Roland Dreier wrote:
I don't understand: your results below don't seem to have good results
without X86_PAT? In any case I don't see how you can get the lowest
latency (from WC mapping to userspace) without X86_PAT, since
pgprot_writecombine() will fall back to pgprot_noncached() is X86_PAT=n.
Again, the facts are that on my latest round of testing I got a latency
of 1.4us with and without X86_PAT which went down to 1.1us when applying
your patch on top of a X86_PAT config. Even if I'm doing something
wrong, I still want to see that we understand each other... so please
let me know if you still don't understand me.
Or.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
