Wake-on-LAN is broken for certain machines
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
linux-source-2.6.10 (Ubuntu) |
Invalid
|
Medium
|
Unassigned |
Bug Description
It seems that Wake-on-LAN is broken for some of my machines and not others.
They're all running 2.6.12-10, and they're all AMD's.
The WORKING case is an MSI K7T Turbo 2 (VIA chipset, Phoenix Award BIOS version
W6330MS V3.0 092601 10:49:47); I can wake this machine from poweroff at all
times. (That machine -still- suffers from bug #22544, where it won't -reboot-
without turning itself off, but that's a different story.)
The FAILING case is a pair of MSI K7N2 Delta-L machines (nVidia chipset, Phoenix
Award BIOS version W6570MS V5.6 110703 10:38:29). The failure mode here is
interesting: If I boot one of them and halt it in the BIOS (e.g., by holding
down DEL on boot), and then turn off the power with the front-panel
soft-power-off switch, I can wake the machine using Wake-on-LAN. However, if I
let the machine boot Ubuntu, then no matter how it powers down (soft-power-off
button, shutdown -h now, or hard power-off via the power-supply switch), I can't
wake it. I'm guessing that the OS has to leave the NIC in the right state
(e.g., waiting for the magic packet to go by, as documented in
http://
and it's not. The BIOS, on the other hand, is.
[The hard-power-off was followed by turning the hard-power-switch back on with
the BIOS set to -not- power up automatically, of course. All of my other
testing had the BIOS power-on action set to "last state", e.g. if the machine's
soft-power was on when its hard power was cycled, it would turn itself back on.]
I've tried both ether-wake and wakeonlan; see
http://
for sources. The former sends an Ethernet frame; the latter sends a UDP packet.
If you're debugging this, note carefully that the VIA-based chipsets will wake
if the magic packet is sent straight to them, -or- if it's broadcast. The
nVidia chips, on the other hand, don't seem to notice the packet unless it's
sent to the broadcast address (-b in ether-wake).
Note also that it's not just this kernel version; the working machine was
running 2.6.12-9 until just a few minutes ago, and I meticulously tested its
wake-on-LAN behavior under that kernel before I took the offered upgrade to -10.
Nothing about its behavior (neither the Wake-on-LAN, nor the
halts-instead-
This bug has been open for 8 weeks now with no response whatsoever, including even moving it to some known package. It'd be really nice if this got fixed before the next release came out; does anyone need any more information from me on this?
Thanks!