Re: Current NTPL changes for PPC32/PPC64 M2

Jakub Jelinek writes:

> I think at least atomic_decrement_if_positive should be added to atomic.h
> and sysdeps/linux/ code should use the atomic.h primitives, not some lll_*
> ones.

I tried including atomic.h in lowlevellock.h and that caused all sorts
of problems.  There are things elsewhere in glibc that want
lowlevellock.h but also include atomicity.h, which also defines an
atomic_add (for instance) with somewhat different semantics - it
doesn't return a value and it doesn't include any barrier.  That's why
I defined __lll_add in lowlevellock.h rather than using atomic_add.

Paul.