linux-kernel - Re: [PATCH][RFC] trace: profile likely and unlikely annotations

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Tue, 28 Oct 2008 10:37:20 -0400
From:	Theodore Tso <tytso@....edu>
To:	Steven Rostedt <rostedt@...dmis.org>
Cc:	LKML <linux-kernel@...r.kernel.org>,
	Mike Snitzer <snitzer@...il.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Ingo Molnar <mingo@...e.hu>,
	Thomas Gleixner <tglx@...utronix.de>,
	Peter Zijlstra <peterz@...radead.org>,
	Arjan van de Ven <arjan@...radead.org>,
	Frederic Weisbecker <fweisbec@...il.com>
Subject: Re: [PATCH][RFC] trace: profile likely and unlikely annotations

On Tue, Oct 28, 2008 at 12:12:48AM -0400, Steven Rostedt wrote:
> 
> Andrew Morton recently suggested having an in-kernel way to profile
> likely and unlikely macros. This patch achieves that goal.

Maybe I'm confused, but when I read through the patch, it looks like
that 'hit' is incremented whenever the condition is true, and 'missed'
is incremented whenever the condition is false, correct?

Is that what you intended?  So for profile_unlikely, "missed" is good,
and "hit" is bad, and for profile_likely, "hit" is good, and "missed"
is bad.  That seems horribly confusing.

If that wasn't what you intended, the meaning of "hit" and "missed"
seems to be highly confusing, either way.  Can we perhaps use some
other terminology?  Simply using "True" and "False" would be better,
since there's no possible confusion what the labels mean.   

> +#define unlikely(x) ({							\
> +			int ______r;					\
> +			static struct ftrace_likely_data ______f	\
> +				__attribute__((__aligned__(4)))		\
> +				__attribute__((section("_ftrace_unlikely"))); \
> +			if (unlikely_notrace(!______f.ip))		\
> +				______f.ip = __THIS_IP__;		\
> +			______r = unlikely_notrace(x);			\
> +			ftrace_likely_update(&______f, ______r);	\
> +			______r;					\
> +		})

Note that unlikely(x) calls ftrace_likely_update(), which does this:

> +void ftrace_likely_update(struct ftrace_likely_data *f, int val)
> +{
> +	/* FIXME: Make this atomic! */
> +	if (val)
> +		f->hit++;
> +	else
> +		f->missed++;
> +}
> +EXPORT_SYMBOL(ftrace_likely_update);

So that seems to mean that if unlikely(x) is false, then _____r is 0,
which means we increment f->missed.   Or am I missing something?

I would have thought that if unlikely(x) is false, that's *good*,
since it means the unlikely label was correct.  And normally, when
people think about cache hits vs cache misses, hits are good and
misses are bad.  Which is why I think the terminology is highly
confusing...

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