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

[Nick Piggin <nickpiggin@yahoo.com.au>]

Nicolas Pitre wrote:
> On Wed, 21 Dec 2005, Nick Piggin wrote:
> 
> 
>>Nicolas Pitre wrote:
>>
>>>On Tue, 20 Dec 2005, Nick Piggin 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() ?
> 
> 
> Sure, and we're now more costly than the current implementation with irq 
> disabling.
> 

Why? It is just a plain increment of a location that will almost certainly
be in cache. I can't see how it would be more than half the cost of the
irq implementation (based on looking at your measurements). How do you
figure?

Also, preempt_disable is a very frequent operation on preempt kernels so
if you have CONFIG_PREEMPT then you don't care about preempt_disable in
down() (and if you do then you are calling down too often).

> If we go with simple mutexes that's because there is a gain, even a huge 
> one on ARM, especially for the fast uncontended case.  If you guys 
> insist on making things so generic and rigid then there is no gain 
> anymore worth the bother.
> 

I guess there is no bother for you, but maintaining code for 1 generic
platform versus two dozen architectures is a huge win for many.

-- 
SUSE Labs, Novell Inc.

Send instant messages to your online friends http://au.messenger.yahoo.com