[fm-discuss] many fmd errors on T2000

[Mike Gerdts]

On 6/20/06, Gavin Maltby <Gavin.Maltby at sun.com> wrote:
> My pcie-enabled colleague says:
>
>         These are correctable errors on the PCIe link (receiver errors and Bad TLPs).
>         Sounds like a bad link. The transactions are being retried successfully (for now).
>
>         Actually the spec specifically says that if the Physical Layer detects a
>         receiver error, then the Link Layer must not also report the error as a bad TLP,
>         so there looks to be a chip error handling bug too.
>
> All those ereports should have resulted in some diagnosis.  What does
> 'fmadm faulty' report, and what does 'fmdump -v' show?

Could it be related to this?

http://sunsolve.sun.com/search/document.do?assetkey=1-26-102099-1

To avoid the described issues, the following mandatory patches must be applied:

    * 118822-23 or later
    * 119578-10 or later
    * 121236-01 or later
    * 121265-01 or later
    * 119981-05 or later
    * 120824-03 or later
    * 120849-02 or later
    * 118918-09 or later

along with the following entries that must be set in the "/etc/system" file:

    set segkmem_lpsize=0x400000
    set pcie:pcie_aer_ce_mask=0x1

Presumably the patches aren't the issue, but the /etc/system entries could be.

Mike

-- 
Mike Gerdts
http://mgerdts.blogspot.com/