Re: sgimips, wdsc / scsi timeouts and kernel panics

On Mon, 17 Nov 2025 at 04:17, Frank Scheiner <[email protected]> wrote:

> Hi Adrian,
>
> On 16.11.25 23:44, Adrian Chadd wrote:
> > I may have slipped and bought a couple of SGI indy's and I wanted to
> > install netbsd on one.
>
> Good decision, I'd say.
>

It's such a good decision that I slipped and acquired two (as I wanted an
R5000.)


>
> > However! It looks like the wdsc/wd33c93 driver does not handle the device
> > going away on the bus during a transfer.
> >
> > [...]
>
> Could it be helpful to boot the Indys via network (incl. root FS on NFS)
> when further debugging this or will this still lead to a panic when the
> described transfers happen to the zuluscsi which then doesn't host the
> actually used root FS?
>

I'm netbooting the inst image so i can run newfs and dd and other stuff to
test :-)

If I do 64k transfers then the zuluscsi seems to error out and detach.
The wdsc chipset code handles this very poorly ;-)

I'm experimenting with limiting IO to smaller amounts, but now i'm hitting
some
bus_dmamap_sync call panics in wdsc. I'll go figure it out; i have a
feeling that's
just due to what i'm doing locally. (I was playing with limiting the map
setup,
but now I think I'm going to just limit it using a custom minphys() and
leave the
dmamap setup to MAXPHYS.

I'll tinker some more later and report back.

>
> I don't know how to network boot these with NetBSD, but if it's the same
> or similar to how OpenBSD does it, you can find a recent boot log on [1],
> together with the needed PROM configuration. Requires DHCP, TFTP (for
> initial load of elf-to-ecoff loader and kernel by the PROM), RARP,
> Bootparams, NFS (for getting the mount information for and mounting the
> root FS) and possibly DNS services on the server side.


I need to do a write-up of the shenanigans i needed to do in 2025 to get
this to work
as bootp is both not clear in dnsmasq and freebsd's tftpd needed options
negotiation
disabled for it to work.


-adrian