kern/119487: page fault in em taskq with polling enabled

[Åke Wallebom]

>Number:         119487
>Category:       kern
>Synopsis:       page fault in em taskq with polling enabled
>Confidential:   no
>Severity:       serious
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Jan 09 12:10:02 UTC 2008
>Closed-Date:
>Last-Modified:
>Originator:     Åke Wallebom
>Release:        7.0 RC1
>Organization:
>Environment:
FreeBSD dunder 7.0-RC1 FreeBSD 7.0-RC1 #2: Tue Jan  8 22:29:24 CET 2008     removed at dunder:/usr/obj/usr/src/sys/CUSTOM  amd64

>Description:
I got a page fault with 'current process' being 'em0 taskq' while transferring about 200 GB of data from an ftp server to my FreeBSD box via Gigabit LAN.

I had compiled my kernel with DEVICE_POLLING and HZ=2000, and kern.polling.enable was set to 1 (I noticed later that this was deprecated).

The files were transfered onto a ZFS pool on a GELI-device (1TB SATA2-drive).

After transferring 150 GB i got the above error and the kernel panic'ed.

Sadly, I can't remember the whole error message.

More about the machine:

Core 2 Duo E6750
4 GB RAM
Three Intel PRO/1000 GT Desktop NICs (Single port cards)
>How-To-Repeat:
I really don't know...

Hasn't happened again - but I have now recompiled the kernel with just 'POLLING' set, using default HZ now. And now i don't enable polling via sysctl but via ifconfig xxx polling.

Transfered 180 GB and did not crash. So I guess I'm just reporting this as a curiosity for the em driver developer(s).
>Fix:
Decrease the HZ value? 
Set polling via ifconfig instead of kern.polling.enable?

>Release-Note:
>Audit-Trail:
>Unformatted: