Re: [PATCH] cpuset semaphore depth check optimize

[Paul Jackson <pj@sgi.com>]

Linus wrote:
> and then operate on _that_ level instead. 

As I noted in my reply a few minutes ago, the one unusual rule that
this scheme imposes is that all up's and down's on cpuset_sem must be
done via the wrappers.

So I have continued to strive to have the lock and unlock calls have as
literal substrings "up(&cpuset_sem)" and "down (&cpuset_sem)", such
as with "cpuset_up(&cpuset_sem)" and "cpuset_down(&cpuset_sem)".

This serves as a clear visual reminder of this extra wrapper rule.

The usual "best practices" of:
 1) consistent API's (referring to Nikita's suggestion that these
    routines have "void" arguments instead of "&cpuset_sem"), 
 2) encapsulating related data (your suggestion here), and
 3) [in my inbox] Nikita's cpuset_lock/cpuset_unlock hiding, echoing
    an earlier suggestion of Linus's

are appropriate and desirable mechanisms for building clean abstraction
layers.

I am more of a mind to code this as a thinly veiled hack for use just
within cpuset.c, not another abstraction layer.

I can certainly code this as a proper layer, if you like.  My intuition
is that, in this case, doing so would slightly increase the mental load
on the reader, not decrease it.

In actuality, I don't code for elegance so much as I code to minimize
the time it takes the typical reader to -correctly- understand what's
going on.

But if after all my eloquence of the last hour, Linus, Nikita and
Andrew are all in agreement that cpuset_lock/cpuset_unlock with
struct encapuslation of the 3 data items is preferrable, I'll gladly
code that up.  Well, actually, just a single clear "make it so"
from Linus or Andrew would likely be sufficient.

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401