_IO_IS_ISA question

Paul Mackerras paulus at samba.org
Fri Mar 26 08:22:09 EST 2004


linas at austin.ibm.com writes:

> I take back what I took back and put forth my initial argument:  isync does
> do what I thought it did.  It waits for completion (which is what we want)
> but it also causes a refetch (which we don't want).  The refetch can be
> something like 200 or 300 clock cycles on power5 I think (antonb knows for
> sure) which is one hell of a price to pay.

POWER4 and later cpus use a scoreboard to record whether any changes
are made that would cause a refetch to be necessary, and then only do
the refetch on isync if necessary.

Paul.

** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/





More information about the Linuxppc64-dev mailing list