sparc64: revert pause instruction patch for atomic backoff and cpu_relax()

sparc64: revert pause instruction patch for atomic backoff and cpu_relax()

This patch reverts the commit e9b9eb59ffcdee09ec96b040f85c919618f4043e
("sparc64: Use pause instruction when available").

This all started with our TPCC results on UEK4. During T7 testing, the TPCC results
were much lower compared to UEK2. The atomic calls like atomic_add and atomic_sub
were showing top on perf results. Karl found out that this was caused by
the upstream commit e9b9eb59ffcdee09ec96b040f85c919618f4043e
(sparc64: Use pause instruction when available). After reverting this commit on UEK4,
the TPCC numbers were back to UEK2 level. However, things changed after Atish's
scheduler fixes on UEK4. The TPCC numbers improved and the upstream commit
(sparc64: Use pause instruction when available) did not seem make any difference.
So, Karl's "revert pause instruction" patch was removed from UEK4.

Now again with T8 testing, we are seeing the same old behaviour. The atomic calls
like atomic_add and atomic_sub are showing top on perf results. After trying with
Karl's patch(revert pause instruction patch for atomic backoff) the TPCC numbers
improved(about %25 better than T7) and atomic calls are not showing on top in perf.

So, we are adding this patch back again. This is a temporary fix. Long term solution
is still in the discussion. The original patch is from Karl.
http://ca-git.us.oracle.com/?p=linux-uek-sparc.git;a=commit;h=f214eebf2223d23a2b1499be5b54719bdd7651e3

All the credit should go to Karl. Rebased it on latest sparc tree.

Orabug: 26306832

Signed-off-by: Karl Volz <karl.volz@Oracle.com>
Reviewed-by: Atish Patra <atish.patra@oracle.com>
Signed-off-by: Henry Willard <henry.willard@oracle.com>
Signed-off-by: Babu Moger <babu.moger@oracle.com>
Reviewed-by: Karl Volz <karl.volz@Oracle.com>
Signed-off-by: Allen Pais <allen.pais@oracle.com>