diff options
author | Artem Bityutskiy <Artem.Bityutskiy@nokia.com> | 2010-07-25 07:29:22 -0400 |
---|---|---|
committer | Jens Axboe <jaxboe@fusionio.com> | 2010-08-07 12:53:56 -0400 |
commit | 6467716a37673e8d47b4984eb19839bdad0a8353 (patch) | |
tree | 8b2bfd38d53e31c47654162d4ce9220c4424a31d /mm | |
parent | 253c34e9b10c30d3064be654b5b78fbc1a8b1896 (diff) |
writeback: optimize periodic bdi thread wakeups
Whe the first inode for a bdi is marked dirty, we wake up the bdi thread which
should take care of the periodic background write-out. However, the write-out
will actually start only 'dirty_writeback_interval' centisecs later, so we can
delay the wake-up.
This change was requested by Nick Piggin who pointed out that if we delay the
wake-up, we weed out 2 unnecessary contex switches, which matters because
'__mark_inode_dirty()' is a hot-path function.
This patch introduces a new function - 'bdi_wakeup_thread_delayed()', which
sets up a timer to wake-up the bdi thread and returns. So the wake-up is
delayed.
We also delete the timer in bdi threads just before writing-back. And
synchronously delete it when unregistering bdi. At the unregister point the bdi
does not have any users, so no one can arm it again.
Since now we take 'bdi->wb_lock' in the timer, which can execute in softirq
context, we have to use 'spin_lock_bh()' for 'bdi->wb_lock'. This patch makes
this change as well.
This patch also moves the 'bdi_wb_init()' function down in the file to avoid
forward-declaration of 'bdi_wakeup_thread_delayed()'.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Diffstat (limited to 'mm')
-rw-r--r-- | mm/backing-dev.c | 73 |
1 files changed, 57 insertions, 16 deletions
diff --git a/mm/backing-dev.c b/mm/backing-dev.c index a9a08d88a74..cfff7225138 100644 --- a/mm/backing-dev.c +++ b/mm/backing-dev.c | |||
@@ -248,17 +248,6 @@ static int __init default_bdi_init(void) | |||
248 | } | 248 | } |
249 | subsys_initcall(default_bdi_init); | 249 | subsys_initcall(default_bdi_init); |
250 | 250 | ||
251 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) | ||
252 | { | ||
253 | memset(wb, 0, sizeof(*wb)); | ||
254 | |||
255 | wb->bdi = bdi; | ||
256 | wb->last_old_flush = jiffies; | ||
257 | INIT_LIST_HEAD(&wb->b_dirty); | ||
258 | INIT_LIST_HEAD(&wb->b_io); | ||
259 | INIT_LIST_HEAD(&wb->b_more_io); | ||
260 | } | ||
261 | |||
262 | int bdi_has_dirty_io(struct backing_dev_info *bdi) | 251 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
263 | { | 252 | { |
264 | return wb_has_dirty_io(&bdi->wb); | 253 | return wb_has_dirty_io(&bdi->wb); |
@@ -316,6 +305,43 @@ static void sync_supers_timer_fn(unsigned long unused) | |||
316 | bdi_arm_supers_timer(); | 305 | bdi_arm_supers_timer(); |
317 | } | 306 | } |
318 | 307 | ||
308 | static void wakeup_timer_fn(unsigned long data) | ||
309 | { | ||
310 | struct backing_dev_info *bdi = (struct backing_dev_info *)data; | ||
311 | |||
312 | spin_lock_bh(&bdi->wb_lock); | ||
313 | if (bdi->wb.task) { | ||
314 | wake_up_process(bdi->wb.task); | ||
315 | } else { | ||
316 | /* | ||
317 | * When bdi tasks are inactive for long time, they are killed. | ||
318 | * In this case we have to wake-up the forker thread which | ||
319 | * should create and run the bdi thread. | ||
320 | */ | ||
321 | wake_up_process(default_backing_dev_info.wb.task); | ||
322 | } | ||
323 | spin_unlock_bh(&bdi->wb_lock); | ||
324 | } | ||
325 | |||
326 | /* | ||
327 | * This function is used when the first inode for this bdi is marked dirty. It | ||
328 | * wakes-up the corresponding bdi thread which should then take care of the | ||
329 | * periodic background write-out of dirty inodes. Since the write-out would | ||
330 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | ||
331 | * set up a timer which wakes the bdi thread up later. | ||
332 | * | ||
333 | * Note, we wouldn't bother setting up the timer, but this function is on the | ||
