]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
6aa8b732 | 54 | |
e495606d | 55 | #include <asm/processor.h> |
2ea75be3 | 56 | #include <asm/ioctl.h> |
7c0f6ba6 | 57 | #include <linux/uaccess.h> |
3e021bf5 | 58 | #include <asm/pgtable.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
3c3c29fd | 62 | #include "vfio.h" |
5f94c174 | 63 | |
229456fc MT |
64 | #define CREATE_TRACE_POINTS |
65 | #include <trace/events/kvm.h> | |
66 | ||
536a6f88 JF |
67 | /* Worst case buffer size needed for holding an integer. */ |
68 | #define ITOA_MAX_LEN 12 | |
69 | ||
6aa8b732 AK |
70 | MODULE_AUTHOR("Qumranet"); |
71 | MODULE_LICENSE("GPL"); | |
72 | ||
920552b2 | 73 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 74 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 75 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 76 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 77 | |
aca6ff29 | 78 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 79 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 80 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 81 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 82 | |
49113d36 NW |
83 | /* The start value to grow halt_poll_ns from */ |
84 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
85 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
86 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
87 | ||
aca6ff29 | 88 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 89 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 90 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 91 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 92 | |
fa40a821 MT |
93 | /* |
94 | * Ordering of locks: | |
95 | * | |
b7d409de | 96 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
97 | */ |
98 | ||
0d9ce162 | 99 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 100 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 101 | LIST_HEAD(vm_list); |
133de902 | 102 | |
7f59f492 | 103 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 104 | static int kvm_usage_count; |
10474ae8 | 105 | static atomic_t hardware_enable_failed; |
1b6c0168 | 106 | |
aaba298c | 107 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 108 | |
15ad7146 AK |
109 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
110 | ||
76f7c879 | 111 | struct dentry *kvm_debugfs_dir; |
e23a808b | 112 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 113 | |
536a6f88 | 114 | static int kvm_debugfs_num_entries; |
09cbcef6 | 115 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 116 | |
bccf2150 AK |
117 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
118 | unsigned long arg); | |
de8e5d74 | 119 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
120 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
121 | unsigned long arg); | |
7ddfd3e0 MZ |
122 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
123 | #else | |
9cb09e7c MZ |
124 | /* |
125 | * For architectures that don't implement a compat infrastructure, | |
126 | * adopt a double line of defense: | |
127 | * - Prevent a compat task from opening /dev/kvm | |
128 | * - If the open has been done by a 64bit task, and the KVM fd | |
129 | * passed to a compat task, let the ioctls fail. | |
130 | */ | |
7ddfd3e0 MZ |
131 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
132 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
133 | |
134 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
135 | { | |
136 | return is_compat_task() ? -ENODEV : 0; | |
137 | } | |
138 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
139 | .open = kvm_no_compat_open | |
1dda606c | 140 | #endif |
10474ae8 AG |
141 | static int hardware_enable_all(void); |
142 | static void hardware_disable_all(void); | |
bccf2150 | 143 | |
e93f8a0f | 144 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 145 | |
bc009e43 | 146 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 147 | |
52480137 | 148 | __visible bool kvm_rebooting; |
b7c4145b | 149 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 150 | |
54dee993 MT |
151 | static bool largepages_enabled = true; |
152 | ||
286de8f6 CI |
153 | #define KVM_EVENT_CREATE_VM 0 |
154 | #define KVM_EVENT_DESTROY_VM 1 | |
155 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
156 | static unsigned long long kvm_createvm_count; | |
157 | static unsigned long long kvm_active_vms; | |
158 | ||
93065ac7 MH |
159 | __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
160 | unsigned long start, unsigned long end, bool blockable) | |
b1394e74 | 161 | { |
93065ac7 | 162 | return 0; |
b1394e74 RK |
163 | } |
164 | ||
a78986aa SC |
165 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
166 | { | |
167 | /* | |
168 | * The metadata used by is_zone_device_page() to determine whether or | |
169 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
170 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
171 | * page_count() is zero to help detect bad usage of this helper. | |
172 | */ | |
173 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
174 | return false; | |
175 | ||
176 | return is_zone_device_page(pfn_to_page(pfn)); | |
177 | } | |
178 | ||
ba049e93 | 179 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 180 | { |
a78986aa SC |
181 | /* |
182 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
183 | * perspective they are "normal" pages, albeit with slightly different | |
184 | * usage rules. | |
185 | */ | |
11feeb49 | 186 | if (pfn_valid(pfn)) |
a78986aa SC |
187 | return PageReserved(pfn_to_page(pfn)) && |
188 | !kvm_is_zone_device_pfn(pfn); | |
cbff90a7 BAY |
189 | |
190 | return true; | |
191 | } | |
192 | ||
bccf2150 AK |
193 | /* |
194 | * Switches to specified vcpu, until a matching vcpu_put() | |
195 | */ | |
ec7660cc | 196 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 197 | { |
ec7660cc | 198 | int cpu = get_cpu(); |
15ad7146 | 199 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 200 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 201 | put_cpu(); |
6aa8b732 | 202 | } |
2f1fe811 | 203 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 204 | |
313a3dc7 | 205 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 206 | { |
15ad7146 | 207 | preempt_disable(); |
313a3dc7 | 208 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
209 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
210 | preempt_enable(); | |
6aa8b732 | 211 | } |
2f1fe811 | 212 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 213 | |
7a97cec2 PB |
214 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
215 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
216 | { | |
217 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
218 | ||
219 | /* | |
220 | * We need to wait for the VCPU to reenable interrupts and get out of | |
221 | * READING_SHADOW_PAGE_TABLES mode. | |
222 | */ | |
223 | if (req & KVM_REQUEST_WAIT) | |
224 | return mode != OUTSIDE_GUEST_MODE; | |
225 | ||
226 | /* | |
227 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
228 | */ | |
229 | return mode == IN_GUEST_MODE; | |
230 | } | |
231 | ||
d9e368d6 AK |
232 | static void ack_flush(void *_completed) |
233 | { | |
d9e368d6 AK |
234 | } |
235 | ||
b49defe8 PB |
236 | static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait) |
237 | { | |
238 | if (unlikely(!cpus)) | |
239 | cpus = cpu_online_mask; | |
240 | ||
241 | if (cpumask_empty(cpus)) | |
242 | return false; | |
243 | ||
244 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
245 | return true; | |
246 | } | |
247 | ||
7053df4e VK |
248 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
249 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) | |
d9e368d6 | 250 | { |
597a5f55 | 251 | int i, cpu, me; |
d9e368d6 | 252 | struct kvm_vcpu *vcpu; |
7053df4e | 253 | bool called; |
6ef7a1bc | 254 | |
3cba4130 | 255 | me = get_cpu(); |
7053df4e | 256 | |
988a2cae | 257 | kvm_for_each_vcpu(i, vcpu, kvm) { |
a812297c | 258 | if (vcpu_bitmap && !test_bit(i, vcpu_bitmap)) |
7053df4e VK |
259 | continue; |
260 | ||
3cba4130 | 261 | kvm_make_request(req, vcpu); |
d9e368d6 | 262 | cpu = vcpu->cpu; |
6b7e2d09 | 263 | |
178f02ff RK |
264 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
265 | continue; | |
6c6e8360 | 266 | |
7053df4e | 267 | if (tmp != NULL && cpu != -1 && cpu != me && |
7a97cec2 | 268 | kvm_request_needs_ipi(vcpu, req)) |
7053df4e | 269 | __cpumask_set_cpu(cpu, tmp); |
49846896 | 270 | } |
7053df4e VK |
271 | |
272 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); | |
3cba4130 | 273 | put_cpu(); |
7053df4e VK |
274 | |
275 | return called; | |
276 | } | |
277 | ||
278 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) | |
279 | { | |
280 | cpumask_var_t cpus; | |
281 | bool called; | |
7053df4e VK |
282 | |
283 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
284 | ||
a812297c | 285 | called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus); |
7053df4e | 286 | |
6ef7a1bc | 287 | free_cpumask_var(cpus); |
49846896 | 288 | return called; |
d9e368d6 AK |
289 | } |
290 | ||
a6d51016 | 291 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 292 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 293 | { |
4ae3cb3a LT |
294 | /* |
295 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
296 | * kvm_make_all_cpus_request. | |
297 | */ | |
298 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
299 | ||
300 | /* | |
301 | * We want to publish modifications to the page tables before reading | |
302 | * mode. Pairs with a memory barrier in arch-specific code. | |
303 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
304 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
305 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
306 | * | |
307 | * There is already an smp_mb__after_atomic() before | |
308 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
309 | * barrier here. | |
310 | */ | |
b08660e5 TL |
311 | if (!kvm_arch_flush_remote_tlb(kvm) |
312 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
49846896 | 313 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 314 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 315 | } |
2ba9f0d8 | 316 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 317 | #endif |
2e53d63a | 318 | |
49846896 RR |
319 | void kvm_reload_remote_mmus(struct kvm *kvm) |
320 | { | |
445b8236 | 321 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 322 | } |
2e53d63a | 323 | |
8bd826d6 | 324 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 325 | { |
fb3f0f51 RR |
326 | mutex_init(&vcpu->mutex); |
327 | vcpu->cpu = -1; | |
fb3f0f51 RR |
328 | vcpu->kvm = kvm; |
329 | vcpu->vcpu_id = id; | |
34bb10b7 | 330 | vcpu->pid = NULL; |
8577370f | 331 | init_swait_queue_head(&vcpu->wq); |
af585b92 | 332 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 333 | |
bf9f6ac8 FW |
334 | vcpu->pre_pcpu = -1; |
335 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
336 | ||
4c088493 R |
337 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
338 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 339 | vcpu->preempted = false; |
d73eb57b | 340 | vcpu->ready = false; |
d5c48deb | 341 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fb3f0f51 | 342 | } |
fb3f0f51 | 343 | |
4543bdc0 SC |
344 | void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
345 | { | |
346 | kvm_arch_vcpu_destroy(vcpu); | |
e529ef66 | 347 | |
9941d224 SC |
348 | /* |
349 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
350 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
351 | * are already gone. | |
352 | */ | |
353 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
354 | ||
8bd826d6 | 355 | free_page((unsigned long)vcpu->run); |
e529ef66 | 356 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 SC |
357 | } |
358 | EXPORT_SYMBOL_GPL(kvm_vcpu_destroy); | |
359 | ||
e930bffe AA |
360 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
361 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
362 | { | |
363 | return container_of(mn, struct kvm, mmu_notifier); | |
364 | } | |
365 | ||
3da0dd43 IE |
366 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
367 | struct mm_struct *mm, | |
368 | unsigned long address, | |
369 | pte_t pte) | |
370 | { | |
371 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 372 | int idx; |
3da0dd43 | 373 | |
bc6678a3 | 374 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
375 | spin_lock(&kvm->mmu_lock); |
376 | kvm->mmu_notifier_seq++; | |
0cf853c5 LT |
377 | |
378 | if (kvm_set_spte_hva(kvm, address, pte)) | |
379 | kvm_flush_remote_tlbs(kvm); | |
380 | ||
3da0dd43 | 381 | spin_unlock(&kvm->mmu_lock); |
bc6678a3 | 382 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
383 | } |
384 | ||
93065ac7 | 385 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
5d6527a7 | 386 | const struct mmu_notifier_range *range) |
e930bffe AA |
387 | { |
388 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 389 | int need_tlb_flush = 0, idx; |
93065ac7 | 390 | int ret; |
e930bffe | 391 | |
bc6678a3 | 392 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
393 | spin_lock(&kvm->mmu_lock); |
394 | /* | |
395 | * The count increase must become visible at unlock time as no | |
396 | * spte can be established without taking the mmu_lock and | |
397 | * count is also read inside the mmu_lock critical section. | |
398 | */ | |
399 | kvm->mmu_notifier_count++; | |
5d6527a7 | 400 | need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end); |
a4ee1ca4 | 401 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
402 | /* we've to flush the tlb before the pages can be freed */ |
403 | if (need_tlb_flush) | |
404 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
405 | |
406 | spin_unlock(&kvm->mmu_lock); | |
b1394e74 | 407 | |
5d6527a7 | 408 | ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start, |
dfcd6660 JG |
409 | range->end, |
410 | mmu_notifier_range_blockable(range)); | |
b1394e74 | 411 | |
565f3be2 | 412 | srcu_read_unlock(&kvm->srcu, idx); |
93065ac7 MH |
413 | |
414 | return ret; | |
e930bffe AA |
415 | } |
416 | ||
417 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
5d6527a7 | 418 | const struct mmu_notifier_range *range) |
e930bffe AA |
419 | { |
420 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
421 | ||
422 | spin_lock(&kvm->mmu_lock); | |
423 | /* | |
424 | * This sequence increase will notify the kvm page fault that | |
425 | * the page that is going to be mapped in the spte could have | |
426 | * been freed. | |
427 | */ | |
428 | kvm->mmu_notifier_seq++; | |
a355aa54 | 429 | smp_wmb(); |
e930bffe AA |
430 | /* |
431 | * The above sequence increase must be visible before the | |
a355aa54 PM |
432 | * below count decrease, which is ensured by the smp_wmb above |
433 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
434 | */ |
435 | kvm->mmu_notifier_count--; | |
436 | spin_unlock(&kvm->mmu_lock); | |
437 | ||
438 | BUG_ON(kvm->mmu_notifier_count < 0); | |
439 | } | |
440 | ||
441 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
442 | struct mm_struct *mm, | |
57128468 ALC |
443 | unsigned long start, |
444 | unsigned long end) | |
e930bffe AA |
445 | { |
446 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 447 | int young, idx; |
e930bffe | 448 | |
bc6678a3 | 449 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 450 | spin_lock(&kvm->mmu_lock); |
e930bffe | 451 | |
57128468 | 452 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
453 | if (young) |
454 | kvm_flush_remote_tlbs(kvm); | |
455 | ||
565f3be2 TY |
456 | spin_unlock(&kvm->mmu_lock); |
457 | srcu_read_unlock(&kvm->srcu, idx); | |
458 | ||
e930bffe AA |
459 | return young; |
460 | } | |
461 | ||
1d7715c6 VD |
462 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
463 | struct mm_struct *mm, | |
464 | unsigned long start, | |
465 | unsigned long end) | |
466 | { | |
467 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
468 | int young, idx; | |
469 | ||
470 | idx = srcu_read_lock(&kvm->srcu); | |
471 | spin_lock(&kvm->mmu_lock); | |
472 | /* | |
473 | * Even though we do not flush TLB, this will still adversely | |
474 | * affect performance on pre-Haswell Intel EPT, where there is | |
475 | * no EPT Access Bit to clear so that we have to tear down EPT | |
476 | * tables instead. If we find this unacceptable, we can always | |
477 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
478 | * do anything on clear_young. | |
479 | * | |
480 | * Also note that currently we never issue secondary TLB flushes | |
481 | * from clear_young, leaving this job up to the regular system | |
482 | * cadence. If we find this inaccurate, we might come up with a | |
483 | * more sophisticated heuristic later. | |
484 | */ | |
485 | young = kvm_age_hva(kvm, start, end); | |
486 | spin_unlock(&kvm->mmu_lock); | |
487 | srcu_read_unlock(&kvm->srcu, idx); | |
488 | ||
489 | return young; | |
490 | } | |
491 | ||
8ee53820 AA |
492 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
493 | struct mm_struct *mm, | |
494 | unsigned long address) | |
495 | { | |
496 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
497 | int young, idx; | |
498 | ||
499 | idx = srcu_read_lock(&kvm->srcu); | |
500 | spin_lock(&kvm->mmu_lock); | |
501 | young = kvm_test_age_hva(kvm, address); | |
502 | spin_unlock(&kvm->mmu_lock); | |
503 | srcu_read_unlock(&kvm->srcu, idx); | |
504 | ||
505 | return young; | |
506 | } | |
507 | ||
85db06e5 MT |
508 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
509 | struct mm_struct *mm) | |
510 | { | |
511 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
512 | int idx; |
513 | ||
514 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 515 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 516 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
517 | } |
518 | ||
e930bffe | 519 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e930bffe AA |
520 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
521 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
522 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 523 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 524 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 525 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 526 | .release = kvm_mmu_notifier_release, |
e930bffe | 527 | }; |
4c07b0a4 AK |
528 | |
529 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
530 | { | |
531 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
532 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
533 | } | |
534 | ||
535 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
536 | ||
537 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
538 | { | |
539 | return 0; | |
540 | } | |
541 | ||
e930bffe AA |
542 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
543 | ||
a47d2b07 | 544 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
545 | { |
546 | int i; | |
a47d2b07 | 547 | struct kvm_memslots *slots; |
bf3e05bc | 548 | |
b12ce36a | 549 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
550 | if (!slots) |
551 | return NULL; | |
552 | ||
bf3e05bc | 553 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 554 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
555 | |
556 | return slots; | |
557 | } | |
558 | ||
559 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
560 | { | |
561 | if (!memslot->dirty_bitmap) | |
562 | return; | |
563 | ||
564 | kvfree(memslot->dirty_bitmap); | |
565 | memslot->dirty_bitmap = NULL; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Free any memory in @free but not in @dont. | |
570 | */ | |
571 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
572 | struct kvm_memory_slot *dont) | |
573 | { | |
574 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
575 | kvm_destroy_dirty_bitmap(free); | |
576 | ||
577 | kvm_arch_free_memslot(kvm, free, dont); | |
578 | ||
579 | free->npages = 0; | |
580 | } | |
581 | ||
582 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
583 | { | |
584 | struct kvm_memory_slot *memslot; | |
585 | ||
586 | if (!slots) | |
587 | return; | |
588 | ||
