Re: Cyrix 6x86 - wierd problem

Mike Jagdis (
Mon, 16 Dec 1996 16:12:32 +0000 (GMT/BST)

On Fri, 13 Dec 1996, Nigel Metheringham wrote:

> However if run from the rc files on boot, the script consistantly manages
> to produce a zero length output (which for some reason is not picked up by
> the supervisory script).
> One of the other machines has shown up odd perl problems - script fails to
> compile, but then works OK.

With VSPM disabled the immediate candidates appear to be DTE_EN (directory
table entry caching), WT_ALLOC (allocate cache lines on write misses
as well as read misses), ~SLOP (allow loops to be optimised) and
the use of branch prediction for far targets.

> stepping : 1 rev 5

What are the revisions on the other boxes?

> bogomips : 52.33

This is way too low. This seems to suggest that __delay() is
being inordinately slow. This should be running straight out
of the 256 byte icache so looks like a chip problem. Is it
possible that the chip is overheating? Once the system is
running it will sleep when idle so will tend to keep its
temperature down however under CPU intensive activity or
during the boot its temperature will tend to skyrocket if
there isn't sufficient cooling. I seem to remember that
the Cyrix uses clock skipping if it starts to cook. Perhaps
interestingly 52.33 is about half what you get if you build
the kernel with 486 alignment optimizations rather than 586.
This could be an indication that chip has throttled back to
half speed and is powering down on-chip caches.

The other possibility is that the chip has problems with
branch prediction but this seems unlikely in the absence
of other bug reports.


|  Mike Jagdis                  |  Internet:   |
|  Roan Technology Ltd.         |                                      |
|  54A Peach Street, Wokingham  |  Telephone:  +44 118 989 0403        |
|  RG40 1XG, ENGLAND            |  Fax:        +44 118 989 1195        |