Re: [patch 04/15] Generic Mutex Subsystem, add-atomic-call-func-x86_64.patch

[Ingo Molnar <mingo@elte.hu>]

* Nick Piggin <nickpiggin@yahoo.com.au> wrote:

> >>Considering that on UP, the arm should not need to disable interrupts
> >>for this function (or has someone refuted Linus?), how about:
> >
> >Kernel preemption.
> 
> preempt_disable() ?

please take a look at kernel/mutex.c, there's a define at the top of the 
file:

// #define MUTEX_IRQ_SAFE

which, if off, makes the mutex code use preempt_disable() and 
preempt_enable() to make it preemption-safe. If it's on, the mutex 
implementation uses IRQ flags.

in my current tree i've already eliminated this define, and have 
switched the code to use preempt_disable()/preempt_enable() exclusively, 
because preempt_*() is equally fast on all platforms, while IRQ disable 
costs vary largely. (and they are rarely faster than preempt_disable()).

my current tree also provides a mechanism for architectures to hand-code 
the mutex lock and unlock fastpath, if they choose to do so. So i think 
we can really stop the cycle counting.

	Ingo