300716 | 3615 | sizeof(struct kvm_io_range)), GFP_KERNEL); |
90db1043 DH |
3616 | if (!new_bus) { |
3617 | pr_err("kvm: failed to shrink bus, removing it completely\n"); | |
3618 | goto broken; | |
3619 | } | |
a1300716 AK |
3620 | |
3621 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3622 | new_bus->dev_count--; | |
3623 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3624 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f | 3625 | |
90db1043 | 3626 | broken: |
e93f8a0f MT |
3627 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3628 | synchronize_srcu_expedited(&kvm->srcu); | |
3629 | kfree(bus); | |
90db1043 | 3630 | return; |
2eeb2e94 GH |
3631 | } |
3632 | ||
8a39d006 AP |
3633 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3634 | gpa_t addr) | |
3635 | { | |
3636 | struct kvm_io_bus *bus; | |
3637 | int dev_idx, srcu_idx; | |
3638 | struct kvm_io_device *iodev = NULL; | |
3639 | ||
3640 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
3641 | ||
3642 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
3643 | if (!bus) |
3644 | goto out_unlock; | |
8a39d006 AP |
3645 | |
3646 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
3647 | if (dev_idx < 0) | |
3648 | goto out_unlock; | |
3649 | ||
3650 | iodev = bus->range[dev_idx].dev; | |
3651 | ||
3652 | out_unlock: | |
3653 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
3654 | ||
3655 | return iodev; | |
3656 | } | |
3657 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
3658 | ||
536a6f88 JF |
3659 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
3660 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
3661 | const char *fmt) | |
3662 | { | |
3663 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3664 | inode->i_private; | |
3665 | ||
3666 | /* The debugfs files are a reference to the kvm struct which | |
3667 | * is still valid when kvm_destroy_vm is called. | |
3668 | * To avoid the race between open and the removal of the debugfs | |
3669 | * directory we test against the users count. | |
3670 | */ | |
e3736c3e | 3671 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
3672 | return -ENOENT; |
3673 | ||
3674 | if (simple_attr_open(inode, file, get, set, fmt)) { | |
3675 | kvm_put_kvm(stat_data->kvm); | |
3676 | return -ENOMEM; | |
3677 | } | |
3678 | ||
3679 | return 0; | |
3680 | } | |
3681 | ||
3682 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
3683 | { | |
3684 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3685 | inode->i_private; | |
3686 | ||
3687 | simple_attr_release(inode, file); | |
3688 | kvm_put_kvm(stat_data->kvm); | |
3689 | ||
3690 | return 0; | |
3691 | } | |
3692 | ||
3693 | static int vm_stat_get_per_vm(void *data, u64 *val) | |
3694 | { | |
3695 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3696 | ||
8a7e75d4 | 3697 | *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset); |
536a6f88 JF |
3698 | |
3699 | return 0; | |
3700 | } | |
3701 | ||
ce35ef27 SJS |
3702 | static int vm_stat_clear_per_vm(void *data, u64 val) |
3703 | { | |
3704 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3705 | ||
3706 | if (val) | |
3707 | return -EINVAL; | |
3708 | ||
3709 | *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0; | |
3710 | ||
3711 | return 0; | |
3712 | } | |
3713 | ||
536a6f88 JF |
3714 | static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file) |
3715 | { | |
3716 | __simple_attr_check_format("%llu\n", 0ull); | |
3717 | return kvm_debugfs_open(inode, file, vm_stat_get_per_vm, | |
ce35ef27 | 3718 | vm_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
3719 | } |
3720 | ||
3721 | static const struct file_operations vm_stat_get_per_vm_fops = { | |
3722 | .owner = THIS_MODULE, | |
3723 | .open = vm_stat_get_per_vm_open, | |
3724 | .release = kvm_debugfs_release, | |
3725 | .read = simple_attr_read, | |
3726 | .write = simple_attr_write, | |
3bed8888 | 3727 | .llseek = no_llseek, |
536a6f88 JF |
3728 | }; |
3729 | ||
3730 | static int vcpu_stat_get_per_vm(void *data, u64 *val) | |
3731 | { | |
3732 | int i; | |
3733 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3734 | struct kvm_vcpu *vcpu; | |
3735 | ||
3736 | *val = 0; | |
3737 | ||
3738 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
8a7e75d4 | 3739 | *val += *(u64 *)((void *)vcpu + stat_data->offset); |
536a6f88 JF |
3740 | |
3741 | return 0; | |
3742 | } | |
3743 | ||
ce35ef27 SJS |
