Re: 5.1_RC2 panics during install on indy

["Stephen M. Rumble" <rumble%cs.stanford.edu@localhost>]

On May 29, 2010, at 10:50 AM, George Harvey wrote:

> On Fri, 28 May 2010 18:36:50 -0700
> "Stephen M. Rumble" <rumble%cs.stanford.edu@localhost> wrote:
> 
>> On May 28, 2010, at 3:45 PM, George Harvey wrote:
> 
>>> My Indy is a 150MHz R5000SC model with 256Mb RAM and the disk is a 4.3GB
>>> IBM DCAS-34330. Does this look like a disk error or a kernel problem?
>> 
>> What prior version were you running? I made several small changes between 
>> NetBSD-4 and NetBSD-5:
>>    - Added fast scsi support to WD33C93B.
>>    - Set the appropriate (I'm pretty sure) clock divisor for the scsi 
>> chipset.
>>    - Started using burst dma mode, rather than dma mode.
>> 
>> Could be something else entirely, but if you're interested in building an 
>> install kernel I could provide a two line patch to effectively revert the 
>> above.
> 
> I suspect it's something specific to R5000 Indys. I moved the disk onto
> another Indy, this time with a R4600, and the install went through with
> no problems. I'm currently building bash from pkgsrc as a quick(ish)
> functional test on the R4600 box. I run 5.0 on a R4400 Challenge S but
> when I tried 5.0 on the R5000 Indy last year, I got a lot of random
> hangs which I never resolved. Going back further, I did have 4.0
> running successfully on the R5000.
> 
> I'm happy to build a patched kernel and give it a try.

It's quicker for me just to tell you than to generate a patch. In 
src/sys/arch/sgimips/hpc/wdsc.c, if you change the following
    sc->sc_clkfreq = 200;
    sc->sc_dmamode = SBIC_CTL_BURST_DMA;
to
    sc->sc_clkfreq = 100;
    sc->sc_dmamode = SBIC_CTL_DMA;
you should get about the 4.0 behaviour (no fast scsi, wrong clock, no burst 
dma).

Steve