Commit | Line | Data |
---|---|---|
215e262f KO |
1 | /* |
2 | * Percpu refcounts: | |
3 | * (C) 2012 Google, Inc. | |
4 | * Author: Kent Overstreet <koverstreet@google.com> | |
5 | * | |
6 | * This implements a refcount with similar semantics to atomic_t - atomic_inc(), | |
7 | * atomic_dec_and_test() - but percpu. | |
8 | * | |
9 | * There's one important difference between percpu refs and normal atomic_t | |
10 | * refcounts; you have to keep track of your initial refcount, and then when you | |
11 | * start shutting down you call percpu_ref_kill() _before_ dropping the initial | |
12 | * refcount. | |
13 | * | |
14 | * The refcount will have a range of 0 to ((1U << 31) - 1), i.e. one bit less | |
15 | * than an atomic_t - this is because of the way shutdown works, see | |
eecc16ba | 16 | * percpu_ref_kill()/PERCPU_COUNT_BIAS. |
215e262f KO |
17 | * |
18 | * Before you call percpu_ref_kill(), percpu_ref_put() does not check for the | |
19 | * refcount hitting 0 - it can't, if it was in percpu mode. percpu_ref_kill() | |
20 | * puts the ref back in single atomic_t mode, collecting the per cpu refs and | |
21 | * issuing the appropriate barriers, and then marks the ref as shutting down so | |
22 | * that percpu_ref_put() will check for the ref hitting 0. After it returns, | |
23 | * it's safe to drop the initial ref. | |
24 | * | |
25 | * USAGE: | |
26 | * | |
27 | * See fs/aio.c for some example usage; it's used there for struct kioctx, which | |
28 | * is created when userspaces calls io_setup(), and destroyed when userspace | |
29 | * calls io_destroy() or the process exits. | |
30 | * | |
31 | * In the aio code, kill_ioctx() is called when we wish to destroy a kioctx; it | |
d7cdb968 | 32 | * calls percpu_ref_kill(), then hlist_del_rcu() and synchronize_rcu() to remove |
215e262f KO |
33 | * the kioctx from the proccess's list of kioctxs - after that, there can't be |
34 | * any new users of the kioctx (from lookup_ioctx()) and it's then safe to drop | |
35 | * the initial ref with percpu_ref_put(). | |
36 | * | |
37 | * Code that does a two stage shutdown like this often needs some kind of | |
38 | * explicit synchronization to ensure the initial refcount can only be dropped | |
39 | * once - percpu_ref_kill() does this for you, it returns true once and false if | |
40 | * someone else already called it. The aio code uses it this way, but it's not | |
41 | * necessary if the code has some other mechanism to synchronize teardown. | |
42 | * around. | |
43 | */ | |
44 | ||
45 | #ifndef _LINUX_PERCPU_REFCOUNT_H | |
46 | #define _LINUX_PERCPU_REFCOUNT_H | |
47 | ||
48 | #include <linux/atomic.h> | |
49 | #include <linux/kernel.h> | |
50 | #include <linux/percpu.h> | |
51 | #include <linux/rcupdate.h> | |
a34375ef | 52 | #include <linux/gfp.h> |
215e262f KO |
53 | |
54 | struct percpu_ref; | |
ac899061 | 55 | typedef void (percpu_ref_func_t)(struct percpu_ref *); |
215e262f | 56 | |
9e804d1f TH |
57 | /* flags set in the lower bits of percpu_ref->percpu_count_ptr */ |
58 | enum { | |
59 | __PERCPU_REF_ATOMIC = 1LU << 0, /* operating in atomic mode */ | |
27344a90 TH |
60 | __PERCPU_REF_DEAD = 1LU << 1, /* (being) killed */ |
61 | __PERCPU_REF_ATOMIC_DEAD = __PERCPU_REF_ATOMIC | __PERCPU_REF_DEAD, | |
62 | ||
63 | __PERCPU_REF_FLAG_BITS = 2, | |
9e804d1f TH |
64 | }; |
65 | ||
2aad2a86 TH |
66 | /* @flags for percpu_ref_init() */ |
67 | enum { | |
68 | /* | |
69 | * Start w/ ref == 1 in atomic mode. Can be switched to percpu | |
1cae13e7 TH |
70 | * operation using percpu_ref_switch_to_percpu(). If initialized |
