Re: Synchronizing Bit operations V2

From: Nick Piggin
Date: Fri Mar 31 2006 - 21:54:01 EST


Christoph Lameter wrote:
On Fri, 31 Mar 2006, Nick Piggin wrote:

You acknowledge that you have to fix ia64 to match current semantics
first, right?


Right. I believe I have done so by making both smb_mb_* full barriers.

All bitop and atomic test_and_set, inc_return, etc etc (ie. everything
that modifies the operand and returns something) needs to be a full
barrier before and after too.

Now people seem to be worried about the performance impact that will
have, so I simply suggest that adding two or three new macros for the
important cases to give you a 90% solution.


We could transition some key locations of core code to use _mode bitops
if there are performance problems.


I think Documentation/atomic_ops.txt isn't bad. smp_mb__* really
is a smp_mb, which can be optimised sometimes.


Ok. Then we are on the same page and the solution I presented may be acceptable. I have a new rev here that changes the naming a bit but I think we are okay right?

Not sure, to be honest. I think it is probably something which needs
input from all the other arch people, and Linus, if you intend to use
it to introduce new types of barriers.

--
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com -
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/