Re: RFC: threads PREPARE_TO_PROCEED patch

[Daniel Jacobowitz <drow at mvista dot com>]

On Sat, Mar 15, 2003 at 09:39:14PM +0100, Mark Kettenis wrote:
>    Date: Fri, 14 Mar 2003 13:35:29 -0500
>    From: Daniel Jacobowitz <drow at mvista dot com>
> 
>    > I'm not quite sure whether changing the gdbarch default is a good
>    > idea, but replacing lin_lwp_prepare_to_proceed with
>    > generic_prepare_to_proceed has been the intention all along.
> 
>    Well, let me describe the problem I'm trying to solve; I'd like your
>    opinion on how to approach it.  When using gdbserver, we need to have
>    generic_prepare_to_proceed.  Not the lin_lwp version, and not the
>    "default" one from arch-utils.  The former won't work and the latter
>    doesn't do enough.  So cross debuggers need to pick this up.
> 
> OK, but generic_prepare_to_proceed() is perfectly usable on a native
> GNU/Linux GDB too, isn't it?

Yes, exactly.

>    Note that this is a property of the target.  Not of the architecture. 
>    I'm not sure PREPARE_TO_PROCEED belongs in gdbarch at all.  It's only
>    defined by Mach3, HP/UX, and Linux; it's undefined for x86-64-linux
>    (why???).  I could set it in all the Linux gdbarch init functions that
>    I care about, but that doesn't seem like much of a solution.
> 
> It seems to be a property of the OS to me.  In its current
> incarnation, gdbarch does includes details of both the architecture
> (ISA) and the OS (OS/ABI).  So gdbarch seems to be the correct place
> for PREPARE_TO_PROCEED to me.  So yes, I think you should add it to
> all relevant Linux gdbarch init functions.

I can do that; I'll put a patch together.

But I must admit that I don't really agree.  It seems to be a property
of the threads implementation for the target instead. Consider this
case: if someone wanted to write a remote protocol stub for HP/UX. 
They wouldn't want the HP/UX version of PREPARE_TO_PROCEED naturally,
since that's native-only.  They'd want most likely
generic_prepare_to_proceed.  The default function isn't useful; it
doesn't support switching threads correctly.

(Incidentally, from reading the HP/UX implementation, I believe that
using generic_prepare_to_proceed would work there too.  It wouldn't
work for the Mach 3.0 implementation as-is but I think it could be made
to work.  I'm not volunteering; are either hppa*-*-osf* or
mips*-*-mach3* still living?  Perhaps we should deprecate them next
release.)

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer