Re: [PATCH] gdb: nat/x86-dregs.c: add and use x86_dr_low_can_get_status

[John Baldwin <jhb@FreeBSD.org>]

On 7/15/21 12:53 PM, Simon Marchi via Gdb-patches wrote:
> I built and tried gdb on OpenBSD, and it immediately segfaults when
> running a program.  I tracked down the problem to x86_dr_low.get_status
> being nullptr at this point:

I've looked, and another way to possibly fix this is to make the x86_bsd_nat_target
class not inherit from x86_nat_target<> when x86 debug registers aren't supported.
Something like this in x86-bsd-nat.h:

/* A prototype *BSD/x86 target.  */

#ifdef HAVE_PT_GETDBREGS
template<typename BaseTarget>
class x86bsd_nat_target : public x86_nat_target<BaseTarget>
{
   using base_class = x86_nat_target<BaseTarget>
public:
   void mourn_inferior () override
   {
     x86_cleanup_dregs ();
     base_class::mourn_inferior ();
   }
};
#else /* !HAVE_PT_GETDBREGS */
template<typename BaseTarget>
class x86bsd_nat_target : public BaseTarget
{
};
#endif /* HAVE_PT_GETDBREGS */

I would be further tempted to then remove x86-bsd-nat.o from OpenBSD's configure.nat
lines entirely sine it shouldn't be needed at that point, and maybe adjust x86-bsd-nat.c
to start with something like:

#ifndef HAVE_PT_GETDBREGS
#error "x86-bsd-nat.c requires PT_GETDBREGS"
#endif

My other series that divorces FreeBSD from using i386-bsd-nat.c and
amd64-bsd-nat.c almost makes it so that we could further clean those
files up perhaps so that NetBSD explicitly used x86bsd_nat_target as the
Base class passed to the templates defined in i386-bsd-nat.c and
amd64-bsd-nat.c so that we wouldn't even need an empty version of
x86bsd_nat_target for the OpenBSD case, but I think we should try to
fix the immediate issue on OpenBSD first and maybe do that cleanup
later.

I wouldn't be opposed to this fix also going in, but I kind of think it's
cleaner overall for the OpenBSD target to not use x86-nat.c at all if
it doesn't support the x86 debug registers.  I can take a stab at a
patch to do what I described above if that is easier (I don't currently
have an OpenBSD VM around, but it shouldn't be too hard for me to spin
one up to test this)

-- 
John Baldwin