Messages in this thread | | | Date | Thu, 19 Jan 2012 06:59:23 -0800 | From | "H. Peter Anvin" <> | Subject | Re: [PATCH 2/2] ix86: atomic64 assembly improvements |
| |
On 01/19/2012 06:50 AM, Jan Beulich wrote: >> >> It doesn't have anything to do with caches on or off. > > How does it not? If any part of the bus topology is only 32 bits wide, > a 64-bit read or write simply can't be executed atomically without > asserting LOCK#. >
Furthermore, if you're going down that rathole then we can't even trust MOV (nor, for that matter, can you trust LOCK in a lot of circumstances.) In short, you need to be extremely careful about what you do to uncached memory *at all* (and you need to know exactly what is behind this bus) but pessimizing these kinds of construct for that reason is wrong in the extreme.
-hpa
| |