This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [rfc] Handle broken CFI for signal trampolines in libc on amd64-linux
- From: "Mark Kettenis" <mark dot kettenis at xs4all dot nl>
- To: "Ulrich Weigand" <uweigand at de dot ibm dot com>, gdb-patches at sourceware dot org
- Date: Fri, 21 Nov 2008 12:39:59 +0100 (CET)
- Subject: Re: [rfc] Handle broken CFI for signal trampolines in libc on amd64-linux
> On Fri, Nov 21, 2008 at 02:33:29AM +0100, Ulrich Weigand wrote:
> > Anyway, while it is certainly good that this is fixed, I'm still
> > wondering why we should rely on that when we have a hard-coded
> > sigtramp detector that should be working just fine under any
> > circumstances.
>
> I think that one reason was the extra work of the signal handler
> sniffer. The amd64 one doesn't do much for named functions, though,
> and functions with CFI are likely to be named. I suggest asking
> Mark Kettenis's opinion.
My memory is a bit hazy on this, but I think the idea was that the signal
frame unwinder would only be used for older versions of linux/glibc that
don't provide the necessary CFI, and that newer versions would provide
correct CFI which would give the kernel/glibc people complete freedom on
how to implement signal frames. As such, I'm inclined to say "no" to your
diff.