Re: isp driver on amd64?

[Matthew Jacob <Matthew.Jacob%Sun.COM@localhost>]

Thor Lancelot Simon wrote:
> On Wed, Sep 10, 2008 at 10:29:54AM -0700, Matthew Jacob wrote:
>   
> > The first error you saw:
> >
> > isp0: bad underrun for 0.0 (count 32768 resid 2147473407, status not marked)
> > isp0: BOTCHED cmd for 0.0.0 cmd datalen 32768
> > sd0(isp0:0:0:0): generic HBA error
> >
> > That's /usually/ a botched target problem. They aren't correctly marking  
> > a residual in the FCP response frame.
> > If you're absolutely sure the target is good, you can comment out the  
> > check in isp_intr.
> >     
>
> I'm not certain I trust the target.  It is somewhat old and was bought
> from the lowest bidder.  On the other hand, I was using this target with
> a QLA21xx and NetBSD/i386 (1.6 or thereabouts) a few years ago and
> didn't see this error.
>
> The actual requested transfer size was in fact 32768.  Is this a useful
> clue one way or the other?
>   

Yes. It says "I asked for 32k". The target filled in bits in the 
response frame that said that the residual field is valid, but the 
residual field is some wrongo large number.

It could be a 2300 firmware bug, and the fact that this didn't occur on 
21xx or 22xx is only because this part of the frame handling went into 
hardware in the 23XX.

The linux Qlogic supplied driver ignores this field entirely, which is 
just plain wrong.

> Is there an adapter feature I could adjust in the BIOS which might have
> any effect on this?  The only things changed from the defaults are
> Fast Command Posting (turned on) and command throttle changed from 16 to
> 64.
>
>   

No, it's not part of any of the settings.

Try commenting out this clause in isp_parse_status:

#if 0 <<< ADDED
                               isp_prt(isp, ISP_LOGWARN, bun, XS_TGT(xs),
                                   XS_LUN(xs), XS_XFRLEN(xs), 
sp->req_resid,
                                   (ru_marked)? "marked" : "not marked");
                               if (XS_NOERR(xs)) {
                                       XS_SETERR(xs, HBA_BOTCH);
                               }
#endif

and see if life gets better.