Re: Panics going multiuser with NetBSD 8, 9, current

[John Klos <john%ziaspace.com@localhost>]

> So I did a fresh install with 9.2 via miniroot.fs.
>
> The installer shows a few problems, where I think I didn't see them in the
> past:
>
> 1. Seems that the INSTALL kernel has no longer any support for keymaps. When
> I select '0' for "DIN keymap" during installation I get:
> ITEIOCSKMAP: Function not implemented
> The problem seems gone after installing the GENERIC kernel, though.
> Maybe the INSTALL kernel uses WSCONS? Then we should either remove the
> keymap selection from the install script or use an ITE-kernel for INSTALL.

This seems easy to address.

> 2. I couldn't configure my network during installation. I have an X-Surf
> board, which is supported by the ne(4) driver and appeared as ne0.
> When the installer asked for adding media or link-layer arguments I just hit
> enter for "none" and the following error appeared:
> SIOCGIFMEDIA: inappropriate iotctl for device
> After that I escaped into the shell and could set up my network by
> ifconfig(8) without problems.

I think all cards that don't offer physical media options give that 
message because the configuration tries to set media to autoconfig. Not 
sure if it's better to ignore it or if some logic should be added to not 
set media type in these cases.

> 3. I chose to install the archives via FTP from my NAS. I could navigate to
> the right directory and select the archives to install, but downloading
> failed because the paths were suddenly wrong ('/' separators were missing).
> So I had to install them from a local disk partition.

That's clearly a bug. I've had the opposite, where entering an extra '/' 
leads to paths with extra slashes, but that doesn't stop things from 
working. It does make me think the Amiga code is different from other 
ports, which should be fixed or shared.

> While transfering the files from my NAS to the local disk I got several
> errors from the ne(4) driver:
> ne0: NIC memory corrupt - invalid packet length
>
> I never had any issues with my X-Surf before, so this is strange. Anyway,
> seems the data was transfered correctly.

I get those messages on several older systems. It seems to happen much 
more often on networks with lots of traffic and never happens when 
directly connected to another system.

> > The moment I try to chroot to a filesystem, either on an IDE wd disk or
> > on a SCSI disk on the CyberStorm Mk3 SCSI bus, it panics.
>
> I couldn't reproduce that. My 9.2 installation seems to work, using the same
> cbiiisc(4) driver. My dmesg:
>
> [     1.000005] p5bus0: CyberStorm PPC 604e (sn DC50xxx)
> [     1.000005] cbiiisc0 at p5bus0: CyberStorm PPC/Mk-III SCSI host adapter
> [     1.000005] cbiiisc0: siopng type 770 id 16 reset V1
> [     1.000005] scsibus1 at cbiiisc0: 16 targets, 8 luns per target
> [     1.000005] p5pb0 at p5bus0: no PCI bridges detected
> [     5.004812] cbiiisc0: target 1 now wide 1
> [     6.414673] cbiiisc0: target 1 now synchronous, period=50ns, offset=14
> [     6.434715] sd0 at scsibus1 target 1 lun 0: <ATA, SAMSUNG HD502HI, 1AG0>
> disk fixed
> [     6.454679] sd0: 465 GB, 242255 cyl, 64 head, 63 sec, 512 bytes/sect x
> 976773168 sectors

Yes, about the same for me (no CyberVision).

> (Did I mention that I hate these timestamps, which are useless in 99.99% of
> all boots? I have to build a new kernel with KLOG_NOTIMESTAMP again... :( )

;)

> Did you try FFSv1? I'm not sure, but I may be running v1. dumpfs(8) reports
> "format: FFSv1", although the sblock is FFSv2...

That's next.

> Your panics seem related to disk operations.

Indeed, which is why I've tried an IDE drive on the IDE bus, with nothing 
on the SCSI bus, and a CF card on the IDE bus, and a SATA drive on the 
SCSI bus (with SCSI-IDE, IDE-SATA). No changes.

I can fsck, mount, copy, et cetera, just fine, so it's really an issue of 
trying to run executables.

Will report back after testing FFSv1.

Thanks,
John