More about 2.4.3 timer problems

[Eric Gillespie]

First off, CC: back to me, as my machine can't handle an estimated 200
messages a day for me to sign up to the list 8-( - Anyway..

I updated my kernel to 2.4.3 when the patch was released. As I said earlier, I
noticed my timer losing a few seconds over the space of a couple of
hours.  Well, I have since found out that the amount of time lost is
proportional to the load on the machine: the heavier the load, the more time
lost.

I found this out while I was compiling a kernel, incidentally:

kernel-2.4.3 (and I am trying this without the TSC enabled, but it hasn't
made any difference that I've noticed), other details as before.
I really DON'T want to be running a cron job to re-read the CMOS
clock every minute or so - that's a patch, not a fix.

Output from "time make -j 3 bzImage"

VESA fb mode 1280x1024x16, clock lost 1m 35s in time
996.52 user 166.82 system 23:19.40 elapsed 83% CPU 
(0 avgtext+0 avgdata 0 maxresident)k 
0 inputs+0 outputs (471222 major+583276 minor) pagefaults 0swaps

Same command, fbmode 640x480x24, clock lost 19s in time
751.42 user 89.47 system 18:40.79 elapsed 75% CPU 
(0 avgtext+0 avgdata 0 maxresident)k
0inputs+0outputs (468926 major+578877 minor) pagefaults 0 swaps

Same command, normal (non-fb mode) lost no time off clock.
991.13 user 160.44 system 22:38.15 elapsed 84% CPU 
(0 avgtext+0 avgdata 0 maxresident)k 0 inputs+0 outputs 
(468051 major+577233 minor) pagefaults 0 swaps

Hope this helps out, and by the way, thank you to Mark Hahn & Gerhard Mack for
their suggestions.

-- 
 /|   _,.:*^*:.,   |\           Cheers from the Viking family, 
| |_/'  viking@ `\_| |            including Pippin, our cat
|    flying-brick    | $FunnyMail  Bilbo   : Now far ahead the Road has gone,
 \_.caverock.net.nz_/     5.39    in LOTR  : Let others follow it who can!