[CRIS] Backtrace past clone

[CRIS] Backtrace past clone

[Orjan Friberg]

I'm having trouble with backtrace not stopping at clone (in the child). 
  Normally I rely on the Dwarf-2 CFI, but clone doesn't have any, so 
this falls back on the manual unwinder.

I understand that using frame pointer register != 0 is dubious (since 
the code we may end up analyzing in the manual unwinder could be 
compiled without a frame pointer), although the glibc clone function for 
CRIS does set the frame pointer to 0.  Other than that, I can't see how 
to detect that the call stack ends for the child.

I see this has been discussed for MIPS in the thread starting at 
http://sources.redhat.com/ml/gdb-patches/2004-03/msg00128.html, which 
also references missing Dwarf-2 information.  Besides this, "grep clone" 
in src/gdb gave surprisingly few hits, as did searching for "clone 
backtrace" in the gdb-patches archive.  The lack of architecture 
-specific search results me wonder if this is something that should be 
handled by the Dwarf-2 unwinder (and that what needs to be fixed is the 
missing Dwarf-2 CFI for clone).

Any suggestions?

-- 
Orjan Friberg
Axis Communications

Re: [CRIS] Backtrace past clone

[Andrew Cagney]

> I'm having trouble with backtrace not stopping at clone (in the child).  Normally I rely on the Dwarf-2 CFI, but clone doesn't have any, so this falls back on the manual unwinder.
> 
> I understand that using frame pointer register != 0 is dubious (since the code we may end up analyzing in the manual unwinder could be compiled without a frame pointer), although the glibc clone function for CRIS does set the frame pointer to 0.  Other than that, I can't see how to detect that the call stack ends for the child.
> 
> I see this has been discussed for MIPS in the thread starting at http://sources.redhat.com/ml/gdb-patches/2004-03/msg00128.html, which also references missing Dwarf-2 information.  Besides this, "grep clone" in src/gdb gave surprisingly few hits, as did searching for "clone backtrace" in the gdb-patches archive.  The lack of architecture -specific search results me wonder if this is something that should be handled by the Dwarf-2 unwinder (and that what needs to be fixed is the missing Dwarf-2 CFI for clone).
> 
> Any suggestions?

Have a look at http://sources.redhat.com/ml/gdb/2004-08/msg00027.html

Andrew

Re: [CRIS] Backtrace past clone

[Orjan Friberg]

Andrew Cagney wrote:
> 
> Have a look at http://sources.redhat.com/ml/gdb/2004-08/msg00027.html

Thanks. (Note to self: search other gdb-related mailing list archives 
than just gdb-patches.)

-- 
Orjan Friberg
Axis Communications