AMD Platforms May Experience ACPI and P-state CPU Errors

Asset ID:	1-72-1329381.1
Update Date:	2012-03-07
Keywords:

Solution Type Problem Resolution Sure

Solution 1329381.1 : AMD Platforms May Experience ACPI and P-state CPU Errors

Applies to:

Sun Fire X4600 M2 Server - Version: Not Applicable and later   [Release: N/A and later ]
Sun Fire X4500 Server - Version: Not Applicable and later    [Release: N/A and later]
Sun Fire X4600 Server - Version: Not Applicable and later    [Release: N/A and later]
Sun Blade X8420 Server Module - Version: Not Applicable and later    [Release: N/A and later]
Sun Fire X2100 Server - Version: Not Applicable and later    [Release: N/A and later]
Information in this document applies to any platform.

Symptoms

Customer may experience CPU error messages is/var/adm/messages that resemble the following:

May 22 16:27:24 HOST unix: [ID 388705 kern.warning] WARNING: cpu_acpi: error parsing _PSS for CPU 0
May 22 16:27:24 HOST unix: [ID 196753 kern.warning] WARNING: cpupm_init: processor 0: unable to initialize P-state support
May 22 16:27:30 HOST unix: [ID 388705 kern.warning] WARNING: cpu_acpi: error parsing _PSS for CPU 1
May 22 16:27:30 HOST unix: [ID 196753 kern.warning] WARNING: cpupm_init: processor 1: unable to initialize P-state support

These errors look like a CPU problem but may be seen across multiple CPU sockets and cores.

Changes

This problem may occur after a BIOS update or after a BIOS factory reset.
Clearing CMOS may also reveal this issue.

Cause

The error messages are due to the BIOS "PowerNow" feature being disabled in system BIOS.
If this feature is disabled, when the system boots the kernel will try to access the advanced power features of each CPU core and will output an error because the function is are disabled.

Some older dual core revision 'E' AMD CPU's have a design fault where they may produce memory errors whilst trying to power manage cores. Therefore in many late BIOS releases, the "PowerNow" features is disabled.
However, more modern revision 'F' AMD CPU's and above do not experience this problem and as such the BIOS options should be re-enabled to support full power management.

In most platforms, the BIOS default can be applied but in modular platforms such as the Sun Fire X4600 and X4600 M2, the CPU's can be interchanged with older models highlighting this issue.

To verify which revision CPU a customer system has:

Get the CPU info from the O/S e.g. psrinfo -vp in Solaris, cat /proc/cpu in Linux and msinfo32.exe in Windows will all give the family and stepping information needed to identify which model and revision of processor they have.

Example output:

psrinfo -vp

 The physical processor has 1 virtual processor (0)
   x86 (AuthenticAMD family 15 model 37 step 1 clock 2210 MHz)
         AMD Opteron(tm) Processor 848

You can get same info while system performs POST or from BIOS.

The model number tells if it's a rev E or rev F.

"AMD Opteron(tm) Processor 848"
or

"AMD Opteron Dual-Core 8216"

3 digit numbers like the 848 in the above output are rev E CPU's or below.
4 digit numbers like for example the 8216 are rev F CPU's or above.

Solution

Check the status of the "PowerNow" feature in BIOS by navigating to:

BIOS --> Advanced --> AMD PowerNow configuration

If the platform using revision 'E' CPU's then leave this feature disabled and disregard any boot messages.

The location of this BIOS option may differ depending on platform and BIOS revision.

Attachments

This solution has no attachment