3744 | static int vcpu_stat_clear_per_vm(void *data, u64 val) |
3745 | { | |
3746 | int i; | |
3747 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
3748 | struct kvm_vcpu *vcpu; | |
3749 | ||
3750 | if (val) | |
3751 | return -EINVAL; | |
3752 | ||
3753 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
3754 | *(u64 *)((void *)vcpu + stat_data->offset) = 0; | |
3755 | ||
3756 | return 0; | |
3757 | } | |
3758 | ||
536a6f88 JF |
3759 | static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file) |
3760 | { | |
3761 | __simple_attr_check_format("%llu\n", 0ull); | |
3762 | return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm, | |
ce35ef27 | 3763 | vcpu_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
3764 | } |
3765 | ||
3766 | static const struct file_operations vcpu_stat_get_per_vm_fops = { | |
3767 | .owner = THIS_MODULE, | |
3768 | .open = vcpu_stat_get_per_vm_open, | |
3769 | .release = kvm_debugfs_release, | |
3770 | .read = simple_attr_read, | |
3771 | .write = simple_attr_write, | |
3bed8888 | 3772 | .llseek = no_llseek, |
536a6f88 JF |
3773 | }; |
3774 | ||
3775 | static const struct file_operations *stat_fops_per_vm[] = { | |
3776 | [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops, | |
3777 | [KVM_STAT_VM] = &vm_stat_get_per_vm_fops, | |
3778 | }; | |
3779 | ||
8b88b099 | 3780 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
3781 | { |
3782 | unsigned offset = (long)_offset; | |
ba1389b7 | 3783 | struct kvm *kvm; |
536a6f88 JF |
3784 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
3785 | u64 tmp_val; | |
ba1389b7 | 3786 | |
8b88b099 | 3787 | *val = 0; |
2f303b74 | 3788 | spin_lock(&kvm_lock); |
536a6f88 JF |
3789 | list_for_each_entry(kvm, &vm_list, vm_list) { |
3790 | stat_tmp.kvm = kvm; | |
3791 | vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
3792 | *val += tmp_val; | |
3793 | } | |
2f303b74 | 3794 | spin_unlock(&kvm_lock); |
8b88b099 | 3795 | return 0; |
ba1389b7 AK |
3796 | } |
3797 | ||
ce35ef27 SJS |
3798 | static int vm_stat_clear(void *_offset, u64 val) |
3799 | { | |
3800 | unsigned offset = (long)_offset; | |
3801 | struct kvm *kvm; | |
3802 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
3803 | ||
3804 | if (val) | |
3805 | return -EINVAL; | |
3806 | ||
3807 | spin_lock(&kvm_lock); | |
3808 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3809 | stat_tmp.kvm = kvm; | |
3810 | vm_stat_clear_per_vm((void *)&stat_tmp, 0); | |
3811 | } | |
3812 | spin_unlock(&kvm_lock); | |
3813 | ||
3814 | return 0; | |
3815 | } | |
3816 | ||
3817 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 3818 | |
8b88b099 | 3819 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
3820 | { |
3821 | unsigned offset = (long)_offset; | |
1165f5fe | 3822 | struct kvm *kvm; |
536a6f88 JF |
3823 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
3824 | u64 tmp_val; | |
1165f5fe | 3825 | |
8b88b099 | 3826 | *val = 0; |
2f303b74 | 3827 | spin_lock(&kvm_lock); |
536a6f88 JF |
3828 | list_for_each_entry(kvm, &vm_list, vm_list) { |
3829 | stat_tmp.kvm = kvm; | |
3830 | vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
3831 | *val += tmp_val; | |
3832 | } | |
2f303b74 | 3833 | spin_unlock(&kvm_lock); |
8b88b099 | 3834 | return 0; |
1165f5fe AK |
3835 | } |
3836 | ||
ce35ef27 SJS |
3837 | static int vcpu_stat_clear(void *_offset, u64 val) |
3838 | { | |
3839 | unsigned offset = (long)_offset; | |
3840 | struct kvm *kvm; | |
3841 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
3842 | ||
3843 | if (val) | |
3844 | return -EINVAL; | |
3845 | ||
3846 | spin_lock(&kvm_lock); | |
3847 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3848 | stat_tmp.kvm = kvm; | |
3849 | vcpu_stat_clear_per_vm((void *)&stat_tmp, 0); | |
3850 | } | |
3851 | spin_unlock(&kvm_lock); | |
3852 | ||
3853 | return 0; | |
3854 | } | |
3855 | ||
3856 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
3857 | "%llu\n"); | |
ba1389b7 | 3858 | |
828c0950 | 3859 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
3860 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
3861 | [KVM_STAT_VM] = &vm_stat_fops, | |
3862 | }; | |
1165f5fe | 3863 | |
286de8f6 CI |
3864 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
3865 | { | |
3866 | struct kobj_uevent_env *env; | |
286de8f6 CI |
3867 | unsigned long long created, active; |
3868 | ||
3869 | if (!kvm_dev.this_device || !kvm) | |
