blob: 3d463a39e0f7d39448fa0b7dbbf3bed6f738e68e [file] [log] [blame]
Kent Overstreet215e2622013-05-31 15:26:45 -07001/*
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
Tejun Heoeecc16b2014-09-24 13:31:48 -040016 * percpu_ref_kill()/PERCPU_COUNT_BIAS.
Kent Overstreet215e2622013-05-31 15:26:45 -070017 *
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
32 * calls percpu_ref_kill(), then hlist_del_rcu() and sychronize_rcu() to remove
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>
Tejun Heoa34375e2014-09-08 09:51:30 +090052#include <linux/gfp.h>
Kent Overstreet215e2622013-05-31 15:26:45 -070053
54struct percpu_ref;
Tejun Heoac899062013-06-12 20:43:06 -070055typedef void (percpu_ref_func_t)(struct percpu_ref *);
Kent Overstreet215e2622013-05-31 15:26:45 -070056
57struct percpu_ref {
Tejun Heoe6253052014-09-20 01:27:25 -040058 atomic_long_t count;
Kent Overstreet215e2622013-05-31 15:26:45 -070059 /*
60 * The low bit of the pointer indicates whether the ref is in percpu
Tejun Heo9a1049d2014-06-28 08:10:14 -040061 * mode; if set, then get/put will manipulate the atomic_t.
Kent Overstreet215e2622013-05-31 15:26:45 -070062 */
Tejun Heoeecc16b2014-09-24 13:31:48 -040063 unsigned long percpu_count_ptr;
Tejun Heoac899062013-06-12 20:43:06 -070064 percpu_ref_func_t *release;
Tejun Heodbece3a2013-06-13 19:23:53 -070065 percpu_ref_func_t *confirm_kill;
Kent Overstreet215e2622013-05-31 15:26:45 -070066 struct rcu_head rcu;
67};
68
Tejun Heoacac7882013-06-12 20:52:01 -070069int __must_check percpu_ref_init(struct percpu_ref *ref,
Tejun Heoa34375e2014-09-08 09:51:30 +090070 percpu_ref_func_t *release, gfp_t gfp);
Tejun Heo9a1049d2014-06-28 08:10:14 -040071void percpu_ref_exit(struct percpu_ref *ref);
Tejun Heodbece3a2013-06-13 19:23:53 -070072void percpu_ref_kill_and_confirm(struct percpu_ref *ref,
73 percpu_ref_func_t *confirm_kill);
Tejun Heoa2237372014-09-24 13:31:48 -040074void percpu_ref_reinit(struct percpu_ref *ref);
Tejun Heodbece3a2013-06-13 19:23:53 -070075
76/**
77 * percpu_ref_kill - drop the initial ref
78 * @ref: percpu_ref to kill
79 *
80 * Must be used to drop the initial ref on a percpu refcount; must be called
81 * precisely once before shutdown.
82 *
83 * Puts @ref in non percpu mode, then does a call_rcu() before gathering up the
84 * percpu counters and dropping the initial ref.
85 */
86static inline void percpu_ref_kill(struct percpu_ref *ref)
87{
88 return percpu_ref_kill_and_confirm(ref, NULL);
89}
Kent Overstreet215e2622013-05-31 15:26:45 -070090
Tejun Heoeecc16b2014-09-24 13:31:48 -040091#define __PERCPU_REF_DEAD 1
Kent Overstreet215e2622013-05-31 15:26:45 -070092
Tejun Heoeae79752014-06-28 08:10:13 -040093/*
94 * Internal helper. Don't use outside percpu-refcount proper. The
95 * function doesn't return the pointer and let the caller test it for NULL
96 * because doing so forces the compiler to generate two conditional
Tejun Heoeecc16b2014-09-24 13:31:48 -040097 * branches as it can't assume that @ref->percpu_count is not NULL.
