Inconsistent CPU speed reported on RX6600

[Pete Des Moulins]

All,

I am attempting to run a discovery-type script to centrally report on all of the HP-UX servers in my estate. (HP-UX 11.00 - 11.23 for both PA-RISC and IA64 architectures)

However, I have noted an inconsistency with all of my RX6600's whereby the CPU speed being reported is incorrect.
The output is the same from all of my RX6600 servers and also (my only) RX3600 server, suggesting a common anomaly in reporting, rather than a rogue system.

Can anyone clarify the cause of this behaviour please - because I have other IA64 RX4640 () servers which report correctly using ADB.

I appreciate that the RX4640's are running processor family 31 = Madison CPU's so is this reporting anomaly restricted to the Montecito chips?


All output below relates to an RX6600 server:
# Output from ADB:
#echo "itick_per_usec/D" |adb -o /stand/vmunix /dev/kmem
# Output:
itick_per_usec:
itick_per_usec: 399

NB: I am unable to get any CPU info out of stm for comparison:
STM output:
cstm>selclass qualifier cpu
cstm>info
^-- (Information) is currently disabled. --


HOWEVER, the CPU speed (as reported by machinfo and also from EFI 'System Information') shows that my CPU's are 1.6 GHz (Montecito) processors

EFI System Info:
================

PROCESSOR MODULE INFORMATION

# of L3 L4 Family/
CPU Logical Cache Cache Model Processor
Module CPUs Speed Size Size (hex.) Rev State
------ ------- -------- ------ ------ ------- --- ------------
0 2 1.6 GHz 12 MB None 20/00 C2 Active
1 2 1.6 GHz 12 MB None 20/00 C2 Active

CPU threads are turned off.


machinfo output:
================
CPU info:
Number of CPUs = 4
Clock speed = 1595 MHz
Bus speed = 532 MT/s
CPUID registers
vendor information = "GenuineIntel"
processor serial number = 0x0000000000000000
processor version info = 0x0000000020000704
architecture revision: 0
processor family: 32 Intel(R) Itanium 2 9000 series
processor model: 0 Intel(R) Itanium 2 9000 series
processor revision: 7 Stepping C2
largest CPUID reg: 4
processor capabilities = 0x0000000000000005
implements long branch: 1
implements 16-byte atomic operations: 1

[Pete Des Moulins]

OK - I have just found a previous forum question that covers the same issue that I posed:

Link: http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1232229

[Pete Des Moulins]

Closed - see previous update

[Duncan Edmonstone]

Pete,

Actually itick_per_usec *never* showed the processors core frequency, it always showed the processors itimer frequency. It's just that up until the Montecito processors these values were always the same. machinfo is the correct method for getting this info.

This just goes to show that you should never assume that anything that involves poking around in the kernel is going to have consistent behaviour between systems. (yes I *know* there are plenty of ITRC docs which indicate this as a method for finding out info - doesn't make it right though!) The problem with all this "echo whatever/D |adb -o /stand/vmunix /dev/kmem" stuff is that we reserve the right to change anything in the kernel without telling you anyway...

HTH

Duncan