[RFC PATCH 0/4] powerpc/qspinlock: make slowpath accesses more efficient

[Nicholas Piggin]

This series (at least the first 3 patches) will get folded into the main
series, but I will to post incremental changes first to make it easier
to see the change.

The 3rd patch is now quite an ugly mess, but once it is folded in then
everything should become simpler because that special case will never be
added in the first place. So don't review it with a fine comb, all it
really does is remove that branch.

There were a couple of cases where I tried to make things a bit simpler
but less efficient in contended paths. It turns out some microbenchmarks
or pathological cases might be limited by this so I'm trying to find
improvements. In one case there is already a better primitive available,
in another case the end result allows a much bigger simplification so it
should be worth the more complicated asm primitive.

Thanks,
Nick

Nicholas Piggin (4):
  powerpc/qspinlock: Avoid cmpxchg pattern in lock stealing
  powerpc/qspinlock: Avoid cmpxchg style patterns in queue head locking
  powerpc/qspinlock: Remove !maybe_waiters special case queue head
    locking
  powerpc/qspinlock: add compile-time tuning adjustments

 arch/powerpc/include/asm/qspinlock.h |  61 ++++++-
 arch/powerpc/lib/qspinlock.c         | 237 ++++++++++++---------------
 2 files changed, 167 insertions(+), 131 deletions(-)

-- 
2.37.2