334 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | ||
335 | * by delaying the wake-up. | ||
336 | */ | ||
337 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | ||
338 | { | ||
339 | unsigned long timeout; | ||
340 | |||
341 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | ||
342 | mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout); | ||
343 | } | ||
344 | |||
319 | /* | 345 | /* |
320 | * Calculate the longest interval (jiffies) bdi threads are allowed to be | 346 | * Calculate the longest interval (jiffies) bdi threads are allowed to be |
321 | * inactive. | 347 | * inactive. |
@@ -353,8 +379,10 @@ static int bdi_forker_thread(void *ptr) | |||
353 | * Temporary measure, we want to make sure we don't see | 379 | * Temporary measure, we want to make sure we don't see |
354 | * dirty data on the default backing_dev_info | 380 | * dirty data on the default backing_dev_info |
355 | */ | 381 | */ |
356 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) | 382 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) { |
383 | del_timer(&me->wakeup_timer); | ||
357 | wb_do_writeback(me, 0); | 384 | wb_do_writeback(me, 0); |
385 | } | ||
358 | 386 | ||
359 | spin_lock_bh(&bdi_lock); | 387 | spin_lock_bh(&bdi_lock); |
360 | set_current_state(TASK_INTERRUPTIBLE); | 388 | set_current_state(TASK_INTERRUPTIBLE); |
@@ -386,7 +414,7 @@ static int bdi_forker_thread(void *ptr) | |||
386 | break; | 414 | break; |
387 | } | 415 | } |
388 | 416 | ||
389 | spin_lock(&bdi->wb_lock); | 417 | spin_lock_bh(&bdi->wb_lock); |
390 | /* | 418 | /* |
391 | * If there is no work to do and the bdi thread was | 419 | * If there is no work to do and the bdi thread was |
392 | * inactive long enough - kill it. The wb_lock is taken | 420 | * inactive long enough - kill it. The wb_lock is taken |
@@ -403,7 +431,7 @@ static int bdi_forker_thread(void *ptr) | |||
403 | action = KILL_THREAD; | 431 | action = KILL_THREAD; |
404 | break; | 432 | break; |
405 | } | 433 | } |
406 | spin_unlock(&bdi->wb_lock); | 434 | spin_unlock_bh(&bdi->wb_lock); |
407 | } | 435 | } |
408 | spin_unlock_bh(&bdi_lock); | 436 | spin_unlock_bh(&bdi_lock); |
409 | 437 | ||
@@ -427,9 +455,9 @@ static int bdi_forker_thread(void *ptr) | |||
427 | * The spinlock makes sure we do not lose | 455 | * The spinlock makes sure we do not lose |
428 | * wake-ups when racing with 'bdi_queue_work()'. | 456 | * wake-ups when racing with 'bdi_queue_work()'. |
429 | */ | 457 | */ |
430 | spin_lock(&bdi->wb_lock); | 458 | spin_lock_bh(&bdi->wb_lock); |
431 | bdi->wb.task = task; | 459 | bdi->wb.task = task; |
432 | spin_unlock(&bdi->wb_lock); | 460 | spin_unlock_bh(&bdi->wb_lock); |
433 | } | 461 | } |
434 | break; | 462 | break; |
435 | 463 | ||
@@ -586,6 +614,7 @@ void bdi_unregister(struct backing_dev_info *bdi) | |||
586 | if (bdi->dev) { | 614 | if (bdi->dev) { |
587 | trace_writeback_bdi_unregister(bdi); | 615 | trace_writeback_bdi_unregister(bdi); |
588 | bdi_prune_sb(bdi); | 616 | bdi_prune_sb(bdi); |
617 | del_timer_sync(&bdi->wb.wakeup_timer); | ||
589 | 618 | ||
590 | if (!bdi_cap_flush_forker(bdi)) | 619 | if (!bdi_cap_flush_forker(bdi)) |
591 | bdi_wb_shutdown(bdi); | 620 | bdi_wb_shutdown(bdi); |
@@ -596,6 +625,18 @@ void bdi_unregister(struct backing_dev_info *bdi) | |||
596 | } | 625 | } |
597 | EXPORT_SYMBOL(bdi_unregister); | 626 | EXPORT_SYMBOL(bdi_unregister); |
598 | 627 | ||
628 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) | ||
629 | { | ||
630 | memset(wb, 0, sizeof(*wb)); | ||
631 | |||
632 | wb->bdi = bdi; | ||
633 | wb->last_old_flush = jiffies; | ||
634 | INIT_LIST_HEAD(&wb->b_dirty); | ||
635 | INIT_LIST_HEAD(&wb->b_io); | ||
636 | INIT_LIST_HEAD(&wb->b_more_io); | ||
637 | setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi); | ||
638 | } | ||
639 | |||
599 | int bdi_init(struct backing_dev_info *bdi) | 640 | int bdi_init(struct backing_dev_info *bdi) |
600 | { | 641 | { |
601 | int i, err; | 642 | int i, err; |