Re: RAID1 might_sleep() warning on 3.19-rc7

On Fri, 13 Feb 2015 15:48:03 +0100 Peter Zijlstra [off-list ref]
wrote:

On Fri, Feb 13, 2015 at 11:27:46AM +0100, Peter Zijlstra wrote:

quoted

quoted

I've moved blk_flush_plug to the beginning of the function.

quoted

I wondered if it really make sense to call blk_flush_plug with nr_iowait
elevated and delayacct_blkio active.  blk_flush_plug() could call schedule()
for non-"io" reasons and maybe that could upset stuff???

Yeah, good question that. Lemme ponder that a bit.

Yes, I thikn your version makes most sense as, you say, even regular
schedule() call nested in my version would go towards blk delayacct --
and I doubt that was the intent; even though the current kernel works
that way.

I'll move the now rudimentary io_schedule() into sched.h as an inline.

Thanks.

Are you OK with this going to -stable for 3.19?

NeilBrown