diff options
Diffstat (limited to 'kernel/slow-work.c')
| -rw-r--r-- | kernel/slow-work.c | 640 |
1 files changed, 640 insertions, 0 deletions
diff --git a/kernel/slow-work.c b/kernel/slow-work.c new file mode 100644 index 00000000000..cf2bc01186e --- /dev/null +++ b/kernel/slow-work.c | |||
| @@ -0,0 +1,640 @@ | |||
| 1 | /* Worker thread pool for slow items, such as filesystem lookups or mkdirs | ||
| 2 | * | ||
| 3 | * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved. | ||
| 4 | * Written by David Howells (dhowells@redhat.com) | ||
| 5 | * | ||
| 6 | * This program is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU General Public Licence | ||
| 8 | * as published by the Free Software Foundation; either version | ||
| 9 | * 2 of the Licence, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * See Documentation/slow-work.txt | ||
| 12 | */ | ||
| 13 | |||
| 14 | #include <linux/module.h> | ||
| 15 | #include <linux/slow-work.h> | ||
| 16 | #include <linux/kthread.h> | ||
| 17 | #include <linux/freezer.h> | ||
| 18 | #include <linux/wait.h> | ||
| 19 | |||
| 20 | #define SLOW_WORK_CULL_TIMEOUT (5 * HZ) /* cull threads 5s after running out of | ||
| 21 | * things to do */ | ||
| 22 | #define SLOW_WORK_OOM_TIMEOUT (5 * HZ) /* can't start new threads for 5s after | ||
| 23 | * OOM */ | ||
| 24 | |||
| 25 | static void slow_work_cull_timeout(unsigned long); | ||
| 26 | static void slow_work_oom_timeout(unsigned long); | ||
| 27 | |||
| 28 | #ifdef CONFIG_SYSCTL | ||
| 29 | static int slow_work_min_threads_sysctl(struct ctl_table *, int, struct file *, | ||
| 30 | void __user *, size_t *, loff_t *); | ||
| 31 | |||
| 32 | static int slow_work_max_threads_sysctl(struct ctl_table *, int , struct file *, | ||
| 33 | void __user *, size_t *, loff_t *); | ||
| 34 | #endif | ||
| 35 | |||
| 36 | /* | ||
| 37 | * The pool of threads has at least min threads in it as long as someone is | ||
| 38 | * using the facility, and may have as many as max. | ||
| 39 | * | ||
| 40 | * A portion of the pool may be processing very slow operations. | ||
| 41 | */ | ||
| 42 | static unsigned slow_work_min_threads = 2; | ||
| 43 | static unsigned slow_work_max_threads = 4; | ||
| 44 | static unsigned vslow_work_proportion = 50; /* % of threads that may process | ||
| 45 | * very slow work */ | ||
| 46 | |||
| 47 | #ifdef CONFIG_SYSCTL | ||
| 48 | static const int slow_work_min_min_threads = 2; | ||
| 49 | static int slow_work_max_max_threads = 255; | ||
| 50 | static const int slow_work_min_vslow = 1; | ||
| 51 | static const int slow_work_max_vslow = 99; | ||
| 52 | |||
| 53 | ctl_table slow_work_sysctls[] = { | ||
| 54 | { | ||
| 55 | .ctl_name = CTL_UNNUMBERED, | ||
| 56 | .procname = "min-threads", | ||
| 57 | .data = &slow_work_min_threads, | ||
| 58 | .maxlen = sizeof(unsigned), | ||
| 59 | .mode = 0644, | ||
| 60 | .proc_handler = slow_work_min_threads_sysctl, | ||
| 61 | .extra1 = (void *) &slow_work_min_min_threads, | ||
| 62 | .extra2 = &slow_work_max_threads, | ||
| 63 | }, | ||
| 64 | { | ||
| 65 | .ctl_name = CTL_UNNUMBERED, | ||
| 66 | .procname = "max-threads", | ||
| 67 | .data = &slow_work_max_threads, | ||
| 68 | .maxlen = sizeof(unsigned), | ||
| 69 | .mode = 0644, | ||
| 70 | .proc_handler = slow_work_max_threads_sysctl, | ||
