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); | |
210f7cdc | 102 | void percpu_ref_switch_to_atomic_sync(struct percpu_ref *ref); |
f47ad457 | 103 | void percpu_ref_switch_to_percpu(struct percpu_ref *ref); |
dbece3a0 TH |
104 | void percpu_ref_kill_and_confirm(struct percpu_ref *ref, |
105 | percpu_ref_func_t *confirm_kill); | |
f47ad457 | 106 | void percpu_ref_reinit(struct percpu_ref *ref); |
dbece3a0 TH |
107 | |
108 | /** | |
109 | * percpu_ref_kill - drop the initial ref | |
110 | * @ref: percpu_ref to kill | |
111 | * | |
112 | * Must be used to drop the initial ref on a percpu refcount; must be called | |
113 | * precisely once before shutdown. | |
114 | * | |
115 | * Puts @ref in non percpu mode, then does a call_rcu() before gathering up the | |
116 | * percpu counters and dropping the initial ref. | |
117 | */ | |
118 | static inline void percpu_ref_kill(struct percpu_ref *ref) | |
119 | { | |
4d414269 | 120 | percpu_ref_kill_and_confirm(ref, NULL); |
dbece3a0 | 121 | } |
215e262f | 122 | |
eae7975d TH |
123 | /* |
124 | * Internal helper. Don't use outside percpu-refcount proper. The | |
125 | * function doesn't return the pointer and let the caller test it for NULL | |
126 | * because doing so forces the compiler to generate two conditional | |
eecc16ba | 127 | * branches as it can't assume that @ref->percpu_count is not NULL. |
eae7975d | 128 | */ |
9e804d1f TH |
129 | static inline bool __ref_is_percpu(struct percpu_ref *ref, |
130 | unsigned long __percpu **percpu_countp) | |
eae7975d | 131 | { |
6810e4a3 TH |
132 | unsigned long percpu_ptr; |
133 | ||
134 | /* | |
135 | * The value of @ref->percpu_count_ptr is tested for | |
136 | * !__PERCPU_REF_ATOMIC, which may be set asynchronously, and then | |
137 | * used as a pointer. If the compiler generates a separate fetch | |
138 | * when using it as a pointer, __PERCPU_REF_ATOMIC may be set in | |
139 | * between contaminating the pointer value, meaning that | |
ed8ebd1d | 140 | * READ_ONCE() is required when fetching it. |
6810e4a3 | 141 | */ |
ed8ebd1d TH |
142 | percpu_ptr = READ_ONCE(ref->percpu_count_ptr); |
143 | ||
144 | /* paired with smp_store_release() in __percpu_ref_switch_to_percpu() */ | |
145 | smp_read_barrier_depends(); | |
2d722782 | 146 | |
4aab3b5b TH |
147 | /* |
148 | * Theoretically, the following could test just ATOMIC; however, | |
149 | * then we'd have to mask off DEAD separately as DEAD may be | |
150 | * visible without ATOMIC if we race with percpu_ref_kill(). DEAD | |
151 | * implies ATOMIC anyway. Test them together. | |
152 | */ | |
153 | if (unlikely(percpu_ptr & __PERCPU_REF_ATOMIC_DEAD)) | |
eae7975d TH |
154 | return false; |
155 | ||
eecc16ba | 156 | *percpu_countp = (unsigned long __percpu *)percpu_ptr; |
eae7975d TH |
157 | return true; |
158 | } | |
215e262f KO |
159 | |
160 | /** | |
e8ea14cc | 161 | * percpu_ref_get_many - increment a percpu refcount |
ac899061 | 162 | * @ref: percpu_ref to get |
e8ea14cc | 163 | * @nr: number of references to get |
215e262f | 164 | * |
e8ea14cc | 165 | * Analogous to atomic_long_add(). |
6251f997 TH |
166 | * |
167 | * This function is safe to call as long as @ref is between init and exit. | |
