Re: Afterburner and Linux

From: Michael Schmitz <schmitzm_at_uclink4.berkeley.edu>
Date: Fri, 5 Jun 1998 14:51:47 -0700

At 10:46 PM +0200 6/5/98, Thomas O. Kruse wrote:
>>The other basic problem that annoys SCSI is trying to DMA in and out of
>>FastRAM without going through an STRam window.
>Is that true for a kernel loaded to ST-RAM?

Regardless if the kernel is loaded to ST-RAM or FastRAM, it's BS. Linux
recognizes buffers above 16 MB as useless for DMA, and falls back to
copying
these to a ST-RAM 'dribble buffer'. Otherwise, none of the FastRAM solutions
for Linux would have worked. Credits to Guenther Kelleter.

The only problem the AB040 poses on top of that is not doing bus snooping
(plus maybe other weirdness we didn't discover yet). But the ball is back in
your court now - I hope I could prove that Linux works on the AB040 when you
don't cache the RAM. What are your results with that kernel ???

>Any idea, if this (and the cache of the hardware-registers) could be fixed
>in a Linux kernel? Perhaps the other memory area could be cached again and
>the Afterburner would be a quite fast working machine....

The 'don't DMA to FastRAM' has been fixed, long ago. It's done in the same
function that decides if DMA transfer should be used for the current
command, and calculates the number of bytes to transfer via DMA (the Falcon
has some
other bugs WRT DMA).
Hardware registers were always noncacheable-serialized. Maybe we can set
FastRAM cacheable again; that depends if the kernel works reliable on your
system.
Let me know, and I can build a new one with only ST-RAM excluded from caching.
Please also try running the kernel in FastRAM, just for reference.

        Michael
Received on lø. juni 06 1998 - 00:25:00 CEST