589 | kvm_for_each_memslot(memslot, slots) | |
590 | kvm_free_memslot(kvm, memslot, NULL); | |
591 | ||
592 | kvfree(slots); | |
bf3e05bc XG |
593 | } |
594 | ||
536a6f88 JF |
595 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
596 | { | |
597 | int i; | |
598 | ||
599 | if (!kvm->debugfs_dentry) | |
600 | return; | |
601 | ||
602 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
603 | ||
9d5a1dce LC |
604 | if (kvm->debugfs_stat_data) { |
605 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
606 | kfree(kvm->debugfs_stat_data[i]); | |
607 | kfree(kvm->debugfs_stat_data); | |
608 | } | |
536a6f88 JF |
609 | } |
610 | ||
611 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
612 | { | |
613 | char dir_name[ITOA_MAX_LEN * 2]; | |
614 | struct kvm_stat_data *stat_data; | |
615 | struct kvm_stats_debugfs_item *p; | |
616 | ||
617 | if (!debugfs_initialized()) | |
618 | return 0; | |
619 | ||
620 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
929f45e3 | 621 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir); |
536a6f88 JF |
622 | |
623 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
624 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 625 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
626 | if (!kvm->debugfs_stat_data) |
627 | return -ENOMEM; | |
628 | ||
629 | for (p = debugfs_entries; p->name; p++) { | |
b12ce36a | 630 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
631 | if (!stat_data) |
632 | return -ENOMEM; | |
633 | ||
634 | stat_data->kvm = kvm; | |
09cbcef6 | 635 | stat_data->dbgfs_item = p; |
536a6f88 | 636 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; |
09cbcef6 MP |
637 | debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p), |
638 | kvm->debugfs_dentry, stat_data, | |
639 | &stat_fops_per_vm); | |
536a6f88 JF |
640 | } |
641 | return 0; | |
642 | } | |
643 | ||
1aa9b957 JS |
644 | /* |
645 | * Called after the VM is otherwise initialized, but just before adding it to | |
646 | * the vm_list. | |
647 | */ | |
648 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
649 | { | |
650 | return 0; | |
651 | } | |
652 | ||
653 | /* | |
654 | * Called just after removing the VM from the vm_list, but before doing any | |
655 | * other destruction. | |
656 | */ | |
657 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
658 | { | |
659 | } | |
660 | ||
e08b9637 | 661 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 662 | { |
d89f5eff | 663 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
664 | int r = -ENOMEM; |
665 | int i; | |
6aa8b732 | 666 | |
d89f5eff JK |
667 | if (!kvm) |
668 | return ERR_PTR(-ENOMEM); | |
669 | ||
e9ad4ec8 | 670 | spin_lock_init(&kvm->mmu_lock); |
f1f10076 | 671 | mmgrab(current->mm); |
e9ad4ec8 PB |
672 | kvm->mm = current->mm; |
673 | kvm_eventfd_init(kvm); | |
674 | mutex_init(&kvm->lock); | |
675 | mutex_init(&kvm->irq_lock); | |
676 | mutex_init(&kvm->slots_lock); | |
e9ad4ec8 PB |
677 | INIT_LIST_HEAD(&kvm->devices); |
678 | ||
1e702d9a AW |
679 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
680 | ||
8a44119a PB |
681 | if (init_srcu_struct(&kvm->srcu)) |
682 | goto out_err_no_srcu; | |
683 | if (init_srcu_struct(&kvm->irq_srcu)) | |
684 | goto out_err_no_irq_srcu; | |
685 | ||
e2d3fcaf | 686 | refcount_set(&kvm->users_count, 1); |
f481b069 | 687 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 688 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 689 | |
4bd518f1 | 690 | if (!slots) |
a97b0e77 | 691 | goto out_err_no_arch_destroy_vm; |
0e32958e | 692 | /* Generations must be different for each address space. */ |
164bf7e5 | 693 | slots->generation = i; |
4bd518f1 | 694 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 695 | } |
00f034a1 | 696 | |
e93f8a0f | 697 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 698 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 699 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 700 | if (!kvm->buses[i]) |
a97b0e77 | 701 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 702 | } |
e930bffe | 703 | |
e08b9637 | 704 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 705 | if (r) |
a97b0e77 | 706 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
707 | |
708 | r = hardware_enable_all(); | |
709 | if (r) | |
719d93cd | 710 | goto out_err_no_disable; |
10474ae8 | 711 | |
c77dcacb | 712 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 713 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 714 | #endif |
6aa8b732 | 715 | |
74b5c5bf | 716 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
717 | if (r) |
718 | goto out_err_no_mmu_notifier; | |
719 | ||
720 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
721 | if (r) |
722 | goto out_err; | |
723 | ||
0d9ce162 | 724 | mutex_lock(&kvm_lock); |
5e58cfe4 | 725 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 726 | mutex_unlock(&kvm_lock); |
d89f5eff | 727 | |
2ecd9d29 PZ |
728 | preempt_notifier_inc(); |
729 | ||
f17abe9a | 730 | return kvm; |
10474ae8 AG |
731 | |
732 | out_err: | |
1aa9b957 JS |
733 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
734 | if (kvm->mmu_notifier.ops) | |
735 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
736 | #endif | |
737 | out_err_no_mmu_notifier: | |
10474ae8 | 738 | hardware_disable_all(); |
719d93cd | 739 | out_err_no_disable: |
a97b0e77 | 740 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 741 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 742 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 743 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 744 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 745 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 746 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
747 | cleanup_srcu_struct(&kvm->irq_srcu); |
748 | out_err_no_irq_srcu: | |
749 | cleanup_srcu_struct(&kvm->srcu); | |
750 | out_err_no_srcu: | |
d89f5eff | 751 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 752 | mmdrop(current->mm); |
10474ae8 | 753 | return ERR_PTR(r); |
f17abe9a AK |
754 | } |
755 | ||
07f0a7bd SW |
756 | static void kvm_destroy_devices(struct kvm *kvm) |
757 | { | |
e6e3b5a6 | 758 | struct kvm_device *dev, *tmp; |
07f0a7bd | 759 | |
a28ebea2 CD |
760 | /* |
761 | * We do not need to take the kvm->lock here, because nobody else | |
762 | * has a reference to the struct kvm at this point and therefore | |
763 | * cannot access the devices list anyhow. | |
764 | */ | |
e6e3b5a6 GT |
765 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
766 | list_del(&dev->vm_node); | |
07f0a7bd SW |
767 | dev->ops->destroy(dev); |
768 | } | |
769 | } | |
770 | ||
f17abe9a AK |
771 | static void kvm_destroy_vm(struct kvm *kvm) |
772 | { | |
e93f8a0f | 773 | int i; |
6d4e4c4f AK |
774 | struct mm_struct *mm = kvm->mm; |
775 | ||
286de8f6 | 776 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 777 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 778 | kvm_arch_sync_events(kvm); |
0d9ce162 | 779 | mutex_lock(&kvm_lock); |
133de902 | 780 | list_del(&kvm->vm_list); |
0d9ce162 | 781 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
782 | kvm_arch_pre_destroy_vm(kvm); |
783 | ||
399ec807 | 784 | kvm_free_irq_routing(kvm); |
df630b8c | 785 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 786 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 787 | |
4a12f951 CB |
788 | if (bus) |
789 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
790 | kvm->buses[i] = NULL; |
791 | } | |
980da6ce | 792 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
793 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
794 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 795 | #else |
2df72e9b | 796 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 797 | #endif |
d19a9cd2 | 798 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 799 | kvm_destroy_devices(kvm); |
f481b069 | 800 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 801 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 802 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
803 | cleanup_srcu_struct(&kvm->srcu); |
804 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 805 | preempt_notifier_dec(); |
10474ae8 | 806 | hardware_disable_all(); |
6d4e4c4f | 807 | mmdrop(mm); |
f17abe9a AK |
808 | } |
809 | ||
d39f13b0 IE |
810 | void kvm_get_kvm(struct kvm *kvm) |
811 | { | |
e3736c3e | 812 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
813 | } |
814 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
815 | ||
816 | void kvm_put_kvm(struct kvm *kvm) | |
817 | { | |
e3736c3e | 818 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
819 | kvm_destroy_vm(kvm); |
820 | } | |
821 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
822 | ||
149487bd SC |
823 | /* |
824 | * Used to put a reference that was taken on behalf of an object associated | |
825 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
826 | * of the new file descriptor fails and the reference cannot be transferred to | |
827 | * its final owner. In such cases, the caller is still actively using @kvm and | |
828 | * will fail miserably if the refcount unexpectedly hits zero. | |
829 | */ | |
830 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
831 | { | |
832 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
833 | } | |
834 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 835 | |
f17abe9a AK |
836 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
837 | { | |
838 | struct kvm *kvm = filp->private_data; | |
839 | ||
721eecbf GH |
840 | kvm_irqfd_release(kvm); |
841 | ||
d39f13b0 | 842 | kvm_put_kvm(kvm); |
6aa8b732 AK |
843 | return 0; |
844 | } | |
845 | ||
515a0127 TY |
846 | /* |
847 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 848 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 849 | */ |
a36a57b1 TY |
850 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
851 | { | |
515a0127 | 852 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 853 | |
b12ce36a | 854 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
855 | if (!memslot->dirty_bitmap) |
856 | return -ENOMEM; | |
857 | ||
a36a57b1 TY |
858 | return 0; |
859 | } | |
860 | ||
bf3e05bc | 861 | /* |
0e60b079 IM |
862 | * Insert memslot and re-sort memslots based on their GFN, |
863 | * so binary search could be used to lookup GFN. | |
864 | * Sorting algorithm takes advantage of having initially | |
865 | * sorted array and known changed memslot position. | |
bf3e05bc | 866 | */ |
5cc15027 | 867 | static void update_memslots(struct kvm_memslots *slots, |
31fc4f95 WY |
868 | struct kvm_memory_slot *new, |
869 | enum kvm_mr_change change) | |
bf3e05bc | 870 | { |
8593176c PB |
871 | int id = new->id; |
872 | int i = slots->id_to_index[id]; | |
063584d4 | 873 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 874 | |
8593176c | 875 | WARN_ON(mslots[i].id != id); |
31fc4f95 WY |
876 | switch (change) { |
877 | case KVM_MR_CREATE: | |
878 | slots->used_slots++; | |
879 | WARN_ON(mslots[i].npages || !new->npages); | |
880 | break; | |
881 | case KVM_MR_DELETE: | |
882 | slots->used_slots--; | |
883 | WARN_ON(new->npages || !mslots[i].npages); | |
884 | break; | |
885 | default: | |
886 | break; | |
9c1a5d38 | 887 | } |
0e60b079 | 888 | |
7f379cff | 889 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
890 | new->base_gfn <= mslots[i + 1].base_gfn) { |
891 | if (!mslots[i + 1].npages) | |
892 | break; | |
7f379cff IM |
893 | mslots[i] = mslots[i + 1]; |
894 | slots->id_to_index[mslots[i].id] = i; | |
895 | i++; | |
896 | } | |
efbeec70 PB |
897 | |
898 | /* | |
899 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
900 | * so that it moves before all those with base_gfn == npages == 0. | |
901 | * | |
902 | * On the other hand, if new->npages is zero, the above loop has | |
903 | * already left i pointing to the beginning of the empty part of | |
904 | * mslots, and the ">=" would move the hole backwards in this | |
905 | * case---which is wrong. So skip the loop when deleting a slot. | |
906 | */ | |
907 | if (new->npages) { | |
908 | while (i > 0 && | |
909 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
910 | mslots[i] = mslots[i - 1]; | |
911 | slots->id_to_index[mslots[i].id] = i; | |
912 | i--; | |
913 | } | |
dbaff309 PB |
914 | } else |
915 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 916 | |
8593176c PB |
917 | mslots[i] = *new; |
918 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
919 | } |
920 | ||
09170a49 | 921 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 922 | { |
4d8b81ab XG |
923 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
924 | ||
0f8a4de3 | 925 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
926 | valid_flags |= KVM_MEM_READONLY; |
927 | #endif | |
928 | ||
929 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
930 | return -EINVAL; |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
7ec4fb44 | 935 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 936 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 937 | { |
f481b069 | 938 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 939 | u64 gen = old_memslots->generation; |
7ec4fb44 | 940 | |
361209e0 SC |
941 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
942 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 943 | |
f481b069 | 944 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 945 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 946 | |
ee3d1570 | 947 | /* |
361209e0 | 948 | * Increment the new memslot generation a second time, dropping the |
00116795 | 949 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
950 | * the number of address spaces. This provides a unique and easily |
951 | * identifiable generation number while the memslots are in flux. | |
952 | */ | |
953 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
954 | ||
955 | /* | |
4bd518f1 PB |
956 | * Generations must be unique even across address spaces. We do not need |
957 | * a global counter for that, instead the generation space is evenly split | |
958 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
959 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
960 | * use generations 1, 3, 5, ... | |
ee3d1570 | 961 | */ |
164bf7e5 | 962 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 963 | |
15248258 | 964 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 965 | |
15248258 | 966 | slots->generation = gen; |
e59dbe09 TY |
967 | |
968 | return old_memslots; | |
7ec4fb44 GN |
969 | } |
970 | ||
6aa8b732 AK |
971 | /* |
972 | * Allocate some memory and give it an address in the guest physical address | |
973 | * space. | |
974 | * | |
975 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 976 | * |
02d5d55b | 977 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 978 | */ |
f78e0e2e | 979 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 980 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 981 | { |
8234b22e | 982 | int r; |
6aa8b732 | 983 | gfn_t base_gfn; |
28bcb112 | 984 | unsigned long npages; |
a843fac2 | 985 | struct kvm_memory_slot *slot; |
6aa8b732 | 986 | struct kvm_memory_slot old, new; |
b7f69c55 | 987 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 988 | int as_id, id; |
f64c0398 | 989 | enum kvm_mr_change change; |
6aa8b732 | 990 | |
a50d64d6 XG |
991 | r = check_memory_region_flags(mem); |
992 | if (r) | |
993 | goto out; | |
994 | ||
6aa8b732 | 995 | r = -EINVAL; |
f481b069 PB |
996 | as_id = mem->slot >> 16; |
997 | id = (u16)mem->slot; | |
998 | ||
6aa8b732 AK |
999 | /* General sanity checks */ |
1000 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
1001 | goto out; | |
1002 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
1003 | goto out; | |
fa3d315a | 1004 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 1005 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 1006 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
96d4f267 | 1007 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
9e3bb6b6 | 1008 | mem->memory_size))) |
78749809 | 1009 | goto out; |
f481b069 | 1010 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
1011 | goto out; |
1012 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
1013 | goto out; | |
1014 | ||
f481b069 | 1015 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
1016 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1017 | npages = mem->memory_size >> PAGE_SHIFT; | |
1018 | ||
660c22c4 TY |
1019 | if (npages > KVM_MEM_MAX_NR_PAGES) |
1020 | goto out; | |
1021 | ||
a843fac2 | 1022 | new = old = *slot; |
6aa8b732 | 1023 | |
f481b069 | 1024 | new.id = id; |
6aa8b732 AK |
1025 | new.base_gfn = base_gfn; |
1026 | new.npages = npages; | |
1027 | new.flags = mem->flags; | |
1028 | ||
f64c0398 TY |
1029 | if (npages) { |
1030 | if (!old.npages) | |
1031 | change = KVM_MR_CREATE; | |
1032 | else { /* Modify an existing slot. */ | |
1033 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
1034 | (npages != old.npages) || |
1035 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
1036 | goto out; |
1037 | ||
1038 | if (base_gfn != old.base_gfn) | |
1039 | change = KVM_MR_MOVE; | |
1040 | else if (new.flags != old.flags) | |
1041 | change = KVM_MR_FLAGS_ONLY; | |
1042 | else { /* Nothing to change. */ | |
1043 | r = 0; | |
1044 | goto out; | |
1045 | } | |
1046 | } | |
09170a49 PB |
1047 | } else { |
1048 | if (!old.npages) | |
1049 | goto out; | |
1050 | ||
f64c0398 | 1051 | change = KVM_MR_DELETE; |
09170a49 PB |
1052 | new.base_gfn = 0; |
1053 | new.flags = 0; | |
1054 | } | |
6aa8b732 | 1055 | |
f64c0398 | 1056 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
1057 | /* Check for overlaps */ |
1058 | r = -EEXIST; | |
f481b069 | 1059 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
b28676bb | 1060 | if (slot->id == id) |
0a706bee TY |
1061 | continue; |
1062 | if (!((base_gfn + npages <= slot->base_gfn) || | |
1063 | (base_gfn >= slot->base_gfn + slot->npages))) | |
1064 | goto out; | |
1065 | } | |
6aa8b732 | 1066 | } |
6aa8b732 | 1067 | |
6aa8b732 AK |
1068 | /* Free page dirty bitmap if unneeded */ |
1069 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 1070 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
1071 | |
1072 | r = -ENOMEM; | |
f64c0398 | 1073 | if (change == KVM_MR_CREATE) { |
189a2f7b | 1074 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 1075 | |
5587027c | 1076 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 1077 | goto out_free; |
6aa8b732 | 1078 | } |
ec04b260 | 1079 | |
6aa8b732 AK |
1080 | /* Allocate page dirty bitmap if needed */ |
1081 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 1082 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 1083 | goto out_free; |
6aa8b732 AK |
1084 | } |
1085 | ||
b12ce36a | 1086 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
f2a81036 PB |
1087 | if (!slots) |
1088 | goto out_free; | |
f481b069 | 1089 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 1090 | |
f64c0398 | 1091 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 1092 | slot = id_to_memslot(slots, id); |
28a37544 XG |
1093 | slot->flags |= KVM_MEMSLOT_INVALID; |
1094 | ||
f481b069 | 1095 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 1096 | |
12d6e753 MT |
1097 | /* From this point no new shadow pages pointing to a deleted, |
1098 | * or moved, memslot will be created. | |
bc6678a3 MT |
1099 | * |
1100 | * validation of sp->gfn happens in: | |
b7d409de | 1101 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
668effb6 | 1102 | * - kvm_is_visible_gfn (mmu_check_root) |
bc6678a3 | 1103 | */ |
2df72e9b | 1104 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
1105 | |
1106 | /* | |
1107 | * We can re-use the old_memslots from above, the only difference | |
1108 | * from the currently installed memslots is the invalid flag. This | |
1109 | * will get overwritten by update_memslots anyway. | |
1110 | */ | |
b7f69c55 | 1111 | slots = old_memslots; |
bc6678a3 | 1112 | } |
34d4cb8f | 1113 | |
7b6195a9 | 1114 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 1115 | if (r) |
b7f69c55 | 1116 | goto out_slots; |
