On Tue, Mar 09, 1999 at 03:19:14PM -0700, Justin T. Gibbs wrote:
> >I also loaded FreeBSD 3.1 on the system as a test;  it seems to work OK.
> >Not to disparage FreeBSD, NetBSD's what I've got on all my other systems,
> >and by gum, I'd like it here!   So, the hardware itself seems OK.  Just
> >in case it triggers anything for anyone, here's what a FreeBSD kernel
> >with AHC_DEBUG says (though I had to edit some stuff out, as it didn't
> >compile cleanly with AHC_DEBUG):
> 
> Booting with -v would be more useful.  I haven't used the AHC_DEBUG
> stuff in a long time.

Noted for future reference. :)  (It didn't reveal anything particularly
interesting-looking, at least as relates to this)

> 
> >ahc0: <Adaptec aic7880 Ultra SCSI adapter> rev 0x00 int a irq 11 on pci0.11.0
> >ahc0: aic7880 Wide Channel A, SCSI Id=7, 16/255 SCBs
> 					  ^^<= That's the important number.			

Yep.  I did try the obvious "Hardcode 16 SCBs and see if it just works".
(A few different ways, as I realized various things I was missing)  Pretty
much the same problems occur, which indicates to me a more fundamental
problem.

> >So, anyway, I'm at the point now where I'm slowly sifting through the 
> >attachment of the ahc device using GDB, but I'm afraid my skills might
> >not be up to the task.  Anyone got any further pointers of what I might
> >look at?
> 
> Look at the loop in ahc_init that determines the number of hardware
> SCBs.  You should also see if NetBSD is preferring memory mapped I/O
> over PIO as FreeBSD defaults to PIO access of the chip.
> 

Yeah; I've stepped through that loop in gdb under NetBSD.  To my eyes, the
code seems _very_ similar to that in FreeBSD-3.1 (not surprising, given
its provenance ;)


Here's the relevant code snippet, from dev/ic/aic7xxx.c :

*************
        /* Determine the number of SCBs */

        {
                AHC_OUTB(ahc, SCBPTR, 0);
                AHC_OUTB(ahc, SCB_CONTROL, 0);
                for(i = 1; i < AHC_SCB_MAX; i++) {
                        AHC_OUTB(ahc, SCBPTR, i);
                        AHC_OUTB(ahc, SCB_CONTROL, i);
                        if(AHC_INB(ahc, SCB_CONTROL) != i)
 /* XXX */                      break;
                        AHC_OUTB(ahc, SCBPTR, 0);
                        if(AHC_INB(ahc, SCB_CONTROL) != 0)
                                break;
                        /* Clear the control byte. */
                        AHC_OUTB(ahc, SCBPTR, i);
                        AHC_OUTB(ahc, SCB_CONTROL, 0);

                        ahc->qcntmask |= i;     /* Update the count mask. */
                }

                /* Ensure we clear the 0 SCB's control byte. */
                AHC_OUTB(ahc, SCBPTR, 0);
                AHC_OUTB(ahc, SCB_CONTROL, 0);

                ahc->qcntmask |= i;
                ahc->maxhscbs = i;
        }
******************

when i=1, it hits the "break" at /* XXX */, leaving the loop, and resulting
in the "1 SCBs" message.  I'm not really certain how to tell if NetBSD is
preferring memory mapped i/o or not;  if it matters, ahc->sc_st is 
I386_BUS_SPACE_IO during the snippet above.  

Can you tell that devices aren't my forte?  ;)

-- 
Jeff Rizzo                                         http://boogers.sf.ca.us/~riz