]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/pdflush.c - worker threads for writing back filesystem data | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
e1f8e874 | 6 | * 09Apr2002 Andrew Morton |
1da177e4 LT |
7 | * Initial version |
8 | * 29Feb2004 kaos@sgi.com | |
9 | * Move worker thread creation to kthread to avoid chewing | |
10 | * up stack space with nested calls to kernel_thread. | |
11 | */ | |
12 | ||
13 | #include <linux/sched.h> | |
14 | #include <linux/list.h> | |
15 | #include <linux/signal.h> | |
16 | #include <linux/spinlock.h> | |
17 | #include <linux/gfp.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/module.h> | |
f5264481 PM |
20 | #include <linux/fs.h> /* Needed by writeback.h */ |
21 | #include <linux/writeback.h> /* Prototypes pdflush_operation() */ | |
1da177e4 | 22 | #include <linux/kthread.h> |
28a42b9e | 23 | #include <linux/cpuset.h> |
7dfb7103 | 24 | #include <linux/freezer.h> |
1da177e4 LT |
25 | |
26 | ||
27 | /* | |
28 | * Minimum and maximum number of pdflush instances | |
29 | */ | |
30 | #define MIN_PDFLUSH_THREADS 2 | |
31 | #define MAX_PDFLUSH_THREADS 8 | |
32 | ||
33 | static void start_one_pdflush_thread(void); | |
34 | ||
35 | ||
36 | /* | |
37 | * The pdflush threads are worker threads for writing back dirty data. | |
38 | * Ideally, we'd like one thread per active disk spindle. But the disk | |
39 | * topology is very hard to divine at this level. Instead, we take | |
40 | * care in various places to prevent more than one pdflush thread from | |
41 | * performing writeback against a single filesystem. pdflush threads | |
42 | * have the PF_FLUSHER flag set in current->flags to aid in this. | |
43 | */ | |
44 | ||
45 | /* | |
46 | * All the pdflush threads. Protected by pdflush_lock | |
47 | */ | |
48 | static LIST_HEAD(pdflush_list); | |
49 | static DEFINE_SPINLOCK(pdflush_lock); | |
50 | ||
51 | /* | |
52 | * The count of currently-running pdflush threads. Protected | |
53 | * by pdflush_lock. | |
54 | * | |
55 | * Readable by sysctl, but not writable. Published to userspace at | |
56 | * /proc/sys/vm/nr_pdflush_threads. | |
57 | */ | |
58 | int nr_pdflush_threads = 0; | |
59 | ||
60 | /* | |
61 | * The time at which the pdflush thread pool last went empty | |
62 | */ | |
63 | static unsigned long last_empty_jifs; | |
64 | ||
65 | /* | |
66 | * The pdflush thread. | |
67 | * | |
68 | * Thread pool management algorithm: | |
69 | * | |
70 | * - The minimum and maximum number of pdflush instances are bound | |
cd17cbfd | 71 | * by MIN_PDFLUSH_THREADS and MAX_PDFLUSH_THREADS. |
1da177e4 LT |
72 | * |
73 | * - If there have been no idle pdflush instances for 1 second, create | |
74 | * a new one. | |
75 | * | |
76 | * - If the least-recently-went-to-sleep pdflush thread has been asleep | |
77 | * for more than one second, terminate a thread. | |
78 | */ | |
79 | ||
80 | /* | |
81 | * A structure for passing work to a pdflush thread. Also for passing | |
82 | * state information between pdflush threads. Protected by pdflush_lock. | |
83 | */ | |
84 | struct pdflush_work { | |
85 | struct task_struct *who; /* The thread */ | |
86 | void (*fn)(unsigned long); /* A callback function */ | |
87 | unsigned long arg0; /* An argument to the callback */ | |
88 | struct list_head list; /* On pdflush_list, when idle */ | |
