Re: [PATCH 8/8] vm: Add an tuning knob for vm.max_writeback_mb

[Wu Fengguang <fengguang.wu@intel.com>]

On Wed, Sep 30, 2009 at 08:10:30PM +0800, Jens Axboe wrote:
> On Wed, Sep 30 2009, Jan Kara wrote:
> > > +void bdi_writeback_wait(struct backing_dev_info *bdi, long nr_pages)
> > > +{
> > > +	struct dirty_throttle_task tt = {
> > > +		.nr_pages = nr_pages,
> > > +		.complete = COMPLETION_INITIALIZER_ONSTACK(tt.complete),
> > > +	};
> > > +	struct wb_writeback_args args = {
> > > +		.sync_mode	= WB_SYNC_NONE,
> > > +		.nr_pages	= LONG_MAX,
> > > +		.range_cyclic	= 1,
> > > +		.for_background	= 1,
> > > +	};
> > > +	struct bdi_work work;
> > > +
> > > +	bdi_work_init(&work, &args);
> > > +	work.state |= WS_ONSTACK;
> > > +
> > > +	/*
> > > +	 * make sure we will be waken up by someone
> > > +	 */
> > > +	bdi_queue_work(bdi, &work);
> >   This is wrong, you shouldn't submit the work like this because you'll
> > have to wait for completion (wb_clear_pending below is just bogus). You
> > should rather do bdi_start_writeback(bdi, NULL, 0).
> 
> Indeed, the above will die a horrible death fairly soon. But we can add
> some "barrier" like synchronization, if you just wish to wait for
> previously submitted work to have been completed.

Thanks, I just purged that hack and go for bdi_start_writeback :)

Thanks,
Fengguang