Hi Chris,
On Wed, 6 Mar 2019, Chris Jones wrote:
> The setup: Mac SE/30, standard but with 80MB of RAM and a 2GB CF card in
> an Adtron SCSI-to-PCMCIA adapter. Appears to work flawlessly as a Mac
> under System 7.1, hasn't given any hint of trouble.
>
> Install process: CF card partitioned on mac with LIDO, then removed and
> put in to card reader on i386 machine to have files copied on to it
> (kernel on to HFS partition for Penguin, rootfs in to ext3 partition).
>
> Kernel: 4.20.13 crossbuilt from source on kernel.org,with default mac
> config. Earlier kernels all had unusable 5380 SCSI drivers.
Well, there were no relevant changes to the mac_scsi driver since about
v4.9. Therefore, it seems likely that v4.20 has the same driver bug,
albeit an intermittent one.
> Root filesystem set up from image at
> https://cdimage.debian.org/cdimage/ports/9.0/m68k/iso-cd/debian-9.0-m68k-NETINST-1.iso
> using debootstrap --arch m68k --foreign --no-check-gpg sid ... a few
> tweaks to /dev and /etc to get a shell prompt
>
> Boots OK, rootfs mounts OK, kernel finds /bin/sh and I can remount / rw
> and run debootstrap --second-stage. It starts to work, glacially slowly
> of course, but after a few hours (and unpacking a dozen or so packages)
> things start to fail with segfaults and illegal instructions.
>
> After this, the filesystem is hopelessly corrupted, with thousands of
> errors and in one case was unrecoverable by fsck.
>
> My first suspect is bad RAM. For the last 12 hours or so the mac has
> been running memtest 64M (from Woody, I think) under the above kernel.
> It has revealed no errors but hasn't completed a full cycle yet (!).
>
That's possible too, but it probably doesn't explain the "unusable 5380
SCSI drivers" that you observed.
Anyway, I've been working on 5380 driver patches lately so I can easily
send you a new kernel to test if you like.
BTW, you can get a slow but stable 5380 SCSI driver by passing
'mac_scsi.setup_use_pdma=0' in the kernel parameters.
I've now tried this, with no other changes, and debootstrap runs to completion (in about 10 hours) without any errors or filesystem corruption, so it's pretty safe to say that the problem was in the kernel SCSI driver and not the install process.
Chris