AMD Opteron Power Now X86_64 + Red Hat 5.2 = Time Drift

[Roger Lewinton]

I have a pair of BL460c's with fresh installs of Red Hat 5.2 Both of these systems are suffering from extremely fast system clock drift.

System info:
CPU = Quad-Core AMD Opteron(tm) Processor 8354
Kernel = 2.6.18-92.el5

I have queried the hwclock (hwclock --show) and that is keeping time correctly, but the system cock (date) is running approx twice as fast.

Using ntp to correct the system clock fails. BUT i can use ntpdate set time/date correctly. If then left to ntpd to correct, it rejects any ntp time source and the system clock continues to drift as before.

If i boot the system with the kernel parameter notsc, everything works OK system time and hwclocks match and ntp will synch with and use any ntp time source I provide.
See HP case:
http://h18004.www1.hp.com/products/servers/linux/powernow-notes.html
http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01075682〈=en&cc=us&taskId=101&prodSeriesId=428936&prodTypeId=15351

According to which Red Hat 5.2 is not subject to this problem as it does not use TSC as a time source.

So what am I doing wrong?

[John Daley]

Roger,

Sure about that model number? AMD-based ProLiant model names usually end in "5". Are the systems G5s?

Anyway, did you also happen to notice if BogoMips was half what it is on working systems of the same model? What's the cpu speed value in /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq? Are there errors when PowerNow! loads?

You're right, the customer advisory says the issue fixed by "notsc" won't happen on RHEL 5. Also says it won't happen on Generation 5 ProLiants. Have you opened a case with HP yet?

[John Daley]

Roger,

Check Redhat Bug here:
https://bugzilla.redhat.com/show_bug.cgi?id=483300