Way back in November, in 
<http://sourceforge.net/mailarchive/forum.php?thread_name=87k4kfq1at.fsf%40spindle.srvr.nix&forum_name=e1000-devel>,
I reported a problem with the 82754 in one of my machines freezing up at
random. This problem continues in 2.6.37, and bisection has still failed
because the fault is so intermittent (averaging three days apart and
sometimes taking as long as a week to freeze up, with many registers suddenly
reset to 0xff: but sometimes it freezes in only half an hour).

I moaned about it in an LWN thread as well: <http://lwn.net/Articles/416758/>
and hmh suggested I come here, but I decided to hold off until I knew a
bit more. Since then, I've been able to characterize it a bit. (All the
conclusions below are tentative: perhaps I was just lucky in some cases
and the fault happened not to kick in before I tried something else.)

It happens with both the in-kernel and out-of-tree drivers in 2.6.36 and
above, but does not affect 2.6.35 with either driver. It is *not*
suppressed by turning off MSI-X, nor by turning off jumbo frames (both
of which are working in 2.6.35 anyway). It is apparently suppressed by
switching it out of gigabit mode, by turning off every machine attached
to the subnet on which it is transmitting (though this may simply be an
artefact caused by its not needing to send anything down the link when
that is done), and, oddly, by pingflooding the machine (with the packets
entering via the NIC that fails). (I've been pingflooding it for three
weeks now, and no halts have happened. I stopped for three hours and the
NIC locked up.)

I wonder if this has something to do with PCI ASPM? The driver turns
ASPM off at least partially for this NIC, but if the NIC is being
flipped into some sort of low-power state when transmission ceases for a
while, then perhaps there is a low probability of it not coming out of
it again properly. That would explain the symptoms I see (but so would
many other things, I suppoe).

------------------------------------------------------------------------------
Special Offer-- Download ArcSight Logger for FREE (a $49 USD value)!
Finally, a world-class log management solution at an even better price-free!
Download using promo code Free_Logger_4_Dev2Dev. Offer expires 
February 28th, so secure your free ArcSight Logger TODAY! 
http://p.sf.net/sfu/arcsight-sfd2d
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel&#174; Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to