Re: PATCH, IDE corruption, 2.4.18

[Mike Fedyk]

On Sun, May 05, 2002 at 02:54:17AM +0100, Neil Conway wrote:
> Hi...
> 
>  --- Mike Fedyk <[e-mail]> wrote: > On Sat, May 04, 2002 at
> 01:15:20PM +0100, Neil Conway wrote:
> > > -	byte stat;
> > > +	byte stat,unit;
> > 
> > [snip]
> > 
> > >  #if defined(CONFIG_BLK_DEV_IDEDMA) && !defined(CONFIG_DMA_NONPCI)
> > > -	byte unit = (drive->select.b.unit & 0x01);
> > > +	unit = (drive->select.b.unit & 0x01);
> > 
> > Why are you moving the init of "unit" out of that ifdef?
> 
> Basically to make it compile.  I'm only moving the declaration, and
> that's only because I need to abort the routine BEFORE the second line
> of the ifdef switches off DMA on that unit.  I did compile a version
> with the ifdef split into two bits but I decided that was a little
> messy. (It won't compile if my check goes in before the first line of
> the ifdef as originally written as it declares a variable and C won't
> let declarations follow plain code.)
>

Right.  Duh on my part, sorry.

> > Can you see if this problem is still in 2.5 also? 
> 
> I haven't got a 2.5.13 tree but I found 2.5.7 on a source browser
> online and verified that, back then at least, ide-features.c was still
> basically the same.  Of course, the routines in between
> ide_register_subdriver and ide_config_drive_speed might have been
> different.  If someone can look at the code-path between these two
> routines in 2.5.13 to see if there is any check on whether or not the
> hwgroup is busy (or simply whether or not DMA is in progress) that
> would clear it up.  I'll probably download 2.5.13 sometime soon anyway.
> 

Andre cleaims that the situation is worse on 2.5.13 or so.  I wouldn't hurt
to test-run the code though.