3870 | return; | |
3871 | ||
3872 | spin_lock(&kvm_lock); | |
3873 | if (type == KVM_EVENT_CREATE_VM) { | |
3874 | kvm_createvm_count++; | |
3875 | kvm_active_vms++; | |
3876 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
3877 | kvm_active_vms--; | |
3878 | } | |
3879 | created = kvm_createvm_count; | |
3880 | active = kvm_active_vms; | |
3881 | spin_unlock(&kvm_lock); | |
3882 | ||
3883 | env = kzalloc(sizeof(*env), GFP_KERNEL); | |
3884 | if (!env) | |
3885 | return; | |
3886 | ||
3887 | add_uevent_var(env, "CREATED=%llu", created); | |
3888 | add_uevent_var(env, "COUNT=%llu", active); | |
3889 | ||
fdeaf7e3 | 3890 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 3891 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
3892 | kvm->userspace_pid = task_pid_nr(current); |
3893 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 3894 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
3895 | } |
3896 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 CI |
3897 | |
3898 | if (kvm->debugfs_dentry) { | |
fdeaf7e3 CI |
3899 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL); |
3900 | ||
3901 | if (p) { | |
3902 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
3903 | if (!IS_ERR(tmp)) | |
3904 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
3905 | kfree(p); | |
286de8f6 CI |
3906 | } |
3907 | } | |
3908 | /* no need for checks, since we are adding at most only 5 keys */ | |
3909 | env->envp[env->envp_idx++] = NULL; | |
3910 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
3911 | kfree(env); | |
286de8f6 CI |
3912 | } |
3913 | ||
4f69b680 | 3914 | static int kvm_init_debug(void) |
6aa8b732 | 3915 | { |
0c8eb04a | 3916 | int r = -EEXIST; |
6aa8b732 AK |
3917 | struct kvm_stats_debugfs_item *p; |
3918 | ||
76f7c879 | 3919 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 H |
3920 | if (kvm_debugfs_dir == NULL) |
3921 | goto out; | |
3922 | ||
536a6f88 JF |
3923 | kvm_debugfs_num_entries = 0; |
3924 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
ce35ef27 | 3925 | if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir, |
4bd33b56 JF |
3926 | (void *)(long)p->offset, |
3927 | stat_fops[p->kind])) | |
4f69b680 H |
3928 | goto out_dir; |
3929 | } | |
3930 | ||
3931 | return 0; | |
3932 | ||
3933 | out_dir: | |
3934 | debugfs_remove_recursive(kvm_debugfs_dir); | |
3935 | out: | |
3936 | return r; | |
6aa8b732 AK |
3937 | } |
3938 | ||
fb3600cc | 3939 | static int kvm_suspend(void) |
59ae6c6b | 3940 | { |
10474ae8 | 3941 | if (kvm_usage_count) |
75b7127c | 3942 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
3943 | return 0; |
3944 | } | |
3945 | ||
fb3600cc | 3946 | static void kvm_resume(void) |
59ae6c6b | 3947 | { |
ca84d1a2 | 3948 | if (kvm_usage_count) { |
4a937f96 | 3949 | WARN_ON(raw_spin_is_locked(&kvm_count_lock)); |
75b7127c | 3950 | hardware_enable_nolock(NULL); |
ca84d1a2 | 3951 | } |
59ae6c6b AK |
3952 | } |
3953 | ||
fb3600cc | 3954 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
3955 | .suspend = kvm_suspend, |
3956 | .resume = kvm_resume, | |
3957 | }; | |
3958 | ||
15ad7146 AK |
3959 | static inline |
3960 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
3961 | { | |
3962 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
3963 | } | |
3964 | ||
3965 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
3966 | { | |
3967 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 3968 | |
3a08a8f9 R |
3969 | if (vcpu->preempted) |
3970 | vcpu->preempted = false; | |
15ad7146 | 3971 | |
e790d9ef RK |
3972 | kvm_arch_sched_in(vcpu, cpu); |
3973 | ||
e9b11c17 | 3974 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
3975 | } |
3976 | ||
3977 | static void kvm_sched_out(struct preempt_notifier *pn, | |
3978 | struct task_struct *next) | |
3979 | { | |
3980 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
3981 | ||
3a08a8f9 R |
3982 | if (current->state == TASK_RUNNING) |
3983 | vcpu->preempted = true; | |
e9b11c17 | 3984 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
3985 | } |
3986 | ||
0ee75bea | 3987 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 3988 | struct module *module) |
6aa8b732 AK |
3989 | { |
3990 | int r; | |
002c7f7c | 3991 | int cpu; |
6aa8b732 | 3992 | |
f8c16bba ZX |
3993 | r = kvm_arch_init(opaque); |
3994 | if (r) | |
d2308784 | 3995 | goto out_fail; |
cb498ea2 | 3996 | |
7dac16c3 AH |
3997 | /* |