89 | unsigned long when_i_went_to_sleep; | |
90 | }; | |
91 | ||
92 | static int __pdflush(struct pdflush_work *my_work) | |
93 | { | |
930d9152 | 94 | current->flags |= PF_FLUSHER | PF_SWAPWRITE; |
83144186 | 95 | set_freezable(); |
1da177e4 LT |
96 | my_work->fn = NULL; |
97 | my_work->who = current; | |
98 | INIT_LIST_HEAD(&my_work->list); | |
99 | ||
100 | spin_lock_irq(&pdflush_lock); | |
1da177e4 LT |
101 | for ( ; ; ) { |
102 | struct pdflush_work *pdf; | |
103 | ||
104 | set_current_state(TASK_INTERRUPTIBLE); | |
105 | list_move(&my_work->list, &pdflush_list); | |
106 | my_work->when_i_went_to_sleep = jiffies; | |
107 | spin_unlock_irq(&pdflush_lock); | |
1da177e4 | 108 | schedule(); |
d616e09a | 109 | try_to_freeze(); |
1da177e4 LT |
110 | spin_lock_irq(&pdflush_lock); |
111 | if (!list_empty(&my_work->list)) { | |
d616e09a AM |
112 | /* |
113 | * Someone woke us up, but without removing our control | |
114 | * structure from the global list. swsusp will do this | |
115 | * in try_to_freeze()->refrigerator(). Handle it. | |
116 | */ | |
1da177e4 LT |
117 | my_work->fn = NULL; |
118 | continue; | |
119 | } | |
120 | if (my_work->fn == NULL) { | |
d616e09a | 121 | printk("pdflush: bogus wakeup\n"); |
1da177e4 LT |
122 | continue; |
123 | } | |
124 | spin_unlock_irq(&pdflush_lock); | |
125 | ||
126 | (*my_work->fn)(my_work->arg0); | |
127 | ||
a56ed663 PM |
128 | spin_lock_irq(&pdflush_lock); |
129 | ||
1da177e4 LT |
130 | /* |
131 | * Thread creation: For how long have there been zero | |
132 | * available threads? | |
a56ed663 PM |
133 | * |
134 | * To throttle creation, we reset last_empty_jifs. | |
1da177e4 | 135 | */ |
2b4bc460 | 136 | if (time_after(jiffies, last_empty_jifs + 1 * HZ)) { |
cd17cbfd JA |
137 | if (list_empty(&pdflush_list)) { |
138 | if (nr_pdflush_threads < MAX_PDFLUSH_THREADS) { | |
139 | last_empty_jifs = jiffies; | |
140 | nr_pdflush_threads++; | |
141 | spin_unlock_irq(&pdflush_lock); | |
142 | start_one_pdflush_thread(); | |
143 | spin_lock_irq(&pdflush_lock); | |
144 | } | |
1da177e4 LT |
145 | } |
146 | } | |
147 | ||
1da177e4 LT |
148 | my_work->fn = NULL; |
149 | ||
150 | /* | |
151 | * Thread destruction: For how long has the sleepiest | |
152 | * thread slept? | |
153 | */ | |
154 | if (list_empty(&pdflush_list)) | |
155 | continue; | |
cd17cbfd | 156 | if (nr_pdflush_threads <= MIN_PDFLUSH_THREADS) |
1da177e4 LT |
157 | continue; |
158 | pdf = list_entry(pdflush_list.prev, struct pdflush_work, list); | |
2b4bc460 | 159 | if (time_after(jiffies, pdf->when_i_went_to_sleep + 1 * HZ)) { |
1da177e4 LT |
160 | /* Limit exit rate */ |
161 | pdf->when_i_went_to_sleep = jiffies; | |
162 | break; /* exeunt */ | |
163 | } | |
164 | } | |
165 | nr_pdflush_threads--; | |
166 | spin_unlock_irq(&pdflush_lock); | |
167 | return 0; | |
168 | } | |
169 | ||
170 | /* | |
171 | * Of course, my_work wants to be just a local in __pdflush(). It is | |
172 | * separated out in this manner to hopefully prevent the compiler from | |
173 | * performing unfortunate optimisations against the auto variables. Because | |
174 | * these are visible to other tasks and CPUs. (No problem has actually | |
175 | * been observed. This is just paranoia). | |
176 | */ | |
177 | static int pdflush(void *dummy) | |
