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 /fs/fs-writeback.c | |
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 'fs/fs-writeback.c')
-rw-r--r-- | fs/fs-writeback.c | 36 |
1 files changed, 11 insertions, 25 deletions
diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c index 55f6e46e06f1..bfa2df2c7ce2 100644 --- a/fs/fs-writeback.c +++ b/fs/fs-writeback.c | |||
@@ -76,7 +76,7 @@ static void bdi_queue_work(struct backing_dev_info *bdi, | |||
76 | { | 76 | { |
77 | trace_writeback_queue(bdi, work); | 77 | trace_writeback_queue(bdi, work); |
78 | 78 | ||
79 | spin_lock(&bdi->wb_lock); | 79 | spin_lock_bh(&bdi->wb_lock); |
80 | list_add_tail(&work->list, &bdi->work_list); | 80 | list_add_tail(&work->list, &bdi->work_list); |
81 | if (bdi->wb.task) { | 81 | if (bdi->wb.task) { |
82 | wake_up_process(bdi->wb.task); | 82 | wake_up_process(bdi->wb.task); |
@@ -88,7 +88,7 @@ static void bdi_queue_work(struct backing_dev_info *bdi, | |||
88 | trace_writeback_nothread(bdi, work); | 88 | trace_writeback_nothread(bdi, work); |
89 | wake_up_process(default_backing_dev_info.wb.task); | 89 | wake_up_process(default_backing_dev_info.wb.task); |
90 | } | 90 | } |
91 | spin_unlock(&bdi->wb_lock); | 91 | spin_unlock_bh(&bdi->wb_lock); |
92 | } | 92 | } |
93 | 93 | ||
94 | static void | 94 | static void |
@@ -704,13 +704,13 @@ get_next_work_item(struct backing_dev_info *bdi) | |||
704 | { | 704 | { |
705 | struct wb_writeback_work *work = NULL; | 705 | struct wb_writeback_work *work = NULL; |
706 | 706 | ||
707 | spin_lock(&bdi->wb_lock); | 707 | spin_lock_bh(&bdi->wb_lock); |
708 | if (!list_empty(&bdi->work_list)) { | 708 | if (!list_empty(&bdi->work_list)) { |
709 | work = list_entry(bdi->work_list.next, | 709 | work = list_entry(bdi->work_list.next, |
710 | struct wb_writeback_work, list); | 710 | struct wb_writeback_work, list); |
711 | list_del_init(&work->list); | 711 | list_del_init(&work->list); |
712 | } | 712 | } |
713 | spin_unlock(&bdi->wb_lock); | 713 | spin_unlock_bh(&bdi->wb_lock); |
714 | return work; | 714 | return work; |
715 | } | 715 | } |
716 | 716 | ||
@@ -810,6 +810,12 @@ int bdi_writeback_thread(void *data) | |||
810 | trace_writeback_thread_start(bdi); | 810 | trace_writeback_thread_start(bdi); |
811 | 811 | ||
812 | while (!kthread_should_stop()) { | 812 | while (!kthread_should_stop()) { |
813 | /* | ||
814 | * Remove own delayed wake-up timer, since we are already awake | ||
815 | * and we'll take care of the preriodic write-back. | ||
816 | */ | ||
817 | del_timer(&wb->wakeup_timer); | ||
818 | |||
813 | pages_written = wb_do_writeback(wb, 0); | 819 | pages_written = wb_do_writeback(wb, 0); |
814 | 820 | ||
815 | trace_writeback_pages_written(pages_written); | 821 | trace_writeback_pages_written(pages_written); |
@@ -868,26 +874,6 @@ void wakeup_flusher_threads(long nr_pages) | |||
868 | rcu_read_unlock(); | 874 | rcu_read_unlock(); |
869 | } | 875 | } |
870 | 876 | ||
871 | /* | ||
872 | * This function is used when the first inode for this bdi is marked dirty. It | ||
873 | * wakes-up the corresponding bdi thread which should then take care of the | ||
874 | * periodic background write-out of dirty inodes. | ||
875 | */ | ||
876 | static void wakeup_bdi_thread(struct backing_dev_info *bdi) | ||
877 | { | ||
878 | spin_lock(&bdi->wb_lock); | ||
879 | if (bdi->wb.task) | ||
880 | wake_up_process(bdi->wb.task); | ||
881 | else | ||
882 | /* | ||
883 | * When bdi tasks are inactive for long time, they are killed. | ||
884 | * In this case we have to wake-up the forker thread which | ||
885 | * should create and run the bdi thread. | ||
886 | */ | ||
887 | wake_up_process(default_backing_dev_info.wb.task); | ||
888 | spin_unlock(&bdi->wb_lock); | ||
889 | } | ||
890 | |||
891 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | 877 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
892 | { | 878 | { |
893 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | 879 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
@@ -1019,7 +1005,7 @@ out: | |||
1019 | spin_unlock(&inode_lock); | 1005 | spin_unlock(&inode_lock); |
1020 | 1006 | ||
1021 | if (wakeup_bdi) | 1007 | if (wakeup_bdi) |
1022 | wakeup_bdi_thread(bdi); | 1008 | bdi_wakeup_thread_delayed(bdi); |
1023 | } | 1009 | } |
1024 | EXPORT_SYMBOL(__mark_inode_dirty); | 1010 | EXPORT_SYMBOL(__mark_inode_dirty); |
1025 | 1011 | ||