f7784b8e | 1117 | |
a47d2b07 | 1118 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 1119 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 1120 | new.dirty_bitmap = NULL; |
db3fe4eb | 1121 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
1122 | } |
1123 | ||
31fc4f95 | 1124 | update_memslots(slots, &new, change); |
f481b069 | 1125 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 1126 | |
f36f3f28 | 1127 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 1128 | |
a47d2b07 | 1129 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 1130 | kvfree(old_memslots); |
6aa8b732 AK |
1131 | return 0; |
1132 | ||
e40f193f | 1133 | out_slots: |
74496134 | 1134 | kvfree(slots); |
f78e0e2e | 1135 | out_free: |
a47d2b07 | 1136 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
1137 | out: |
1138 | return r; | |
210c7c4d | 1139 | } |
f78e0e2e SY |
1140 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1141 | ||
1142 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1143 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1144 | { |
1145 | int r; | |
1146 | ||
79fac95e | 1147 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1148 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1149 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1150 | return r; |
1151 | } | |
210c7c4d IE |
1152 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1153 | ||
7940876e SH |
1154 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1155 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1156 | { |
f481b069 | 1157 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1158 | return -EINVAL; |
09170a49 | 1159 | |
47ae31e2 | 1160 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1161 | } |
1162 | ||
5bb064dc ZX |
1163 | int kvm_get_dirty_log(struct kvm *kvm, |
1164 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 1165 | { |
9f6b8029 | 1166 | struct kvm_memslots *slots; |
6aa8b732 | 1167 | struct kvm_memory_slot *memslot; |
843574a3 | 1168 | int i, as_id, id; |
87bf6e7d | 1169 | unsigned long n; |
6aa8b732 AK |
1170 | unsigned long any = 0; |
1171 | ||
f481b069 PB |
1172 | as_id = log->slot >> 16; |
1173 | id = (u16)log->slot; | |
1174 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1175 | return -EINVAL; |
6aa8b732 | 1176 | |
f481b069 PB |
1177 | slots = __kvm_memslots(kvm, as_id); |
1178 | memslot = id_to_memslot(slots, id); | |
6aa8b732 | 1179 | if (!memslot->dirty_bitmap) |
843574a3 | 1180 | return -ENOENT; |
6aa8b732 | 1181 | |
87bf6e7d | 1182 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 1183 | |
cd1a4a98 | 1184 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1185 | any = memslot->dirty_bitmap[i]; |
1186 | ||
6aa8b732 | 1187 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
843574a3 | 1188 | return -EFAULT; |
6aa8b732 | 1189 | |
5bb064dc ZX |
1190 | if (any) |
1191 | *is_dirty = 1; | |
843574a3 | 1192 | return 0; |
6aa8b732 | 1193 | } |
2ba9f0d8 | 1194 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1195 | |
ba0513b5 MS |
1196 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1197 | /** | |
b8b00220 | 1198 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1199 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1200 | * @kvm: pointer to kvm instance |
1201 | * @log: slot id and address to which we copy the log | |
b8b00220 | 1202 | * @flush: true if TLB flush is needed by caller |
ba0513b5 MS |
1203 | * |
1204 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1205 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1206 | * following order: | |
1207 | * | |
1208 | * 1. Take a snapshot of the bit and clear it if needed. | |
1209 | * 2. Write protect the corresponding page. | |
1210 | * 3. Copy the snapshot to the userspace. | |
1211 | * 4. Upon return caller flushes TLB's if needed. | |
1212 | * | |
1213 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1214 | * entry. This is not a problem because the page is reported dirty using | |
1215 | * the snapshot taken before and step 4 ensures that writes done after | |
1216 | * exiting to userspace will be logged for the next call. | |
1217 | * | |
1218 | */ | |
1219 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
8fe65a82 | 1220 | struct kvm_dirty_log *log, bool *flush) |
ba0513b5 | 1221 | { |
9f6b8029 | 1222 | struct kvm_memslots *slots; |
ba0513b5 | 1223 | struct kvm_memory_slot *memslot; |
58d6db34 | 1224 | int i, as_id, id; |
ba0513b5 MS |
1225 | unsigned long n; |
1226 | unsigned long *dirty_bitmap; | |
1227 | unsigned long *dirty_bitmap_buffer; | |
1228 | ||
f481b069 PB |
1229 | as_id = log->slot >> 16; |
1230 | id = (u16)log->slot; | |
1231 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1232 | return -EINVAL; |
ba0513b5 | 1233 | |
f481b069 PB |
1234 | slots = __kvm_memslots(kvm, as_id); |
1235 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1236 | |
1237 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1238 | if (!dirty_bitmap) |
58d6db34 | 1239 | return -ENOENT; |
ba0513b5 MS |
1240 | |
1241 | n = kvm_dirty_bitmap_bytes(memslot); | |
2a31b9db PB |
1242 | *flush = false; |
1243 | if (kvm->manual_dirty_log_protect) { | |
1244 | /* | |
1245 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1246 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1247 | * is some code duplication between this function and | |
1248 | * kvm_get_dirty_log, but hopefully all architecture | |
1249 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1250 | * can be eliminated. | |
1251 | */ | |
1252 | dirty_bitmap_buffer = dirty_bitmap; | |
1253 | } else { | |
1254 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1255 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1256 | |
2a31b9db PB |
1257 | spin_lock(&kvm->mmu_lock); |
1258 | for (i = 0; i < n / sizeof(long); i++) { | |
1259 | unsigned long mask; | |
1260 | gfn_t offset; | |
ba0513b5 | 1261 | |
2a31b9db PB |
1262 | if (!dirty_bitmap[i]) |
1263 | continue; | |
1264 | ||
1265 | *flush = true; | |
1266 | mask = xchg(&dirty_bitmap[i], 0); | |
1267 | dirty_bitmap_buffer[i] = mask; | |
1268 | ||
a67794ca LT |
1269 | offset = i * BITS_PER_LONG; |
1270 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1271 | offset, mask); | |
2a31b9db PB |
1272 | } |
1273 | spin_unlock(&kvm->mmu_lock); | |
1274 | } | |
1275 | ||
1276 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1277 | return -EFAULT; | |
1278 | return 0; | |
1279 | } | |
1280 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1281 | ||
1282 | /** | |
1283 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1284 | * and reenable dirty page tracking for the corresponding pages. | |
1285 | * @kvm: pointer to kvm instance | |
1286 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
b8b00220 | 1287 | * @flush: true if TLB flush is needed by caller |
2a31b9db PB |
1288 | */ |
1289 | int kvm_clear_dirty_log_protect(struct kvm *kvm, | |
1290 | struct kvm_clear_dirty_log *log, bool *flush) | |
1291 | { | |
1292 | struct kvm_memslots *slots; | |
1293 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1294 | int as_id, id; |
2a31b9db | 1295 | gfn_t offset; |
98938aa8 | 1296 | unsigned long i, n; |
2a31b9db PB |
1297 | unsigned long *dirty_bitmap; |
1298 | unsigned long *dirty_bitmap_buffer; | |
1299 | ||
1300 | as_id = log->slot >> 16; | |
1301 | id = (u16)log->slot; | |
1302 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1303 | return -EINVAL; | |
1304 | ||
76d58e0f | 1305 | if (log->first_page & 63) |
2a31b9db PB |
1306 | return -EINVAL; |
1307 | ||
1308 | slots = __kvm_memslots(kvm, as_id); | |
1309 | memslot = id_to_memslot(slots, id); | |
1310 | ||
1311 | dirty_bitmap = memslot->dirty_bitmap; | |
1312 | if (!dirty_bitmap) | |
1313 | return -ENOENT; | |
1314 | ||
4ddc9204 | 1315 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
1316 | |
1317 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
1318 | log->num_pages > memslot->npages - log->first_page || |
1319 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
1320 | return -EINVAL; | |
98938aa8 | 1321 | |
8fe65a82 | 1322 | *flush = false; |
2a31b9db PB |
1323 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
1324 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
1325 | return -EFAULT; | |
ba0513b5 | 1326 | |
2a31b9db | 1327 | spin_lock(&kvm->mmu_lock); |
53eac7a8 PX |
1328 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
1329 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
1330 | i++, offset += BITS_PER_LONG) { |
1331 | unsigned long mask = *dirty_bitmap_buffer++; | |
1332 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
1333 | if (!mask) | |
ba0513b5 MS |
1334 | continue; |
1335 | ||
2a31b9db | 1336 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 1337 | |
2a31b9db PB |
1338 | /* |
1339 | * mask contains the bits that really have been cleared. This | |
1340 | * never includes any bits beyond the length of the memslot (if | |
1341 | * the length is not aligned to 64 pages), therefore it is not | |
1342 | * a problem if userspace sets them in log->dirty_bitmap. | |
1343 | */ | |
58d2930f | 1344 | if (mask) { |
2a31b9db | 1345 | *flush = true; |
58d2930f TY |
1346 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
1347 | offset, mask); | |
1348 | } | |
ba0513b5 | 1349 | } |
ba0513b5 | 1350 | spin_unlock(&kvm->mmu_lock); |
2a31b9db | 1351 | |
58d6db34 | 1352 | return 0; |
ba0513b5 | 1353 | } |
2a31b9db | 1354 | EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect); |
ba0513b5 MS |
1355 | #endif |
1356 | ||
db3fe4eb TY |
1357 | bool kvm_largepages_enabled(void) |
1358 | { | |
1359 | return largepages_enabled; | |
1360 | } | |
1361 | ||
54dee993 MT |
1362 | void kvm_disable_largepages(void) |
1363 | { | |
1364 | largepages_enabled = false; | |
1365 | } | |
1366 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1367 | ||
49c7754c GN |
1368 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1369 | { | |
1370 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1371 | } | |
a1f4d395 | 1372 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1373 | |
8e73485c PB |
1374 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1375 | { | |
1376 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1377 | } | |
1378 | ||
33e94154 | 1379 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1380 | { |
bf3e05bc | 1381 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1382 | |
bbacc0c1 | 1383 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc | 1384 | memslot->flags & KVM_MEMSLOT_INVALID) |
33e94154 | 1385 | return false; |
e0d62c7f | 1386 | |
33e94154 | 1387 | return true; |
e0d62c7f IE |
1388 | } |
1389 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1390 | ||
8f0b1ab6 JR |
1391 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1392 | { | |
1393 | struct vm_area_struct *vma; | |
1394 | unsigned long addr, size; | |
1395 | ||
1396 | size = PAGE_SIZE; | |
1397 | ||
1398 | addr = gfn_to_hva(kvm, gfn); | |
1399 | if (kvm_is_error_hva(addr)) | |
1400 | return PAGE_SIZE; | |
1401 | ||
1402 | down_read(¤t->mm->mmap_sem); | |
1403 | vma = find_vma(current->mm, addr); | |
1404 | if (!vma) | |
1405 | goto out; | |
1406 | ||
1407 | size = vma_kernel_pagesize(vma); | |
1408 | ||
1409 | out: | |
1410 | up_read(¤t->mm->mmap_sem); | |
1411 | ||
1412 | return size; | |
1413 | } | |
1414 | ||
4d8b81ab XG |
1415 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1416 | { | |
1417 | return slot->flags & KVM_MEM_READONLY; | |
1418 | } | |
1419 | ||
4d8b81ab XG |
1420 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1421 | gfn_t *nr_pages, bool write) | |
539cb660 | 1422 | { |
bc6678a3 | 1423 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1424 | return KVM_HVA_ERR_BAD; |
48987781 | 1425 | |
4d8b81ab XG |
1426 | if (memslot_is_readonly(slot) && write) |
1427 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1428 | |
1429 | if (nr_pages) | |
1430 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1431 | ||
4d8b81ab | 1432 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1433 | } |
48987781 | 1434 | |
4d8b81ab XG |
1435 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1436 | gfn_t *nr_pages) | |
1437 | { | |
1438 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1439 | } |
48987781 | 1440 | |
4d8b81ab | 1441 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1442 | gfn_t gfn) |
4d8b81ab XG |
1443 | { |
1444 | return gfn_to_hva_many(slot, gfn, NULL); | |
1445 | } | |
1446 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1447 | ||
48987781 XG |
1448 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1449 | { | |
49c7754c | 1450 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1451 | } |
0d150298 | 1452 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1453 | |
8e73485c PB |
1454 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1455 | { | |
1456 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1457 | } | |
1458 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1459 | ||
86ab8cff | 1460 | /* |
970c0d4b WY |
1461 | * Return the hva of a @gfn and the R/W attribute if possible. |
1462 | * | |
1463 | * @slot: the kvm_memory_slot which contains @gfn | |
1464 | * @gfn: the gfn to be translated | |
1465 | * @writable: used to return the read/write attribute of the @slot if the hva | |
1466 | * is valid and @writable is not NULL | |
86ab8cff | 1467 | */ |
64d83126 CD |
1468 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1469 | gfn_t gfn, bool *writable) | |
86ab8cff | 1470 | { |
a2ac07fe GN |
1471 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1472 | ||
1473 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1474 | *writable = !memslot_is_readonly(slot); |
1475 | ||
a2ac07fe | 1476 | return hva; |
86ab8cff XG |
1477 | } |
1478 | ||
64d83126 CD |
1479 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1480 | { | |
1481 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1482 | ||
1483 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1484 | } | |
1485 | ||
8e73485c PB |
1486 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1487 | { | |
1488 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1489 | ||
1490 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1491 | } | |
1492 | ||
fafc3dba HY |
1493 | static inline int check_user_page_hwpoison(unsigned long addr) |
1494 | { | |
0d731759 | 1495 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1496 | |
0d731759 | 1497 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1498 | return rc == -EHWPOISON; |
1499 | } | |
1500 | ||
2fc84311 | 1501 | /* |
b9b33da2 PB |
1502 | * The fast path to get the writable pfn which will be stored in @pfn, |
1503 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 1504 | * only part that runs if we can in atomic context. |
2fc84311 | 1505 | */ |
b9b33da2 PB |
1506 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
1507 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 1508 | { |
8d4e1288 | 1509 | struct page *page[1]; |
2fc84311 | 1510 | int npages; |
954bbbc2 | 1511 | |
12ce13fe XG |
1512 | /* |
1513 | * Fast pin a writable pfn only if it is a write fault request | |
1514 | * or the caller allows to map a writable pfn for a read fault | |
1515 | * request. | |
1516 | */ | |
1517 | if (!(write_fault || writable)) | |
1518 | return false; | |
612819c3 | 1519 | |
2fc84311 XG |
1520 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1521 | if (npages == 1) { | |
1522 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1523 | |
2fc84311 XG |
1524 | if (writable) |
1525 | *writable = true; | |
1526 | return true; | |
1527 | } | |
af585b92 | 1528 | |
2fc84311 XG |
1529 | return false; |
1530 | } | |
612819c3 | 1531 | |
2fc84311 XG |
1532 | /* |
1533 | * The slow path to get the pfn of the specified host virtual address, | |
1534 | * 1 indicates success, -errno is returned if error is detected. | |
1535 | */ | |
1536 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1537 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 1538 | { |
ce53053c AV |
1539 | unsigned int flags = FOLL_HWPOISON; |
1540 | struct page *page; | |
2fc84311 | 1541 | int npages = 0; |
612819c3 | 1542 | |
2fc84311 XG |
1543 | might_sleep(); |
1544 | ||
1545 | if (writable) | |
1546 | *writable = write_fault; | |
1547 | ||
ce53053c AV |
1548 | if (write_fault) |
1549 | flags |= FOLL_WRITE; | |
1550 | if (async) | |
1551 | flags |= FOLL_NOWAIT; | |
d4944b0e | 1552 | |
ce53053c | 1553 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
1554 | if (npages != 1) |
1555 | return npages; | |
1556 | ||
1557 | /* map read fault as writable if possible */ | |
12ce13fe | 1558 | if (unlikely(!write_fault) && writable) { |
ce53053c | 1559 | struct page *wpage; |
2fc84311 | 1560 | |
ce53053c | 1561 | if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) { |
2fc84311 | 1562 | *writable = true; |
ce53053c AV |
1563 | put_page(page); |
1564 | page = wpage; | |
612819c3 | 1565 | } |
887c08ac | 1566 | } |
ce53053c | 1567 | *pfn = page_to_pfn(page); |
2fc84311 XG |
1568 | return npages; |
1569 | } | |
539cb660 | 1570 | |
4d8b81ab XG |
1571 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1572 | { | |
1573 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1574 | return false; | |
2e2e3738 | 1575 | |
4d8b81ab XG |
1576 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1577 | return false; | |
887c08ac | 1578 | |
4d8b81ab XG |
1579 | return true; |
1580 | } | |
bf998156 | 1581 | |
92176a8e PB |
1582 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1583 | unsigned long addr, bool *async, | |
a340b3e2 KA |
1584 | bool write_fault, bool *writable, |
1585 | kvm_pfn_t *p_pfn) | |
92176a8e | 1586 | { |
add6a0cd PB |
1587 | unsigned long pfn; |
1588 | int r; | |
1589 | ||
1590 | r = follow_pfn(vma, addr, &pfn); | |
1591 | if (r) { | |
1592 | /* | |
1593 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1594 | * not call the fault handler, so do it here. | |
1595 | */ | |
1596 | bool unlocked = false; | |
1597 | r = fixup_user_fault(current, current->mm, addr, | |
1598 | (write_fault ? FAULT_FLAG_WRITE : 0), | |
1599 | &unlocked); | |
1600 | if (unlocked) | |
1601 | return -EAGAIN; | |
1602 | if (r) | |
1603 | return r; | |
1604 | ||
1605 | r = follow_pfn(vma, addr, &pfn); | |
1606 | if (r) | |
1607 | return r; | |
1608 | ||
1609 | } | |
1610 | ||
a340b3e2 KA |
1611 | if (writable) |
1612 | *writable = true; | |
add6a0cd PB |
1613 | |
1614 | /* | |
1615 | * Get a reference here because callers of *hva_to_pfn* and | |
1616 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1617 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1618 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1619 | * simply do nothing for reserved pfns. | |
1620 | * | |
1621 | * Whoever called remap_pfn_range is also going to call e.g. | |
1622 | * unmap_mapping_range before the underlying pages are freed, | |
1623 | * causing a call to our MMU notifier. | |
1624 | */ | |
1625 | kvm_get_pfn(pfn); | |
1626 | ||
1627 | *p_pfn = pfn; | |
92176a8e PB |
1628 | return 0; |
1629 | } | |
1630 | ||
12ce13fe XG |
1631 | /* |
1632 | * Pin guest page in memory and return its pfn. | |
1633 | * @addr: host virtual address which maps memory to the guest | |
1634 | * @atomic: whether this function can sleep | |
1635 | * @async: whether this function need to wait IO complete if the | |
1636 | * host page is not in the memory | |
1637 | * @write_fault: whether we should get a writable host page | |
1638 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1639 | * | |
1640 | * The function will map a writable host page for these two cases: | |
1641 | * 1): @write_fault = true | |
1642 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1643 | * whether the mapping is writable. | |
1644 | */ | |
ba049e93 | 1645 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1646 | bool write_fault, bool *writable) |
1647 | { | |
1648 | struct vm_area_struct *vma; | |
ba049e93 | 1649 | kvm_pfn_t pfn = 0; |
92176a8e | 1650 | int npages, r; |
2e2e3738 | 1651 | |
2fc84311 XG |
1652 | /* we can do it either atomically or asynchronously, not both */ |
