On Sun, Jun 10, 2012 at 03:24:14PM +0800, Fengguang Wu wrote:
>On Sun, Jun 10, 2012 at 12:54:03PM +0800, Wanpeng Li wrote:
>> On Sun, Jun 10, 2012 at 12:36:41PM +0800, Fengguang Wu wrote:
>> >Wanpeng,
>> >
>> >Sorry this I won't take this: it don't really improve anything. Even
>> >with the changed test, the real intervals are still some random values
>> >above (and not far away from) 200ms.. We are saying about 200ms
>> >intervals just for convenience.
>> >
>> But some parts like:
>>
>> __bdi_update_bandwidth which bdi_update_bandwidth will call:
>>
>> if(elapsed < BANDWIDTH_INTERVAL)
>> return;
>>
>> or
>>
>> global_update_bandwidth:
>>
>> if(time_before(now, update_time + BANDWIDTH_INTERVAL))
>> return;
>>
>> You me just ignore this disunion ?
>
>Not a problem for me. But if that consistency makes you feel happy,
>you might revise the changelog and resend. But it's not that simple
>for the below reason..
>
>> >On Sun, Jun 10, 2012 at 12:20:05PM +0800, Wanpeng Li wrote:
>> >> From: Wanpneg Li <liwp@xxxxxxxxxxxxxxxxxx>
>> >>
>> >> Since bdi_update_bandwidth function should estimate write bandwidth at 200ms intervals,
>
>The above line represents a wrong assumption. It's normal for the
>re-estimate intervals to be >= 200ms.
>
>> >> so the time is bdi->bw_time_stamp + BANDWIDTH_INTERVAL == jiffies, but
>> >> if use time_is_after_eq_jiffies intervals will be bdi->bw_time_stamp +
>> >> BANDWIDTH_INTERVAL + 1.
>
>Strictly speaking, to ensure that ">= 200ms" is true, we'll have to
>skip the "jiffies == bw_time_stamp + BANDWIDTH_INTERVAL" case. For
>example, when HZ=100, the bw_time_stamp may actually be recorded in
>the very last ms of a 10ms range, and jiffies may be in the very first
>ms of the current 10ms range. So if using ">=" comparisons, it may
>actually let less than 200ms intervals go though.
>
>We can only reliably ensure "> 200ms", but no way for ">= 200ms".
>
static void global_update_bandwidth(unsigned long thresh,
unsigned long dirty,
unsigned long now)
{
static DEFINE_SPINLOCK(dirty_lock);
static unsigned long update_time;
/*
* check locklessly first to optimize away locking for the most time
*/
if (time_before(now, update_time + BANDWIDTH_INTERVAL))
return;
spin_lock(&dirty_lock);
if (time_after_eq(now, update_time + BANDWIDTH_INTERVAL)) {
update_dirty_limit(thresh, dirty);
update_time = now;
}
spin_unlock(&dirty_lock);
}
So time_after_eq in global_update_bandwidth function should also change
to time_after, or just ignore this disunion?
>
>> >> Signed-off-by: Wanpeng Li <liwp@xxxxxxxxxxxxxxxxxx>
>> >> ---
>> >> mm/page-writeback.c | 2 +-
>> >> 1 file changed, 1 insertion(+), 1 deletion(-)
>> >>
>> >> diff --git a/mm/page-writeback.c b/mm/page-writeback.c
>> >> index c833bf0..099e225 100644
>> >> --- a/mm/page-writeback.c
>> >> +++ b/mm/page-writeback.c
>> >> @@ -1032,7 +1032,7 @@ static void bdi_update_bandwidth(struct backing_dev_info *bdi,
>> >> unsigned long bdi_dirty,
>> >> unsigned long start_time)
>> >> {
>> >> - if (time_is_after_eq_jiffies(bdi->bw_time_stamp + BANDWIDTH_INTERVAL))
>> >> + if (time_is_after_jiffies(bdi->bw_time_stamp + BANDWIDTH_INTERVAL))
>> >> return;
>> >> spin_lock(&bdi->wb.list_lock);
>> >> __bdi_update_bandwidth(bdi, thresh, bg_thresh, dirty,
>> >> --
>> >> 1.7.9.5
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>