24 SPARC Enterprise T5440 Server Service Manual • June 2011
■“Detecting Faults Using POST” on page 46
■SPARC Enterprise T5440 Server Installation and Setup Guide
■SPARC Enterprise T5440 Server Administration Guide
Memory Fault Handling Overview
A variety of features plays a role in how the memory subsystem is configured and
how memory faults are handled. Understanding the underlying features helps you
identify and repair memory problems. This section describes how the server deals
with memory faults.
Note – For memory configuration information, see “FB-DIMM Configuration” on
page 113.
The server uses advanced ECC technology that corrects up to 4-bits in error on nibble
boundaries, as long as the bits are all in the same DRAM. On 4 GB FB-DIMMs, if a
DRAM fails, the DIMM continues to function.
The following server features independently manage memory faults:
■POST – Based on ILOM configuration variables, POST runs when the server is
powered on.
For correctable memory errors (CEs), POST forwards the error to the Predictive
Self-Healing (PSH) daemon for error handling. If an uncorrectable memory fault is
detected, POST displays the fault with the device name of the faulty FB-DIMMs,
and logs the fault. POST then disables the faulty FB-DIMMs. Depending on the
memory configuration and the location of the faulty FB-DIMM, POST disables half
of physical memory in the system, or half the physical memory and half the
processor threads. When this offlining process occurs in normal operation, you
must replace the faulty FB-DIMMs based on the fault message and enable the
disabled FB-DIMMs with the ILOM command set device component_state=
enabled where device is the name of the FB-DIMM being enabled (for example,
set /SYS/MB/CPU0/CMP0/BR0/CH0/D0 component_state=enabled).
■Predictive Self-Healing (PSH) technology – A feature of the Oracle Solaris OS, PSH
uses the Fault Manager daemon (fmd) to watch for various kinds of faults. When
a fault occurs, the fault is assigned a unique fault ID (UUID), and logged. PSH
reports the fault and identifies the locations of the faulty FB-DIMMs.
If you suspect that the server has a memory problem, follow the flowchart (see
FIGURE: Diagnostic Flowchart on page 14). Run the ILOM show faulty command.
The show faulty command lists memory faults and lists the specific FB-DIMMs
that are associated with the fault.