On Fri, Sep 11, 2015 at 7:15 PM, Chris Mason <clm@xxxxxx> wrote:
>
> But flushing on schedule is a little different. It ends up calling
> blk_schedule_flush_plug() which will hand off work to kblockd through
> blk_run_queue_async()
I was more worried about some actual deadlock from the changes. And
blk_schedule_flush_plug() is fine in that it doesn't actually remove
the plug, it just schedules the currently plugged pending IO, so the
IO will start from waiting on the inode, but the plug will still
remain for the rest of the writeback, and it all looks like it should
be fine.
And the reason we use kblockd is simple: stack usage. The reschedule
can happen pretty deep on the stack, we don't actually want to
necessarily then cause much more stack use through things like md/raid
allocating new requests etc.
So it all looks fine to me.
Btw, very tangentially related: grepping a bit shows that
"blk_flush_plug()" isn't actually used anywhere any more. Can we get
rid of that?
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
