Re: [linux-mac68k] Results of trying 2.6.7 and 2.6.8 on Macintosh Quadra 650
On Wed, 2004-08-25 at 22:40, Erik C.J. Laan wrote:
> Kernel 2.6.8 progresses even further, and even
> mounts the rootfs, but dies soon after with a BAD KERNEL TRAP.
> Because these logs show alot of messages about SCSI-stuff and quite some
> about sde, I tried booting without the Syquest on, but this time both
> kernels hang at eth0 initialisation again (see 2.6.-mac_run5*.txt)
> Next and last try was with the Syquest turned on again, and a cartridge
> in it. I used my MacOS to make 1 partition on it of type A/UX data. It
> was to small for A/UX Root&Usr, and mac-fdisk on my stable 2.2.25 kernel
> could not read the partition table, and so refused to make a new one.
> With this test both kernels fail when the eth0 driver starts (see
> I'm wondering if the SCSI messages from the "2.6.-mac_run4*txt" logs
> are any indication whether the fix to disable SCSI reselects is actually
> in these kernels. This fix is activated by the mac53c9x=1,0 kernel
> commandline argument and Kars de Jong wrote on Aug 19 in a message on
> the debian-68k-maillist (thread "light weight desktop for browsing" ;-)
> that he thought this fix had not been entered in the general Linux-m68k
> CVS, while it has been in the Linux-mac68k CVS since Ray Knight wrote it
> for 2.2.20 in June of 2001. My experience with multiple SCSI drives (HD,
> CD) on a 68040 Mac is that this fix is essential.
This fix is not in yet. Since you appear to have a working serial
console, and I'm the maintainer (sort of) of the NCR53C9x driver, it
would be nice if you could work with me to get this fixed.
It seems to me by the way that most drives work even without the fix,
it's only your SCSI 1 target (the Syquest) that seems to break. So it
seems a bit of a shame to disable disconnects globally.
I actually don't understand why this kernel message:
esp0: Forcing async and disabling disconnect for target 5
keeps coming up. It should only be printed once. It seems to not
actually disable disconnects for that target...
Also, there's code in the driver already to stop non-tape SCSI 1 targets
<some checking later>
Oh right, someone broke that for the Mac driver it seems, because it
contains a nice section of code with a 'goto' which constantly enables
I don't really understand the Mac CVS repository either. It looks like
the disconnect stuff (which was apparently written by David D. Kilzer)
has been added and removed again from the driver at least 2 times.
Looks like main tree imports weren't done on the vendor branch, so all
local changes were lost every time someone synced with the main tree...
Anyway, I'll try to incorporate this stuff cleanly in the driver, and I
hope you can test it for me, Erik :)