1653 | BUG_ON(atomic && async); | |
8d4e1288 | 1654 | |
b9b33da2 | 1655 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
1656 | return pfn; |
1657 | ||
1658 | if (atomic) | |
1659 | return KVM_PFN_ERR_FAULT; | |
1660 | ||
1661 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1662 | if (npages == 1) | |
1663 | return pfn; | |
8d4e1288 | 1664 | |
2fc84311 XG |
1665 | down_read(¤t->mm->mmap_sem); |
1666 | if (npages == -EHWPOISON || | |
1667 | (!async && check_user_page_hwpoison(addr))) { | |
1668 | pfn = KVM_PFN_ERR_HWPOISON; | |
1669 | goto exit; | |
1670 | } | |
1671 | ||
add6a0cd | 1672 | retry: |
2fc84311 XG |
1673 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
1674 | ||
1675 | if (vma == NULL) | |
1676 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 1677 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 1678 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
add6a0cd PB |
1679 | if (r == -EAGAIN) |
1680 | goto retry; | |
92176a8e PB |
1681 | if (r < 0) |
1682 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 1683 | } else { |
4d8b81ab | 1684 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1685 | *async = true; |
1686 | pfn = KVM_PFN_ERR_FAULT; | |
1687 | } | |
1688 | exit: | |
1689 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1690 | return pfn; |
35149e21 AL |
1691 | } |
1692 | ||
ba049e93 DW |
1693 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
1694 | bool atomic, bool *async, bool write_fault, | |
1695 | bool *writable) | |
887c08ac | 1696 | { |
4d8b81ab XG |
1697 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1698 | ||
b2740d35 PB |
1699 | if (addr == KVM_HVA_ERR_RO_BAD) { |
1700 | if (writable) | |
1701 | *writable = false; | |
4d8b81ab | 1702 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 1703 | } |
4d8b81ab | 1704 | |
b2740d35 PB |
1705 | if (kvm_is_error_hva(addr)) { |
1706 | if (writable) | |
1707 | *writable = false; | |
81c52c56 | 1708 | return KVM_PFN_NOSLOT; |
b2740d35 | 1709 | } |
4d8b81ab XG |
1710 | |
1711 | /* Do not map writable pfn in the readonly memslot. */ | |
1712 | if (writable && memslot_is_readonly(slot)) { | |
1713 | *writable = false; | |
1714 | writable = NULL; | |
1715 | } | |
1716 | ||
1717 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1718 | writable); | |
887c08ac | 1719 | } |
3520469d | 1720 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1721 | |
ba049e93 | 1722 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
1723 | bool *writable) |
1724 | { | |
e37afc6e PB |
1725 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1726 | write_fault, writable); | |
612819c3 MT |
1727 | } |
1728 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1729 | ||
ba049e93 | 1730 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1731 | { |
4d8b81ab | 1732 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1733 | } |
e37afc6e | 1734 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1735 | |
ba049e93 | 1736 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1737 | { |
4d8b81ab | 1738 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1739 | } |
037d92dc | 1740 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1741 | |
ba049e93 | 1742 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1743 | { |
1744 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1745 | } | |
1746 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1747 | ||
ba049e93 | 1748 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1749 | { |
1750 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1751 | } | |
1752 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1753 | ||
ba049e93 | 1754 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1755 | { |
1756 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1757 | } | |
1758 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1759 | ||
ba049e93 | 1760 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1761 | { |
1762 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1763 | } | |
1764 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1765 | ||
d9ef13c2 PB |
1766 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1767 | struct page **pages, int nr_pages) | |
48987781 XG |
1768 | { |
1769 | unsigned long addr; | |
076b925d | 1770 | gfn_t entry = 0; |
48987781 | 1771 | |
d9ef13c2 | 1772 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1773 | if (kvm_is_error_hva(addr)) |
1774 | return -1; | |
1775 | ||
1776 | if (entry < nr_pages) | |
1777 | return 0; | |
1778 | ||
1779 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1780 | } | |
1781 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1782 | ||
ba049e93 | 1783 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 1784 | { |
81c52c56 | 1785 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1786 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1787 | |
bf4bea8e | 1788 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1789 | WARN_ON(1); |
6cede2e6 | 1790 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1791 | } |
a2766325 XG |
1792 | |
1793 | return pfn_to_page(pfn); | |
1794 | } | |
1795 | ||
35149e21 AL |
1796 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1797 | { | |
ba049e93 | 1798 | kvm_pfn_t pfn; |
2e2e3738 AL |
1799 | |
1800 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1801 | |
a2766325 | 1802 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1803 | } |
1804 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1805 | ||
e45adf66 KA |
1806 | static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn, |
1807 | struct kvm_host_map *map) | |
1808 | { | |
1809 | kvm_pfn_t pfn; | |
1810 | void *hva = NULL; | |
1811 | struct page *page = KVM_UNMAPPED_PAGE; | |
1812 | ||
1813 | if (!map) | |
1814 | return -EINVAL; | |
1815 | ||
1816 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
1817 | if (is_error_noslot_pfn(pfn)) | |
1818 | return -EINVAL; | |
1819 | ||
1820 | if (pfn_valid(pfn)) { | |
1821 | page = pfn_to_page(pfn); | |
1822 | hva = kmap(page); | |
d30b214d | 1823 | #ifdef CONFIG_HAS_IOMEM |
e45adf66 KA |
1824 | } else { |
1825 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); | |
d30b214d | 1826 | #endif |
e45adf66 KA |
1827 | } |
1828 | ||
1829 | if (!hva) | |
1830 | return -EFAULT; | |
1831 | ||
1832 | map->page = page; | |
1833 | map->hva = hva; | |
1834 | map->pfn = pfn; | |
1835 | map->gfn = gfn; | |
1836 | ||
1837 | return 0; | |
1838 | } | |
1839 | ||
1840 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) | |
1841 | { | |
1842 | return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map); | |
1843 | } | |
1844 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
1845 | ||
1846 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, | |
1847 | bool dirty) | |
1848 | { | |
1849 | if (!map) | |
1850 | return; | |
1851 | ||
1852 | if (!map->hva) | |
1853 | return; | |
1854 | ||
b614c602 | 1855 | if (map->page != KVM_UNMAPPED_PAGE) |
e45adf66 | 1856 | kunmap(map->page); |
eb1f2f38 | 1857 | #ifdef CONFIG_HAS_IOMEM |
e45adf66 KA |
1858 | else |
1859 | memunmap(map->hva); | |
eb1f2f38 | 1860 | #endif |
e45adf66 KA |
1861 | |
1862 | if (dirty) { | |
1863 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); | |
1864 | kvm_release_pfn_dirty(map->pfn); | |
1865 | } else { | |
1866 | kvm_release_pfn_clean(map->pfn); | |
1867 | } | |
1868 | ||
1869 | map->hva = NULL; | |
1870 | map->page = NULL; | |
1871 | } | |
1872 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
1873 | ||
8e73485c PB |
1874 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1875 | { | |
ba049e93 | 1876 | kvm_pfn_t pfn; |
8e73485c PB |
1877 | |
1878 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1879 | ||
1880 | return kvm_pfn_to_page(pfn); | |
1881 | } | |
1882 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1883 | ||
b4231d61 IE |
1884 | void kvm_release_page_clean(struct page *page) |
1885 | { | |
32cad84f XG |
1886 | WARN_ON(is_error_page(page)); |
1887 | ||
35149e21 | 1888 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1889 | } |
1890 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1891 | ||
ba049e93 | 1892 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 1893 | { |
bf4bea8e | 1894 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1895 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1896 | } |
1897 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1898 | ||
b4231d61 | 1899 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1900 | { |
a2766325 XG |
1901 | WARN_ON(is_error_page(page)); |
1902 | ||
35149e21 AL |
1903 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1904 | } | |
1905 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1906 | ||
f7a6509f | 1907 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
1908 | { |
1909 | kvm_set_pfn_dirty(pfn); | |
1910 | kvm_release_pfn_clean(pfn); | |
1911 | } | |
f7a6509f | 1912 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 1913 | |
ba049e93 | 1914 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 1915 | { |
d29c03a5 ML |
1916 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
1917 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 1918 | } |
35149e21 AL |
1919 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1920 | ||
ba049e93 | 1921 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 1922 | { |
a78986aa | 1923 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 1924 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1925 | } |
1926 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1927 | ||
ba049e93 | 1928 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 1929 | { |
bf4bea8e | 1930 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1931 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1932 | } |
1933 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1934 | |
195aefde IE |
1935 | static int next_segment(unsigned long len, int offset) |
1936 | { | |
1937 | if (len > PAGE_SIZE - offset) | |
1938 | return PAGE_SIZE - offset; | |
1939 | else | |
1940 | return len; | |
1941 | } | |
1942 | ||
8e73485c PB |
1943 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1944 | void *data, int offset, int len) | |
195aefde | 1945 | { |
e0506bcb IE |
1946 | int r; |
1947 | unsigned long addr; | |
195aefde | 1948 | |
8e73485c | 1949 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1950 | if (kvm_is_error_hva(addr)) |
1951 | return -EFAULT; | |
3180a7fc | 1952 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1953 | if (r) |
195aefde | 1954 | return -EFAULT; |
195aefde IE |
1955 | return 0; |
1956 | } | |
8e73485c PB |
1957 | |
1958 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1959 | int len) | |
1960 | { | |
1961 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1962 | ||
1963 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1964 | } | |
195aefde IE |
1965 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1966 | ||
8e73485c PB |
1967 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1968 | int offset, int len) | |
1969 | { | |
1970 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1971 | ||
1972 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1973 | } | |
1974 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1975 | ||
195aefde IE |
1976 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1977 | { | |
1978 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
1979 | int seg; | |
1980 | int offset = offset_in_page(gpa); | |
1981 | int ret; | |
1982 | ||
1983 | while ((seg = next_segment(len, offset)) != 0) { | |
1984 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
1985 | if (ret < 0) | |
1986 | return ret; | |
1987 | offset = 0; | |
1988 | len -= seg; | |
1989 | data += seg; | |
1990 | ++gfn; | |
1991 | } | |
1992 | return 0; | |
1993 | } | |
1994 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
1995 | ||
8e73485c | 1996 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 1997 | { |
7ec54588 | 1998 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 1999 | int seg; |
7ec54588 | 2000 | int offset = offset_in_page(gpa); |
8e73485c PB |
2001 | int ret; |
2002 | ||
2003 | while ((seg = next_segment(len, offset)) != 0) { | |
2004 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2005 | if (ret < 0) | |
2006 | return ret; | |
2007 | offset = 0; | |
2008 | len -= seg; | |
2009 | data += seg; | |
2010 | ++gfn; | |
2011 | } | |
2012 | return 0; | |
2013 | } | |
2014 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2015 | |
8e73485c PB |
2016 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2017 | void *data, int offset, unsigned long len) | |
2018 | { | |
2019 | int r; | |
2020 | unsigned long addr; | |
2021 | ||
2022 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2023 | if (kvm_is_error_hva(addr)) |
2024 | return -EFAULT; | |
0aac03f0 | 2025 | pagefault_disable(); |
3180a7fc | 2026 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2027 | pagefault_enable(); |
7ec54588 MT |
2028 | if (r) |
2029 | return -EFAULT; | |
2030 | return 0; | |
2031 | } | |
7ec54588 | 2032 | |
8e73485c PB |
2033 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2034 | void *data, unsigned long len) | |
2035 | { | |
2036 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2037 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2038 | int offset = offset_in_page(gpa); | |
2039 | ||
2040 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2041 | } | |
2042 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2043 | ||
2044 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
2045 | const void *data, int offset, int len) | |
195aefde | 2046 | { |
e0506bcb IE |
2047 | int r; |
2048 | unsigned long addr; | |
195aefde | 2049 | |
251eb841 | 2050 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2051 | if (kvm_is_error_hva(addr)) |
2052 | return -EFAULT; | |
8b0cedff | 2053 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2054 | if (r) |
195aefde | 2055 | return -EFAULT; |
bc009e43 | 2056 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
2057 | return 0; |
2058 | } | |
8e73485c PB |
2059 | |
2060 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2061 | const void *data, int offset, int len) | |
2062 | { | |
2063 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2064 | ||
2065 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
2066 | } | |
195aefde IE |
2067 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2068 | ||
8e73485c PB |
2069 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2070 | const void *data, int offset, int len) | |
2071 | { | |
2072 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2073 | ||
2074 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
2075 | } | |
2076 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2077 | ||
195aefde IE |
2078 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2079 | unsigned long len) | |
2080 | { | |
2081 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2082 | int seg; | |
2083 | int offset = offset_in_page(gpa); | |
2084 | int ret; | |
2085 | ||
2086 | while ((seg = next_segment(len, offset)) != 0) { | |
2087 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2088 | if (ret < 0) | |
2089 | return ret; | |
2090 | offset = 0; | |
2091 | len -= seg; | |
2092 | data += seg; | |
2093 | ++gfn; | |
2094 | } | |
2095 | return 0; | |
2096 | } | |
ff651cb6 | 2097 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2098 | |
8e73485c PB |
2099 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2100 | unsigned long len) | |
2101 | { | |
2102 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2103 | int seg; | |
2104 | int offset = offset_in_page(gpa); | |
2105 | int ret; | |
2106 | ||
2107 | while ((seg = next_segment(len, offset)) != 0) { | |
2108 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2109 | if (ret < 0) | |
2110 | return ret; | |
2111 | offset = 0; | |
2112 | len -= seg; | |
2113 | data += seg; | |
2114 | ++gfn; | |
2115 | } | |
2116 | return 0; | |
2117 | } | |
2118 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2119 | ||
5a2d4365 PB |
2120 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2121 | struct gfn_to_hva_cache *ghc, | |
2122 | gpa_t gpa, unsigned long len) | |
49c7754c | 2123 | { |
49c7754c | 2124 | int offset = offset_in_page(gpa); |
8f964525 AH |
2125 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2126 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2127 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2128 | gfn_t nr_pages_avail; | |
f1b9dd5e | 2129 | int r = start_gfn <= end_gfn ? 0 : -EINVAL; |
49c7754c GN |
2130 | |
2131 | ghc->gpa = gpa; | |
2132 | ghc->generation = slots->generation; | |
8f964525 | 2133 | ghc->len = len; |
f1b9dd5e JM |
2134 | ghc->hva = KVM_HVA_ERR_BAD; |
2135 | ||
2136 | /* | |
2137 | * If the requested region crosses two memslots, we still | |
2138 | * verify that the entire region is valid here. | |
2139 | */ | |
2140 | while (!r && start_gfn <= end_gfn) { | |
2141 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); | |
2142 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2143 | &nr_pages_avail); | |
2144 | if (kvm_is_error_hva(ghc->hva)) | |
2145 | r = -EFAULT; | |
2146 | start_gfn += nr_pages_avail; | |
2147 | } | |
2148 | ||
2149 | /* Use the slow path for cross page reads and writes. */ | |
2150 | if (!r && nr_pages_needed == 1) | |
49c7754c | 2151 | ghc->hva += offset; |
f1b9dd5e | 2152 | else |
8f964525 | 2153 | ghc->memslot = NULL; |
f1b9dd5e JM |
2154 | |
2155 | return r; | |
49c7754c | 2156 | } |
5a2d4365 | 2157 | |
4e335d9e | 2158 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2159 | gpa_t gpa, unsigned long len) |
2160 | { | |
4e335d9e | 2161 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2162 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2163 | } | |
4e335d9e | 2164 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2165 | |
4e335d9e | 2166 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2167 | void *data, unsigned int offset, |
2168 | unsigned long len) | |
49c7754c | 2169 | { |
4e335d9e | 2170 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2171 | int r; |
4ec6e863 | 2172 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2173 | |
4ec6e863 | 2174 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2175 | |
49c7754c | 2176 | if (slots->generation != ghc->generation) |
5a2d4365 | 2177 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
2178 | |
2179 | if (unlikely(!ghc->memslot)) | |
4e335d9e | 2180 | return kvm_write_guest(kvm, gpa, data, len); |
49c7754c GN |
2181 | |
2182 | if (kvm_is_error_hva(ghc->hva)) | |
2183 | return -EFAULT; | |
2184 | ||
4ec6e863 | 2185 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2186 | if (r) |
2187 | return -EFAULT; | |
4ec6e863 | 2188 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2189 | |
2190 | return 0; | |
2191 | } | |
4e335d9e | 2192 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 2193 | |
4e335d9e PB |
2194 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2195 | void *data, unsigned long len) | |
4ec6e863 | 2196 | { |
4e335d9e | 2197 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 2198 | } |
4e335d9e | 2199 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 2200 | |
4e335d9e PB |
2201 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2202 | void *data, unsigned long len) | |
e03b644f | 2203 | { |
4e335d9e | 2204 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f GN |
2205 | int r; |
2206 | ||
8f964525 AH |
2207 | BUG_ON(len > ghc->len); |
2208 | ||
e03b644f | 2209 | if (slots->generation != ghc->generation) |
5a2d4365 | 2210 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
2211 | |
2212 | if (unlikely(!ghc->memslot)) | |
4e335d9e | 2213 | return kvm_read_guest(kvm, ghc->gpa, data, len); |
e03b644f GN |
2214 | |
2215 | if (kvm_is_error_hva(ghc->hva)) | |
2216 | return -EFAULT; | |
2217 | ||
2218 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
2219 | if (r) | |
2220 | return -EFAULT; | |
2221 | ||
2222 | return 0; | |
2223 | } | |
4e335d9e | 2224 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 2225 | |
