On Thursday 16 August 2012, Will Deacon wrote:

On Wed, Aug 15, 2012 at 04:07:36PM +0100, Arnd Bergmann wrote:

quoted

On Tuesday 14 August 2012, Catalin Marinas wrote:

quoted

From what I can tell, there is no support for 32 bit processes debugging
64 bit ones. Is that something you plan to add in the future, or do you
consider that out of scope? In either case, a comment would be helpful.

That can't really work because the debugger won't be able to manipulate
child pointers properly without us adding a new ptrace interface (and then,
I still wonder about how feasible it really is). I can add a comment.

You can already have a 32 bit gdb that is able to do remote debugging of
64 bit processes using a gdb server process. I guess it wouldn't be
too strange to have a ptrace extension to allow the native case as well.
I agree it's not a high priority.

quoted

quoted

+long arch_ptrace(struct task_struct *child, long request,
+		 unsigned long addr, unsigned long data)
+{
+	int ret;
+	unsigned long *datap = (unsigned long __user *)data;
+
+	switch (request) {
+		case PTRACE_GET_THREAD_AREA:
+			ret = put_user(child->thread.tp_value, datap);
+			break;
+
+#ifdef CONFIG_HAVE_HW_BREAKPOINT
+		case PTRACE_GETHBPREGS:
+			ret = ptrace_gethbpregs(child, addr, datap);
+			break;
+
+		case PTRACE_SETHBPREGS:
+			ret = ptrace_sethbpregs(child, addr, datap);
+			break;
+#endif
+
+		default:
+			ret = ptrace_request(child, request, addr, data);
+			break;
+	}
+
+	return ret;
+}

Is there a reaons why these are not regsets but have their own ptrace
commands? I believe new architectures should generally not add ptrace
commands any more.

I could probably add some regset wrappers about the hbp accessors (which we
have to keep for the compat ptrace interface). I'll have a think as it might
even make sense to have different regsets for breakpoints and watchpoints.

As for the the tls, is it worth having a regset with only one register?

Better ask the gdb folks. I'm adding Uli to Cc, maybe he has some insight.

	Arnd