Re: [PATCH 09/14] perf ftrace: Add 'record' sub-command

From: David Ahern
Date: Sun May 05 2013 - 23:22:43 EST


On 5/5/13 7:57 PM, Namhyung Kim wrote:
Hi David,

On Fri, 26 Apr 2013 07:44:18 -0600, David Ahern wrote:
On 4/25/13 12:24 AM, Namhyung Kim wrote:
But it looks there's a race between cond_wait() and cond_broatcast().
I'll take a look at that.

Why not use eventfd or a pipe for the signalling instead?

Thanks for your feedback!

I didn't have an idea of using eventfd for this case. Does it helpful
to prevent such a deadlock like this?

I have used it to replace pthread_cond_wait / pthread_cond_{signal,broadcast} -- lighter and faster. It was added in 2.6.27 I believe, so will be available for any system that supports perf (record side). For people compiling perf against say RHEL5 (e.g. analysis) you will need a check as to whether it exists.

David

--
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/