lkml.org 
[lkml]   [2007]   [Oct]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [bug] block subsystem related crash with latest -git
On Thu, Oct 18 2007, David Miller wrote:
> From: Jens Axboe <jens.axboe@oracle.com>
> Date: Thu, 18 Oct 2007 12:55:17 +0200
>
> > Things have progressed a lot since, see my recent posting based on
> > Davem's proposal. Will post another patch soonish, that is also
> > tested.
>
> One core issue here is that we need to decide whether this thing to be
> iterated like an array or like a linked list. It's trying to be both.
>
> If we decide upon a looping construct for consumers and stick to it,
> we'll be in much better shape than we are now and bugs will be eaiser
> to spot. It would be so much simpler to audit if all we saw in the
> consumers were things like:
>
> while (sg) {
> do_stuff(sg);
> sg = sg_next(sg);
> }
>
> I would suggest that we just get it over with and convert the whole
> tree now rather than trying to do this kind of thing in stages.
> Because then we can say that ever scatterlist creator has to set
> the "end" bit and therefore you use well established patterns
> for scatterlist iteration such as "traverse sg_next() until NULL"
> as shown above.

The above should work now, PROVIDED that the end has been marked
properly. I have added some helpers to init an sglist (or sg entry).
You can still use

for_each_sg()

and pass in the number of entries, that'll work even with an end marker.
So, it does both.

> I also noticed that there is the issue of on-stack and embedded
> scatterlist users. We'll need some sort of "DECLARE_SCATTERLIST"
> and a "scatterlist_init()" thing so that we can keep DEBUG_SG
> working even in those cases. But for all I know Jens could be
> working on that already :-)

Heh, got some of it covered! The stack usage is generally just converted
to either use sg_init_one() if we can, or sg_init_table(). Adding some
soft of DECLARE_SCATTERLIST() sounds like a good idea, though. I'll take
patches :-)

> The only other real option if we don't convert the whole tree now to
> the "end" marker stuff, is to enforce that every scatterlist iterator
> only traverse the number of entries there were told are in the one
> given to them.

Driver that do all their own sg management typically still use a manual
loop over the number of entries they know are there. I didn't change
those, and the original sg chaining didn't convert those either. Some
manual labor is required in utilizing chained sg lists, unless you are
part of a subsystem (like SCSI) that allocates and inits themf or you.
So I still think that baby steps are a good idea - only convert things
that matter. Leave the rest to janitors or others willing to dive in.

--
Jens Axboe

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2007-10-18 14:31    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site