Re: Possibly wrong BIO usage in ide_multwrite

[Christophe Saout <christophe@saout.de>]

Am Mo, den 05.01.2004 schrieb Bartlomiej Zolnierkiewicz um 17:12:

> > The IDE_TASKFILE_IO gets things right (from my point of view) and is
> > also much cleaner. (I would personally vote for dropping the non
> > TASKFILE_IO code, it would make my problem go away :D - why is it still
> > marked as experimental BTW? I've been using it since it was introduced,
> > without any problems)
> 
> There are still some issues to be resolved:
> - hangs during reading /proc/ide/<cdrom>/identify on some drives
>   (workaround is now known thanks to debugging done by Andi+BenH+Andre)
> - unexplained fs corruption on x86-64 with AMD IDE chipsets
>   (the real showstopper)
> - somebody needs to test taskfile code on old Promise PDC4030 controller
>   (low priority)

Unexplained corruptions. Coder's nightmare. ;) Yes, that's always really
bad. Unfortunately I don't have such a machine so I can't help trying to
nail it down.

> > Perhaps I can think of something else. It's really tricky...
> 
> I would like to remove non CONFIG_IDE_TASKFILE_IO paths in 2.6.x
> (after issues are resolved) instead of trying to fix them.

Sure, we agree on that point. I think everyone does.

BTW, I've found a not too complicated workaround for my particular
original problem so the bi_idx issue isn't a showstopper for my
device-mapper target. But apart from that the rewinding bi_idx to zero
thing still gives me headaches just being there. A small non-invasive
workaround won't make it much worse, it's hopefully going to die anyway.

Thanks for being patient with me. :-)
(and not raising new paranoia theories ;-))