Re: [patch 1/3] mutex subsystem: trylock

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

Nicolas Pitre wrote:

> I provided you with the demonstration last week:
> 

I didn't really find it convincing.

> - the non SMP (ARM version < 6) is using xchg.
> 
> - xchg on ARM version < 6 is always faster and smaller than any 
>   preemption disable.
> 

Maybe true, but as I said, if you have preemption enabled, then there
are far more preempt_xxx operations in other places than semaphores,
which impact both size and speed.

> - xchg on ARM is the same size as the smallest solution you may think of
>   when preemption is disabled and never slower (prove me otherwise? if 
>   you wish).
> 

I was going from your numbers where you said it was IIRC a cycle faster.

> - all xchg based primitives are "generic" code already.
> 

And yet there is a need for architecture specific code. Also having
xchg and a cmpxchg variants mean that you have two different sets of
possible interleavings of instructions, and xchg has much more subtle
cases as you demonstrated.

> And I think you didn't look at the overall patch set before talking 
> about "lots of ugliness", did you?  The fact is that the code, 

Yes I did and I think it is more ugly than my proposal would be.

> especially the core code, is much cleaner now than it was when 
> everything was seemingly "generic" since the current work on 
> architecture abstractions still allows optimizations in a way that is so 
> much cleaner and controlled than what happened with the semaphore code, 
> and the debugging code can even take advantage of it without polluting 
> the core code.
> 
> It happens that i386, x86_64 and ARM (if v6 or above) currently have 
> their own tweaks to save space and/or cycles in a pretty strictly 
> defined way.  The effort is currently spent on making sure if other 
> architectures want to use one of their own tricks for those they can do 
> it without affecting the core code which remains 95% of the whole thing 
> (which is not the case for semaphores), and the currently provided 
> architecture specific versions are _never_ bigger nor slower than any 
> generic version would be.  Otherwise what would be the point?  
> 

I don't think size is a great argument, because the operations should
be out of line anyway to save icache (your numbers showed a pretty
large increase when they're inlined)

And as for speed, I'm not arguing that you can't save a couple of
cycles, but I'm weighing that against the maintainability of a generic
implementation which has definite advantages. Wheras I don't think you
could demonstrate any real speed improvement for saving a few cycles
from slightly faster semaphore operations on CONFIG_PREEMPT kernels?

If someone ever did find the need for an arch specific variant that
really offers advantages, then there is nothing to stop tha being
added.

-- 
SUSE Labs, Novell Inc.

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