195aefde IE |
2226 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
2227 | { | |
8a3caa6d HC |
2228 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
2229 | ||
2230 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
2231 | } |
2232 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
2233 | ||
2234 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
2235 | { | |
2236 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2237 | int seg; | |
2238 | int offset = offset_in_page(gpa); | |
2239 | int ret; | |
2240 | ||
bfda0e84 | 2241 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
2242 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
2243 | if (ret < 0) | |
2244 | return ret; | |
2245 | offset = 0; | |
2246 | len -= seg; | |
2247 | ++gfn; | |
2248 | } | |
2249 | return 0; | |
2250 | } | |
2251 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2252 | ||
bc009e43 | 2253 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 2254 | gfn_t gfn) |
6aa8b732 | 2255 | { |
7e9d619d RR |
2256 | if (memslot && memslot->dirty_bitmap) { |
2257 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 2258 | |
b74ca3b3 | 2259 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
2260 | } |
2261 | } | |
2262 | ||
49c7754c GN |
2263 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2264 | { | |
2265 | struct kvm_memory_slot *memslot; | |
2266 | ||
2267 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 2268 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 2269 | } |
2ba9f0d8 | 2270 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2271 | |
8e73485c PB |
2272 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2273 | { | |
2274 | struct kvm_memory_slot *memslot; | |
2275 | ||
2276 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2277 | mark_page_dirty_in_slot(memslot, gfn); | |
2278 | } | |
2279 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2280 | ||
20b7035c JS |
2281 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
2282 | { | |
2283 | if (!vcpu->sigset_active) | |
2284 | return; | |
2285 | ||
2286 | /* | |
2287 | * This does a lockless modification of ->real_blocked, which is fine | |
2288 | * because, only current can change ->real_blocked and all readers of | |
2289 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
2290 | * of ->blocked. | |
2291 | */ | |
2292 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
2293 | } | |
2294 | ||
2295 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
2296 | { | |
2297 | if (!vcpu->sigset_active) | |
2298 | return; | |
2299 | ||
2300 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
2301 | sigemptyset(¤t->real_blocked); | |
2302 | } | |
2303 | ||
aca6ff29 WL |
2304 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
2305 | { | |
dee339b5 | 2306 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 2307 | |
2cbd7824 | 2308 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 2309 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 2310 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
2311 | if (!grow) |
2312 | goto out; | |
2313 | ||
dee339b5 NW |
2314 | val *= grow; |
2315 | if (val < grow_start) | |
2316 | val = grow_start; | |
aca6ff29 | 2317 | |
313f636d DM |
2318 | if (val > halt_poll_ns) |
2319 | val = halt_poll_ns; | |
2320 | ||
aca6ff29 | 2321 | vcpu->halt_poll_ns = val; |
7fa08e71 | 2322 | out: |
2cbd7824 | 2323 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2324 | } |
2325 | ||
2326 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
2327 | { | |
6b6de68c | 2328 | unsigned int old, val, shrink; |
aca6ff29 | 2329 | |
2cbd7824 | 2330 | old = val = vcpu->halt_poll_ns; |
6b6de68c CB |
2331 | shrink = READ_ONCE(halt_poll_ns_shrink); |
2332 | if (shrink == 0) | |
aca6ff29 WL |
2333 | val = 0; |
2334 | else | |
6b6de68c | 2335 | val /= shrink; |
aca6ff29 WL |
2336 | |
2337 | vcpu->halt_poll_ns = val; | |
2cbd7824 | 2338 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2339 | } |
2340 | ||
f7819512 PB |
2341 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
2342 | { | |
50c28f21 JS |
2343 | int ret = -EINTR; |
2344 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
2345 | ||
f7819512 PB |
2346 | if (kvm_arch_vcpu_runnable(vcpu)) { |
2347 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 2348 | goto out; |
f7819512 PB |
2349 | } |
2350 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 2351 | goto out; |
f7819512 | 2352 | if (signal_pending(current)) |
50c28f21 | 2353 | goto out; |
f7819512 | 2354 | |
50c28f21 JS |
2355 | ret = 0; |
2356 | out: | |
2357 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
2358 | return ret; | |
f7819512 PB |
2359 | } |
2360 | ||
b6958ce4 ED |
2361 | /* |
2362 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
2363 | */ | |
8776e519 | 2364 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 2365 | { |
f7819512 | 2366 | ktime_t start, cur; |
8577370f | 2367 | DECLARE_SWAITQUEUE(wait); |
f7819512 | 2368 | bool waited = false; |
aca6ff29 | 2369 | u64 block_ns; |
f7819512 | 2370 | |
07ab0f8d MZ |
2371 | kvm_arch_vcpu_blocking(vcpu); |
2372 | ||
f7819512 | 2373 | start = cur = ktime_get(); |
cdd6ad3a | 2374 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 2375 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 2376 | |
62bea5bf | 2377 | ++vcpu->stat.halt_attempted_poll; |
f7819512 PB |
2378 | do { |
2379 | /* | |
2380 | * This sets KVM_REQ_UNHALT if an interrupt | |
2381 | * arrives. | |
2382 | */ | |
2383 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
2384 | ++vcpu->stat.halt_successful_poll; | |
3491caf2 CB |
2385 | if (!vcpu_valid_wakeup(vcpu)) |
2386 | ++vcpu->stat.halt_poll_invalid; | |
f7819512 PB |
2387 | goto out; |
2388 | } | |
2389 | cur = ktime_get(); | |
2390 | } while (single_task_running() && ktime_before(cur, stop)); | |
2391 | } | |
e5c239cf MT |
2392 | |
2393 | for (;;) { | |
b3dae109 | 2394 | prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
e5c239cf | 2395 | |
f7819512 | 2396 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
2397 | break; |
2398 | ||
f7819512 | 2399 | waited = true; |
b6958ce4 | 2400 | schedule(); |
b6958ce4 | 2401 | } |
d3bef15f | 2402 | |
8577370f | 2403 | finish_swait(&vcpu->wq, &wait); |
f7819512 | 2404 | cur = ktime_get(); |
f7819512 | 2405 | out: |
07ab0f8d | 2406 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
2407 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
2408 | ||
44551b2f WL |
2409 | if (!kvm_arch_no_poll(vcpu)) { |
2410 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 2411 | shrink_halt_poll_ns(vcpu); |
44551b2f WL |
2412 | } else if (halt_poll_ns) { |
2413 | if (block_ns <= vcpu->halt_poll_ns) | |
2414 | ; | |
2415 | /* we had a long block, shrink polling */ | |
2416 | else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns) | |
2417 | shrink_halt_poll_ns(vcpu); | |
2418 | /* we had a short halt and our poll time is too small */ | |
2419 | else if (vcpu->halt_poll_ns < halt_poll_ns && | |
2420 | block_ns < halt_poll_ns) | |
2421 | grow_halt_poll_ns(vcpu); | |
2422 | } else { | |
2423 | vcpu->halt_poll_ns = 0; | |
2424 | } | |
2425 | } | |
aca6ff29 | 2426 | |
3491caf2 CB |
2427 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
2428 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 2429 | } |
2ba9f0d8 | 2430 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 2431 | |
178f02ff | 2432 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2433 | { |
8577370f | 2434 | struct swait_queue_head *wqp; |
b6d33834 CD |
2435 | |
2436 | wqp = kvm_arch_vcpu_wq(vcpu); | |
5e0018b3 | 2437 | if (swq_has_sleeper(wqp)) { |
b3dae109 | 2438 | swake_up_one(wqp); |
d73eb57b | 2439 | WRITE_ONCE(vcpu->ready, true); |
b6d33834 | 2440 | ++vcpu->stat.halt_wakeup; |
178f02ff | 2441 | return true; |
b6d33834 CD |
2442 | } |
2443 | ||
178f02ff | 2444 | return false; |
dd1a4cc1 RK |
2445 | } |
2446 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2447 | ||
0266c894 | 2448 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
2449 | /* |
2450 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2451 | */ | |
2452 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2453 | { | |
2454 | int me; | |
2455 | int cpu = vcpu->cpu; | |
2456 | ||
178f02ff RK |
2457 | if (kvm_vcpu_wake_up(vcpu)) |
2458 | return; | |
2459 | ||
b6d33834 CD |
2460 | me = get_cpu(); |
2461 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2462 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2463 | smp_send_reschedule(cpu); | |
2464 | put_cpu(); | |
2465 | } | |
a20ed54d | 2466 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 2467 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2468 | |
fa93384f | 2469 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2470 | { |
2471 | struct pid *pid; | |
2472 | struct task_struct *task = NULL; | |
fa93384f | 2473 | int ret = 0; |
41628d33 KW |
2474 | |
2475 | rcu_read_lock(); | |
2476 | pid = rcu_dereference(target->pid); | |
2477 | if (pid) | |
27fbe64b | 2478 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2479 | rcu_read_unlock(); |
2480 | if (!task) | |
c45c528e | 2481 | return ret; |
c45c528e | 2482 | ret = yield_to(task, 1); |
41628d33 | 2483 | put_task_struct(task); |
c45c528e R |
2484 | |
2485 | return ret; | |
41628d33 KW |
2486 | } |
2487 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2488 | ||
06e48c51 R |
2489 | /* |
2490 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2491 | * Most eligible candidate to yield is decided by following heuristics: | |
2492 | * | |
2493 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2494 | * (preempted lock holder), indicated by @in_spin_loop. | |
2495 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2496 | * | |
2497 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2498 | * chance last time (mostly it has become eligible now since we have probably | |
2499 | * yielded to lockholder in last iteration. This is done by toggling | |
2500 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2501 | * | |
2502 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2503 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2504 | * burning. Giving priority for a potential lock-holder increases lock | |
2505 | * progress. | |
2506 | * | |
2507 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2508 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2509 | * and continue with next VCPU and so on. | |
2510 | */ | |
7940876e | 2511 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2512 | { |
4a55dd72 | 2513 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2514 | bool eligible; |
2515 | ||
2516 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2517 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2518 | |
2519 | if (vcpu->spin_loop.in_spin_loop) | |
2520 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2521 | ||
2522 | return eligible; | |
4a55dd72 SW |
2523 | #else |
2524 | return true; | |
06e48c51 | 2525 | #endif |
4a55dd72 | 2526 | } |
c45c528e | 2527 | |
17e433b5 WL |
2528 | /* |
2529 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
2530 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
2531 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
2532 | */ | |
2533 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
2534 | { | |
2535 | return kvm_arch_vcpu_runnable(vcpu); | |
2536 | } | |
2537 | ||
2538 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
2539 | { | |
2540 | if (kvm_arch_dy_runnable(vcpu)) | |
2541 | return true; | |
2542 | ||
2543 | #ifdef CONFIG_KVM_ASYNC_PF | |
2544 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
2545 | return true; | |
2546 | #endif | |
2547 | ||
2548 | return false; | |
2549 | } | |
2550 | ||
199b5763 | 2551 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 2552 | { |
217ece61 RR |
2553 | struct kvm *kvm = me->kvm; |
2554 | struct kvm_vcpu *vcpu; | |
2555 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2556 | int yielded = 0; | |
c45c528e | 2557 | int try = 3; |
217ece61 RR |
2558 | int pass; |
2559 | int i; | |
d255f4f2 | 2560 | |
4c088493 | 2561 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2562 | /* |
2563 | * We boost the priority of a VCPU that is runnable but not | |
2564 | * currently running, because it got preempted by something | |
2565 | * else and called schedule in __vcpu_run. Hopefully that | |
2566 | * VCPU is holding the lock that we need and will release it. | |
2567 | * We approximate round-robin by starting at the last boosted VCPU. | |
2568 | */ | |
c45c528e | 2569 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2570 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2571 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2572 | i = last_boosted_vcpu; |
2573 | continue; | |
2574 | } else if (pass && i > last_boosted_vcpu) | |
2575 | break; | |
d73eb57b | 2576 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 2577 | continue; |
217ece61 RR |
2578 | if (vcpu == me) |
2579 | continue; | |
17e433b5 | 2580 | if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 2581 | continue; |
046ddeed WL |
2582 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
2583 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 2584 | continue; |
06e48c51 R |
2585 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2586 | continue; | |
c45c528e R |
2587 | |
2588 | yielded = kvm_vcpu_yield_to(vcpu); | |
2589 | if (yielded > 0) { | |
217ece61 | 2590 | kvm->last_boosted_vcpu = i; |
217ece61 | 2591 | break; |
c45c528e R |
2592 | } else if (yielded < 0) { |
2593 | try--; | |
2594 | if (!try) | |
2595 | break; | |
217ece61 | 2596 | } |
217ece61 RR |
2597 | } |
2598 | } | |
4c088493 | 2599 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2600 | |
2601 | /* Ensure vcpu is not eligible during next spinloop */ | |
2602 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2603 | } |
2604 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2605 | ||
1499fa80 | 2606 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 2607 | { |
11bac800 | 2608 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
2609 | struct page *page; |
2610 | ||
e4a533a4 | 2611 | if (vmf->pgoff == 0) |
039576c0 | 2612 | page = virt_to_page(vcpu->run); |
09566765 | 2613 | #ifdef CONFIG_X86 |
e4a533a4 | 2614 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2615 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 2616 | #endif |
4b4357e0 | 2617 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
2618 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
2619 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2620 | #endif |
039576c0 | 2621 | else |
5b1c1493 | 2622 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2623 | get_page(page); |
e4a533a4 NP |
2624 | vmf->page = page; |
2625 | return 0; | |
9a2bb7f4 AK |
2626 | } |
2627 | ||
f0f37e2f | 2628 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2629 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2630 | }; |
2631 | ||
2632 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2633 | { | |
2634 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2635 | return 0; | |
2636 | } | |
2637 | ||
bccf2150 AK |
2638 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2639 | { | |
2640 | struct kvm_vcpu *vcpu = filp->private_data; | |
2641 | ||
45b5939e | 2642 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
66c0b394 | 2643 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2644 | return 0; |
2645 | } | |
2646 | ||
3d3aab1b | 2647 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2648 | .release = kvm_vcpu_release, |
2649 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 2650 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2651 | .llseek = noop_llseek, |
7ddfd3e0 | 2652 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
2653 | }; |
2654 | ||
2655 | /* | |
2656 | * Allocates an inode for the vcpu. | |
2657 | */ | |
2658 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2659 | { | |
e46b4692 MY |
2660 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
2661 | ||
2662 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
2663 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
2664 | } |
2665 | ||
3e7093d0 | 2666 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 2667 | { |
741cbbae | 2668 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
45b5939e | 2669 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 2670 | |
45b5939e | 2671 | if (!debugfs_initialized()) |
3e7093d0 | 2672 | return; |
45b5939e LC |
2673 | |
2674 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
2675 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, | |
3e7093d0 | 2676 | vcpu->kvm->debugfs_dentry); |
45b5939e | 2677 | |
3e7093d0 | 2678 | kvm_arch_create_vcpu_debugfs(vcpu); |
741cbbae | 2679 | #endif |
45b5939e LC |
2680 | } |
2681 | ||
c5ea7660 AK |
2682 | /* |
2683 | * Creates some virtual cpus. Good luck creating more than one. | |
2684 | */ | |
73880c80 | 2685 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2686 | { |
2687 | int r; | |
e09fefde | 2688 | struct kvm_vcpu *vcpu; |
8bd826d6 | 2689 | struct page *page; |
c5ea7660 | 2690 | |
0b1b1dfd | 2691 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
2692 | return -EINVAL; |
2693 | ||
6c7caebc PB |
2694 | mutex_lock(&kvm->lock); |
2695 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
2696 | mutex_unlock(&kvm->lock); | |
2697 | return -EINVAL; | |
2698 | } | |
2699 | ||
2700 | kvm->created_vcpus++; | |
2701 | mutex_unlock(&kvm->lock); | |
2702 | ||
897cc38e SC |
2703 | r = kvm_arch_vcpu_precreate(kvm, id); |
2704 | if (r) | |
2705 | goto vcpu_decrement; | |
2706 | ||
e529ef66 SC |
2707 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
2708 | if (!vcpu) { | |
2709 | r = -ENOMEM; | |
6c7caebc PB |
2710 | goto vcpu_decrement; |
2711 | } | |
c5ea7660 | 2712 | |
fcd97ad5 | 2713 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
8bd826d6 SC |
2714 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
2715 | if (!page) { | |
2716 | r = -ENOMEM; | |
e529ef66 | 2717 | goto vcpu_free; |
8bd826d6 SC |
2718 | } |
2719 | vcpu->run = page_address(page); | |
2720 | ||
2721 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
2722 | |
2723 | r = kvm_arch_vcpu_create(vcpu); | |
2724 | if (r) | |
8bd826d6 | 2725 | goto vcpu_free_run_page; |
e529ef66 | 2726 | |
3e7093d0 | 2727 | kvm_create_vcpu_debugfs(vcpu); |
45b5939e | 2728 | |
11ec2804 | 2729 | mutex_lock(&kvm->lock); |
e09fefde DH |
2730 | if (kvm_get_vcpu_by_id(kvm, id)) { |
2731 | r = -EEXIST; | |
2732 | goto unlock_vcpu_destroy; | |
2733 | } | |
73880c80 | 2734 | |
8750e72a RK |
2735 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
2736 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 2737 | |
fb3f0f51 | 2738 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2739 | kvm_get_kvm(kvm); |
bccf2150 | 2740 | r = create_vcpu_fd(vcpu); |
73880c80 | 2741 | if (r < 0) { |
149487bd | 2742 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 2743 | goto unlock_vcpu_destroy; |
73880c80 GN |
2744 | } |
2745 | ||
8750e72a | 2746 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
2747 | |
2748 | /* | |
2749 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2750 | * before kvm->online_vcpu's incremented value. | |
2751 | */ | |
73880c80 GN |
2752 | smp_wmb(); |
2753 | atomic_inc(&kvm->online_vcpus); | |
2754 | ||
73880c80 | 2755 | mutex_unlock(&kvm->lock); |
42897d86 | 2756 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2757 | return r; |
39c3b86e | 2758 | |
d780592b | 2759 | unlock_vcpu_destroy: |
7d8fece6 | 2760 | mutex_unlock(&kvm->lock); |
45b5939e | 2761 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
d40ccc62 | 2762 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
2763 | vcpu_free_run_page: |
2764 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
2765 | vcpu_free: |
2766 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
2767 | vcpu_decrement: |
2768 | mutex_lock(&kvm->lock); | |
2769 | kvm->created_vcpus--; | |
2770 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
