Re: RFA: Support libthread_db xregs interface

[Andrew Cagney <cagney at gnu dot org>]

> This patch adds support to GDB for the libthread_db 'xregs' functions,
> which allow registers not present in 'gregset_t' or 'fpregset_t' to be
> passed through libthread_db.  The contents of the 'xregs' register set
> is not specified by GDB; the libthread_db client can specify whatever
> size it likes, and libthread_db will pass it through unchanged.
>
> I'll post a second patch which defines an 'xregs' register set for the
> PowerPC E500 SPE registers.
>
> At present, the GNU C library's libthread_db's xregs functions are
> stubbed out; I also have a patch to glibc to implement them, which
> I'll be submitting.  If GDB loads a libthread_db that does not
> implement the xregs functions, this patch handles that case.
>
> These have all been tested on E500 powerpc-unknown-linux-gnu and on
> i686-pc-linux-gnu (where they should have no effect).
>
> This patch should allow us to remove the call to dummy_sse_values in
> supply_fpregset.  Not that that call is especially necessary or
> helpful at the moment anyway, as far as I can tell.
>
> This patch was originally posted here:
> http://sources.redhat.com/ml/gdb-patches/2004-03/msg00571.html
>
> I believe it's been revised to implement the suggestions made there.

[...]

> + v:=:const struct regset *:xregs_regset:::0

Mark's regset change added both the "regset.h" object and the 
regset_from_core_section architecture method.  They, together, replace 
the old *-nat.c:fill_regset et.al. calls.

Can we implement the equivalent here for ptrace/thread-db?

--

Instead of:
> +           if (! warned_xregs_not_implemented)
> +             {
> +               warning ("thread debugging library is too old to access "
> +                        "%s registers.",
> +                        gdbarch_xregs_name (current_gdbarch));
can we use "complaint".  That "thread" should be "Thread".

--

> + v:=:int:xregs_size:::0
> + v:=:const char *:xregs_name:::0

I gather these were fields in the original xreg_desc object but are 
missing from the "regset"?  Should these, instead be added to the 
regset, or a new object extending regset created?

Andrew