Tejun Heoeae79752014-06-28 08:10:13 -040098 */
Tejun Heoeecc16b2014-09-24 13:31:48 -040099static inline bool __percpu_ref_alive(struct percpu_ref *ref,
100 unsigned long __percpu **percpu_countp)
Tejun Heoeae79752014-06-28 08:10:13 -0400101{
Tejun Heoeecc16b2014-09-24 13:31:48 -0400102 unsigned long percpu_ptr = ACCESS_ONCE(ref->percpu_count_ptr);
Tejun Heoeae79752014-06-28 08:10:13 -0400103
Tejun Heo2d722782014-06-28 08:10:14 -0400104 /* paired with smp_store_release() in percpu_ref_reinit() */
105 smp_read_barrier_depends();
106
Tejun Heoeecc16b2014-09-24 13:31:48 -0400107 if (unlikely(percpu_ptr & __PERCPU_REF_DEAD))
Tejun Heoeae79752014-06-28 08:10:13 -0400108 return false;
109
Tejun Heoeecc16b2014-09-24 13:31:48 -0400110 *percpu_countp = (unsigned long __percpu *)percpu_ptr;
Tejun Heoeae79752014-06-28 08:10:13 -0400111 return true;
112}
Kent Overstreet215e2622013-05-31 15:26:45 -0700113
114/**
115 * percpu_ref_get - increment a percpu refcount
Tejun Heoac899062013-06-12 20:43:06 -0700116 * @ref: percpu_ref to get
Kent Overstreet215e2622013-05-31 15:26:45 -0700117 *
Tejun Heo6251f992014-09-24 13:31:48 -0400118 * Analagous to atomic_long_inc().
119 *
120 * This function is safe to call as long as @ref is between init and exit.
121 */
Kent Overstreet215e2622013-05-31 15:26:45 -0700122static inline void percpu_ref_get(struct percpu_ref *ref)
123{
Tejun Heoeecc16b2014-09-24 13:31:48 -0400124 unsigned long __percpu *percpu_count;
Kent Overstreet215e2622013-05-31 15:26:45 -0700125
Tejun Heoa4244452013-06-16 16:12:26 -0700126 rcu_read_lock_sched();
Kent Overstreet215e2622013-05-31 15:26:45 -0700127
Tejun Heoeecc16b2014-09-24 13:31:48 -0400128 if (__percpu_ref_alive(ref, &percpu_count))
129 this_cpu_inc(*percpu_count);
Kent Overstreet215e2622013-05-31 15:26:45 -0700130 else
Tejun Heoe6253052014-09-20 01:27:25 -0400131 atomic_long_inc(&ref->count);
Kent Overstreet215e2622013-05-31 15:26:45 -0700132
Tejun Heoa4244452013-06-16 16:12:26 -0700133 rcu_read_unlock_sched();
Kent Overstreet215e2622013-05-31 15:26:45 -0700134}
135
136/**
Tejun Heo4fb6e252014-05-09 15:11:53 -0400137 * percpu_ref_tryget - try to increment a percpu refcount
138 * @ref: percpu_ref to try-get
139 *
140 * Increment a percpu refcount unless its count already reached zero.
141 * Returns %true on success; %false on failure.
142 *
Tejun Heo6251f992014-09-24 13:31:48 -0400143 * This function is safe to call as long as @ref is between init and exit.
Tejun Heo4fb6e252014-05-09 15:11:53 -0400144 */
145static inline bool percpu_ref_tryget(struct percpu_ref *ref)
146{
Tejun Heoeecc16b2014-09-24 13:31:48 -0400147 unsigned long __percpu *percpu_count;
Tejun Heo6251f992014-09-24 13:31:48 -0400148 int ret;
Tejun Heo4fb6e252014-05-09 15:11:53 -0400149
150 rcu_read_lock_sched();
151
Tejun Heoeecc16b2014-09-24 13:31:48 -0400152 if (__percpu_ref_alive(ref, &percpu_count)) {
153 this_cpu_inc(*percpu_count);
Tejun Heo4fb6e252014-05-09 15:11:53 -0400154 ret = true;
155 } else {
Tejun Heoe6253052014-09-20 01:27:25 -0400156 ret = atomic_long_inc_not_zero(&ref->count);
Tejun Heo4fb6e252014-05-09 15:11:53 -0400157 }
158
159 rcu_read_unlock_sched();
160
161 return ret;
162}
163
164/**
Tejun Heo2070d502014-05-09 15:11:53 -0400165 * percpu_ref_tryget_live - try to increment a live percpu refcount
Tejun Heodbece3a2013-06-13 19:23:53 -0700166 * @ref: percpu_ref to try-get
167 *
168 * Increment a percpu refcount unless it has already been killed. Returns
169 * %true on success; %false on failure.
170 *
Tejun Heo6251f992014-09-24 13:31:48 -0400171 * Completion of percpu_ref_kill() in itself doesn't guarantee that this
172 * function will fail. For such guarantee, percpu_ref_kill_and_confirm()
173 * should be used. After the confirm_kill callback is invoked, it's
174 * guaranteed that no new reference will be given out by
175 * percpu_ref_tryget_live().