2771 | return r; |
2772 | } | |
2773 | ||
1961d276 AK |
2774 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2775 | { | |
2776 | if (sigset) { | |
2777 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2778 | vcpu->sigset_active = 1; | |
2779 | vcpu->sigset = *sigset; | |
2780 | } else | |
2781 | vcpu->sigset_active = 0; | |
2782 | return 0; | |
2783 | } | |
2784 | ||
bccf2150 AK |
2785 | static long kvm_vcpu_ioctl(struct file *filp, |
2786 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2787 | { |
bccf2150 | 2788 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2789 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2790 | int r; |
fa3795a7 DH |
2791 | struct kvm_fpu *fpu = NULL; |
2792 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2793 | |
6d4e4c4f AK |
2794 | if (vcpu->kvm->mm != current->mm) |
2795 | return -EIO; | |
2122ff5e | 2796 | |
2ea75be3 DM |
2797 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2798 | return -EINVAL; | |
2799 | ||
2122ff5e | 2800 | /* |
5cb0944c PB |
2801 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
2802 | * execution; mutex_lock() would break them. | |
2122ff5e | 2803 | */ |
5cb0944c PB |
2804 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
2805 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 2806 | return r; |
2122ff5e | 2807 | |
ec7660cc CD |
2808 | if (mutex_lock_killable(&vcpu->mutex)) |
2809 | return -EINTR; | |
6aa8b732 | 2810 | switch (ioctl) { |
0e4524a5 CB |
2811 | case KVM_RUN: { |
2812 | struct pid *oldpid; | |
f0fe5108 AK |
2813 | r = -EINVAL; |
2814 | if (arg) | |
2815 | goto out; | |
0e4524a5 | 2816 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 2817 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 2818 | /* The thread running this VCPU changed. */ |
bd2a6394 | 2819 | struct pid *newpid; |
f95ef0cd | 2820 | |
bd2a6394 CD |
2821 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
2822 | if (r) | |
2823 | break; | |
2824 | ||
2825 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
2826 | rcu_assign_pointer(vcpu->pid, newpid); |
2827 | if (oldpid) | |
2828 | synchronize_rcu(); | |
2829 | put_pid(oldpid); | |
2830 | } | |
b6c7a5dc | 2831 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2832 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2833 | break; |
0e4524a5 | 2834 | } |
6aa8b732 | 2835 | case KVM_GET_REGS: { |
3e4bb3ac | 2836 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2837 | |
3e4bb3ac | 2838 | r = -ENOMEM; |
b12ce36a | 2839 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 2840 | if (!kvm_regs) |
6aa8b732 | 2841 | goto out; |
3e4bb3ac XZ |
2842 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2843 | if (r) | |
2844 | goto out_free1; | |
6aa8b732 | 2845 | r = -EFAULT; |
3e4bb3ac XZ |
2846 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2847 | goto out_free1; | |
6aa8b732 | 2848 | r = 0; |
3e4bb3ac XZ |
2849 | out_free1: |
2850 | kfree(kvm_regs); | |
6aa8b732 AK |
2851 | break; |
2852 | } | |
2853 | case KVM_SET_REGS: { | |
3e4bb3ac | 2854 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2855 | |
3e4bb3ac | 2856 | r = -ENOMEM; |
ff5c2c03 SL |
2857 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2858 | if (IS_ERR(kvm_regs)) { | |
2859 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2860 | goto out; |
ff5c2c03 | 2861 | } |
3e4bb3ac | 2862 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2863 | kfree(kvm_regs); |
6aa8b732 AK |
2864 | break; |
2865 | } | |
2866 | case KVM_GET_SREGS: { | |
b12ce36a BG |
2867 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
2868 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
2869 | r = -ENOMEM; |
2870 | if (!kvm_sregs) | |
2871 | goto out; | |
2872 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2873 | if (r) |
2874 | goto out; | |
2875 | r = -EFAULT; | |
fa3795a7 | 2876 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2877 | goto out; |
2878 | r = 0; | |
2879 | break; | |
2880 | } | |
2881 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2882 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2883 | if (IS_ERR(kvm_sregs)) { | |
2884 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2885 | kvm_sregs = NULL; |
6aa8b732 | 2886 | goto out; |
ff5c2c03 | 2887 | } |
fa3795a7 | 2888 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2889 | break; |
2890 | } | |
62d9f0db MT |
2891 | case KVM_GET_MP_STATE: { |
2892 | struct kvm_mp_state mp_state; | |
2893 | ||
2894 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2895 | if (r) | |
2896 | goto out; | |
2897 | r = -EFAULT; | |
893bdbf1 | 2898 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2899 | goto out; |
2900 | r = 0; | |
2901 | break; | |
2902 | } | |
2903 | case KVM_SET_MP_STATE: { | |
2904 | struct kvm_mp_state mp_state; | |
2905 | ||
2906 | r = -EFAULT; | |
893bdbf1 | 2907 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2908 | goto out; |
2909 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2910 | break; |
2911 | } | |
6aa8b732 AK |
2912 | case KVM_TRANSLATE: { |
2913 | struct kvm_translation tr; | |
2914 | ||
2915 | r = -EFAULT; | |
893bdbf1 | 2916 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2917 | goto out; |
8b006791 | 2918 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2919 | if (r) |
2920 | goto out; | |
2921 | r = -EFAULT; | |
893bdbf1 | 2922 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2923 | goto out; |
2924 | r = 0; | |
2925 | break; | |
2926 | } | |
d0bfb940 JK |
2927 | case KVM_SET_GUEST_DEBUG: { |
2928 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2929 | |
2930 | r = -EFAULT; | |
893bdbf1 | 2931 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2932 | goto out; |
d0bfb940 | 2933 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2934 | break; |
2935 | } | |
1961d276 AK |
2936 | case KVM_SET_SIGNAL_MASK: { |
2937 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2938 | struct kvm_signal_mask kvm_sigmask; | |
2939 | sigset_t sigset, *p; | |
2940 | ||
2941 | p = NULL; | |
2942 | if (argp) { | |
2943 | r = -EFAULT; | |
2944 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2945 | sizeof(kvm_sigmask))) |
1961d276 AK |
2946 | goto out; |
2947 | r = -EINVAL; | |
893bdbf1 | 2948 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2949 | goto out; |
2950 | r = -EFAULT; | |
2951 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2952 | sizeof(sigset))) |
1961d276 AK |
2953 | goto out; |
2954 | p = &sigset; | |
2955 | } | |
376d41ff | 2956 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2957 | break; |
2958 | } | |
b8836737 | 2959 | case KVM_GET_FPU: { |
b12ce36a | 2960 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
2961 | r = -ENOMEM; |
2962 | if (!fpu) | |
2963 | goto out; | |
2964 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2965 | if (r) |
2966 | goto out; | |
2967 | r = -EFAULT; | |
fa3795a7 | 2968 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2969 | goto out; |
2970 | r = 0; | |
2971 | break; | |
2972 | } | |
2973 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2974 | fpu = memdup_user(argp, sizeof(*fpu)); |
2975 | if (IS_ERR(fpu)) { | |
2976 | r = PTR_ERR(fpu); | |
18595411 | 2977 | fpu = NULL; |
b8836737 | 2978 | goto out; |
ff5c2c03 | 2979 | } |
fa3795a7 | 2980 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2981 | break; |
2982 | } | |
bccf2150 | 2983 | default: |
313a3dc7 | 2984 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
2985 | } |
2986 | out: | |
ec7660cc | 2987 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
2988 | kfree(fpu); |
2989 | kfree(kvm_sregs); | |
bccf2150 AK |
2990 | return r; |
2991 | } | |
2992 | ||
de8e5d74 | 2993 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
2994 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
2995 | unsigned int ioctl, unsigned long arg) | |
2996 | { | |
2997 | struct kvm_vcpu *vcpu = filp->private_data; | |
2998 | void __user *argp = compat_ptr(arg); | |
2999 | int r; | |
3000 | ||
3001 | if (vcpu->kvm->mm != current->mm) | |
3002 | return -EIO; | |
3003 | ||
3004 | switch (ioctl) { | |
3005 | case KVM_SET_SIGNAL_MASK: { | |
3006 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3007 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3008 | sigset_t sigset; |
3009 | ||
3010 | if (argp) { | |
3011 | r = -EFAULT; | |
3012 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3013 | sizeof(kvm_sigmask))) |
1dda606c AG |
3014 | goto out; |
3015 | r = -EINVAL; | |
3968cf62 | 3016 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3017 | goto out; |
3018 | r = -EFAULT; | |
3968cf62 | 3019 | if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset)) |
1dda606c | 3020 | goto out; |
760a9a30 AC |
3021 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3022 | } else | |
3023 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3024 | break; |
3025 | } | |
3026 | default: | |
3027 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3028 | } | |
3029 | ||
3030 | out: | |
3031 | return r; | |
3032 | } | |
3033 | #endif | |
3034 | ||
a1cd3f08 CLG |
3035 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3036 | { | |
3037 | struct kvm_device *dev = filp->private_data; | |
3038 | ||
3039 | if (dev->ops->mmap) | |
3040 | return dev->ops->mmap(dev, vma); | |
3041 | ||
3042 | return -ENODEV; | |
3043 | } | |
3044 | ||
852b6d57 SW |
3045 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3046 | int (*accessor)(struct kvm_device *dev, | |
3047 | struct kvm_device_attr *attr), | |
3048 | unsigned long arg) | |
3049 | { | |
3050 | struct kvm_device_attr attr; | |
3051 | ||
3052 | if (!accessor) | |
3053 | return -EPERM; | |
3054 | ||
3055 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
3056 | return -EFAULT; | |
3057 | ||
3058 | return accessor(dev, &attr); | |
3059 | } | |
3060 | ||
3061 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
3062 | unsigned long arg) | |
3063 | { | |
3064 | struct kvm_device *dev = filp->private_data; | |
3065 | ||
ddba9180 SC |
3066 | if (dev->kvm->mm != current->mm) |
3067 | return -EIO; | |
3068 | ||
852b6d57 SW |
3069 | switch (ioctl) { |
3070 | case KVM_SET_DEVICE_ATTR: | |
3071 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
3072 | case KVM_GET_DEVICE_ATTR: | |
3073 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
3074 | case KVM_HAS_DEVICE_ATTR: | |
3075 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
3076 | default: | |
3077 | if (dev->ops->ioctl) | |
3078 | return dev->ops->ioctl(dev, ioctl, arg); | |
3079 | ||
3080 | return -ENOTTY; | |
3081 | } | |
3082 | } | |
3083 | ||
852b6d57 SW |
3084 | static int kvm_device_release(struct inode *inode, struct file *filp) |
3085 | { | |
3086 | struct kvm_device *dev = filp->private_data; | |
3087 | struct kvm *kvm = dev->kvm; | |
3088 | ||
2bde9b3e CLG |
3089 | if (dev->ops->release) { |
3090 | mutex_lock(&kvm->lock); | |
3091 | list_del(&dev->vm_node); | |
3092 | dev->ops->release(dev); | |
3093 | mutex_unlock(&kvm->lock); | |
3094 | } | |
3095 | ||
852b6d57 SW |
3096 | kvm_put_kvm(kvm); |
3097 | return 0; | |
3098 | } | |
3099 | ||
3100 | static const struct file_operations kvm_device_fops = { | |
3101 | .unlocked_ioctl = kvm_device_ioctl, | |
3102 | .release = kvm_device_release, | |
7ddfd3e0 | 3103 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 3104 | .mmap = kvm_device_mmap, |
852b6d57 SW |
3105 | }; |
3106 | ||
3107 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
3108 | { | |
3109 | if (filp->f_op != &kvm_device_fops) | |
3110 | return NULL; | |
3111 | ||
3112 | return filp->private_data; | |
3113 | } | |
3114 | ||
8538cb22 | 3115 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 3116 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
3117 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
3118 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 3119 | #endif |
d60eacb0 WD |
3120 | }; |
3121 | ||
8538cb22 | 3122 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
3123 | { |
3124 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
3125 | return -ENOSPC; | |
3126 | ||
3127 | if (kvm_device_ops_table[type] != NULL) | |
3128 | return -EEXIST; | |
3129 | ||
3130 | kvm_device_ops_table[type] = ops; | |
3131 | return 0; | |
3132 | } | |
3133 | ||
571ee1b6 WL |
3134 | void kvm_unregister_device_ops(u32 type) |
3135 | { | |
3136 | if (kvm_device_ops_table[type] != NULL) | |
3137 | kvm_device_ops_table[type] = NULL; | |
3138 | } | |
3139 | ||
852b6d57 SW |
3140 | static int kvm_ioctl_create_device(struct kvm *kvm, |
3141 | struct kvm_create_device *cd) | |
3142 | { | |
8538cb22 | 3143 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
3144 | struct kvm_device *dev; |
3145 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 3146 | int type; |
852b6d57 SW |
3147 | int ret; |
3148 | ||
d60eacb0 WD |
3149 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
3150 | return -ENODEV; | |
3151 | ||
1d487e9b PB |
3152 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
3153 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 3154 | if (ops == NULL) |
852b6d57 | 3155 | return -ENODEV; |
852b6d57 SW |
3156 | |
3157 | if (test) | |
3158 | return 0; | |
3159 | ||
b12ce36a | 3160 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
3161 | if (!dev) |
3162 | return -ENOMEM; | |
3163 | ||
3164 | dev->ops = ops; | |
3165 | dev->kvm = kvm; | |
852b6d57 | 3166 | |
a28ebea2 | 3167 | mutex_lock(&kvm->lock); |
1d487e9b | 3168 | ret = ops->create(dev, type); |
852b6d57 | 3169 | if (ret < 0) { |
a28ebea2 | 3170 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
3171 | kfree(dev); |
3172 | return ret; | |
3173 | } | |
a28ebea2 CD |
3174 | list_add(&dev->vm_node, &kvm->devices); |
3175 | mutex_unlock(&kvm->lock); | |
852b6d57 | 3176 | |
023e9fdd CD |
3177 | if (ops->init) |
3178 | ops->init(dev); | |
3179 | ||
cfa39381 | 3180 | kvm_get_kvm(kvm); |
24009b05 | 3181 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 3182 | if (ret < 0) { |
149487bd | 3183 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
3184 | mutex_lock(&kvm->lock); |
3185 | list_del(&dev->vm_node); | |
3186 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 3187 | ops->destroy(dev); |
852b6d57 SW |
3188 | return ret; |
3189 | } | |
3190 | ||
852b6d57 SW |
3191 | cd->fd = ret; |
3192 | return 0; | |
3193 | } | |
3194 | ||
92b591a4 AG |
3195 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
3196 | { | |
3197 | switch (arg) { | |
3198 | case KVM_CAP_USER_MEMORY: | |
3199 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
3200 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
3201 | case KVM_CAP_INTERNAL_ERROR_DATA: |
3202 | #ifdef CONFIG_HAVE_KVM_MSI | |
3203 | case KVM_CAP_SIGNAL_MSI: | |
3204 | #endif | |
297e2105 | 3205 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 3206 | case KVM_CAP_IRQFD: |
92b591a4 AG |
3207 | case KVM_CAP_IRQFD_RESAMPLE: |
3208 | #endif | |
e9ea5069 | 3209 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 3210 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 3211 | case KVM_CAP_ENABLE_CAP_VM: |
2a31b9db | 3212 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
d7547c55 | 3213 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
2a31b9db | 3214 | #endif |
92b591a4 | 3215 | return 1; |
4b4357e0 | 3216 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
3217 | case KVM_CAP_COALESCED_MMIO: |
3218 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
3219 | case KVM_CAP_COALESCED_PIO: |
3220 | return 1; | |
30422558 | 3221 | #endif |
92b591a4 AG |
3222 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3223 | case KVM_CAP_IRQ_ROUTING: | |
3224 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
3225 | #endif |
3226 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
3227 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
3228 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 3229 | #endif |
c110ae57 PB |
3230 | case KVM_CAP_NR_MEMSLOTS: |
3231 | return KVM_USER_MEM_SLOTS; | |
92b591a4 AG |
3232 | default: |
3233 | break; | |
3234 | } | |
3235 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
3236 | } | |
3237 | ||
e5d83c74 PB |
3238 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
3239 | struct kvm_enable_cap *cap) | |
3240 | { | |
3241 | return -EINVAL; | |
3242 | } | |
3243 | ||
3244 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
3245 | struct kvm_enable_cap *cap) | |
3246 | { | |
3247 | switch (cap->cap) { | |
2a31b9db | 3248 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
d7547c55 | 3249 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
2a31b9db PB |
3250 | if (cap->flags || (cap->args[0] & ~1)) |
3251 | return -EINVAL; | |
3252 | kvm->manual_dirty_log_protect = cap->args[0]; | |
3253 | return 0; | |
3254 | #endif | |
e5d83c74 PB |
3255 | default: |
3256 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
3257 | } | |
3258 | } | |
3259 | ||
bccf2150 AK |
3260 | static long kvm_vm_ioctl(struct file *filp, |
3261 | unsigned int ioctl, unsigned long arg) | |
3262 | { | |
3263 | struct kvm *kvm = filp->private_data; | |
3264 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 3265 | int r; |
bccf2150 | 3266 | |
6d4e4c4f AK |
3267 | if (kvm->mm != current->mm) |
3268 | return -EIO; | |
bccf2150 AK |
3269 | switch (ioctl) { |
3270 | case KVM_CREATE_VCPU: | |
3271 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 3272 | break; |
e5d83c74 PB |
3273 | case KVM_ENABLE_CAP: { |
3274 | struct kvm_enable_cap cap; | |
3275 | ||
3276 | r = -EFAULT; | |
3277 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
3278 | goto out; | |
3279 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
3280 | break; | |
3281 | } | |
6fc138d2 IE |
3282 | case KVM_SET_USER_MEMORY_REGION: { |
3283 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
3284 | ||
3285 | r = -EFAULT; | |
3286 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 3287 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
3288 | goto out; |
3289 | ||
47ae31e2 | 3290 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
3291 | break; |
3292 | } | |
3293 | case KVM_GET_DIRTY_LOG: { | |
3294 | struct kvm_dirty_log log; | |
3295 | ||
3296 | r = -EFAULT; | |
893bdbf1 | 3297 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 3298 | goto out; |
2c6f5df9 | 3299 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
3300 | break; |
3301 | } | |
2a31b9db PB |
3302 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3303 | case KVM_CLEAR_DIRTY_LOG: { | |
3304 | struct kvm_clear_dirty_log log; | |
3305 | ||
3306 | r = -EFAULT; | |
3307 | if (copy_from_user(&log, argp, sizeof(log))) | |
3308 | goto out; | |
3309 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
3310 | break; | |
3311 | } | |
3312 | #endif | |
4b4357e0 | 3313 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3314 | case KVM_REGISTER_COALESCED_MMIO: { |
3315 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3316 | |
5f94c174 | 3317 | r = -EFAULT; |
893bdbf1 | 3318 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3319 | goto out; |
5f94c174 | 3320 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3321 | break; |
3322 | } | |
3323 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
3324 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3325 | |
5f94c174 | 3326 | r = -EFAULT; |
893bdbf1 | 3327 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3328 | goto out; |
5f94c174 | 3329 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3330 | break; |
3331 | } | |
3332 | #endif | |
721eecbf GH |
3333 | case KVM_IRQFD: { |
3334 | struct kvm_irqfd data; | |
3335 | ||
3336 | r = -EFAULT; | |
893bdbf1 | 3337 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3338 | goto out; |
d4db2935 | 3339 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3340 | break; |
3341 | } | |
d34e6b17 GH |
3342 | case KVM_IOEVENTFD: { |
3343 | struct kvm_ioeventfd data; | |
3344 | ||
3345 | r = -EFAULT; | |
893bdbf1 | 3346 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3347 | goto out; |
3348 | r = kvm_ioeventfd(kvm, &data); | |
3349 | break; | |
