[Crash-utility] Corrupted tee in crash gdb

[Johan.Erlandsson at sony.com]

> Hi,
> 
> I have several crash dump files that reliably crash gdb inside crash.
> 
> This is with compressed dumps and recent kernels
> 
> It appears there is an error during the symbol processing (usually
> when looking up runqueue, but even if you disable that it fails on
> something else) and then gdb crashes because the tee
> structure it uses to print error messages is corrupted. The tee fputs
> vector is NULL and gdb jumps to zero while trying to print the
> error message. So somehow crash doesn't set this up properly.
> 
> Using -minimal works, but that's too limiting.
> 
> I'm using this gdb patch to work around it by disabling the gdb errors.
> That makes crash work well enough to look at most things.
> 
> It's probably not the correct fix, but at least it works for me.

Hi Andi

Sorry for late response. But I noticed this after making the post
below:

[PATCH] GDB started inside crash should not read any .gdbinit
https://listman.redhat.com/archives/crash-utility/2021-March/msg00014.html

Your behavior and patch match with what is described there. So could
it be that you have a .gdbinit file with 'set logging on'? Anyway you
might want to double check without a .gdbinit file.

Regards
Johan