168 | */ | |
e8ea14cc | 169 | static inline void percpu_ref_get_many(struct percpu_ref *ref, unsigned long nr) |
215e262f | 170 | { |
eecc16ba | 171 | unsigned long __percpu *percpu_count; |
215e262f | 172 | |
a4244454 | 173 | rcu_read_lock_sched(); |
215e262f | 174 | |
9e804d1f | 175 | if (__ref_is_percpu(ref, &percpu_count)) |
e8ea14cc | 176 | this_cpu_add(*percpu_count, nr); |
215e262f | 177 | else |
e8ea14cc | 178 | atomic_long_add(nr, &ref->count); |
215e262f | 179 | |
a4244454 | 180 | rcu_read_unlock_sched(); |
215e262f KO |
181 | } |
182 | ||
e8ea14cc JW |
183 | /** |
184 | * percpu_ref_get - increment a percpu refcount | |
185 | * @ref: percpu_ref to get | |
186 | * | |
187 | * Analagous to atomic_long_inc(). | |
188 | * | |
189 | * This function is safe to call as long as @ref is between init and exit. | |
190 | */ | |
191 | static inline void percpu_ref_get(struct percpu_ref *ref) | |
192 | { | |
193 | percpu_ref_get_many(ref, 1); | |
194 | } | |
195 | ||
4fb6e250 TH |
196 | /** |
197 | * percpu_ref_tryget - try to increment a percpu refcount | |
198 | * @ref: percpu_ref to try-get | |
199 | * | |
200 | * Increment a percpu refcount unless its count already reached zero. | |
201 | * Returns %true on success; %false on failure. | |
202 | * | |
6251f997 | 203 | * This function is safe to call as long as @ref is between init and exit. |
4fb6e250 TH |
204 | */ |
205 | static inline bool percpu_ref_tryget(struct percpu_ref *ref) | |
206 | { | |
eecc16ba | 207 | unsigned long __percpu *percpu_count; |
966d2b04 | 208 | bool ret; |
4fb6e250 TH |
209 | |
210 | rcu_read_lock_sched(); | |
211 | ||
9e804d1f | 212 | if (__ref_is_percpu(ref, &percpu_count)) { |
eecc16ba | 213 | this_cpu_inc(*percpu_count); |
4fb6e250 TH |
214 | ret = true; |
215 | } else { | |
e625305b | 216 | ret = atomic_long_inc_not_zero(&ref->count); |
4fb6e250 TH |
217 | } |
218 | ||
219 | rcu_read_unlock_sched(); | |
220 | ||
221 | return ret; | |
222 | } | |
223 | ||
dbece3a0 | 224 | /** |
2070d50e | 225 | * percpu_ref_tryget_live - try to increment a live percpu refcount |
dbece3a0 TH |
226 | * @ref: percpu_ref to try-get |
227 | * | |
228 | * Increment a percpu refcount unless it has already been killed. Returns | |
229 | * %true on success; %false on failure. | |
230 | * | |
6251f997 TH |
231 | * Completion of percpu_ref_kill() in itself doesn't guarantee that this |
232 | * function will fail. For such guarantee, percpu_ref_kill_and_confirm() | |
233 | * should be used. After the confirm_kill callback is invoked, it's | |
234 | * guaranteed that no new reference will be given out by | |
235 | * percpu_ref_tryget_live(). | |
4fb6e250 | 236 | * |
6251f997 | 237 | * This function is safe to call as long as @ref is between init and exit. |
dbece3a0 | 238 | */ |
2070d50e | 239 | static inline bool percpu_ref_tryget_live(struct percpu_ref *ref) |
dbece3a0 | 240 | { |
eecc16ba | 241 | unsigned long __percpu *percpu_count; |
966d2b04 | 242 | bool ret = false; |
dbece3a0 | 243 | |
a4244454 | 244 | rcu_read_lock_sched(); |
dbece3a0 | 245 | |
9e804d1f | 246 | if (__ref_is_percpu(ref, &percpu_count)) { |
