From: Christoph Hellwig on
I haven't reviewed this in detail, but what ensures the timer is
synchronously removed when the forker goes away? I don't see a
del_timer_sync call anywhere. For now it might be easier to just
skip this patch and leave it for later.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Artem Bityutskiy on
On Fri, 2010-07-23 at 12:28 -0400, Christoph Hellwig wrote:
> I haven't reviewed this in detail, but what ensures the timer is
> synchronously removed when the forker goes away?

Good point, thanks.

> I don't see a
> del_timer_sync call anywhere. For now it might be easier to just
> skip this patch and leave it for later.

Well, my tests showed that with this patch the flushers wake up
considerably less. So I'll try to come up with a better patch.

I will set-up better testing. Will hack things so that the background
dirty writeout timeout is something like 1-3 jiffies and the bdi thread
inactive timeout is something like 3-5 jiffies. Then will write a script
which forks many tasks each of each creates a loop-back device, mounts
it, does some I/O, unmounts, removes the loop-back device, and so on. If
run for long time, it should give good stress to the code paths I'm
working on. I have a 2-way 4-core (total 8) amd64 testbox to test.

--
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Artem Bityutskiy on
On Sun, 2010-07-25 at 11:29 +0300, Artem Bityutskiy wrote:
> +static void wakeup_timer_fn(unsigned long data)
> +{
> + struct backing_dev_info *bdi = (struct backing_dev_info *)data;
> +
> + spin_lock(&bdi->wb_lock);
> + if (bdi->wb.task) {
> + wake_up_process(bdi->wb.task);
> + } else {
> + /*
> + * When bdi tasks are inactive for long time, they are killed.
> + * In this case we have to wake-up the forker thread which
> + * should create and run the bdi thread.
> + */
> + wake_up_process(default_backing_dev_info.wb.task);
> + }
> + spin_unlock(&bdi->wb_lock);
> +}

OK, since now we use this lock in a timer, this should become a
spin_lock_bh().

--
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/