3998 | * kvm_arch_init makes sure there's at most one caller | |
3999 | * for architectures that support multiple implementations, | |
4000 | * like intel and amd on x86. | |
36343f6e PB |
4001 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
4002 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 4003 | */ |
36343f6e PB |
4004 | r = kvm_irqfd_init(); |
4005 | if (r) | |
4006 | goto out_irqfd; | |
7dac16c3 | 4007 | |
8437a617 | 4008 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
4009 | r = -ENOMEM; |
4010 | goto out_free_0; | |
4011 | } | |
4012 | ||
e9b11c17 | 4013 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 4014 | if (r < 0) |
7f59f492 | 4015 | goto out_free_0a; |
6aa8b732 | 4016 | |
002c7f7c YS |
4017 | for_each_online_cpu(cpu) { |
4018 | smp_call_function_single(cpu, | |
e9b11c17 | 4019 | kvm_arch_check_processor_compat, |
8691e5a8 | 4020 | &r, 1); |
002c7f7c | 4021 | if (r < 0) |
d2308784 | 4022 | goto out_free_1; |
002c7f7c YS |
4023 | } |
4024 | ||
73c1b41e | 4025 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 4026 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 4027 | if (r) |
d2308784 | 4028 | goto out_free_2; |
6aa8b732 AK |
4029 | register_reboot_notifier(&kvm_reboot_notifier); |
4030 | ||
c16f862d | 4031 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
4032 | if (!vcpu_align) |
4033 | vcpu_align = __alignof__(struct kvm_vcpu); | |
4034 | kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align, | |
46bea48a | 4035 | SLAB_ACCOUNT, NULL); |
c16f862d RR |
4036 | if (!kvm_vcpu_cache) { |
4037 | r = -ENOMEM; | |
fb3600cc | 4038 | goto out_free_3; |
c16f862d RR |
4039 | } |
4040 | ||
af585b92 GN |
4041 | r = kvm_async_pf_init(); |
4042 | if (r) | |
4043 | goto out_free; | |
4044 | ||
6aa8b732 | 4045 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
4046 | kvm_vm_fops.owner = module; |
4047 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
4048 | |
4049 | r = misc_register(&kvm_dev); | |
4050 | if (r) { | |
1170adc6 | 4051 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 4052 | goto out_unreg; |
6aa8b732 AK |
4053 | } |
4054 | ||
fb3600cc RW |
4055 | register_syscore_ops(&kvm_syscore_ops); |
4056 | ||
15ad7146 AK |
4057 | kvm_preempt_ops.sched_in = kvm_sched_in; |
4058 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
4059 | ||
4f69b680 H |
4060 | r = kvm_init_debug(); |
4061 | if (r) { | |
1170adc6 | 4062 | pr_err("kvm: create debugfs files failed\n"); |
4f69b680 H |
4063 | goto out_undebugfs; |
4064 | } | |
0ea4ed8e | 4065 | |
3c3c29fd PB |
4066 | r = kvm_vfio_ops_init(); |
4067 | WARN_ON(r); | |
4068 | ||
c7addb90 | 4069 | return 0; |
6aa8b732 | 4070 | |
4f69b680 H |
4071 | out_undebugfs: |
4072 | unregister_syscore_ops(&kvm_syscore_ops); | |
afc2f792 | 4073 | misc_deregister(&kvm_dev); |
af585b92 GN |
4074 | out_unreg: |
4075 | kvm_async_pf_deinit(); | |
6aa8b732 | 4076 | out_free: |
c16f862d | 4077 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4078 | out_free_3: |
6aa8b732 | 4079 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4080 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4081 | out_free_2: |
d2308784 | 4082 | out_free_1: |
e9b11c17 | 4083 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
4084 | out_free_0a: |
4085 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 4086 | out_free_0: |
a0f155e9 | 4087 | kvm_irqfd_exit(); |
36343f6e | 4088 | out_irqfd: |
7dac16c3 AH |
4089 | kvm_arch_exit(); |
4090 | out_fail: | |
6aa8b732 AK |
4091 | return r; |
4092 | } | |
cb498ea2 | 4093 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 4094 | |
cb498ea2 | 4095 | void kvm_exit(void) |
6aa8b732 | 4096 | { |
4bd33b56 | 4097 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 4098 | misc_deregister(&kvm_dev); |
c16f862d | 4099 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 4100 | kvm_async_pf_deinit(); |
fb3600cc | 4101 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 4102 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4103 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 4104 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 4105 | kvm_arch_hardware_unsetup(); |
f8c16bba | 4106 | kvm_arch_exit(); |
a0f155e9 | 4107 | kvm_irqfd_exit(); |
7f59f492 | 4108 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 4109 | kvm_vfio_ops_exit(); |
6aa8b732 | 4110 | } |
cb498ea2 | 4111 | EXPORT_SYMBOL_GPL(kvm_exit); |