Tejun Heo4fb6e252014-05-09 15:11:53 -0400176 *
Tejun Heo6251f992014-09-24 13:31:48 -0400177 * This function is safe to call as long as @ref is between init and exit.
Tejun Heodbece3a2013-06-13 19:23:53 -0700178 */
Tejun Heo2070d502014-05-09 15:11:53 -0400179static inline bool percpu_ref_tryget_live(struct percpu_ref *ref)
Tejun Heodbece3a2013-06-13 19:23:53 -0700180{
Tejun Heoeecc16b2014-09-24 13:31:48 -0400181 unsigned long __percpu *percpu_count;
Tejun Heodbece3a2013-06-13 19:23:53 -0700182 int ret = false;
183
Tejun Heoa4244452013-06-16 16:12:26 -0700184 rcu_read_lock_sched();
Tejun Heodbece3a2013-06-13 19:23:53 -0700185
Tejun Heoeecc16b2014-09-24 13:31:48 -0400186 if (__percpu_ref_alive(ref, &percpu_count)) {
187 this_cpu_inc(*percpu_count);
Tejun Heodbece3a2013-06-13 19:23:53 -0700188 ret = true;
189 }
190
Tejun Heoa4244452013-06-16 16:12:26 -0700191 rcu_read_unlock_sched();
Tejun Heodbece3a2013-06-13 19:23:53 -0700192
193 return ret;
194}
195
196/**
Kent Overstreet215e2622013-05-31 15:26:45 -0700197 * percpu_ref_put - decrement a percpu refcount
Tejun Heoac899062013-06-12 20:43:06 -0700198 * @ref: percpu_ref to put
Kent Overstreet215e2622013-05-31 15:26:45 -0700199 *
200 * Decrement the refcount, and if 0, call the release function (which was passed
201 * to percpu_ref_init())
Tejun Heo6251f992014-09-24 13:31:48 -0400202 *
203 * This function is safe to call as long as @ref is between init and exit.
Kent Overstreet215e2622013-05-31 15:26:45 -0700204 */
205static inline void percpu_ref_put(struct percpu_ref *ref)
206{
Tejun Heoeecc16b2014-09-24 13:31:48 -0400207 unsigned long __percpu *percpu_count;
Kent Overstreet215e2622013-05-31 15:26:45 -0700208
Tejun Heoa4244452013-06-16 16:12:26 -0700209 rcu_read_lock_sched();
Kent Overstreet215e2622013-05-31 15:26:45 -0700210
Tejun Heoeecc16b2014-09-24 13:31:48 -0400211 if (__percpu_ref_alive(ref, &percpu_count))
212 this_cpu_dec(*percpu_count);
Tejun Heoe6253052014-09-20 01:27:25 -0400213 else if (unlikely(atomic_long_dec_and_test(&ref->count)))
Kent Overstreet215e2622013-05-31 15:26:45 -0700214 ref->release(ref);
215
Tejun Heoa4244452013-06-16 16:12:26 -0700216 rcu_read_unlock_sched();
Kent Overstreet215e2622013-05-31 15:26:45 -0700217}
218
Tejun Heo2d722782014-06-28 08:10:14 -0400219/**
220 * percpu_ref_is_zero - test whether a percpu refcount reached zero
221 * @ref: percpu_ref to test
222 *
223 * Returns %true if @ref reached zero.
Tejun Heo6251f992014-09-24 13:31:48 -0400224 *
225 * This function is safe to call as long as @ref is between init and exit.
Tejun Heo2d722782014-06-28 08:10:14 -0400226 */
227static inline bool percpu_ref_is_zero(struct percpu_ref *ref)
228{
Tejun Heoeecc16b2014-09-24 13:31:48 -0400229 unsigned long __percpu *percpu_count;
Tejun Heo2d722782014-06-28 08:10:14 -0400230
Tejun Heoeecc16b2014-09-24 13:31:48 -0400231 if (__percpu_ref_alive(ref, &percpu_count))
Tejun Heo2d722782014-06-28 08:10:14 -0400232 return false;
Tejun Heoe6253052014-09-20 01:27:25 -0400233 return !atomic_long_read(&ref->count);
Tejun Heo2d722782014-06-28 08:10:14 -0400234}
235
Kent Overstreet215e2622013-05-31 15:26:45 -0700236#endif