Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 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 |
| 16 | * percpu_ref_kill()/PCPU_COUNT_BIAS. |
| 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 |
| 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> |
| 52 | |
| 53 | struct percpu_ref; |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 54 | typedef void (percpu_ref_func_t)(struct percpu_ref *); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 55 | |
| 56 | struct percpu_ref { |
| 57 | atomic_t count; |
| 58 | /* |
| 59 | * The low bit of the pointer indicates whether the ref is in percpu |
Tejun Heo | 9a1049d | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 60 | * mode; if set, then get/put will manipulate the atomic_t. |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 61 | */ |
Tejun Heo | 7d74207 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 62 | unsigned long pcpu_count_ptr; |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 63 | percpu_ref_func_t *release; |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 64 | percpu_ref_func_t *confirm_kill; |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 65 | struct rcu_head rcu; |
| 66 | }; |
| 67 | |
Tejun Heo | acac788 | 2013-06-12 20:52:01 -0700 | [diff] [blame] | 68 | int __must_check percpu_ref_init(struct percpu_ref *ref, |
| 69 | percpu_ref_func_t *release); |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 70 | void percpu_ref_reinit(struct percpu_ref *ref); |
Tejun Heo | 9a1049d | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 71 | void percpu_ref_exit(struct percpu_ref *ref); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 72 | void percpu_ref_kill_and_confirm(struct percpu_ref *ref, |
| 73 | percpu_ref_func_t *confirm_kill); |
| 74 | |
| 75 | /** |
| 76 | * percpu_ref_kill - drop the initial ref |
| 77 | * @ref: percpu_ref to kill |
| 78 | * |
| 79 | * Must be used to drop the initial ref on a percpu refcount; must be called |
| 80 | * precisely once before shutdown. |
| 81 | * |
| 82 | * Puts @ref in non percpu mode, then does a call_rcu() before gathering up the |
| 83 | * percpu counters and dropping the initial ref. |
| 84 | */ |
| 85 | static inline void percpu_ref_kill(struct percpu_ref *ref) |
| 86 | { |
| 87 | return percpu_ref_kill_and_confirm(ref, NULL); |
| 88 | } |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 89 | |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 90 | #define PCPU_REF_DEAD 1 |
| 91 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 92 | /* |
| 93 | * Internal helper. Don't use outside percpu-refcount proper. The |
| 94 | * function doesn't return the pointer and let the caller test it for NULL |
| 95 | * because doing so forces the compiler to generate two conditional |
| 96 | * branches as it can't assume that @ref->pcpu_count is not NULL. |
| 97 | */ |
| 98 | static inline bool __pcpu_ref_alive(struct percpu_ref *ref, |
| 99 | unsigned __percpu **pcpu_countp) |
| 100 | { |
Tejun Heo | 7d74207 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 101 | unsigned long pcpu_ptr = ACCESS_ONCE(ref->pcpu_count_ptr); |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 102 | |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 103 | /* paired with smp_store_release() in percpu_ref_reinit() */ |
| 104 | smp_read_barrier_depends(); |
| 105 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 106 | if (unlikely(pcpu_ptr & PCPU_REF_DEAD)) |
| 107 | return false; |
| 108 | |
| 109 | *pcpu_countp = (unsigned __percpu *)pcpu_ptr; |
| 110 | return true; |
| 111 | } |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 112 | |
| 113 | /** |
| 114 | * percpu_ref_get - increment a percpu refcount |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 115 | * @ref: percpu_ref to get |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 116 | * |
| 117 | * Analagous to atomic_inc(). |
| 118 | */ |
| 119 | static inline void percpu_ref_get(struct percpu_ref *ref) |
| 120 | { |
| 121 | unsigned __percpu *pcpu_count; |
| 122 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 123 | rcu_read_lock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 124 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 125 | if (__pcpu_ref_alive(ref, &pcpu_count)) |
Sebastian Ott | 0c36b39 | 2014-06-04 15:58:24 +0200 | [diff] [blame] | 126 | this_cpu_inc(*pcpu_count); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 127 | else |
| 128 | atomic_inc(&ref->count); |
| 129 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 130 | rcu_read_unlock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 131 | } |
| 132 | |
