em0 82579LM loses link under load

Paul Mather freebsd-lists at gromit.dlib.vt.edu
Wed May 29 16:11:15 UTC 2019


I'm running a FreeBSD/amd64 12-STABLE r348161 GENERIC kernel.  Yesterday,  
when doing bulk data transfer from the system to offsite I noticed that the  
system would drop off the network.  The em0 NIC would regularly report this  
to the console:

em0: TX(0) desc avail = 1024, pidx = 0
em0: TX(0) desc avail = 1024, pidx = 0
em0: TX(0) desc avail = 1024, pidx = 0
[[etc.]]


An "ifconfig em0" would show "no carrier" and ifconfig down/up would not  
bring the system back onto the network.  To do that a reboot was necessary.

(Speeds weren't terribly high either, load-wise, topping out at ~25  
MBytes/s on a gigabit link.)

This issue appears to have been reported as a regression in 12.0-RELEASE  
(e.g., https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=234766).  A  
comment in another bug report concerning em performance  
(https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=235031#c22) mentions  
Errata Note updates being in the works to address em regressions post-iflib  
restructuring of the driver.

Does anyone know whether this issue is fixed in em, at least for "older"  
devices like the 82579LM??  This used to be rock solid and fast for me  
under FreeBSD 11.  For now, I've adopted the route of the bug reporters and  
am using the net/intel-em-kmod driver from ports.  I can't afford to have  
this box drop off the network due to "load" and so stability is paramount.

Cheers,

Paul.


More information about the freebsd-stable mailing list