Re: [PATCH v3 2/4] tracing: Make syscall_(un)regfunc arch-specific

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Aug 24, 2009 at 01:00:23PM -0700, Josh Stone wrote:
> On 08/24/2009 12:58 PM, Frederic Weisbecker wrote:
> > On Mon, Aug 24, 2009 at 12:31:26PM -0700, Josh Stone wrote:
> >> On 08/23/2009 02:14 PM, Frederic Weisbecker wrote:
> >>> I really don't like that.
> >>> See how the s390 and x86 version of the above code are completely
> >>> identical?
> >>>
> >>> Please put this in kernel/ptrace.c
> >>
> >> Yes, I see your point, and I think kernel/ptrace.c is a fine place for
> >> it.  Making it conditional on CONFIG_TRACEPOINTS and
> >> CONFIG_HAVE_FTRACE_SYSCALLS is probably best too, though I think the
> >> latter should now be HAVE_SYSCALL_TRACEPOINTS.
> > 
> > 
> > As you prefer, this new name can be indeed more verbose.
> 
> Actually, now I'm second-guessing the need to move these at all.  Since
> they only make sense for CONFIG_TRACEPOINTS, can't they stay in
> kernel/tracepoint.c and just be conditional on HAVE_SYSCALL_TRACEPOINTS?
>  The only real change needed is for the tracepoint declarations to also
> be conditional.
> 
> Josh
> 

Both ways make sense to me, although I generally see the role of
kernel/tracepoint.c to only host the general core tracepoints mechanism.

And here these two callbacks are more about specific tracepoints coverage,
somewhat tied to the ptrace background because we are using a ptrace
bridge to reach these tracepoints.

Well, either ways look good:

- tracepoint.c: to solve the lack of a functionnality in very
  specific cases.

- ptrace.c: because it's part of a ptrace mechanism.


I don't feel strongly about that :-)