This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: LSDA unwind information is off by one (in __gcc_personality_v0)
On 10/20/2016 08:37 PM, Jeff Law wrote:
Obviously if the PC value is actually pointing at the proper
instruction, then the decrement is highly undesirable as it could put
the adjusted PC into a different exception region.
Yes, that's what's happening with the glibc bug.
The question is how to fix this. We can adjust the unwind tables in
glibc to the current behavior easily enough. The instructions involved
are longer than one byte, so we should be able to compensate for a
potential future libgcc which implements non-call unwinding correctly.
The other question is whether we really want to fix libgcc in this way,
after all these years, and change the interpretation of unwind tables
basically over night. We could perhaps add another personality routine
to get a cleaner transition.
And the final question is we can trust current GCC to emit correct
unwind tables for code which involves non-call exceptions (such as
cancellation handlers in glibc, when we don't write them manually in
assembler).
Thanks,
Florian