Re: cache control primitives

From: Andrew Valencia <vandys_at_nospam.org>
Date: Mon Nov 22 1993 - 10:35:12 PST

[larz@world.std.com (Mike A Larson) writes:]

>[cache control for DMA drivers not needed...]
> 1) the VSTa kernel handles this automatically
> 2) not needed for the 386 based PC's.

The latter. All PC's I have ever worked on run with physical cache,
write-through, which flushes itself when it sees DMA to its addresses.
Therefore, there is no need to even abstract the concept; when I do a
port to a machine with a non-snooping or virtual cache, I will have to
add a port-specific interface to provide such control.

A virtual cache will add the additional unpleasantness of having the
hat_* layer in mach/ have to avoid aliasing. I believe all the hooks
are in place for a hat_* layer to do this, but again (due to i386/ISA
architecture) I've never had to actually implement this.

                                                Regards,
                                                Andy
Received on Mon Nov 22 10:41:53 1993

This archive was generated by hypermail 2.1.8 : Wed Sep 21 2005 - 21:01:53 PDT