Paul E. McKenney wrote:
> The big long test is over, so if you can send me a patch against some
> Linus tree, I will see about testing your code.  Assuming someone else
> doesn't jump on the machines first, that is...  :-/
>   
Cool.
Attached is the latest patch I have, it should be stable and perform 
about as good as the current code.
The only known regressions are:
- a lack of a force_quiescent_state() implementation. Shouldn't matter 
since no_hz cpus are bypassed.
- the irq entry of no_hz cpus takes a global spinlock. For some test 
setups, this will probably kill the performance.

Which test do you intend to run?
An oprofile with an opannotate asm file would be nice. Preferably with 
inlined spinlocks - that would immediately show the hot spots.

Do you have an opannotate output from your patch?

--
    Manfred