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 |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 16 | * percpu_ref_kill()/PERCPU_COUNT_BIAS. |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 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> |
Tejun Heo | a34375e | 2014-09-08 09:51:30 +0900 | [diff] [blame] | 52 | #include <linux/gfp.h> |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 53 | |
| 54 | struct percpu_ref; |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 55 | typedef void (percpu_ref_func_t)(struct percpu_ref *); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 56 | |
| 57 | struct percpu_ref { |
Tejun Heo | e625305 | 2014-09-20 01:27:25 -0400 | [diff] [blame] | 58 | atomic_long_t count; |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 59 | /* |
| 60 | * 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] | 61 | * mode; if set, then get/put will manipulate the atomic_t. |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 62 | */ |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 63 | unsigned long percpu_count_ptr; |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 64 | percpu_ref_func_t *release; |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 65 | percpu_ref_func_t *confirm_kill; |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 66 | struct rcu_head rcu; |
| 67 | }; |
| 68 | |
Tejun Heo | acac788 | 2013-06-12 20:52:01 -0700 | [diff] [blame] | 69 | int __must_check percpu_ref_init(struct percpu_ref *ref, |
Tejun Heo | a34375e | 2014-09-08 09:51:30 +0900 | [diff] [blame] | 70 | percpu_ref_func_t *release, gfp_t gfp); |
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); |
Tejun Heo | a223737 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 74 | void percpu_ref_reinit(struct percpu_ref *ref); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 75 | |
| 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 | */ |
| 86 | static inline void percpu_ref_kill(struct percpu_ref *ref) |
| 87 | { |
| 88 | return percpu_ref_kill_and_confirm(ref, NULL); |
| 89 | } |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 90 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 91 | #define __PERCPU_REF_DEAD 1 |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 92 | |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 93 | /* |
| 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 Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 97 | * branches as it can't assume that @ref->percpu_count is not NULL. |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 98 | */ |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 99 | static inline bool __percpu_ref_alive(struct percpu_ref *ref, |
| 100 | unsigned long __percpu **percpu_countp) |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 101 | { |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 102 | unsigned long percpu_ptr = ACCESS_ONCE(ref->percpu_count_ptr); |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 103 | |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 104 | /* paired with smp_store_release() in percpu_ref_reinit() */ |
| 105 | smp_read_barrier_depends(); |
| 106 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 107 | if (unlikely(percpu_ptr & __PERCPU_REF_DEAD)) |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 108 | return false; |
| 109 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 110 | *percpu_countp = (unsigned long __percpu *)percpu_ptr; |
Tejun Heo | eae7975 | 2014-06-28 08:10:13 -0400 | [diff] [blame] | 111 | return true; |
| 112 | } |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 113 | |
| 114 | /** |
| 115 | * percpu_ref_get - increment a percpu refcount |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 116 | * @ref: percpu_ref to get |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 117 | * |
Tejun Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 118 | * Analagous to atomic_long_inc(). |
| 119 | * |
| 120 | * This function is safe to call as long as @ref is between init and exit. |
| 121 | */ |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 122 | static inline void percpu_ref_get(struct percpu_ref *ref) |
| 123 | { |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 124 | unsigned long __percpu *percpu_count; |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 125 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 126 | rcu_read_lock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 127 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 128 | if (__percpu_ref_alive(ref, &percpu_count)) |
| 129 | this_cpu_inc(*percpu_count); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 130 | else |
Tejun Heo | e625305 | 2014-09-20 01:27:25 -0400 | [diff] [blame] | 131 | atomic_long_inc(&ref->count); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 132 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 133 | rcu_read_unlock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 134 | } |
| 135 | |
| 136 | /** |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 137 | * 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 Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 143 | * This function is safe to call as long as @ref is between init and exit. |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 144 | */ |