eecc16ba | 247 | this_cpu_inc(*percpu_count); |
dbece3a0 | 248 | ret = true; |
6810e4a3 | 249 | } else if (!(ref->percpu_count_ptr & __PERCPU_REF_DEAD)) { |
490c79a6 | 250 | ret = atomic_long_inc_not_zero(&ref->count); |
dbece3a0 TH |
251 | } |
252 | ||
a4244454 | 253 | rcu_read_unlock_sched(); |
dbece3a0 TH |
254 | |
255 | return ret; | |
256 | } | |
257 | ||
215e262f | 258 | /** |
e8ea14cc | 259 | * percpu_ref_put_many - decrement a percpu refcount |
ac899061 | 260 | * @ref: percpu_ref to put |
e8ea14cc | 261 | * @nr: number of references to put |
215e262f KO |
262 | * |
263 | * Decrement the refcount, and if 0, call the release function (which was passed | |
264 | * to percpu_ref_init()) | |
6251f997 TH |
265 | * |
266 | * This function is safe to call as long as @ref is between init and exit. | |
215e262f | 267 | */ |
e8ea14cc | 268 | static inline void percpu_ref_put_many(struct percpu_ref *ref, unsigned long nr) |
215e262f | 269 | { |
eecc16ba | 270 | unsigned long __percpu *percpu_count; |
215e262f | 271 | |
a4244454 | 272 | rcu_read_lock_sched(); |
215e262f | 273 | |
9e804d1f | 274 | if (__ref_is_percpu(ref, &percpu_count)) |
e8ea14cc JW |
275 | this_cpu_sub(*percpu_count, nr); |
276 | else if (unlikely(atomic_long_sub_and_test(nr, &ref->count))) | |
215e262f KO |
277 | ref->release(ref); |
278 | ||
a4244454 | 279 | rcu_read_unlock_sched(); |
215e262f KO |
280 | } |
281 | ||
e8ea14cc JW |
282 | /** |
283 | * percpu_ref_put - decrement a percpu refcount | |
284 | * @ref: percpu_ref to put | |
285 | * | |
286 | * Decrement the refcount, and if 0, call the release function (which was passed | |
287 | * to percpu_ref_init()) | |
288 | * | |
289 | * This function is safe to call as long as @ref is between init and exit. | |
290 | */ | |
291 | static inline void percpu_ref_put(struct percpu_ref *ref) | |
292 | { | |
293 | percpu_ref_put_many(ref, 1); | |
294 | } | |
295 | ||
4c907baf TH |
296 | /** |
297 | * percpu_ref_is_dying - test whether a percpu refcount is dying or dead | |
298 | * @ref: percpu_ref to test | |
299 | * | |
300 | * Returns %true if @ref is dying or dead. | |
301 | * | |
302 | * This function is safe to call as long as @ref is between init and exit | |
303 | * and the caller is responsible for synchronizing against state changes. | |
304 | */ | |
305 | static inline bool percpu_ref_is_dying(struct percpu_ref *ref) | |
306 | { | |
307 | return ref->percpu_count_ptr & __PERCPU_REF_DEAD; | |
308 | } | |
309 | ||
2d722782 TH |
310 | /** |
311 | * percpu_ref_is_zero - test whether a percpu refcount reached zero | |
312 | * @ref: percpu_ref to test | |
313 | * | |
314 | * Returns %true if @ref reached zero. | |
6251f997 TH |
315 | * |
316 | * This function is safe to call as long as @ref is between init and exit. | |
2d722782 TH |
317 | */ |
318 | static inline bool percpu_ref_is_zero(struct percpu_ref *ref) | |
319 | { | |
eecc16ba | 320 | unsigned long __percpu *percpu_count; |
2d722782 | 321 | |
9e804d1f | 322 | if (__ref_is_percpu(ref, &percpu_count)) |
2d722782 | 323 | return false; |
e625305b | 324 | return !atomic_long_read(&ref->count); |
2d722782 TH |
325 | } |
326 | ||
215e262f | 327 | #endif |