| 133 | /** |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 134 | * percpu_ref_tryget - try to increment a percpu refcount |
| 135 | * @ref: percpu_ref to try-get |
| 136 | * |
| 137 | * Increment a percpu refcount unless its count already reached zero. |
| 138 | * Returns %true on success; %false on failure. |
| 139 | * |
| 140 | * The caller is responsible for ensuring that @ref stays accessible. |
| 141 | */ |
| 142 | static inline bool percpu_ref_tryget(struct percpu_ref *ref) |
| 143 | { |
| 144 | unsigned __percpu *pcpu_count; |
| 145 | int ret = false; |
| 146 | |
| 147 | rcu_read_lock_sched(); |
| 148 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 149 | if (__pcpu_ref_alive(ref, &pcpu_count)) { |
Tejun Heo | 315c555 | 2014-06-04 12:50:14 -0400 | [diff] [blame] | 150 | this_cpu_inc(*pcpu_count); |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 151 | ret = true; |
| 152 | } else { |
| 153 | ret = atomic_inc_not_zero(&ref->count); |
| 154 | } |
| 155 | |
| 156 | rcu_read_unlock_sched(); |
| 157 | |
| 158 | return ret; |
| 159 | } |
| 160 | |
| 161 | /** |
Tejun Heo | 2070d50 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 162 | * percpu_ref_tryget_live - try to increment a live percpu refcount |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 163 | * @ref: percpu_ref to try-get |
| 164 | * |
| 165 | * Increment a percpu refcount unless it has already been killed. Returns |
| 166 | * %true on success; %false on failure. |
| 167 | * |
| 168 | * Completion of percpu_ref_kill() in itself doesn't guarantee that tryget |
| 169 | * will fail. For such guarantee, percpu_ref_kill_and_confirm() should be |
| 170 | * used. After the confirm_kill callback is invoked, it's guaranteed that |
| 171 | * no new reference will be given out by percpu_ref_tryget(). |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 172 | * |
| 173 | * The caller is responsible for ensuring that @ref stays accessible. |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 174 | */ |
Tejun Heo | 2070d50 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 175 | static inline bool percpu_ref_tryget_live(struct percpu_ref *ref) |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 176 | { |
| 177 | unsigned __percpu *pcpu_count; |
| 178 | int ret = false; |
| 179 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 180 | rcu_read_lock_sched(); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 181 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 182 | if (__pcpu_ref_alive(ref, &pcpu_count)) { |
Sebastian Ott | 0c36b39 | 2014-06-04 15:58:24 +0200 | [diff] [blame] | 183 | this_cpu_inc(*pcpu_count); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 184 | ret = true; |
| 185 | } |
| 186 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 187 | rcu_read_unlock_sched(); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 188 | |
| 189 | return ret; |
| 190 | } |
| 191 | |
| 192 | /** |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 193 | * percpu_ref_put - decrement a percpu refcount |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 194 | * @ref: percpu_ref to put |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 195 | * |
| 196 | * Decrement the refcount, and if 0, call the release function (which was passed |
| 197 | * to percpu_ref_init()) |
| 198 | */ |
| 199 | static inline void percpu_ref_put(struct percpu_ref *ref) |
| 200 | { |
| 201 | unsigned __percpu *pcpu_count; |
| 202 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 203 | rcu_read_lock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 204 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 205 | if (__pcpu_ref_alive(ref, &pcpu_count)) |
Sebastian Ott | 0c36b39 | 2014-06-04 15:58:24 +0200 | [diff] [blame] | 206 | this_cpu_dec(*pcpu_count); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 207 | else if (unlikely(atomic_dec_and_test(&ref->count))) |
| 208 | ref->release(ref); |
| 209 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 210 | rcu_read_unlock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 211 | } |
| 212 | |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 213 | /** |
| 214 | * percpu_ref_is_zero - test whether a percpu refcount reached zero |
| 215 | * @ref: percpu_ref to test |
| 216 | * |
| 217 | * Returns %true if @ref reached zero. |
| 218 | */ |
| 219 | static inline bool percpu_ref_is_zero(struct percpu_ref *ref) |
| 220 | { |
| 221 | unsigned __percpu *pcpu_count; |
| 222 | |
| 223 | if (__pcpu_ref_alive(ref, &pcpu_count)) |
| 224 | return false; |
| 225 | return !atomic_read(&ref->count); |
| 226 | } |
| 227 | |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 228 | #endif |