| 71 | .extra1 = &slow_work_min_threads, | ||
| 72 | .extra2 = (void *) &slow_work_max_max_threads, | ||
| 73 | }, | ||
| 74 | { | ||
| 75 | .ctl_name = CTL_UNNUMBERED, | ||
| 76 | .procname = "vslow-percentage", | ||
| 77 | .data = &vslow_work_proportion, | ||
| 78 | .maxlen = sizeof(unsigned), | ||
| 79 | .mode = 0644, | ||
| 80 | .proc_handler = &proc_dointvec_minmax, | ||
| 81 | .extra1 = (void *) &slow_work_min_vslow, | ||
| 82 | .extra2 = (void *) &slow_work_max_vslow, | ||
| 83 | }, | ||
| 84 | { .ctl_name = 0 } | ||
| 85 | }; | ||
| 86 | #endif | ||
| 87 | |||
| 88 | /* | ||
| 89 | * The active state of the thread pool | ||
| 90 | */ | ||
| 91 | static atomic_t slow_work_thread_count; | ||
| 92 | static atomic_t vslow_work_executing_count; | ||
| 93 | |||
| 94 | static bool slow_work_may_not_start_new_thread; | ||
| 95 | static bool slow_work_cull; /* cull a thread due to lack of activity */ | ||
| 96 | static DEFINE_TIMER(slow_work_cull_timer, slow_work_cull_timeout, 0, 0); | ||
| 97 | static DEFINE_TIMER(slow_work_oom_timer, slow_work_oom_timeout, 0, 0); | ||
| 98 | static struct slow_work slow_work_new_thread; /* new thread starter */ | ||
| 99 | |||
| 100 | /* | ||
| 101 | * The queues of work items and the lock governing access to them. These are | ||
| 102 | * shared between all the CPUs. It doesn't make sense to have per-CPU queues | ||
| 103 | * as the number of threads bears no relation to the number of CPUs. | ||
| 104 | * | ||
| 105 | * There are two queues of work items: one for slow work items, and one for | ||
| 106 | * very slow work items. | ||
| 107 | */ | ||
| 108 | static LIST_HEAD(slow_work_queue); | ||
| 109 | static LIST_HEAD(vslow_work_queue); | ||
| 110 | static DEFINE_SPINLOCK(slow_work_queue_lock); | ||
| 111 | |||
| 112 | /* | ||
| 113 | * The thread controls. A variable used to signal to the threads that they | ||
| 114 | * should exit when the queue is empty, a waitqueue used by the threads to wait | ||
| 115 | * for signals, and a completion set by the last thread to exit. | ||
| 116 | */ | ||
| 117 | static bool slow_work_threads_should_exit; | ||
| 118 | static DECLARE_WAIT_QUEUE_HEAD(slow_work_thread_wq); | ||
| 119 | static DECLARE_COMPLETION(slow_work_last_thread_exited); | ||
| 120 | |||
| 121 | /* | ||
| 122 | * The number of users of the thread pool and its lock. Whilst this is zero we | ||
| 123 | * have no threads hanging around, and when this reaches zero, we wait for all | ||
| 124 | * active or queued work items to complete and kill all the threads we do have. | ||
| 125 | */ | ||
| 126 | static int slow_work_user_count; | ||
| 127 | static DEFINE_MUTEX(slow_work_user_lock); | ||
| 128 | |||
| 129 | /* | ||
| 130 | * Calculate the maximum number of active threads in the pool that are | ||
| 131 | * permitted to process very slow work items. | ||
| 132 | * | ||
| 133 | * The answer is rounded up to at least 1, but may not equal or exceed the | ||
| 134 | * maximum number of the threads in the pool. This means we always have at | ||
| 135 | * least one thread that can process slow work items, and we always have at | ||
| 136 | * least one thread that won't get tied up doing so. | ||
| 137 | */ | ||
| 138 | static unsigned slow_work_calc_vsmax(void) | ||
