Re: Re: An interesting intermittent problem

On Wed, 10 Mar 2004, Brad Stockdale wrote:

>     Well, this has just not been my week for using eepro100's... I spent 
> several days earlier this week tracking down an intermittent problem with 
> one of my Penguin Computing servers which uses an Intel eepro100...

Note that since Scyld is now "a Penguin Computing company", Penguin
Support has access to the Scyld bug tracking system.  You can have them
log a bug directly into my category.

> I believe it's problem turned out to be sleep mode on the ethernet
> chipset...  Turned it off, and it seems to be ok now...

That's very likely the problem.  Intel didn't come out and say "Sleep
mode is borken, 'orribly, 'orribly borken" thus many vendors shipped the
default configuration.

> Basically, this server works fine... For a while. Then all of the 
> sudden it will cease talking to anything else EXCEPT computers on it's 
> local subnet/segment. If I logon to a box connected to the same switch as 
> the server in question and ping it with an ICMP packet, all of the sudden 
> everything is fine and dandy again... For a while.

Hmmm, this my be a power-management problem.  Do you have all save-power
modes in the BIOS disabled?

Which driver version are you using?  My unmodified releases have a Rx
monitor that resets the receiver when no packets have been received for
a few seconds.

>     It seems to occur when the box is sitting idle -- the longer it is not 
> receiving traffic, the more of a chance that it will cease to respond to 
> network traffic outside of the local subnet...

That does sound like a power management problem.

>     I used eepro100-diag to verify that sleep mode was not enabled on this 
> server -- it was not. It also reports all the checksums as being correct... 
> Here's the output:
> 
> eepro100-diag.c:v2.12 4/15/2003 Donald Becker ([email protected])
>   http://www.scyld.com/diag/index.html
> Index #1: Found a Intel i82557/8/9 EtherExpressPro100 adapter at 0x2800.
> i82557 chip registers at 0x2800:
>    00000050 379b6578 00000000 00080002 183f0000 000005f0
>    No interrupt sources are pending.
>     The transmit unit state is 'Suspended'.
>     The receive unit state is 'Ready'.
>    This status is normal for an activated but idle interface.
> EEPROM contents, size 64x16:
>      00: d000 89b7 bf46 0c03 0003 0201 4701 0000  ____F________G__
>    0x08: 0000 0000 40a0 3000 8086 0000 0000 0000  _____@_0________
>        ...
>    0x38: 0000 0000 0000 0000 0000 0000 0000 5b8f  _______________[
>   The EEPROM checksum is correct.
> Intel EtherExpress Pro 10/100 EEPROM contents:
>    Station address 00:D0:B7:89:46:BF.
>    Board assembly 000000-000, Physical connectors present: RJ45
>    Primary interface chip i82555 PHY #1.

Yup, looks normal.

>     I have tried both the Becker driver, and the Intel driver with the same 
> result.

Ahh, useful info.  The Intel e100 driver is completely different
different that my driver, and has a different set of bug work-arounds.
That points to something else in the system.

-- 
Donald Becker				[email protected]
Scyld Computing Corporation		http://www.scyld.com
914 Bay Ridge Road, Suite 220		Scyld Beowulf cluster systems
Annapolis MD 21403			410-990-9993