1 |
Kyle Liddell <kyle@××××××××××××××××.net> posted |
2 |
20090401060645.GA4182@athlon, excerpted below, on Wed, 01 Apr 2009 |
3 |
02:06:45 -0400: |
4 |
|
5 |
> After upgrading the kernel from 2.6.28-gentoo-r3 to 2.6.29, I'm having a |
6 |
> weird problem: If I make "light" use of the network (ssh, email, http, |
7 |
> etc), no trouble. But, if I make "heavy" use of the network (say, |
8 |
> copying a few GB from a local NFS server to /dev/null), the network |
9 |
> interface stops working almost immediately. ifconfig, mii-tool, the |
10 |
> NIC's led, and the switch's LED all say that the link is still up. But, |
11 |
> all my connections time out, ping gets no replies, etc. (I'm using the |
12 |
> via-velocity driver for my motherboard's built-in 1GB NIC.) |
13 |
|
14 |
I'm unsure if your hardware is affected, but there's a known NAPI and |
15 |
interrupt related regression in 2.6.29. It was due to a commit trying to |
16 |
fix another bug just a few hours before release, and the discovery and |
17 |
subsequent bug trace was one of two subthreads quickly attached to Linus' |
18 |
2.6.29 official announcement. (FWIW, the other subthread was even more |
19 |
fallout from the earlier ext4 data loss over improper shutdown, where |
20 |
ext3 was fine. That has and continues to generate HUGE subthreads all |
21 |
over the place, from the Ubuntu bug filing that was one of the first |
22 |
reports of it to LKML to /. to LWN. ext4 may have had the dev- removed |
23 |
from the name, but it's not exactly stable as this demonstrates, and |
24 |
what's worse, many people including Ted Tso seem to think the problem |
25 |
really is NOTABUG.) |
26 |
|
27 |
Assuming your NIC hardware is indeed affected as it looks to be, if |
28 |
there's nothing in 2.6.29 that you are really needing, consider reverting |
29 |
to the 2.6.28.x stable series until at least 2.6.29.1 (and check its |
30 |
changelog but it should be there). If you do need something in 2.6.29, |
31 |
consider using the git snapshot from a day or two pre-release or go find |
32 |
the thread on LKML and see if there's yet a good patch. (They were still |
33 |
testing patches when I happened across the thread in my browsing, with |
34 |
several proposals not quite fixing it, and it's significantly more |
35 |
complicated than a simple revert of the problem commit, as other 2.6.29 |
36 |
changes triggered IRQ processing issues that weren't there in the |
37 |
2.6.28.x stable series.) |
38 |
|
39 |
-- |
40 |
Duncan - List replies preferred. No HTML msgs. |
41 |
"Every nonfree program has a lord, a master -- |
42 |
and if you use the program, he is your master." Richard Stallman |