Hi Johnathan!
Re: build a new kernel?

i've just been using precompiled binaries of netbsd kernels. A while
back fsck thrashed an internal gig and zip drive, so an old 160MB IBM
harddrive's been the home for the netbsd os. my original post included
partition info (mkfs.out), disklabel, and dmesg to show context for the
lost files (actually the whole target directory gets lost) until reboot.

Is it really necessary to recompile the kernel? can that be done given
the current bugginess of my os? Would it be easier with a mac-side
compiler? (Codewarrier9 and Think C7 live somewhere on my harddrive). On
the other hand i've got two O'Reilly books: "porting with gcc" and
"imake". would those help me do a recomile? Machine: Q700, 20MB RAM.

On Wed, 2 Jul 1997 18:11:56 -0700 (PDT) Jonathan O'Brien wrote...
> Brian,
> 
> The problem you're seeing is similar to something I experienced
> about a year ago. Some scsi drives don't handle residual data
> pointers correctly (I don't fully understand the problem, just
> know of a potential fix). On my drive, I would create a file
> on the drive and the partition would no longer be accessable like
> you describe. The solution was to add an entry into the quirk
> table to force the "SDEV_AUTOSAVE" then build a new kernel with
> the change.
> 
> Here's an example entry in src/sys/scsi/scsiconf.c:
> 
>         {{T_DIRECT, T_FIXED,
>          "DEC     ", "RZ55     (C) DEC", ""},     SDEV_AUTOSAVE},
>
[from original post]
> 
> > mount -u /
> > # ls -l .p*
> > -rw-r--r-- 2 0 wheel 230 Jun 7 04:06 .profile
> > -rw-r--r-- 1 0 wheel 230 Jun 29 11:02 .profile_test
> > # cp .profile_test  .profile_test2
> > # ls -l .p*
> > ls: .p*: No such file or directory
> > # ls -l
> > # <no files listed; boo hiss!>
> > # sync
> > # reboot
> > booting....
> > #cd root
> > # ls -l .p*
> > -rw-r--r-- 2 0 wheel 230 Jun  7 04:06 .profile
> > -rw-r--r-- 1 0 wheel 230 Jun 29 11:02 .profile_test
> > -rw-r--r-- 1 0 wheel 230 Jun 29 11:32 .profile_test2
> >

brian w.
bwildasi@csulb.edu