[PATCH 6/9] Use proper membar instructions on SPARC

[David Miller]

From: Matt Turner <mattst88 at gmail.com>
Date: Thu, 13 Aug 2009 16:17:23 -0400

>>> What you want to check for is __sparc_v9__
>>
>> That's not going to be CPP defined either, unfortunately.
>>
>> I really think, based upon this, that the hard-coded opcodes have to
>> stay.  It's the only way to cover all cases.
> 
> If that's the case, this patch can be ignored.
> 
> Does the hard-coded opcode account for the branch misprediction
> errata? Does it need to?

No, you're need to add that.

Does it need to?  Well, I certainly wouldn't want to have to
track down a bug caused by the branch not being there.  ;-)

BTW, you'd need to use hard-coded opcodes for the branch too.

> I assume it's a write barrier, and if so I'd prefer to change the name
> of the macro from barrier() to write_mem_barrier() to match with the
> others.

I'm personally ambivalent about this.