mounts via fstab...

[Thomas Fröhlich]

Am Samstag, 2. September 2006 18:51 schrieb Rob Landley:
> I asked because devfs was already deprecated back when 2.6.0 shipped,
> back in 2003.  Your previous "stable" deployed something that had been
> deprecated for a dozen stable releases.

sometimes it's easier you will not change a running system, so we make 
little updates without cleaning devfs. :)

> > arm-linux-uclibc-gcc (GCC) 3.3.3
> > xscale, pxa 255
>
> So most likely something's doing an unaligned access.  (Does that
> platform have an mmu?)

yes, the pxa255 has an mmu

> Is ash what's calling "mount -a"?  Presumably there's a script somewhere
> that's calling mount -a.  It's possible that mount - a is failing, but
> unlikely if it's working when called from the command line.  We haven't
> established that the script is getiting far enough to actually make the
> call yet.  What's most likely is something in the script is aborting the
> script. (Something is segfaulting.  It would be nice to track down
> what.)
>
> Your init script has an rcS file you didn't send me.  Don't bother, I
> haven't got an xscale here to run it on.  What I'd do is stick some
> "echo" statements into it to see how far through your script you're
> getting before the segfault occurs.

=======================
[/etc/init.d/rcS]
#! /bin/sh

/bin/mount -a
=======================

> > ok, I get bb-1.1.3 and bb 1.2.0 from the hp, this two versions have no
> > problem to start without seg fault. only the svn version have this
> > feature ;)
>
> If you could trace it down to the changeset that actually caused the
> problem, that would be really useful.  (Do an anonymous checkout and
> then select specific versions with "svn update -r 12345", and binary
> search until you find the changeset that caused the breakage.)

Ok, I will search!

> Or if you don't want to mess with svn, binary search through our nightly
> snapshots...  Actually, Erik seems to have gone through and deleted the
> older ones of those recently.  Unlikely to be useful. :(

I make now a svn update to changeset 16046. there happens on cmd line:

xscale-nfs:/ # mount -a
Segmentation fault

You have the right idea!

> Hmmm, if I get a mercurial version of the repository up, the mercurial
> web viewer can be configured to let you download a tarball of an
> arbitrary version...  (Alas, my conversion script is in pieces again. 
> It turns out that 'sed "stuff" | while read i; do
> blah[${#blah[@]}]="$i"; done' doesn't actually work right because
> everything to the right of the pipe is in a subshell and thus even
> exported variables don't get changed in the parent's environment.  This
> is somewhat frustrating, because I have yet to find a shell script
> alternative to that construct that can take each line (even with
> embedded spaces, which filenames have been known to have) and process it
> as a unit.  Hmmm...)

oohhh... :(
now I will search the changeset
regards
Thomas