Re: [PATCH RFC] perf lock: Add new event "lock_acquired_waittime"for contention analysis

From: Hitoshi Mitake
Date: Tue Aug 03 2010 - 03:37:14 EST


On 08/02/10 21:23, Peter Zijlstra wrote:
On Sun, 2010-08-01 at 14:36 +0900, Hitoshi Mitake wrote:
I want to make lockdep production kernel feature with dynamic
patching.

Not really feasible. Note that enabling lockdep grows the size of
spinlock_t (and others) quite significantly.

Then there is the problem that you need to pass all tasks through some
quiesent state in order to enable lockdep (there must be a guarantee of
no locks held).

Now, we could possibly make it all work, but I'm not at all convinced we
want to pay the price, which is a much _much_ more complex
infrastructure.



I have to agree with the problem of huge cost you mentioned above.

But, I believe that
- lock validation
- lock tracing
- in kernel light weight lock usage statistics
should be divided into individual features, at least.
And tracing and statistics should be available for production kernel, ideally.

How do you think about this point?
--
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/