On 31/08/16 15:28, Peter Zijlstra wrote:

On Wed, Aug 31, 2016 at 02:36:26PM +0100, Matt Redfearn wrote:

quoted

The code previously had 0x10 as a magic number, this patch just replaces
that with a #defined name. The value is documented in the MIPS64 instruction
set manual, https://imgtec.com/?do-download=4302, table 6.5.

This sync type has been standard since MIPSr2. That document also states
that "If an implementation does not use one of these non-zero values to
define a different synchronization behavior, then that non-zero value of
stype must act the same as stype zero completion barrier." As such,
stype_ordering can always be set to this sync type rather than setting it
only for certain CPUs.

Hi Peter,

Right. We all had a bunch of fun trying to decode that manual a while
back, and IIRC were left with a bunch of questions on what it all meant
in 3+ CPU scenarios.

Yes, I remember that fun....

In anycase, not sure why I was Cc'ed to this patch, but in general I

Patman decided to CC you as you've touched arch/mips/include/barrier.h I 
suppose.

have low confidence in barrier patches that lack lots of detail. And the
code in question has woefully inadequate comments:

                 /* Ordering barrier */
                 uasm_i_sync(&p, stype_ordering);

Order what against what and why? Is my first question. A comment really
should explain.

Fair enough - we'll put something together to improve the comments.

In any case, you've removed the only (runtime) assignment to the
variable, it can become 'const'.

True enough.

Thanks,
Matt