71 | * with this flag, the ref will stay in atomic mode until | |
72 | * percpu_ref_switch_to_percpu() is invoked on it. | |
2aad2a86 TH |
73 | */ |
74 | PERCPU_REF_INIT_ATOMIC = 1 << 0, | |
75 | ||
76 | /* | |
77 | * Start dead w/ ref == 0 in atomic mode. Must be revived with | |
78 | * percpu_ref_reinit() before used. Implies INIT_ATOMIC. | |
79 | */ | |
80 | PERCPU_REF_INIT_DEAD = 1 << 1, | |
81 | }; | |
82 | ||
215e262f | 83 | struct percpu_ref { |
e625305b | 84 | atomic_long_t count; |
215e262f KO |
85 | /* |
86 | * The low bit of the pointer indicates whether the ref is in percpu | |
9a1049da | 87 | * mode; if set, then get/put will manipulate the atomic_t. |
215e262f | 88 | */ |
eecc16ba | 89 | unsigned long percpu_count_ptr; |
ac899061 | 90 | percpu_ref_func_t *release; |
9e804d1f | 91 | percpu_ref_func_t *confirm_switch; |
1cae13e7 | 92 | bool force_atomic:1; |
215e262f KO |
93 | struct rcu_head rcu; |
94 | }; | |
95 | ||
acac7883 | 96 | int __must_check percpu_ref_init(struct percpu_ref *ref, |
2aad2a86 TH |
97 | percpu_ref_func_t *release, unsigned int flags, |
98 | gfp_t gfp); | |
9a1049da | 99 | void percpu_ref_exit(struct percpu_ref *ref); |
490c79a6 TH |
100 | void percpu_ref_switch_to_atomic(struct percpu_ref *ref, |
101 | percpu_ref_func_t *confirm_switch); | |
f47ad457 | 102 | void percpu_ref_switch_to_percpu(struct percpu_ref *ref); |
dbece3a0 TH |
103 | void percpu_ref_kill_and_confirm(struct percpu_ref *ref, |
104 | percpu_ref_func_t *confirm_kill); | |
f47ad457 | 105 | void percpu_ref_reinit(struct percpu_ref *ref); |
dbece3a0 TH |
106 | |
107 | /** | |
108 | * percpu_ref_kill - drop the initial ref | |
109 | * @ref: percpu_ref to kill | |
110 | * | |
111 | * Must be used to drop the initial ref on a percpu refcount; must be called | |
112 | * precisely once before shutdown. | |
113 | * | |
114 | * Puts @ref in non percpu mode, then does a call_rcu() before gathering up the | |
115 | * percpu counters and dropping the initial ref. | |
116 | */ | |
117 | static inline void percpu_ref_kill(struct percpu_ref *ref) | |
118 | { | |
4d414269 | 119 | percpu_ref_kill_and_confirm(ref, NULL); |
dbece3a0 | 120 | } |
215e262f | 121 | |
eae7975d TH |
122 | /* |
123 | * Internal helper. Don't use outside percpu-refcount proper. The | |
124 | * function doesn't return the pointer and let the caller test it for NULL | |
125 | * because doing so forces the compiler to generate two conditional | |
eecc16ba | 126 | * branches as it can't assume that @ref->percpu_count is not NULL. |
eae7975d | 127 | */ |
9e804d1f TH |
128 | static inline bool __ref_is_percpu(struct percpu_ref *ref, |
129 | unsigned long __percpu **percpu_countp) | |
eae7975d | 130 | { |
6810e4a3 TH |
131 | unsigned long percpu_ptr; |
132 | ||
133 | /* | |
134 | * The value of @ref->percpu_count_ptr is tested for | |
135 | * !__PERCPU_REF_ATOMIC, which may be set asynchronously, and then | |
136 | * used as a pointer. If the compiler generates a separate fetch | |
137 | * when using it as a pointer, __PERCPU_REF_ATOMIC may be set in | |
138 | * between contaminating the pointer value, meaning that | |
139 | * ACCESS_ONCE() is required when fetching it. | |
140 | * | |
141 | * Also, we need a data dependency barrier to be paired with | |
142 | * smp_store_release() in __percpu_ref_switch_to_percpu(). | |
143 | * | |
144 | * Use lockless deref which contains both. | |
145 | */ | |
146 | percpu_ptr = lockless_dereference(ref->percpu_count_ptr); | |
2d722782 | 147 | |
4aab3b5b TH |
148 | /* |
149 | * Theoretically, the following could test just ATOMIC; however, | |
