Re: AIM7 40% regression with 2.6.26-rc1

[Ingo Molnar <mingo@elte.hu>]

* Matthew Wilcox <matthew@wil.cx> wrote:

> > down(), down_interruptible() and down_try() should use 
> > spin_lock_irq(), not irqsave.
> 
> We talked about this ... the BKL actually requires that you be able to 
> acquire it with interrupts disabled. [...]

hm, where does it require it, besides the early bootup code? (which 
should just be fixed)

down_trylock() is OK as irqsave/irqrestore for legacy reasons, but that 
is fundamentally atomic anyway.

> > up() seems to be doing wake-one, FIFO which is nice.  Did the 
> > implementation which we just removed also do that?  Was it perhaps 
> > accidentally doing LIFO or something like that?
> 
> That's a question for someone who knows x86 assembler, I think.

the assembly is mostly just for the fastpath - and a 40% regression 
cannot be about fastpath differences. In the old code the scheduling 
happens in lib/semaphore-sleeper.c, and from the looks of it it appears 
to be a proper FIFO as well. (plus this small wakeup weirdness it has)

i reviewed the new code in kernel/semaphore.c as well and can see 
nothing bad in it - it does proper wake-up, FIFO queueing, like the 
mutex code.

	Ingo