🚀 go-pugleaf

In <5clq15$i1l$1@newz.oit.unc.edu> mitchell@auger.radonc.unc.edu (Mike Mitchell) writes:

>     With increasing frequency, I'm seeing ecc memory errors on my old
>SPARCserver 470 and SPARCserver 490.  These memory errors show up
>in the messages file as:

>    vmunix: mem0: soft ecc addr 1738c80 syn 5b<S16,S4,S2,S0,SX> bit 38 U2026
>    vmunix: resetting ecc handling

>When I have similar messages on a small SparcStation, I just reseat the
>memory and all is well for a while.  On the SPARCservers, however, there
>are several cards holding memory.  Is there anything in the above error
>messages that would tell me which memory card to pull and investigate?

Ok, on the 4/4xx series (like the 4/2xx, 3/4xx and 3/2xx), which use ECC
(Error-Correcting Control) RAM boards, the RAM is managed by dedicated
controllers on each VME card. These controllers handle and report on all
RAM errors, and the messages you're seeing are the result of one of the
DRAM controllers recovering from a soft (ie. generally non-critical)
error. 

Each RAM board in your machine will have an array of LEDS on the card
panel to indicate it's status, as well as a small jumper block used to set
the ID number of the board. Here's a description of what the LED's mean: 

        There are five LEDs on the upper rear edge of the board. In
        normal operation, only the two green LEDs should be lit.

        UE      Uncorrectable error (when lit)          RED

        CE      Correctable error (when lit)            YELLOW

        DIS     CPU access disabled (when lit)          YELLOW

        CPU     CPU accessing memory                    GREEN
          This LED flickers because it is only lit when the CPU is
          actually accessing the memory on the board. If the LED is not
          flickering, that simply means you have more memory than you
          need at the moment -- the board is not being accessed
          significantly.

        REF     Refresh OK (when lit)                   GREEN
          If this LED is not lit, refresh has failed and the board
          should be repaired or replaced.

If a soft error is detected, the 'CE' LED should be lit, but it's only 
when the 'UE' LED is lit, or when the 'REF' LED is *not* lit that you 
will need to take action to replace the board.

Using the information in the syslog messages, and looking at the LED's to 
determine which board has given you an error, you can, if you have the 
resources, locate the RAM chip mentioned and replace it if it gives 
continual errors.

The desktop Sparcstations use a completely different architecture for RAM.
The error correction ability is basically discarded, and SIMM's are used
instead which can only have simple parity error detection (not used by
most, if not all non-ECC-RAM hardware these days), but this is ok for
99.99999 percent of cases because the reliability of RAM devices has
increased, or so we're led to believe. 

If you'd like to find out all there is to know about older Sun hardware, 
check out James Birdsall's Sun Hardware Reference, which you can FTP from 
ftp.picarefy.com in '/pub/Sun-Hardware-Ref'.

Regards,

Craig.

-- 
        Craig Dewick - craig@zeta.org.au - http://www.zeta.org.au/~craig  
            Professional Train Driver. Located in Sydney, Australia.
          Sun Ripened Kernels - http://www.zeta.org.au/~craig/srk.html
                Surplus and pre-loved Sun Microsystems equipment.