EDAC i5000 NON-FATAL ERRORs

[Roger Heflin]

Jack Howarth wrote:
>    I noticed that after upgrading the kernel on a Fedora 7 x86_64
> box is the latest kernel (the box hadn't been rebooted for some months)
> that I am now seeing the following in my messages log...
> 
> May 25 04:30:56 fourier kernel: EDAC i5000 MC0: NON-FATAL ERRORS Found!!! 1st NON-FATAL Err Reg= 0x10000
> May 25 04:30:56 fourier kernel: EDAC MC0: CE row 1, channel 0, label "": (Branch=0 DRAM-Bank=3 RDWR=Read RAS=14339 CAS=672, CE Err=0x10000)
> 
> These messages always occur on DRAM-Bank 3 and are always NON-FATAL. The messages appear roughly once
> an hour and are rarely repeated immediately. This machine contains a Tyan Tempest i5000XL motherboard
> with ECC memory installed. Does anyone know if the recent kernels had any changes which made these
> motherboard chipset report ECC memory errors which were not reported in the past? I haven't been
> able to reproduce these errors in memtest86 yet with or without ECC. So I am wondering if I am seeing
> noise from the EDAC driver or real ECC errors. Thanks in advance for any insights on this.
>            Jack
> 

Well, until recently the module that supports the i5000 chipset was probably 
*NOT* in the kernel, so it was probably added recently or before it was not 
loading at all.

You could check the older kernels and see if it had the proper i5000 modules 
being loaded.

If memtest86 is new enough and can see the ecc monitoring hardware of the i5000 
you should be able to duplicate it, if the memtest86 is older and does not 
properly detect the i5000 hardware then any correctable ECC errors will be 
silently corrected by the hardware and memtest86 will be none the wiser.

There is an edac list someplace and someone over there can probably interpret 
the error in more detail.

If it was noise, I would have expected the bank to move around, if you have more 
than one dimm you could try moving the dimms around and see if the error 
location changes, there are also some stuff for edac in /sys that gives more 
details and has running counters of the errors since the machine has been up.


                              Roger