16 mutext lock/unlocks per query?

Paul A Vixie paul at vix.com
Wed Dec 5 22:04:07 UTC 2001


> > so the lost processor time is all function call overhead unless your
> > TSI has to empty the cache pipelines rather than snooping them.
> 
> I'm reasonably certain that the instruction does not have to flush the
> caches.

so am i, but the writes that are in the pipe headed out of the cpu into
the memory subsystem might have to be stalled to make this instruction
work.  iirc, on some cpu's the addresses of these writes are CAM-accessable
from within the alu which means TSI won't stall the pipeline. 


More information about the bind-workers mailing list