Re: [RFC PATCH 05/10] ftrace: Drop buffer check in functionprofiler callbacks

From: Frederic Weisbecker
Date: Sat Jan 30 2010 - 15:47:53 EST


On Mon, Jan 25, 2010 at 02:17:29PM +0800, Lai Jiangshan wrote:
> Frederic Weisbecker wrote:
> > Drop the null check on hlist->hash. It is wasteful because
> > we don't register the tracer if the buffer allocation failed,
> > and the memory barrier in register_ftrace_graph() ensure it
> > is visible to the callbacks right away.
> >
>
> When it is on a cpu which is just came up, hlist->hash is not
> allocated either.
>
> See ftrace_profile_init().
>
> function_profile_call(), profile_graph_entry() and profile_graph_return()
> on a just online cpu are wasteful. I think we need call
> ftrace_profile_init_cpu()(if ftrace_profile_enabled=1) when CPU_UP_PREPARE.



May be yeah, or perhaps we should play with cpu_possible_mask()
instead of cpu_online. Whatever, the patch is also buggy for the reasons
given by Steve, I'm going to drop it until we have a sane implementation
of synchronize_preempted_tasks().

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/