request for review: backport of sx and rwlocks from 7.0 to
attilio at freebsd.org
Sat Sep 1 18:04:23 PDT 2007
2007/8/31, Alfred Perlstein <alfred at freebsd.org>:
> Hi guys,
> Some work here at work was approved for sharing with community so
> I'm posting it here in hope of a review.
> We run some pretty good stress testing on our code, so I think it's
> pretty solid.
> My only concern is that I've tried my best to preserve kernel source
> API, but not binary compat though a few simple #defines.
> I can make binary compat, in albeit a somewhat confusing manner, but
> that will require some rototilling and weird renaming of calls to
> the sleepq and turnstile code. In short, I'd rather not, but I will
> if you think it's something that should be done.
> There's also a few placeholders for lock profiling which I will
> very likely be backporting shortly as well.
> Patch is attached.
I started looking at the patch and I have 2 things to say:
- why you backported the allocating patch with UMA in sleepqueues? it
is ortogonhal to this problem and it is not necessary due in this case
- Instead than using the stub __aligned() for struct thread, you
should use what we alredy do for 7.0 as dealing with uma allocation
functions and a separate stub for thread0. you can workaround the
missing of uma functions with a simple macro.
I will try to give a line-by-line revision ASAP.
Peace can only be achieved by understanding - A. Einstein
More information about the freebsd-smp