| 145 | static inline bool percpu_ref_tryget(struct percpu_ref *ref) |
| 146 | { |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 147 | unsigned long __percpu *percpu_count; |
Tejun Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 148 | int ret; |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 149 | |
| 150 | rcu_read_lock_sched(); |
| 151 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 152 | if (__percpu_ref_alive(ref, &percpu_count)) { |
| 153 | this_cpu_inc(*percpu_count); |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 154 | ret = true; |
| 155 | } else { |
Tejun Heo | e625305 | 2014-09-20 01:27:25 -0400 | [diff] [blame] | 156 | ret = atomic_long_inc_not_zero(&ref->count); |
Tejun Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 157 | } |
| 158 | |
| 159 | rcu_read_unlock_sched(); |
| 160 | |
| 161 | return ret; |
| 162 | } |
| 163 | |
| 164 | /** |
Tejun Heo | 2070d50 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 165 | * percpu_ref_tryget_live - try to increment a live percpu refcount |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 166 | * @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 Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 171 | * 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 Heo | 4fb6e25 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 176 | * |
Tejun Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 177 | * This function is safe to call as long as @ref is between init and exit. |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 178 | */ |
Tejun Heo | 2070d50 | 2014-05-09 15:11:53 -0400 | [diff] [blame] | 179 | static inline bool percpu_ref_tryget_live(struct percpu_ref *ref) |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 180 | { |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 181 | unsigned long __percpu *percpu_count; |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 182 | int ret = false; |
| 183 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 184 | rcu_read_lock_sched(); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 185 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 186 | if (__percpu_ref_alive(ref, &percpu_count)) { |
| 187 | this_cpu_inc(*percpu_count); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 188 | ret = true; |
| 189 | } |
| 190 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 191 | rcu_read_unlock_sched(); |
Tejun Heo | dbece3a | 2013-06-13 19:23:53 -0700 | [diff] [blame] | 192 | |
| 193 | return ret; |
| 194 | } |
| 195 | |
| 196 | /** |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 197 | * percpu_ref_put - decrement a percpu refcount |
Tejun Heo | ac89906 | 2013-06-12 20:43:06 -0700 | [diff] [blame] | 198 | * @ref: percpu_ref to put |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 199 | * |
| 200 | * Decrement the refcount, and if 0, call the release function (which was passed |
| 201 | * to percpu_ref_init()) |
Tejun Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 202 | * |
| 203 | * This function is safe to call as long as @ref is between init and exit. |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 204 | */ |
| 205 | static inline void percpu_ref_put(struct percpu_ref *ref) |
| 206 | { |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 207 | unsigned long __percpu *percpu_count; |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 208 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 209 | rcu_read_lock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 210 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 211 | if (__percpu_ref_alive(ref, &percpu_count)) |
| 212 | this_cpu_dec(*percpu_count); |
Tejun Heo | e625305 | 2014-09-20 01:27:25 -0400 | [diff] [blame] | 213 | else if (unlikely(atomic_long_dec_and_test(&ref->count))) |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 214 | ref->release(ref); |
| 215 | |
Tejun Heo | a424445 | 2013-06-16 16:12:26 -0700 | [diff] [blame] | 216 | rcu_read_unlock_sched(); |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 217 | } |
| 218 | |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 219 | /** |
| 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 Heo | 6251f99 | 2014-09-24 13:31:48 -0400 | [diff] [blame] | 224 | * |
| 225 | * This function is safe to call as long as @ref is between init and exit. |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 226 | */ |
| 227 | static inline bool percpu_ref_is_zero(struct percpu_ref *ref) |
| 228 | { |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 229 | unsigned long __percpu *percpu_count; |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 230 | |
Tejun Heo | eecc16b | 2014-09-24 13:31:48 -0400 | [diff] [blame^] | 231 | if (__percpu_ref_alive(ref, &percpu_count)) |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 232 | return false; |
Tejun Heo | e625305 | 2014-09-20 01:27:25 -0400 | [diff] [blame] | 233 | return !atomic_long_read(&ref->count); |
Tejun Heo | 2d72278 | 2014-06-28 08:10:14 -0400 | [diff] [blame] | 234 | } |
| 235 | |
Kent Overstreet | 215e262 | 2013-05-31 15:26:45 -0700 | [diff] [blame] | 236 | #endif |