Date: Thu, 29 May 1997 00:24:06 +0200 From: Stefan Esser <se@FreeBSD.ORG> To: Satoshi Asami <asami@vader.cs.berkeley.edu> Cc: kato@eclogite.eps.nagoya-u.ac.jp, roberto@keltia.freenix.fr, hardware@FreeBSD.ORG, Stefan Esser <se@FreeBSD.ORG> Subject: Re: Intel Pentium II released Message-ID: <19970529002406.34366@x14.mi.uni-koeln.de> In-Reply-To: <199705282105.OAA06447@vader.cs.berkeley.edu>; from Satoshi Asami on Wed, May 28, 1997 at 02:05:21PM -0700 References: <199705281354.WAA19773@gneiss.eps.nagoya-u.ac.jp> <199705282105.OAA06447@vader.cs.berkeley.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
On May 28, Satoshi Asami <asami@vader.cs.berkeley.edu> wrote:
> * > Could it that the motherboard does not support all of K6's new
> * > features? (I heard you need a new BIOS for that if you have an older
> * > version of some motherboards.)
> *
> * Could someone explain how to enable K6's `new feature'?
> The AMD-K6(TM) processor, Socket 7 bus compatible by design, enables
> PC manufacturers and resellers to deliver affordable, top-to-bottom PC
> solutions. For existing Socket 7 motherboards, certain modifications
> may be necessary to ensure proper voltage settings and BIOS support."
> ^^^^^^^^^^^^
> I'm not sure what this means.
Guess this is the write-allocate feature of the primary cache,
which is off by default. There is an additional bit in the TLB,
that indicates a read on some page could be cached. The CPU then
knows that it may cache writes to that page, even if the line is
not currently in cache (-> write-allocate).
There is a base/limit register that allows to specify some address
region (normally system D-RAM) as cacheable, additionally. This
lets all writes in that area be cached.
Both features are off, by default. (There was another feature like
the first one above, but based on a hint found in the cache line,
if I remember correctly. Something like, write-allocate, if the
line had been valid and cacheable, even if it was flushed because
of a bus snoop and is not valid right now ...)
These features can be enabled by a knowledgable BIOS, or from the
OS. I have the docs at home, the details are all in the PDF data
book found on the AMD web site ...
Regards, STefan
PS: Write-allocate should of course make a big difference for block
copies, which would explain the very different results reported.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970529002406.34366>