Very Computer

  I have been using 1.3.25 for quite a while now, cross compiling it
for a number of friends who have 2940 (regular ones, all work fine).
Most machines were P5-100/90 and ran perfectly with reasonalble
meaningless BogoMIPS of 39.94 or thereabouts. The same BogoMIPS number
came up on 1.2.3, 1.2.8 and 1.2.13 for tests. /proc/cpuinfo report
Pentium 90/100 as CPU type.

  Then, as a test I switched to 1.3.30 (and then to .37 and .38). For
some reason, completely unclear to me, the BogoMIPS fell to 39.73 and
/proc/cpuinfo changed to Pentium 75+. Has anyone seen this? And, of
so, where does it come from?

--
---MAV

:   I have been using 1.3.25 for quite a while now, cross compiling it
: for a number of friends who have 2940 (regular ones, all work fine).
: Most machines were P5-100/90 and ran perfectly with reasonalble
: meaningless BogoMIPS of 39.94 or thereabouts. The same BogoMIPS number
: came up on 1.2.3, 1.2.8 and 1.2.13 for tests. /proc/cpuinfo report
: Pentium 90/100 as CPU type.

:   Then, as a test I switched to 1.3.30 (and then to .37 and .38). For
: some reason, completely unclear to me, the BogoMIPS fell to 39.73 and
: /proc/cpuinfo changed to Pentium 75+. Has anyone seen this? And, of
: so, where does it come from?

More recent kernels use a binary approximation to determine the BogoMIPS
value (check out init/main.c). This has the advantage of speeding up the
boot process (by a few seconds :-), but it does lose slightly in accuracy
(my P90 value dropped from 36.08 to 36.04). I guess that the slight
deviation is acceptable to those routines that use it...

As for your /proc/cpuinfo question, have a look at
arch/i386/kernel/setup.c; I would guess that the CPU mask only contains
sufficient information to differentiate between either 60/66 Pentiums or
others (75+).

Hope this has helped.

Julian.
--