>> >You need an up-to-date sys/dev/tc/asc_ioasic.c - either rev 1.15.2.1 of
>> >the 1.4 branch or rev 1.18 for NetBSD-current.  The pre-release 1.4.2
>> >snapshot in pub/arch/pmax/snapshot/1.4.2_ALPHA on ftp.netbsd.org has
>> >this patch applied.  I've included a diff that should apply to a stock
>> >1.4.1 kernel below.
>>
>> Given that I wanted to make a whole new system (in rz3), rather
>> than just copying 'netbsd' I took all the sets of 1.4.2_ALPHA
>> and started from scratch. Everything went fine and the
>> installation was completed successfully. The problem came after
>> when I wanted to copy from the other system (rz0) the directory
>> /usr/users. So I did:
>>
>> bebita# mount /dev/rz0d /mnt
>> bebita# cd /usr
>> bebita# (cd /mnt; tar cf - users) | tar xf -
>>
>> Apparently it completed the copying process, but it came out
>> with:   [[an error message indicating a SCSI problem...]]
>
>You didn't by any chance upgrade user-land (the installation
>binary sets) without upgrading the kernel first, so you ran the
>1.4.1 kernel while doing the above copying?

I first copied "install" from 1.4.2_ALPHA into the boot disk
(rz0). I booted that one with "boot 3/rz0/install" and after it
was finished (with the 1.4.2_ALPHA sets) I booted rz3. So the
attempted copy was done everything with 1.4.2_ALPHA. rz0 was
mounted on /mnt as the source.

I'm puzzled though. It seems that the SCSI problem is related to
the size of the chunk of data that is written on the disk. Why
is that during the installation in which the sets are unpacked
and written on disk the problem doesn't happen? Is it because
they are smaller? In my attempt to copy /usr/users from rz0 to
rz3 there must have been approximately 100MB of data. Swap space
is not a problem because I have more than 300MB.

Thanks in advance.

Dan
>
>Well, the fix mentioned above is obviously part of the kernel, so
>you'd have to install it to get it ;-)
>
>BTW, it's often a necessity that whenever you upgrade a machine
>to first upgrade the kernel, then the user-land.  While you can
>usually get by on "point" releases without doing so, it's usually
>a good idea to follow this practice anyway.