| 139 | { | ||
| 140 | unsigned vsmax; | ||
| 141 | |||
| 142 | vsmax = atomic_read(&slow_work_thread_count) * vslow_work_proportion; | ||
| 143 | vsmax /= 100; | ||
| 144 | vsmax = max(vsmax, 1U); | ||
| 145 | return min(vsmax, slow_work_max_threads - 1); | ||
| 146 | } | ||
| 147 | |||
| 148 | /* | ||
| 149 | * Attempt to execute stuff queued on a slow thread. Return true if we managed | ||
| 150 | * it, false if there was nothing to do. | ||
| 151 | */ | ||
| 152 | static bool slow_work_execute(void) | ||
| 153 | { | ||
| 154 | struct slow_work *work = NULL; | ||
| 155 | unsigned vsmax; | ||
| 156 | bool very_slow; | ||
| 157 | |||
| 158 | vsmax = slow_work_calc_vsmax(); | ||
| 159 | |||
| 160 | /* see if we can schedule a new thread to be started if we're not | ||
| 161 | * keeping up with the work */ | ||
| 162 | if (!waitqueue_active(&slow_work_thread_wq) && | ||
| 163 | (!list_empty(&slow_work_queue) || !list_empty(&vslow_work_queue)) && | ||
| 164 | atomic_read(&slow_work_thread_count) < slow_work_max_threads && | ||
| 165 | !slow_work_may_not_start_new_thread) | ||
| 166 | slow_work_enqueue(&slow_work_new_thread); | ||
| 167 | |||
| 168 | /* find something to execute */ | ||
| 169 | spin_lock_irq(&slow_work_queue_lock); | ||
| 170 | if (!list_empty(&vslow_work_queue) && | ||
| 171 | atomic_read(&vslow_work_executing_count) < vsmax) { | ||
| 172 | work = list_entry(vslow_work_queue.next, | ||
| 173 | struct slow_work, link); | ||
| 174 | if (test_and_set_bit_lock(SLOW_WORK_EXECUTING, &work->flags)) | ||
| 175 | BUG(); | ||
| 176 | list_del_init(&work->link); | ||
| 177 | atomic_inc(&vslow_work_executing_count); | ||
| 178 | very_slow = true; | ||
| 179 | } else if (!list_empty(&slow_work_queue)) { | ||
| 180 | work = list_entry(slow_work_queue.next, | ||
| 181 | struct slow_work, link); | ||
| 182 | if (test_and_set_bit_lock(SLOW_WORK_EXECUTING, &work->flags)) | ||
| 183 | BUG(); | ||
| 184 | list_del_init(&work->link); | ||
| 185 | very_slow = false; | ||
| 186 | } else { | ||
| 187 | very_slow = false; /* avoid the compiler warning */ | ||
| 188 | } | ||
| 189 | spin_unlock_irq(&slow_work_queue_lock); | ||
| 190 | |||
| 191 | if (!work) | ||
| 192 | return false; | ||
| 193 | |||
| 194 | if (!test_and_clear_bit(SLOW_WORK_PENDING, &work->flags)) | ||
| 195 | BUG(); | ||
| 196 | |||
| 197 | work->ops->execute(work); | ||
| 198 | |||
| 199 | if (very_slow) | ||
| 200 | atomic_dec(&vslow_work_executing_count); | ||
| 201 | clear_bit_unlock(SLOW_WORK_EXECUTING, &work->flags); | ||
| 202 | |||
| 203 | /* if someone tried to enqueue the item whilst we were executing it, | ||
| 204 | * then it'll be left unenqueued to avoid multiple threads trying to | ||
| 205 | * execute it simultaneously | ||
| 206 | * | ||
| 207 | * there is, however, a race between us testing the pending flag and | ||
| 208 | * getting the spinlock, and between the enqueuer setting the pending | ||
| 209 | * flag and getting the spinlock, so we use a deferral bit to tell us | ||
| 210 | * if the enqueuer got there first | ||
| 211 | */ | ||
| 212 | if (test_bit(SLOW_WORK_PENDING, &work->flags)) { | ||
| 213 | spin_lock_irq(&slow_work_queue_lock); | ||
| 214 | |||
| 215 | if (!test_bit(SLOW_WORK_EXECUTING, &work->flags) && | ||
| 216 | |||
