From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4.1)
Description of problem:
The e1000 driver appears to lock up several times a day for me on a
Dell Precision 650N with the Intel e1000 gigabit ethernet. The dmesg
output looks like this:
NETDEV WATCHDOG: eth0: transmit timed out
e1000: eth0 NIC Link is Up 100 Mbps Full Duplex
All the FC2 2.6 kernels have exhibited this problem. Some seem worse
than others but I don't know what precisely is triggering the problem
(network load, cpu load, etc). This latest 2.6.2-81 kernel seems a
little better than the precious 2.6.1-65 but it still has the problem.
Version-Release number of selected component (if applicable):
I'm seeing the same. Seems to be worst when copying lots of data
to/from an NFS server.
The other thing that I notice is that initialising eth0 takes a long
time compared with machines that have different cards in.
Machine is a Dell Optiplex SX270.
Ethernet (builtin) from lspci:
0000:01:0c.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
Controller (rev 02)
Kernel is 2.6.3-1.91smp.
I am just experiencing this with the latest 2.6.3 kernel. I have not experienced
this earlier. Essentially the network is useless for transferring large files. I am
getting hundreds of
Feb 21 08:28:18 compsci kernel: e1000: eth0 NIC Link is Up 100 Mbps Full Duplex
Feb 21 08:28:33 compsci kernel: NETDEV WATCHDOG: eth0: transmit timed out
Another note: this seems to happen when transmitting out of the machine only.
I took the driver back to 27-ko version from 2.6.2-rc1 and having still the
same problem. I am sure I did not have this problem at some point with
2.6.2 so I am beginning to think that the problem is also associated with
another part of kernel that was upgraded.
Well, there is another bug 115566, that is reporting the same problem.
Judging from that and the comments here the problem seems to be
happening with SMP kernels. I am on a DELL Precision 350n, single
CPU but hyperthreading turned on to use an smp kernel.
The drivers and what is hapopening are somewhat different with 115566.
There are some patches against 126.96.36.199-k1 that have been posted to
netdev to fix some issues with e1000. I'll attach the patches for
y'all to try. These patches would go into 2.6.4.
Actually, get the patches here:
Apply then in that order.
Could you please attache them as a tar file or text. Some
characters are lost when saved from html.
Created attachment 97995 [details]
tar ball of six patches to apply in order against 2.6.3
Great that FIXED it. By the way, whatever the problem was, it was
only happening with the SMP version of the kernel. Now the SMP
I am using the latest arjanv kernel upgraded to bk6 and very few
bk-patches from the -mm3.
So I don't know what the exit criteria is to close these Bugzillas,
but I do know the attached patches have been submitted to the
upstream netdev-2.6 BK tree. If accepted into netdev-2.6, these
patches will propagate to the downstream kernels, and ultimately to
the RH kernel.
I see that the patches made it into 2.6.3-bk7.
fixed in rawhide