http://bugzilla.suse.com/show_bug.cgi?id=1004579 http://bugzilla.suse.com/show_bug.cgi?id=1004579#c6 Benjamin Poirier changed: What |Removed |Added ---------------------------------------------------------------------------- Flags|needinfo?(bpoirier@suse.com | |) | --- Comment #6 from Benjamin Poirier --- (In reply to Søren Tarp from comment #3)

With 42.1 and older, the network interface worked fine. Now, with Tumbleweed, I have to manually issue a command that restricts the duplex speed of the NIC:

Are you sure that it worked with the in-tree driver in 42.1? The only functional changes to the driver in between these two kernel versions are related to suspend/resume: ben@f1:~/local/src/kernel$ git log --oneline public/openSUSE-42.1..public/stable -- drivers/net/e thernet/jme.* ea01964 drivers/net/ethernet/jme.c: Deinline jme_reset_mac_processor, save 2816 bytes 81422e6 jme: Fix device PM wakeup API usage 0772a99 jme: Do not enable NIC WoL functions on S0 ee50c13 net: jme: fix suspend/resume on JMC260 a188222 net: Rename NETIF_F_ALL_CSUM to NETIF_F_CSUM_MASK e9b5ac2 net: jme: use kzalloc() instead of kmalloc+memset

Other post suggests that all this i due to a bug in the jme driver.

Actually, it seems to be a hardware problem that is worked around in software by the out of tree driver: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/880316/comments/41 It appears that it periodically checks for link failure after autonegotiation and then automatically tries renegotiating with only lower speed(s) enabled. So this is faulty hardware that can't maintain a 1000BASE-T link with some link partners. But since it works with others, disabling 1000BASE-T operation completely would cause a regression for some users. I've prepared a kmp with this patch. It's available in the following repository: http://download.opensuse.org/repositories/home:/benjamin_poirier:/bsc1004579... Please double check about 42.1 and try the kmp. -- You are receiving this mail because: You are on the CC list for the bug.