[ntp:questions] Re: TSC feacture

Piotr Trojanek ptrojane at mion.elka.pw.edu.pl
Mon Sep 1 06:02:58 UTC 2003


In article <birecu$kmc$1 at nntp.webmaster.com>, David Schwartz wrote:
>
>"Xuehuan Li" <libryant at hotmail.com> wrote in message
>news:mailman.20.1062119286.1841.questions at ntp.org...
>
>> my ntpd runs on a PC104 CPU card. the CPU on the card is Cyric 486DX, and
>> the kernel i compiled Do config as 486. but when i run the kernel on the
>> card, it halted and printed
>> kernel panic: kernel compiled for Pentium+, require TSC feature!
>> does it mean the PPS patch can not run on the 486DX? BTW, in the kernel i
>> patch another patch about the M-System's DOC driver(Disk On Chip).
>
>    Without a TSC, you will get tremendously reduced accuracy. So much so
>that there's no reason to run the PPS patch. Just run without the patch and
>you'll get all the accuracy the hardware is capable of. You cannot build a
>real stratum one time server with a CPU that does not have a TSC. If you're
>operating on the 'net, I'd advise you to fudge your reference clock stratum
>to 2 at most.

There is a hack -- some drivers (ie. Motorola OnCore) REQUIRES PPSapi
to work, so there is a reason to patch Linux with PPSkit even for
machines that lacks TSC. Without PPS they would be almost useless for
NTP timestamps.

To say more -- I have i686 board, with TSC present, but it can only
boot linux compiled for 486!, if other CPU was compiled for, it just rebots
without any word:( FreeBSD, QNX and others runs just fine:)

I used to run Linux-2.4.19-NANO on PC104 486DX, with PPSkit patch, DOC
also, with OnCore driver -- it works, but stability and jitter were rather
bad.

On the other side, FreeBSD-4.7 on PC104 386 board, DOC also, with OnCore
driver works really fine for me (unit was dedicated to run ntpd, with almost
none other activity or system load).

I have done tm_basic measurment with ECOS (http://sources.redhat.com/ecos)
on 486 at 100 MHz. Reading hardware timer in the way that linux does takes
15 clock ticks, which is about 12us.

-- 
Piotr Trojanek



More information about the questions mailing list