[PATCH v3 0/1] gdb: avoid conversion of SIGSEGV to SIGTRAP on user breakpoints

[Klaus Gerlicher <klaus.gerlicher@intel.com>]

Hi Andrew,

thanks for the feedback. 

You replied: 

> But, I wonder if there's a different approach that could be used?
> 
> I assume that your out of tree architecture with these imprecise page
> faults has a new gdbarch to represent it.
> 
> You are limiting the SIGSEGV -> SIGTRAP conversion because, I assume,
> you've hit cases where a SIGSEGV occurs and then your target has run on
> and reported the stop at the address of a user breakpoint.  But all
> you're really doing is reducing the scope for errors.  It could be the
> case that the SIGSEGV is reported at the site of an internal breakpoint,
> and then you'll still have issues.
> 
> So what if, instead, you added a new gdbarch method, something like:
> 
>   bool gdbarch_has_imprecise_page_faults (struct gdbarch *gdbarch);
> 
> The default for this, and for all currently in-tree targets, would be to
> return true.  For your target this will return false.
> 
> Then in this code we would do:
> 
>   if (ecs->ws.kind () == TARGET_WAITKIND_STOPPED
>       && (ecs->ws.sig () == GDB_SIGNAL_ILL
> 	  || (ecs->ws.sig () == GDB_SIGNAL_SEGV
>               !gdbarch_has_imprecise_page_faults (gdbarch))
> 	  || ecs->ws.sig () == GDB_SIGNAL_EMT))
> 
> How does this approach sound?

I think for now this would be a viable solution and I implemented it in the V3 
patch.

However, we had a little discussion internally and we were wondering if this now
feels more like a workaround in a workaround.

It seems to me that the conversion of any of these signals should really be
only done for a specific target.  Even though the SIG_ILL and SIG_EMT
conversions  obviously have no side effects for most targets, I would think
these should be avoided.

Do you think there's a way to limit these more specifically or are we unsure
which targets actually need these? Maybe some of the users are already obsolete?

I would assume it would be difficult to limit to a specific target when we used
the remote target since we would then have to update GDB server to also support
this.

I'm of course fine if we fix it with the architecture method but maybe others
would disagree?

Thanks
Klaus

Gerlicher, Klaus (1):
  gdb: avoid conversion of SIGSEGV to SIGTRAP on user breakpoints

 gdb/gdbarch-gen.c         | 22 ++++++++++++++++++++++
 gdb/gdbarch-gen.h         |  7 +++++++
 gdb/gdbarch_components.py | 12 ++++++++++++
 gdb/infrun.c              |  4 +++-
 4 files changed, 44 insertions(+), 1 deletion(-)

-- 
2.34.1

Intel Deutschland GmbH
Registered Address: Am Campeon 10, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928