|
From: Paul P. <ppl...@go...> - 2009-08-12 14:48:28
|
On Wed, Aug 12, 2009 at 7:29 AM, Tom Hughes<to...@co...> wrote:
> On 12/08/09 07:17, Paul Pluzhnikov wrote:
>
>> Is there any reason to overwrite the %{c,d,..}s registers instead of
>> keeping them at whatever values they currently have?
>
> To answer this question, the obvious issue is that we're supposed to be
> setting up the new process to match the guest state - what you're doing is
> assuming that guest segment registers have the same values as the host
> segment registers.
>
> That's probably largely true, but I suspect may not be for FS/GS in all
> cases?
Thanks.
Since VG is (apparently) not tracking FS/GS on amd64 for the guest,
there is no harm in setting them like I did, is there? I believe the
values I set them to are less likely to be wrong then what's happening
now.
> Our whole debugger integration scheme is pretty rubbish and fragile anyway
> and should probably be redone as a gdb debugger stub
That was the next question I was going to ask :-)
Has making VG implement GDB remote stub been considered? I couldn't
find any references in VG mailing lists, and this will *obviously*
make the integration so much better.
[This sounds like an excellent "Google summer of code" or a summer
intern project.]
Thanks,
--
Paul Pluzhnikov
|