Re: [PATCH][noop-iosched] don't reuse a freed request

[Jens Axboe <axboe@suse.de>]

On Mon, Oct 31 2005, Tejun Heo wrote:
> >For now, we should add the former.
> 
> Yeap, also verified oops doesn't happen with the following patch.

Good

> I'll soon post a patch to convert noop such that it does proper 
> dispatching.  BTW, while I was looking at the code, I found something 

Sounds good.

> else, in elv_former/latter_request functions, if the iosched doesn't 
> supply the callbacks, it uses rq->queue_list.prev/next implicitly 
> (without this, this noop bug wouldn't have been triggered).  I think 
> this code is not necessary anymore.  What do you think?

Well that should still work for noop, if it has its own internal
queueing list. But I would be quite fine with removing that implicit
next/prev support and simply require io scheds to supply these functions
always if they require rq-to-rq merging. Noop is the only one that
relied on this in the past, now seems a good time to clean that up as
well.

Besides, as you note, with merging disallowed on the ->queue_head, they
don't work as expected anymore.

-- 
Jens Axboe