178 | { | |
179 | struct pdflush_work my_work; | |
174596a0 RR |
180 | cpumask_var_t cpus_allowed; |
181 | ||
182 | /* | |
183 | * Since the caller doesn't even check kthread_run() worked, let's not | |
184 | * freak out too much if this fails. | |
185 | */ | |
186 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { | |
187 | printk(KERN_WARNING "pdflush failed to allocate cpumask\n"); | |
188 | return 0; | |
189 | } | |
1da177e4 LT |
190 | |
191 | /* | |
192 | * pdflush can spend a lot of time doing encryption via dm-crypt. We | |
193 | * don't want to do that at keventd's priority. | |
194 | */ | |
195 | set_user_nice(current, 0); | |
28a42b9e PJ |
196 | |
197 | /* | |
198 | * Some configs put our parent kthread in a limited cpuset, | |
1a2142af | 199 | * which kthread() overrides, forcing cpus_allowed == cpu_all_mask. |
28a42b9e PJ |
200 | * Our needs are more modest - cut back to our cpusets cpus_allowed. |
201 | * This is needed as pdflush's are dynamically created and destroyed. | |
202 | * The boottime pdflush's are easily placed w/o these 2 lines. | |
203 | */ | |
174596a0 RR |
204 | cpuset_cpus_allowed(current, cpus_allowed); |
205 | set_cpus_allowed_ptr(current, cpus_allowed); | |
206 | free_cpumask_var(cpus_allowed); | |
28a42b9e | 207 | |
1da177e4 LT |
208 | return __pdflush(&my_work); |
209 | } | |
210 | ||
211 | /* | |
212 | * Attempt to wake up a pdflush thread, and get it to do some work for you. | |
213 | * Returns zero if it indeed managed to find a worker thread, and passed your | |
214 | * payload to it. | |
215 | */ | |
216 | int pdflush_operation(void (*fn)(unsigned long), unsigned long arg0) | |
217 | { | |
218 | unsigned long flags; | |
219 | int ret = 0; | |
220 | ||
125e1874 | 221 | BUG_ON(fn == NULL); /* Hard to diagnose if it's deferred */ |
1da177e4 LT |
222 | |
223 | spin_lock_irqsave(&pdflush_lock, flags); | |
224 | if (list_empty(&pdflush_list)) { | |
1da177e4 LT |
225 | ret = -1; |
226 | } else { | |
227 | struct pdflush_work *pdf; | |
228 | ||
229 | pdf = list_entry(pdflush_list.next, struct pdflush_work, list); | |
230 | list_del_init(&pdf->list); | |
231 | if (list_empty(&pdflush_list)) | |
232 | last_empty_jifs = jiffies; | |
233 | pdf->fn = fn; | |
234 | pdf->arg0 = arg0; | |
235 | wake_up_process(pdf->who); | |
1da177e4 | 236 | } |
5aecd559 DC |
237 | spin_unlock_irqrestore(&pdflush_lock, flags); |
238 | ||
1da177e4 LT |
239 | return ret; |
240 | } | |
241 | ||
242 | static void start_one_pdflush_thread(void) | |
243 | { | |
a56ed663 PM |
244 | struct task_struct *k; |
245 | ||
246 | k = kthread_run(pdflush, NULL, "pdflush"); | |
247 | if (unlikely(IS_ERR(k))) { | |
248 | spin_lock_irq(&pdflush_lock); | |
249 | nr_pdflush_threads--; | |
250 | spin_unlock_irq(&pdflush_lock); | |
251 | } | |
1da177e4 LT |
252 | } |
253 | ||
254 | static int __init pdflush_init(void) | |
255 | { | |
256 | int i; | |
257 | ||
a56ed663 PM |
258 | /* |
259 | * Pre-set nr_pdflush_threads... If we fail to create, | |
260 | * the count will be decremented. | |
261 | */ | |
cd17cbfd | 262 | nr_pdflush_threads = MIN_PDFLUSH_THREADS; |
a56ed663 | 263 | |
cd17cbfd | 264 | for (i = 0; i < MIN_PDFLUSH_THREADS; i++) |
1da177e4 LT |
265 | start_one_pdflush_thread(); |
266 | return 0; | |
267 | } | |
268 | ||
269 | module_init(pdflush_init); |