I would argue that this patch is not needed.

Reason:
The loop in /storage/innobase/sync/sync0sync.cc will never be optimized away by gcc because of the code we have in mutex_get_lock_word(mutex), which is part of the loop:

• mutex_get_lock_word() uses a memory barrier
• it references a volatile object

This should make the original code safe.

The other loop also references a volatile argument, so this is also safe.

I suggest we don't take this patch (unless someone can show us an assembler output where the loop is optimized away).

I assume it's ok to add the code to lower cpu priority as long as it's ifdef:ed and only used on power.
Still don't think we need the volatile("":::"memory") part.

Not that instead of using asm instructions directly, it's better to use proper macros:
(this is taken from the linux kernel)

#ifdef powerpc
/* Macros for adjusting thread priority (hardware multi-threading) */
#define HMT_very_low() asm volatile("or 31,31,31 # very low priority")
#define HMT_low() asm volatile("or 1,1,1 # low priority")
#define HMT_medium_low() asm volatile("or 6,6,6 # medium low priority")
#define HMT_medium() asm volatile("or 2,2,2 # medium priority")
#define HMT_medium_high() asm volatile("or 5,5,5 # medium high priority")
#define HMT_high() asm volatile("or 3,3,3 # high priority")
#else
#define HMTL_very_low()
...
#endif

We could add this to a new file include/cpu.h

Looks fine, just a few things for your consideration:

• year in copyrights should be 2014, not 2013
• powerpc is deprecated macro, we should use _powerpc_ instead, see:
  http://nadeausoftware.com/articles/2012/02/c_c_tip_how_detect_processor_type_using_compiler_predefined_macros
• InnoDB uses tabs for indentation (shouldn't we honor that?)
• why this change in rw_lock_s_lock_spin():

  - if (i == SYNC_SPIN_ROUNDS) {

  + if (lock->lock_word <= 0)