Date: Sun, 27 Jun 1999 17:03:46 +0100 (BST) From: Doug Rabson <dfr@nlsystems.com> To: Peter Wemm <peter@netplex.com.au> Cc: alpha@freebsd.org Subject: Re: Hmm!! Message-ID: <Pine.BSF.4.05.9906271702250.80685-100000@herring.nlsystems.com> In-Reply-To: <19990627155406.CCF9181@overcee.netplex.com.au>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 27 Jun 1999, Peter Wemm wrote:
> Peter Wemm wrote:
> > Doug Rabson wrote:
> > > On Sun, 27 Jun 1999, Peter Wemm wrote:
> > [..]
> > > > Anybody got any bright ideas about this? (apart from run the memory test
> er
> > > > in SRM that is..)
> > >
> > > Maybe the kernel doesn't fit in its memory region? You could try enabling
> > > the debug printfs in alpha_init() where it reads the memory cluster
> > > information.
> >
> > Another datapoint:
> >
> > # /bin/cat
> > Jun 27 22:01:19 ashburton /kernel: pid 23490 (cat), uid 0: exited on signal 1
> 1 (core dumped)
> > Segmentation fault - core dumped
> > # /bin/cat
> > foo
> > foo
> > #
> >
> > The interesting thing is that according to the coredumps, it consistantly
> > appears to die in the first getenv() inside setlocale() when walking the
> > environ list. It's being a damn nuisance to check because the slightest
> > change makes it work.. ie: under truss or ktrace, gdb, etc, it works fine.
> > Multiple different cat binaries all do this - including one I compiled with
> > -g - but they only misbehave under /bin/sh. Something funny about the
> > environment passing perhaps?
> >
> > This is on the small kernel BTW, so the size of the kernel seems to affect
> > things, but not eliminate the problems.
> >
> > Cheers,
> > -Peter
>
> This is getting stranger by the minute...
>
> If I do this to cat.c:
> main() {
> ...
> write(1, "A", 1);
> setlocale(LC_CTYPE, "");
> write(1, "A", 1);
> ...
> Then it works. If I take away either added write() syscall, it fails.
> If I change it to this, it works too:
> setlocale(LC_CTYPE, "");
> write(1, "A", 1);
>
> while ((ch = getopt(argc, argv, "benstuv")) != -1) {
> write(1, "A", 1);
> switch (ch) {
> .. even though the second write() is never executed.
>
> If I compile without optimization, cat seems to work always. This is
> bizzare. Why cat? Why with the different kernels does it pick mount_nfs
> or sysctl to screw up, and only then when called from /bin/sh.
>
> I have saved a copy of the (-O) assembler code from a working and non-working
> version of cat.c. (reminder: it only fails when run from sh, eg in /etc/rc
> or under make.)
>
> Oh, also, according to the console:
> pal VMS PALcode V1.20-5, OSF PALcode V1.22-6
> version V5.1-4 Feb 12 1998 15:40:06
>
> If I look at the firmware updates directory, the current 5_4 release is
> labelled:
>
> +++++++++++++++++++++++++++++++++++++++ April 1999
> + AlphaPC 164SX Firmware +
> + README-FIRST +
> +++++++++++++++++++++++++++++++++++++++
> 1. Latest Firmware Update Revisions
> SRM V5.4-1
> VMS PALcode V1.20-6
> Unix PALcode V1.22-6
>
> The Unix palcode is the same version, but the SRM release is a newer. Does
> it seem worthwhile to try to update SRM?
Its worth a try. You should be able to just put the relavent fwupdate.exe
onto a floppy and use SRM's fwupdate command. It will read the floppy
(extremely slowly) then ask you some silly question about option disks
which you should ignore.
--
Doug Rabson Mail: dfr@nlsystems.com
Nonlinear Systems Ltd. Phone: +44 181 442 9037
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.05.9906271702250.80685-100000>