On Thu, Jan 20, 2011 at 07:50:57PM +0100, Tejun Heo wrote:

Hello,

On Tue, Jan 18, 2011 at 09:12:55AM +0800, Shaohua Li wrote:

quoted

This makes sense. would it possible N flush data payloads delay the
first request?

I ended up with a different design.  Still buggy.  It triggers a weird
oops under stress test but other than that things generally seem to
work as expected.  Please read the comment at the top of blk-flush.c
for more info.  I'll post properly after more testing and debugging.

The oops was caused by debugging code I put in (not posted together).
It runs fine without it, so if you have some time, please give it a
spin.

Thanks.

-- 
tejun