3350 | } | |
07975ad3 JK |
3351 | #ifdef CONFIG_HAVE_KVM_MSI |
3352 | case KVM_SIGNAL_MSI: { | |
3353 | struct kvm_msi msi; | |
3354 | ||
3355 | r = -EFAULT; | |
893bdbf1 | 3356 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3357 | goto out; |
3358 | r = kvm_send_userspace_msi(kvm, &msi); | |
3359 | break; | |
3360 | } | |
23d43cf9 CD |
3361 | #endif |
3362 | #ifdef __KVM_HAVE_IRQ_LINE | |
3363 | case KVM_IRQ_LINE_STATUS: | |
3364 | case KVM_IRQ_LINE: { | |
3365 | struct kvm_irq_level irq_event; | |
3366 | ||
3367 | r = -EFAULT; | |
893bdbf1 | 3368 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3369 | goto out; |
3370 | ||
aa2fbe6d YZ |
3371 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3372 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3373 | if (r) |
3374 | goto out; | |
3375 | ||
3376 | r = -EFAULT; | |
3377 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3378 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3379 | goto out; |
3380 | } | |
3381 | ||
3382 | r = 0; | |
3383 | break; | |
3384 | } | |
73880c80 | 3385 | #endif |
aa8d5944 AG |
3386 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3387 | case KVM_SET_GSI_ROUTING: { | |
3388 | struct kvm_irq_routing routing; | |
3389 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3390 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3391 | |
3392 | r = -EFAULT; | |
3393 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3394 | goto out; | |
3395 | r = -EINVAL; | |
5c0aea0e DH |
3396 | if (!kvm_arch_can_set_irq_routing(kvm)) |
3397 | goto out; | |
caf1ff26 | 3398 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3399 | goto out; |
3400 | if (routing.flags) | |
3401 | goto out; | |
f8c1b85b PB |
3402 | if (routing.nr) { |
3403 | r = -ENOMEM; | |
42bc47b3 KC |
3404 | entries = vmalloc(array_size(sizeof(*entries), |
3405 | routing.nr)); | |
f8c1b85b PB |
3406 | if (!entries) |
3407 | goto out; | |
3408 | r = -EFAULT; | |
3409 | urouting = argp; | |
3410 | if (copy_from_user(entries, urouting->entries, | |
3411 | routing.nr * sizeof(*entries))) | |
3412 | goto out_free_irq_routing; | |
3413 | } | |
aa8d5944 AG |
3414 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3415 | routing.flags); | |
a642a175 | 3416 | out_free_irq_routing: |
aa8d5944 AG |
3417 | vfree(entries); |
3418 | break; | |
3419 | } | |
3420 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3421 | case KVM_CREATE_DEVICE: { |
3422 | struct kvm_create_device cd; | |
3423 | ||
3424 | r = -EFAULT; | |
3425 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3426 | goto out; | |
3427 | ||
3428 | r = kvm_ioctl_create_device(kvm, &cd); | |
3429 | if (r) | |
3430 | goto out; | |
3431 | ||
3432 | r = -EFAULT; | |
3433 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3434 | goto out; | |
3435 | ||
3436 | r = 0; | |
3437 | break; | |
3438 | } | |
92b591a4 AG |
3439 | case KVM_CHECK_EXTENSION: |
3440 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3441 | break; | |
f17abe9a | 3442 | default: |
1fe779f8 | 3443 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3444 | } |
3445 | out: | |
3446 | return r; | |
3447 | } | |
3448 | ||
de8e5d74 | 3449 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3450 | struct compat_kvm_dirty_log { |
3451 | __u32 slot; | |
3452 | __u32 padding1; | |
3453 | union { | |
3454 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3455 | __u64 padding2; | |
3456 | }; | |
3457 | }; | |
3458 | ||
3459 | static long kvm_vm_compat_ioctl(struct file *filp, | |
3460 | unsigned int ioctl, unsigned long arg) | |
3461 | { | |
3462 | struct kvm *kvm = filp->private_data; | |
3463 | int r; | |
3464 | ||
3465 | if (kvm->mm != current->mm) | |
3466 | return -EIO; | |
3467 | switch (ioctl) { | |
3468 | case KVM_GET_DIRTY_LOG: { | |
3469 | struct compat_kvm_dirty_log compat_log; | |
3470 | struct kvm_dirty_log log; | |
3471 | ||
6ff5894c AB |
3472 | if (copy_from_user(&compat_log, (void __user *)arg, |
3473 | sizeof(compat_log))) | |
f6a3b168 | 3474 | return -EFAULT; |
6ff5894c AB |
3475 | log.slot = compat_log.slot; |
3476 | log.padding1 = compat_log.padding1; | |
3477 | log.padding2 = compat_log.padding2; | |
3478 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3479 | ||
3480 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
3481 | break; |
3482 | } | |
3483 | default: | |
3484 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
3485 | } | |
6ff5894c AB |
3486 | return r; |
3487 | } | |
3488 | #endif | |
3489 | ||
3d3aab1b | 3490 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
3491 | .release = kvm_vm_release, |
3492 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 3493 | .llseek = noop_llseek, |
7ddfd3e0 | 3494 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
3495 | }; |
3496 | ||
e08b9637 | 3497 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 3498 | { |
aac87636 | 3499 | int r; |
f17abe9a | 3500 | struct kvm *kvm; |
506cfba9 | 3501 | struct file *file; |
f17abe9a | 3502 | |
e08b9637 | 3503 | kvm = kvm_create_vm(type); |
d6d28168 AK |
3504 | if (IS_ERR(kvm)) |
3505 | return PTR_ERR(kvm); | |
4b4357e0 | 3506 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 3507 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
3508 | if (r < 0) |
3509 | goto put_kvm; | |
6ce5a090 | 3510 | #endif |
506cfba9 | 3511 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
3512 | if (r < 0) |
3513 | goto put_kvm; | |
3514 | ||
506cfba9 AV |
3515 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
3516 | if (IS_ERR(file)) { | |
3517 | put_unused_fd(r); | |
78588335 ME |
3518 | r = PTR_ERR(file); |
3519 | goto put_kvm; | |
506cfba9 | 3520 | } |
536a6f88 | 3521 | |
525df861 PB |
3522 | /* |
3523 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
3524 | * already set, with ->release() being kvm_vm_release(). In error | |
3525 | * cases it will be called by the final fput(file) and will take | |
3526 | * care of doing kvm_put_kvm(kvm). | |
3527 | */ | |
536a6f88 | 3528 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
3529 | put_unused_fd(r); |
3530 | fput(file); | |
536a6f88 JF |
3531 | return -ENOMEM; |
3532 | } | |
286de8f6 | 3533 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 3534 | |
506cfba9 | 3535 | fd_install(r, file); |
aac87636 | 3536 | return r; |
78588335 ME |
3537 | |
3538 | put_kvm: | |
3539 | kvm_put_kvm(kvm); | |
3540 | return r; | |
f17abe9a AK |
3541 | } |
3542 | ||
3543 | static long kvm_dev_ioctl(struct file *filp, | |
3544 | unsigned int ioctl, unsigned long arg) | |
3545 | { | |
07c45a36 | 3546 | long r = -EINVAL; |
f17abe9a AK |
3547 | |
3548 | switch (ioctl) { | |
3549 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
3550 | if (arg) |
3551 | goto out; | |
f17abe9a AK |
3552 | r = KVM_API_VERSION; |
3553 | break; | |
3554 | case KVM_CREATE_VM: | |
e08b9637 | 3555 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 3556 | break; |
018d00d2 | 3557 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 3558 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 3559 | break; |
07c45a36 | 3560 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
3561 | if (arg) |
3562 | goto out; | |
adb1ff46 AK |
3563 | r = PAGE_SIZE; /* struct kvm_run */ |
3564 | #ifdef CONFIG_X86 | |
3565 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 3566 | #endif |
4b4357e0 | 3567 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 3568 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 3569 | #endif |
07c45a36 | 3570 | break; |
d4c9ff2d FEL |
3571 | case KVM_TRACE_ENABLE: |
3572 | case KVM_TRACE_PAUSE: | |
3573 | case KVM_TRACE_DISABLE: | |
2023a29c | 3574 | r = -EOPNOTSUPP; |
d4c9ff2d | 3575 | break; |
6aa8b732 | 3576 | default: |
043405e1 | 3577 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
3578 | } |
3579 | out: | |
3580 | return r; | |
3581 | } | |
3582 | ||
6aa8b732 | 3583 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 3584 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 3585 | .llseek = noop_llseek, |
7ddfd3e0 | 3586 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
3587 | }; |
3588 | ||
3589 | static struct miscdevice kvm_dev = { | |
bbe4432e | 3590 | KVM_MINOR, |
6aa8b732 AK |
3591 | "kvm", |
3592 | &kvm_chardev_ops, | |
3593 | }; | |
3594 | ||
75b7127c | 3595 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
3596 | { |
3597 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 3598 | int r; |
1b6c0168 | 3599 | |
7f59f492 | 3600 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3601 | return; |
10474ae8 | 3602 | |
7f59f492 | 3603 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 3604 | |
13a34e06 | 3605 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
3606 | |
3607 | if (r) { | |
3608 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
3609 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 3610 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 3611 | } |
1b6c0168 AK |
3612 | } |
3613 | ||
8c18b2d2 | 3614 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 3615 | { |
4a937f96 | 3616 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3617 | if (kvm_usage_count) |
3618 | hardware_enable_nolock(NULL); | |
4a937f96 | 3619 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3620 | return 0; |
75b7127c TY |
3621 | } |
3622 | ||
3623 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
3624 | { |
3625 | int cpu = raw_smp_processor_id(); | |
3626 | ||
7f59f492 | 3627 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3628 | return; |
7f59f492 | 3629 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 3630 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
3631 | } |
3632 | ||
8c18b2d2 | 3633 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 3634 | { |
4a937f96 | 3635 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3636 | if (kvm_usage_count) |
3637 | hardware_disable_nolock(NULL); | |
4a937f96 | 3638 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3639 | return 0; |
75b7127c TY |
3640 | } |
3641 | ||
10474ae8 AG |
3642 | static void hardware_disable_all_nolock(void) |
3643 | { | |
3644 | BUG_ON(!kvm_usage_count); | |
3645 | ||
3646 | kvm_usage_count--; | |
3647 | if (!kvm_usage_count) | |
75b7127c | 3648 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
3649 | } |
3650 | ||
3651 | static void hardware_disable_all(void) | |
3652 | { | |
4a937f96 | 3653 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3654 | hardware_disable_all_nolock(); |
4a937f96 | 3655 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3656 | } |
3657 | ||
3658 | static int hardware_enable_all(void) | |
3659 | { | |
3660 | int r = 0; | |
3661 | ||
4a937f96 | 3662 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3663 | |
3664 | kvm_usage_count++; | |
3665 | if (kvm_usage_count == 1) { | |
3666 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3667 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3668 | |
3669 | if (atomic_read(&hardware_enable_failed)) { | |
3670 | hardware_disable_all_nolock(); | |
3671 | r = -EBUSY; | |
3672 | } | |
3673 | } | |
3674 | ||
4a937f96 | 3675 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3676 | |
3677 | return r; | |
3678 | } | |
3679 | ||
9a2b85c6 | 3680 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3681 | void *v) |
9a2b85c6 | 3682 | { |
8e1c1815 SY |
3683 | /* |
3684 | * Some (well, at least mine) BIOSes hang on reboot if | |
3685 | * in vmx root mode. | |
3686 | * | |
3687 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3688 | */ | |
1170adc6 | 3689 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3690 | kvm_rebooting = true; |
75b7127c | 3691 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3692 | return NOTIFY_OK; |
3693 | } | |
3694 | ||
3695 | static struct notifier_block kvm_reboot_notifier = { | |
3696 | .notifier_call = kvm_reboot, | |
3697 | .priority = 0, | |
3698 | }; | |
3699 | ||
e93f8a0f | 3700 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3701 | { |
3702 | int i; | |
3703 | ||
3704 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3705 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3706 | |
3707 | kvm_iodevice_destructor(pos); | |
3708 | } | |
e93f8a0f | 3709 | kfree(bus); |
2eeb2e94 GH |
3710 | } |
3711 | ||
c21fbff1 | 3712 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3713 | const struct kvm_io_range *r2) |
743eeb0b | 3714 | { |
8f4216c7 JW |
3715 | gpa_t addr1 = r1->addr; |
3716 | gpa_t addr2 = r2->addr; | |
3717 | ||
3718 | if (addr1 < addr2) | |
743eeb0b | 3719 | return -1; |
8f4216c7 JW |
3720 | |
3721 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
3722 | * accept any overlapping write. Any order is acceptable for | |
3723 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
3724 | * we process all of them. | |
3725 | */ | |
3726 | if (r2->len) { | |
3727 | addr1 += r1->len; | |
3728 | addr2 += r2->len; | |
3729 | } | |
3730 | ||
3731 | if (addr1 > addr2) | |
743eeb0b | 3732 | return 1; |
8f4216c7 | 3733 | |
743eeb0b SL |
3734 | return 0; |
3735 | } | |
3736 | ||
a343c9b7 PB |
3737 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3738 | { | |
c21fbff1 | 3739 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3740 | } |
3741 | ||
39369f7a | 3742 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3743 | gpa_t addr, int len) |
3744 | { | |
3745 | struct kvm_io_range *range, key; | |
3746 | int off; | |
3747 | ||
3748 | key = (struct kvm_io_range) { | |
3749 | .addr = addr, | |
3750 | .len = len, | |
3751 | }; | |
3752 | ||
3753 | range = bsearch(&key, bus->range, bus->dev_count, | |
3754 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3755 | if (range == NULL) | |
3756 | return -ENOENT; | |
3757 | ||
3758 | off = range - bus->range; | |
3759 | ||
c21fbff1 | 3760 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3761 | off--; |
3762 | ||
3763 | return off; | |
3764 | } | |
3765 | ||
e32edf4f | 3766 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3767 | struct kvm_io_range *range, const void *val) |
3768 | { | |
3769 | int idx; | |
3770 | ||
3771 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3772 | if (idx < 0) | |
3773 | return -EOPNOTSUPP; | |
3774 | ||
3775 | while (idx < bus->dev_count && | |
c21fbff1 | 3776 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3777 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3778 | range->len, val)) |
3779 | return idx; | |
3780 | idx++; | |
3781 | } | |
3782 | ||
3783 | return -EOPNOTSUPP; | |
3784 | } | |
3785 | ||
bda9020e | 3786 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3787 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3788 | int len, const void *val) |
2eeb2e94 | 3789 | { |
90d83dc3 | 3790 | struct kvm_io_bus *bus; |
743eeb0b | 3791 | struct kvm_io_range range; |
126a5af5 | 3792 | int r; |
743eeb0b SL |
3793 | |
3794 | range = (struct kvm_io_range) { | |
3795 | .addr = addr, | |
3796 | .len = len, | |
3797 | }; | |
90d83dc3 | 3798 | |
e32edf4f | 3799 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3800 | if (!bus) |
3801 | return -ENOMEM; | |
e32edf4f | 3802 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3803 | return r < 0 ? r : 0; |
3804 | } | |
a2420107 | 3805 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
3806 | |
3807 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3808 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3809 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3810 | { |
3811 | struct kvm_io_bus *bus; | |
3812 | struct kvm_io_range range; | |
3813 | ||
3814 | range = (struct kvm_io_range) { | |
3815 | .addr = addr, | |
3816 | .len = len, | |
3817 | }; | |
3818 | ||
e32edf4f | 3819 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3820 | if (!bus) |
3821 | return -ENOMEM; | |
126a5af5 CH |
3822 | |
3823 | /* First try the device referenced by cookie. */ | |
3824 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3825 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3826 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3827 | val)) |
3828 | return cookie; | |
3829 | ||
3830 | /* | |
3831 | * cookie contained garbage; fall back to search and return the | |
3832 | * correct cookie value. | |
3833 | */ | |
e32edf4f | 3834 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3835 | } |
3836 | ||
e32edf4f NN |
3837 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3838 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3839 | { |
3840 | int idx; | |
3841 | ||
3842 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3843 | if (idx < 0) |
3844 | return -EOPNOTSUPP; | |
3845 | ||
3846 | while (idx < bus->dev_count && | |
c21fbff1 | 3847 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3848 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3849 | range->len, val)) |
3850 | return idx; | |
743eeb0b SL |
3851 | idx++; |
3852 | } | |
3853 | ||
bda9020e MT |
3854 | return -EOPNOTSUPP; |
3855 | } | |
2eeb2e94 | 3856 | |
bda9020e | 3857 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3858 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3859 | int len, void *val) |
bda9020e | 3860 | { |
90d83dc3 | 3861 | struct kvm_io_bus *bus; |
743eeb0b | 3862 | struct kvm_io_range range; |
126a5af5 | 3863 | int r; |
743eeb0b SL |
3864 | |
3865 | range = (struct kvm_io_range) { | |
3866 | .addr = addr, | |
3867 | .len = len, | |
3868 | }; | |
e93f8a0f | 3869 | |
e32edf4f | 3870 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3871 | if (!bus) |
3872 | return -ENOMEM; | |
e32edf4f | 3873 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
3874 | return r < 0 ? r : 0; |
3875 | } | |
743eeb0b | 3876 | |
79fac95e | 3877 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3878 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3879 | int len, struct kvm_io_device *dev) | |
6c474694 | 3880 | { |
d4c67a7a | 3881 | int i; |
e93f8a0f | 3882 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 3883 | struct kvm_io_range range; |
090b7aff | 3884 | |
4a12f951 | 3885 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
3886 | if (!bus) |
3887 | return -ENOMEM; | |
3888 | ||
6ea34c9b AK |
3889 | /* exclude ioeventfd which is limited by maximum fd */ |
3890 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3891 | return -ENOSPC; |
2eeb2e94 | 3892 | |
90952cd3 | 3893 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 3894 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
3895 | if (!new_bus) |
3896 | return -ENOMEM; | |
d4c67a7a GH |
3897 | |
3898 | range = (struct kvm_io_range) { | |
3899 | .addr = addr, | |
3900 | .len = len, | |
3901 | .dev = dev, | |
3902 | }; | |
3903 | ||
3904 | for (i = 0; i < bus->dev_count; i++) | |
3905 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
3906 | break; | |
3907 | ||
3908 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3909 | new_bus->dev_count++; | |
3910 | new_bus->range[i] = range; | |
3911 | memcpy(new_bus->range + i + 1, bus->range + i, | |
3912 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3913 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3914 | synchronize_srcu_expedited(&kvm->srcu); | |
3915 | kfree(bus); | |
090b7aff GH |
3916 | |
3917 | return 0; | |
3918 | } | |
3919 | ||
79fac95e | 3920 | /* Caller must hold slots_lock. */ |
90db1043 DH |
3921 | void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3922 | struct kvm_io_device *dev) | |
090b7aff | 3923 | { |
90db1043 | 3924 | int i; |
e93f8a0f | 3925 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3926 | |
4a12f951 | 3927 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 3928 | if (!bus) |
90db1043 | 3929 | return; |
df630b8c | 3930 | |
a1300716 AK |
3931 | for (i = 0; i < bus->dev_count; i++) |
3932 | if (bus->range[i].dev == dev) { | |
090b7aff GH |
3933 | break; |
3934 | } | |
e93f8a0f | 3935 | |
90db1043 DH |
3936 | if (i == bus->dev_count) |
3937 | return; | |
a1300716 | 3938 | |
90952cd3 | 3939 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 3940 | GFP_KERNEL_ACCOUNT); |
90db1043 DH |
3941 | if (!new_bus) { |
3942 | pr_err("kvm: failed to shrink bus, removing it completely\n"); | |
3943 | goto broken; | |
3944 | } | |
