Re: Undo aic7xxx changes

[Willy Tarreau]

On Fri, May 09, 2003 at 03:46:37PM +0200, Stephan von Krawczynski wrote:
> On Fri, 9 May 2003 15:27:57 +0200
> Willy Tarreau <[e-mail]> wrote:
> 
> > On Fri, May 09, 2003 at 03:02:07PM +0200, Stephan von Krawczynski wrote:
> >  
> > > I cannot say which version of the driver it was, the only thing I can tell
> > > you is that the archive was called aic79xx-linux-2.4-20030410-tar.gz.
> > 
> > That's really interesting, because I got the bug since around this version
> > (20030417 IIRC), and it locked up only on SMP, sometimes during boot, or
> > during heavy disk accesses caused by "updatedb" and "make -j dep". It's
> > fixed in 20030502 from http://people.freebsd.org/~gibbs/linux/SRC/
> 
> I tried to merge the latest aic archive into 2.4.21-rc2, besides the "usual"
> signed/unsigned warnings I got this one:
> 
> aic7xxx_osm.c: In function `ahc_linux_map_seg':
> aic7xxx_osm.c:770: warning: integer constant is too large for "long" type

Good catch, but in fact, it's more this line which worries me :

758:                if ((addr ^ (addr + len - 1)) & ~0xFFFFFFFF) {

I don't see how ~0xFFFFFFFF can be non-null on 32 bits archs, because addr is
a bus_addr_t which is in turn dma_addr_t which itself is u32. So unless I don't
find the trick this would mean that this code should never be executed. Perhaps
~0xFFFFFFFFULL would be more appropriate, or even >0xFFFFFFFF, since this can be
detected with u32 using the carry left by the addition.

Regards,
Willy