150 | * then we'd have to mask off DEAD separately as DEAD may be | |
151 | * visible without ATOMIC if we race with percpu_ref_kill(). DEAD | |
152 | * implies ATOMIC anyway. Test them together. | |
153 | */ | |
154 | if (unlikely(percpu_ptr & __PERCPU_REF_ATOMIC_DEAD)) | |
eae7975d TH |
155 | return false; |
156 | ||
eecc16ba | 157 | *percpu_countp = (unsigned long __percpu *)percpu_ptr; |
eae7975d TH |
158 | return true; |
159 | } | |
215e262f KO |
160 | |
161 | /** | |
e8ea14cc | 162 | * percpu_ref_get_many - increment a percpu refcount |
ac899061 | 163 | * @ref: percpu_ref to get |
e8ea14cc | 164 | * @nr: number of references to get |
215e262f | 165 | * |
e8ea14cc | 166 | * Analogous to atomic_long_add(). |
6251f997 TH |
167 | * |
168 | * This function is safe to call as long as @ref is between init and exit. | |
169 | */ | |
e8ea14cc | 170 | static inline void percpu_ref_get_many(struct percpu_ref *ref, unsigned long nr) |
215e262f | 171 | { |
eecc16ba | 172 | unsigned long __percpu *percpu_count; |
215e262f | 173 | |
a4244454 | 174 | rcu_read_lock_sched(); |
215e262f | 175 | |
9e804d1f | 176 | if (__ref_is_percpu(ref, &percpu_count)) |
e8ea14cc | 177 | this_cpu_add(*percpu_count, nr); |
215e262f | 178 | else |
e8ea14cc | 179 | atomic_long_add(nr, &ref->count); |
215e262f | 180 | |
a4244454 | 181 | rcu_read_unlock_sched(); |
215e262f KO |
182 | } |
183 | ||
e8ea14cc JW |
184 | /** |
185 | * percpu_ref_get - increment a percpu refcount | |
186 | * @ref: percpu_ref to get | |
187 | * | |
188 | * Analagous to atomic_long_inc(). | |
189 | * | |
190 | * This function is safe to call as long as @ref is between init and exit. | |
191 | */ | |
192 | static inline void percpu_ref_get(struct percpu_ref *ref) | |
193 | { | |
194 | percpu_ref_get_many(ref, 1); | |
195 | } | |
196 | ||
4fb6e250 TH |
197 | /** |
198 | * percpu_ref_tryget - try to increment a percpu refcount | |
199 | * @ref: percpu_ref to try-get | |
200 | * | |
201 | * Increment a percpu refcount unless its count already reached zero. | |
202 | * Returns %true on success; %false on failure. | |
203 | * | |
6251f997 | 204 | * This function is safe to call as long as @ref is between init and exit. |
4fb6e250 TH |
205 | */ |
206 | static inline bool percpu_ref_tryget(struct percpu_ref *ref) | |
207 | { | |
eecc16ba | 208 | unsigned long __percpu *percpu_count; |
6251f997 | 209 | int ret; |
4fb6e250 TH |
210 | |
211 | rcu_read_lock_sched(); | |
212 | ||
9e804d1f | 213 | if (__ref_is_percpu(ref, &percpu_count)) { |
eecc16ba | 214 | this_cpu_inc(*percpu_count); |
4fb6e250 TH |
215 | ret = true; |
216 | } else { | |
e625305b | 217 | ret = atomic_long_inc_not_zero(&ref->count); |
4fb6e250 TH |
218 | } |
219 | ||
220 | rcu_read_unlock_sched(); | |
221 | ||
222 | return ret; | |
223 | } | |
224 | ||
dbece3a0 | 225 | /** |
2070d50e | 226 | * percpu_ref_tryget_live - try to increment a live percpu refcount |
dbece3a0 TH |
227 | * @ref: percpu_ref to try-get |
228 | * | |
229 | * Increment a percpu refcount unless it has already been killed. Returns | |
230 | * %true on success; %false on failure. | |
231 | * | |
6251f997 TH |
232 | * Completion of percpu_ref_kill() in itself doesn't guarantee that this |
233 | * function will fail. For such guarantee, percpu_ref_kill_and_confirm() | |
234 | * should be used. After the confirm_kill callback is invoked, it's | |
235 | * guaranteed that no new reference will be given out by | |
236 | * percpu_ref_tryget_live(). | |
4fb6e250 | 237 | * |
6251f997 | 238 | * This function is safe to call as long as @ref is between init and exit. |
