[linux-lvm] Re: [PATCH] 2.5 version of device mapper submission

[Jens Axboe]

On Tue, Oct 15 2002, Joe Thornber wrote:
> On Tue, Oct 15, 2002 at 10:21:52AM +0200, Jens Axboe wrote:
> > On Mon, Oct 14 2002, Joe Thornber wrote:
> > > 10.patch
> > >   [Device-mapper]
> > >   Add call to blk_queue_bounce() at the beginning of the request function.
> > 
> > What on earth for? I also see that you are setting BLK_BOUNCE_HIGH as
> > the bounce limit unconditionally for your queue. Puzzled.
> 
> This is just me stupidly copying loop.c, already found out it doesn't
> work.  Please ignore.

Well it should work, but dm never ever wants to bounce any data on its
own. That should only happen at the target queue, if at all.

> > When does dm even have to touch the data in the bio?
> 
> Tell me; if I'm splitting a bio using bio_clone, and then map the bio
> to a driver that calls blk_queue_bounce.  How can I avoid the
> 
>         BUG_ON((*bio_orig)->bi_idx);
> 
> triggering ?  Or is bio_clone not to be used anymore ?

(btw, do you have a complete patch against a recent kernel?)

Bouncing has just never been used with a cloned bio, so there might be a
corner case or two that needs to be fixed up. But walk me through your
request handling, please. It seems you are always allocating a
clone_info and bio clone for io?

-- 
Jens Axboe