a1300716 AK |
3945 | |
3946 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3947 | new_bus->dev_count--; | |
3948 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3949 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f | 3950 | |
90db1043 | 3951 | broken: |
e93f8a0f MT |
3952 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3953 | synchronize_srcu_expedited(&kvm->srcu); | |
3954 | kfree(bus); | |
90db1043 | 3955 | return; |
2eeb2e94 GH |
3956 | } |
3957 | ||
8a39d006 AP |
3958 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3959 | gpa_t addr) | |
3960 | { | |
3961 | struct kvm_io_bus *bus; | |
3962 | int dev_idx, srcu_idx; | |
3963 | struct kvm_io_device *iodev = NULL; | |
3964 | ||
3965 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
3966 | ||
3967 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
3968 | if (!bus) |
3969 | goto out_unlock; | |
8a39d006 AP |
3970 | |
3971 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
3972 | if (dev_idx < 0) | |
3973 | goto out_unlock; | |
3974 | ||
3975 | iodev = bus->range[dev_idx].dev; | |
3976 | ||
3977 | out_unlock: | |
3978 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
3979 | ||
3980 | return iodev; | |
3981 | } | |
3982 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
3983 | ||
536a6f88 JF |
3984 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
3985 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
3986 | const char *fmt) | |
3987 | { | |
3988 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
3989 | inode->i_private; | |
3990 | ||
3991 | /* The debugfs files are a reference to the kvm struct which | |
3992 | * is still valid when kvm_destroy_vm is called. | |
3993 | * To avoid the race between open and the removal of the debugfs | |
3994 | * directory we test against the users count. | |
3995 | */ | |
e3736c3e | 3996 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
3997 | return -ENOENT; |
3998 | ||
833b45de | 3999 | if (simple_attr_open(inode, file, get, |
09cbcef6 MP |
4000 | KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222 |
4001 | ? set : NULL, | |
4002 | fmt)) { | |
536a6f88 JF |
4003 | kvm_put_kvm(stat_data->kvm); |
4004 | return -ENOMEM; | |
4005 | } | |
4006 | ||
4007 | return 0; | |
4008 | } | |
4009 | ||
4010 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
4011 | { | |
4012 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4013 | inode->i_private; | |
4014 | ||
4015 | simple_attr_release(inode, file); | |
4016 | kvm_put_kvm(stat_data->kvm); | |
4017 | ||
4018 | return 0; | |
4019 | } | |
4020 | ||
09cbcef6 | 4021 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 4022 | { |
09cbcef6 | 4023 | *val = *(ulong *)((void *)kvm + offset); |
536a6f88 | 4024 | |
09cbcef6 MP |
4025 | return 0; |
4026 | } | |
4027 | ||
4028 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
4029 | { | |
4030 | *(ulong *)((void *)kvm + offset) = 0; | |
536a6f88 JF |
4031 | |
4032 | return 0; | |
4033 | } | |
4034 | ||
09cbcef6 | 4035 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 4036 | { |
09cbcef6 MP |
4037 | int i; |
4038 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 4039 | |
09cbcef6 | 4040 | *val = 0; |
ce35ef27 | 4041 | |
09cbcef6 MP |
4042 | kvm_for_each_vcpu(i, vcpu, kvm) |
4043 | *val += *(u64 *)((void *)vcpu + offset); | |
ce35ef27 SJS |
4044 | |
4045 | return 0; | |
4046 | } | |
4047 | ||
09cbcef6 | 4048 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 4049 | { |
09cbcef6 MP |
4050 | int i; |
4051 | struct kvm_vcpu *vcpu; | |
536a6f88 | 4052 | |
09cbcef6 MP |
4053 | kvm_for_each_vcpu(i, vcpu, kvm) |
4054 | *(u64 *)((void *)vcpu + offset) = 0; | |
4055 | ||
4056 | return 0; | |
4057 | } | |
536a6f88 | 4058 | |
09cbcef6 | 4059 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 4060 | { |
09cbcef6 | 4061 | int r = -EFAULT; |
536a6f88 | 4062 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 4063 | |
09cbcef6 MP |
4064 | switch (stat_data->dbgfs_item->kind) { |
4065 | case KVM_STAT_VM: | |
4066 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
4067 | stat_data->dbgfs_item->offset, val); | |
4068 | break; | |
4069 | case KVM_STAT_VCPU: | |
4070 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
4071 | stat_data->dbgfs_item->offset, val); | |
4072 | break; | |
4073 | } | |
536a6f88 | 4074 | |
09cbcef6 | 4075 | return r; |
536a6f88 JF |
4076 | } |
4077 | ||
09cbcef6 | 4078 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 4079 | { |
09cbcef6 | 4080 | int r = -EFAULT; |
ce35ef27 | 4081 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
4082 | |
4083 | if (val) | |
4084 | return -EINVAL; | |
4085 | ||
09cbcef6 MP |
4086 | switch (stat_data->dbgfs_item->kind) { |
4087 | case KVM_STAT_VM: | |
4088 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
4089 | stat_data->dbgfs_item->offset); | |
4090 | break; | |
4091 | case KVM_STAT_VCPU: | |
4092 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
4093 | stat_data->dbgfs_item->offset); | |
4094 | break; | |
4095 | } | |
ce35ef27 | 4096 | |
09cbcef6 | 4097 | return r; |
ce35ef27 SJS |
4098 | } |
4099 | ||
09cbcef6 | 4100 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
4101 | { |
4102 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
4103 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
4104 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
4105 | } |
4106 | ||
09cbcef6 MP |
4107 | static const struct file_operations stat_fops_per_vm = { |
4108 | .owner = THIS_MODULE, | |
4109 | .open = kvm_stat_data_open, | |
536a6f88 | 4110 | .release = kvm_debugfs_release, |
09cbcef6 MP |
4111 | .read = simple_attr_read, |
4112 | .write = simple_attr_write, | |
4113 | .llseek = no_llseek, | |
536a6f88 JF |
4114 | }; |
4115 | ||
8b88b099 | 4116 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
4117 | { |
4118 | unsigned offset = (long)_offset; | |
ba1389b7 | 4119 | struct kvm *kvm; |
536a6f88 | 4120 | u64 tmp_val; |
ba1389b7 | 4121 | |
8b88b099 | 4122 | *val = 0; |
0d9ce162 | 4123 | mutex_lock(&kvm_lock); |
536a6f88 | 4124 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4125 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
4126 | *val += tmp_val; |
4127 | } | |
0d9ce162 | 4128 | mutex_unlock(&kvm_lock); |
8b88b099 | 4129 | return 0; |
ba1389b7 AK |
4130 | } |
4131 | ||
ce35ef27 SJS |
4132 | static int vm_stat_clear(void *_offset, u64 val) |
4133 | { | |
4134 | unsigned offset = (long)_offset; | |
4135 | struct kvm *kvm; | |
ce35ef27 SJS |
4136 | |
4137 | if (val) | |
4138 | return -EINVAL; | |
4139 | ||
0d9ce162 | 4140 | mutex_lock(&kvm_lock); |
ce35ef27 | 4141 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4142 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 4143 | } |
0d9ce162 | 4144 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4145 | |
4146 | return 0; | |
4147 | } | |
4148 | ||
4149 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 4150 | |
8b88b099 | 4151 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
4152 | { |
4153 | unsigned offset = (long)_offset; | |
1165f5fe | 4154 | struct kvm *kvm; |
536a6f88 | 4155 | u64 tmp_val; |
1165f5fe | 4156 | |
8b88b099 | 4157 | *val = 0; |
0d9ce162 | 4158 | mutex_lock(&kvm_lock); |
536a6f88 | 4159 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4160 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
4161 | *val += tmp_val; |
4162 | } | |
0d9ce162 | 4163 | mutex_unlock(&kvm_lock); |
8b88b099 | 4164 | return 0; |
1165f5fe AK |
4165 | } |
4166 | ||
ce35ef27 SJS |
4167 | static int vcpu_stat_clear(void *_offset, u64 val) |
4168 | { | |
4169 | unsigned offset = (long)_offset; | |
4170 | struct kvm *kvm; | |
ce35ef27 SJS |
4171 | |
4172 | if (val) | |
4173 | return -EINVAL; | |
4174 | ||
0d9ce162 | 4175 | mutex_lock(&kvm_lock); |
ce35ef27 | 4176 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4177 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 4178 | } |
0d9ce162 | 4179 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4180 | |
4181 | return 0; | |
4182 | } | |
4183 | ||
4184 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
4185 | "%llu\n"); | |
ba1389b7 | 4186 | |
828c0950 | 4187 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
4188 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
4189 | [KVM_STAT_VM] = &vm_stat_fops, | |
4190 | }; | |
1165f5fe | 4191 | |
286de8f6 CI |
4192 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
4193 | { | |
4194 | struct kobj_uevent_env *env; | |
286de8f6 CI |
4195 | unsigned long long created, active; |
4196 | ||
4197 | if (!kvm_dev.this_device || !kvm) | |
4198 | return; | |
4199 | ||
0d9ce162 | 4200 | mutex_lock(&kvm_lock); |
286de8f6 CI |
4201 | if (type == KVM_EVENT_CREATE_VM) { |
4202 | kvm_createvm_count++; | |
4203 | kvm_active_vms++; | |
4204 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
4205 | kvm_active_vms--; | |
4206 | } | |
4207 | created = kvm_createvm_count; | |
4208 | active = kvm_active_vms; | |
0d9ce162 | 4209 | mutex_unlock(&kvm_lock); |
286de8f6 | 4210 | |
b12ce36a | 4211 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
4212 | if (!env) |
4213 | return; | |
4214 | ||
4215 | add_uevent_var(env, "CREATED=%llu", created); | |
4216 | add_uevent_var(env, "COUNT=%llu", active); | |
4217 | ||
fdeaf7e3 | 4218 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 4219 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
4220 | kvm->userspace_pid = task_pid_nr(current); |
4221 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 4222 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
4223 | } |
4224 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 4225 | |
8ed0579c | 4226 | if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) { |
b12ce36a | 4227 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
4228 | |
4229 | if (p) { | |
4230 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
4231 | if (!IS_ERR(tmp)) | |
4232 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
4233 | kfree(p); | |
286de8f6 CI |
4234 | } |
4235 | } | |
4236 | /* no need for checks, since we are adding at most only 5 keys */ | |
4237 | env->envp[env->envp_idx++] = NULL; | |
4238 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
4239 | kfree(env); | |
286de8f6 CI |
4240 | } |
4241 | ||
929f45e3 | 4242 | static void kvm_init_debug(void) |
6aa8b732 AK |
4243 | { |
4244 | struct kvm_stats_debugfs_item *p; | |
4245 | ||
76f7c879 | 4246 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 4247 | |
536a6f88 JF |
4248 | kvm_debugfs_num_entries = 0; |
4249 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
09cbcef6 MP |
4250 | debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p), |
4251 | kvm_debugfs_dir, (void *)(long)p->offset, | |
929f45e3 | 4252 | stat_fops[p->kind]); |
4f69b680 | 4253 | } |
6aa8b732 AK |
4254 | } |
4255 | ||
fb3600cc | 4256 | static int kvm_suspend(void) |
59ae6c6b | 4257 | { |
10474ae8 | 4258 | if (kvm_usage_count) |
75b7127c | 4259 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
4260 | return 0; |
4261 | } | |
4262 | ||
fb3600cc | 4263 | static void kvm_resume(void) |
59ae6c6b | 4264 | { |
ca84d1a2 | 4265 | if (kvm_usage_count) { |
2eb06c30 WL |
4266 | #ifdef CONFIG_LOCKDEP |
4267 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
4268 | #endif | |
75b7127c | 4269 | hardware_enable_nolock(NULL); |
ca84d1a2 | 4270 | } |
59ae6c6b AK |
4271 | } |
4272 | ||
fb3600cc | 4273 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
4274 | .suspend = kvm_suspend, |
4275 | .resume = kvm_resume, | |
4276 | }; | |
4277 | ||
15ad7146 AK |
4278 | static inline |
4279 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
4280 | { | |
4281 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
4282 | } | |
4283 | ||
4284 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
4285 | { | |
4286 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 4287 | |
046ddeed | 4288 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 4289 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 4290 | |
e790d9ef RK |
4291 | kvm_arch_sched_in(vcpu, cpu); |
4292 | ||
e9b11c17 | 4293 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
4294 | } |
4295 | ||
4296 | static void kvm_sched_out(struct preempt_notifier *pn, | |
4297 | struct task_struct *next) | |
4298 | { | |
4299 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
4300 | ||
d73eb57b | 4301 | if (current->state == TASK_RUNNING) { |
046ddeed | 4302 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
4303 | WRITE_ONCE(vcpu->ready, true); |
4304 | } | |
e9b11c17 | 4305 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
4306 | } |
4307 | ||
f257d6dc SC |
4308 | static void check_processor_compat(void *rtn) |
4309 | { | |
4310 | *(int *)rtn = kvm_arch_check_processor_compat(); | |
4311 | } | |
4312 | ||
0ee75bea | 4313 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 4314 | struct module *module) |
6aa8b732 AK |
4315 | { |
4316 | int r; | |
002c7f7c | 4317 | int cpu; |
6aa8b732 | 4318 | |
f8c16bba ZX |
4319 | r = kvm_arch_init(opaque); |
4320 | if (r) | |
d2308784 | 4321 | goto out_fail; |
cb498ea2 | 4322 | |
7dac16c3 AH |
4323 | /* |
4324 | * kvm_arch_init makes sure there's at most one caller | |
4325 | * for architectures that support multiple implementations, | |
4326 | * like intel and amd on x86. | |
36343f6e PB |
4327 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
4328 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 4329 | */ |
36343f6e PB |
4330 | r = kvm_irqfd_init(); |
4331 | if (r) | |
4332 | goto out_irqfd; | |
7dac16c3 | 4333 | |
8437a617 | 4334 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
4335 | r = -ENOMEM; |
4336 | goto out_free_0; | |
4337 | } | |
4338 | ||
e9b11c17 | 4339 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 4340 | if (r < 0) |
faf0be22 | 4341 | goto out_free_1; |
6aa8b732 | 4342 | |
002c7f7c | 4343 | for_each_online_cpu(cpu) { |
f257d6dc | 4344 | smp_call_function_single(cpu, check_processor_compat, &r, 1); |
002c7f7c | 4345 | if (r < 0) |
faf0be22 | 4346 | goto out_free_2; |
002c7f7c YS |
4347 | } |
4348 | ||
73c1b41e | 4349 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 4350 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 4351 | if (r) |
d2308784 | 4352 | goto out_free_2; |
6aa8b732 AK |
4353 | register_reboot_notifier(&kvm_reboot_notifier); |
4354 | ||
c16f862d | 4355 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
4356 | if (!vcpu_align) |
4357 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
4358 | kvm_vcpu_cache = |
4359 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
4360 | SLAB_ACCOUNT, | |
4361 | offsetof(struct kvm_vcpu, arch), | |
4362 | sizeof_field(struct kvm_vcpu, arch), | |
4363 | NULL); | |
c16f862d RR |
4364 | if (!kvm_vcpu_cache) { |
4365 | r = -ENOMEM; | |
fb3600cc | 4366 | goto out_free_3; |
c16f862d RR |
4367 | } |
4368 | ||
af585b92 GN |
4369 | r = kvm_async_pf_init(); |
4370 | if (r) | |
4371 | goto out_free; | |
4372 | ||
6aa8b732 | 4373 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
4374 | kvm_vm_fops.owner = module; |
4375 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
4376 | |
4377 | r = misc_register(&kvm_dev); | |
4378 | if (r) { | |
1170adc6 | 4379 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 4380 | goto out_unreg; |
6aa8b732 AK |
4381 | } |
4382 | ||
fb3600cc RW |
4383 | register_syscore_ops(&kvm_syscore_ops); |
4384 | ||
15ad7146 AK |
4385 | kvm_preempt_ops.sched_in = kvm_sched_in; |
4386 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
4387 | ||
929f45e3 | 4388 | kvm_init_debug(); |
0ea4ed8e | 4389 | |
3c3c29fd PB |
4390 | r = kvm_vfio_ops_init(); |
4391 | WARN_ON(r); | |
4392 | ||
c7addb90 | 4393 | return 0; |
6aa8b732 | 4394 | |
af585b92 GN |
4395 | out_unreg: |
4396 | kvm_async_pf_deinit(); | |
6aa8b732 | 4397 | out_free: |
c16f862d | 4398 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4399 | out_free_3: |
6aa8b732 | 4400 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4401 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4402 | out_free_2: |
e9b11c17 | 4403 | kvm_arch_hardware_unsetup(); |
faf0be22 | 4404 | out_free_1: |
7f59f492 | 4405 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 4406 | out_free_0: |
a0f155e9 | 4407 | kvm_irqfd_exit(); |
36343f6e | 4408 | out_irqfd: |
7dac16c3 AH |
4409 | kvm_arch_exit(); |
4410 | out_fail: | |
6aa8b732 AK |
4411 | return r; |
4412 | } | |
cb498ea2 | 4413 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 4414 | |
cb498ea2 | 4415 | void kvm_exit(void) |
6aa8b732 | 4416 | { |
4bd33b56 | 4417 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 4418 | misc_deregister(&kvm_dev); |
c16f862d | 4419 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 4420 | kvm_async_pf_deinit(); |
fb3600cc | 4421 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 4422 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4423 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 4424 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 4425 | kvm_arch_hardware_unsetup(); |
f8c16bba | 4426 | kvm_arch_exit(); |
a0f155e9 | 4427 | kvm_irqfd_exit(); |
7f59f492 | 4428 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 4429 | kvm_vfio_ops_exit(); |
6aa8b732 | 4430 | } |
cb498ea2 | 4431 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
4432 | |
4433 | struct kvm_vm_worker_thread_context { | |
4434 | struct kvm *kvm; | |
4435 | struct task_struct *parent; | |
4436 | struct completion init_done; | |
4437 | kvm_vm_thread_fn_t thread_fn; | |
4438 | uintptr_t data; | |
4439 | int err; | |
4440 | }; | |
4441 | ||
4442 | static int kvm_vm_worker_thread(void *context) | |
4443 | { | |
4444 | /* | |
4445 | * The init_context is allocated on the stack of the parent thread, so | |
4446 | * we have to locally copy anything that is needed beyond initialization | |
4447 | */ | |
4448 | struct kvm_vm_worker_thread_context *init_context = context; | |
4449 | struct kvm *kvm = init_context->kvm; | |
4450 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
4451 | uintptr_t data = init_context->data; | |
4452 | int err; | |
4453 | ||
4454 | err = kthread_park(current); | |
4455 | /* kthread_park(current) is never supposed to return an error */ | |
4456 | WARN_ON(err != 0); | |
4457 | if (err) | |
4458 | goto init_complete; | |
4459 | ||
4460 | err = cgroup_attach_task_all(init_context->parent, current); | |
4461 | if (err) { | |
4462 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
4463 | __func__, err); | |
4464 | goto init_complete; | |
4465 | } | |
4466 | ||
4467 | set_user_nice(current, task_nice(init_context->parent)); | |
4468 | ||
4469 | init_complete: | |
4470 | init_context->err = err; | |
4471 | complete(&init_context->init_done); | |
4472 | init_context = NULL; | |
4473 | ||
4474 | if (err) | |
4475 | return err; | |
4476 | ||
4477 | /* Wait to be woken up by the spawner before proceeding. */ | |
4478 | kthread_parkme(); | |
4479 | ||
4480 | if (!kthread_should_stop()) | |
4481 | err = thread_fn(kvm, data); | |
4482 | ||
4483 | return err; | |
4484 | } | |
4485 | ||
4486 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
4487 | uintptr_t data, const char *name, | |
4488 | struct task_struct **thread_ptr) | |
4489 | { | |
4490 | struct kvm_vm_worker_thread_context init_context = {}; | |
4491 | struct task_struct *thread; | |
4492 | ||
4493 | *thread_ptr = NULL; | |
4494 | init_context.kvm = kvm; | |
4495 | init_context.parent = current; | |
4496 | init_context.thread_fn = thread_fn; | |
4497 | init_context.data = data; | |
4498 | init_completion(&init_context.init_done); | |
4499 | ||
4500 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
4501 | "%s-%d", name, task_pid_nr(current)); | |
4502 | if (IS_ERR(thread)) | |
4503 | return PTR_ERR(thread); | |
4504 | ||
4505 | /* kthread_run is never supposed to return NULL */ | |
4506 | WARN_ON(thread == NULL); | |
4507 | ||
4508 | wait_for_completion(&init_context.init_done); | |
4509 | ||
4510 | if (!init_context.err) | |
4511 | *thread_ptr = thread; | |
4512 | ||
4513 | return init_context.err; | |
4514 | } |