dbece3a0 | 239 | */ |
2070d50e | 240 | static inline bool percpu_ref_tryget_live(struct percpu_ref *ref) |
dbece3a0 | 241 | { |
eecc16ba | 242 | unsigned long __percpu *percpu_count; |
dbece3a0 TH |
243 | int ret = false; |
244 | ||
a4244454 | 245 | rcu_read_lock_sched(); |
dbece3a0 | 246 | |
9e804d1f | 247 | if (__ref_is_percpu(ref, &percpu_count)) { |
eecc16ba | 248 | this_cpu_inc(*percpu_count); |
dbece3a0 | 249 | ret = true; |
6810e4a3 | 250 | } else if (!(ref->percpu_count_ptr & __PERCPU_REF_DEAD)) { |
490c79a6 | 251 | ret = atomic_long_inc_not_zero(&ref->count); |
dbece3a0 TH |
252 | } |
253 | ||
a4244454 | 254 | rcu_read_unlock_sched(); |
dbece3a0 TH |
255 | |
256 | return ret; | |
257 | } | |
258 | ||
215e262f | 259 | /** |
e8ea14cc | 260 | * percpu_ref_put_many - decrement a percpu refcount |
ac899061 | 261 | * @ref: percpu_ref to put |
e8ea14cc | 262 | * @nr: number of references to put |
215e262f KO |
263 | * |
264 | * Decrement the refcount, and if 0, call the release function (which was passed | |
265 | * to percpu_ref_init()) | |
6251f997 TH |
266 | * |
267 | * This function is safe to call as long as @ref is between init and exit. | |
215e262f | 268 | */ |
e8ea14cc | 269 | static inline void percpu_ref_put_many(struct percpu_ref *ref, unsigned long nr) |
215e262f | 270 | { |
eecc16ba | 271 | unsigned long __percpu *percpu_count; |
215e262f | 272 | |
a4244454 | 273 | rcu_read_lock_sched(); |
215e262f | 274 | |
9e804d1f | 275 | if (__ref_is_percpu(ref, &percpu_count)) |
e8ea14cc JW |
276 | this_cpu_sub(*percpu_count, nr); |
277 | else if (unlikely(atomic_long_sub_and_test(nr, &ref->count))) | |
215e262f KO |
278 | ref->release(ref); |
279 | ||
a4244454 | 280 | rcu_read_unlock_sched(); |
215e262f KO |
281 | } |
282 | ||
e8ea14cc JW |
283 | /** |
284 | * percpu_ref_put - decrement a percpu refcount | |
285 | * @ref: percpu_ref to put | |
286 | * | |
287 | * Decrement the refcount, and if 0, call the release function (which was passed | |
288 | * to percpu_ref_init()) | |
289 | * | |
290 | * This function is safe to call as long as @ref is between init and exit. | |
291 | */ | |
292 | static inline void percpu_ref_put(struct percpu_ref *ref) | |
293 | { | |
294 | percpu_ref_put_many(ref, 1); | |
295 | } | |
296 | ||
4c907baf TH |
297 | /** |
298 | * percpu_ref_is_dying - test whether a percpu refcount is dying or dead | |
299 | * @ref: percpu_ref to test | |
300 | * | |
301 | * Returns %true if @ref is dying or dead. | |
302 | * | |
303 | * This function is safe to call as long as @ref is between init and exit | |
304 | * and the caller is responsible for synchronizing against state changes. | |
305 | */ | |
306 | static inline bool percpu_ref_is_dying(struct percpu_ref *ref) | |
307 | { | |
308 | return ref->percpu_count_ptr & __PERCPU_REF_DEAD; | |
309 | } | |
310 | ||
2d722782 TH |
311 | /** |
312 | * percpu_ref_is_zero - test whether a percpu refcount reached zero | |
313 | * @ref: percpu_ref to test | |
314 | * | |
315 | * Returns %true if @ref reached zero. | |
6251f997 TH |
316 | * |
317 | * This function is safe to call as long as @ref is between init and exit. | |
2d722782 TH |
318 | */ |
319 | static inline bool percpu_ref_is_zero(struct percpu_ref *ref) | |
320 | { | |
eecc16ba | 321 | unsigned long __percpu *percpu_count; |
2d722782 | 322 | |
9e804d1f | 323 | if (__ref_is_percpu(ref, &percpu_count)) |
2d722782 | 324 | return false; |
e625305b | 325 | return !atomic_long_read(&ref->count); |
2d722782 TH |
326 | } |
327 | ||
215e262f | 328 | #endif |