Re: AIM7 40% regression with 2.6.26-rc1

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> Which is why I'm 100% convinced it's not even worth saving the old 
> code. It needs to use mutexes, or spinlocks. I bet it has *nothing* to 
> do with "slow path" other than the fact that it gets to that slow path 
> much more these days.

i think your theory should be easy to test: Yanmin, could you turn on 
CONFIG_MUTEX_DEBUG=y and check by how much AIM7 regresses?

Because in the CONFIG_MUTEX_DEBUG=y case the mutex debug code does 
exactly that: it doesnt use the single-instruction fastpath [it uses 
asm-generic/mutex-null.h] but always drops into the slowpath (to be able 
to access debug state). That debug code is about as expensive as the 
generic semaphore code's current fastpath. (perhaps even more 
expensive.)

There's far more normal mutex fastpath use during an AIM7 run than any 
BKL use. So if it's due to any direct fastpath overhead and the 
resulting widening of the window for the real slowdown, we should see a 
severe slowdown on AIM7 with CONFIG_MUTEX_DEBUG=y. Agreed?

	Ingo