Prev: [Regression] mm: Huge swapout performance regression on recent kernels
Next: cpufreq: fix memory leak in cpufreq_add_dev
From: Artem Bityutskiy on 20 Jul 2010 08:40
On Sun, 2010-07-18 at 03:02 -0400, Christoph Hellwig wrote:
> > + if (wakeup_default) {
> > trace_writeback_nothread(bdi, work);
> > wake_up_process(default_backing_dev_info.wb.task);
>
> Why not simply do the defaul thread wakeup under wb_lock, too?
> It keeps the code a lot simpler, and this is not a typical path anyway.

Will address.

> > if (dirty_writeback_interval) {
> > + unsigned long wait_jiffies;
> > +
> > wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
> > schedule_timeout(wait_jiffies);
>
> No real need for a local variable here.

Will address.

> > @@ -364,7 +395,7 @@ static int bdi_forker_thread(void *ptr)
> > if (!list_empty(&me->bdi->work_list))
> > __set_current_state(TASK_RUNNING);
> >
> > - if (!fork) {
> > + if (!fork && !kill) {
>
> I think the code here would be a lot cleaner if you implement the
> suggestion I have for the forking restructuring.

As I replied earlier, to fork/kill the the thread from inside list walk
we'd need to drop the spinlock, which is not very nice. So I am keeping
this part intact so far.

--
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 20 Jul 2010 09:10
On Sun, 2010-07-18 at 03:02 -0400, Christoph Hellwig wrote:
> Yes, only killing threads from the caller is much better, that's how
> the kthread API is supposed to be used anyway.
>
> > static void bdi_queue_work(struct backing_dev_info *bdi,
> > struct wb_writeback_work *work)
> > {
> > + bool wakeup_default = false;
> > +
> > trace_writeback_queue(bdi, work);
> >
> > spin_lock(&bdi->wb_lock);
> > list_add_tail(&work->list, &bdi->work_list);
> > - spin_unlock(&bdi->wb_lock);
> > -
> > /*
> > * If the default thread isn't there, make sure we add it. When
> > * it gets created and wakes up, we'll run this work.
> > */
> > - if (unlikely(!bdi->wb.task)) {
> > + if (unlikely(!bdi->wb.task))
> > + wakeup_default = true;
> > + else
> > + wake_up_process(bdi->wb.task);
> > + spin_unlock(&bdi->wb_lock);
> > +
> > + if (wakeup_default) {
> > trace_writeback_nothread(bdi, work);
> > wake_up_process(default_backing_dev_info.wb.task);
>
> Why not simply do the defaul thread wakeup under wb_lock, too?
> It keeps the code a lot simpler, and this is not a typical path anyway.

Hmm, actually, I want to take this lock in __mark_inode_dirty() as well,
so it makes sense to micro-optimize this. Also, can
'trace_writeback_nothread()' be called under a spinlock? If no, then a
variable is needed anyway.


--
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/
 | 
Pages: 1
Prev: [Regression] mm: Huge swapout performance regression on recent kernels
Next: cpufreq: fix memory leak in cpufreq_add_dev