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

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Tue, Dec 20, 2005 at 02:43:30PM -0500, Steven Rostedt wrote:
> So what's wrong with having the generic code, and for those with a fast
> semapore add an arch specific?
> 
> #define mutex_lock down
> #define mutex_unlock up
> #define mutex_trylock(x) (!down_trylock(x))
> 
> Until the mutex code is updated to a fast arch specific implementation.
> 
> Let me restate, that the generic code should not be this, but each arch
> can have this if they already went through great lengths in making a fast
> semaphore.

I have no problem with this since we can then use Nico's swp-based
implementation.  Great!  What seems to be happening though is that
there's a move to make these operations be generic across all
architectures.

What both Nico and myself have demonstrated is that if architectures
are placed into the generic strait-jacket, any alleged performance
benefit of mutexes is completely swamped, which in turn makes the
whole mutex idea entirely pointless.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core