b44 (Broadcom BCM4401 100Base-T) + NETCONSOLE + link up == lockup

b44 (Broadcom BCM4401 100Base-T) + NETCONSOLE + link up == lockup

[Frank van Maarseveen]

Noticed this on 2.6.17.8 and it is easily reproduced. When disconnecting
and connecting the network cable from a BCM4401 100Base-T NIC the usual
"link is down" and "link is up" messages show up unless netconsole
logging is enabled upon boot using (e.g.):

	netconsole=@172.17.1.65/,514@172.17.1.64/00:12:3f:85:17:52

In that case the "link is up" message no longer makes it to the virtual
console/remote machine and the machine seems to lock up. alt-sysrq-p
doesn't show anything on the VC (caps- and numlock are dead too) but
alt-sysrq-b works provided it is not preceded by too many unprinted kernel
messages (so it seems). The link does go up according to the link state
LED and there is some activity (incoming packets only I suppose).

unplugging and plugging again doesn't help.

During boot the kernel says:

	kernel: netconsole: device eth0 not up yet, forcing it
	kernel: b44: eth0: Link is up at 100 Mbps, full duplex.
	kernel: b44: eth0: Flow control is off for TX and off for RX.
	kernel: netconsole: network logging started

-- 
Frank

Re: b44 (Broadcom BCM4401 100Base-T) + NETCONSOLE + link up == lockup

[Michael Buesch]

On Tuesday 22 August 2006 11:20, Frank van Maarseveen wrote:
> Noticed this on 2.6.17.8 and it is easily reproduced. When disconnecting
> and connecting the network cable from a BCM4401 100Base-T NIC the usual
> "link is down" and "link is up" messages show up unless netconsole
> logging is enabled upon boot using (e.g.):
> 
> 	netconsole=@172.17.1.65/,514@172.17.1.64/00:12:3f:85:17:52
> 
> In that case the "link is up" message no longer makes it to the virtual
> console/remote machine and the machine seems to lock up. alt-sysrq-p
> doesn't show anything on the VC (caps- and numlock are dead too) but
> alt-sysrq-b works provided it is not preceded by too many unprinted kernel
> messages (so it seems). The link does go up according to the link state
> LED and there is some activity (incoming packets only I suppose).
> 
> unplugging and plugging again doesn't help.
> 
> During boot the kernel says:
> 
> 	kernel: netconsole: device eth0 not up yet, forcing it
> 	kernel: b44: eth0: Link is up at 100 Mbps, full duplex.
> 	kernel: b44: eth0: Flow control is off for TX and off for RX.
> 	kernel: netconsole: network logging started

I think this is not a b44 but a netconsole problem.
I saw this behaviour with netconsole and sungem in the past, too.
I simply decided to not remove the cable while using netconsole :P
But that's not really a solution, hehe.

-- 
Greetings Michael.

Re: b44 (Broadcom BCM4401 100Base-T) + NETCONSOLE + link up == lockup

[Frank van Maarseveen]

On Tue, Aug 22, 2006 at 11:36:57AM +0200, Michael Buesch wrote:
> On Tuesday 22 August 2006 11:20, Frank van Maarseveen wrote:
> > Noticed this on 2.6.17.8 and it is easily reproduced. When disconnecting
> > and connecting the network cable from a BCM4401 100Base-T NIC the usual
> > "link is down" and "link is up" messages show up unless netconsole
> > logging is enabled upon boot using (e.g.):
> > 
> > 	netconsole=@172.17.1.65/,514@172.17.1.64/00:12:3f:85:17:52
> > 
> > In that case the "link is up" message no longer makes it to the virtual
> > console/remote machine and the machine seems to lock up. alt-sysrq-p
> > doesn't show anything on the VC (caps- and numlock are dead too) but
> > alt-sysrq-b works provided it is not preceded by too many unprinted kernel
> > messages (so it seems). The link does go up according to the link state
> > LED and there is some activity (incoming packets only I suppose).
> > 
> > unplugging and plugging again doesn't help.
> > 
> > During boot the kernel says:
> > 
> > 	kernel: netconsole: device eth0 not up yet, forcing it
> > 	kernel: b44: eth0: Link is up at 100 Mbps, full duplex.
> > 	kernel: b44: eth0: Flow control is off for TX and off for RX.
> > 	kernel: netconsole: network logging started
> 
> I think this is not a b44 but a netconsole problem.
> I saw this behaviour with netconsole and sungem in the past, too.
> I simply decided to not remove the cable while using netconsole :P
> But that's not really a solution, hehe.

I use many other network drivers which do not show this problem. Tested
by repowering a large switch :-). I guess it is a bug (race?) copied
from one driver to another.

-- 
Frank

Re: b44 (Broadcom BCM4401 100Base-T) + NETCONSOLE + link up == lockup

[John W. Linville]

On Tue, Aug 22, 2006 at 11:36:57AM +0200, Michael Buesch wrote:
> On Tuesday 22 August 2006 11:20, Frank van Maarseveen wrote:
> > Noticed this on 2.6.17.8 and it is easily reproduced. When disconnecting
> > and connecting the network cable from a BCM4401 100Base-T NIC the usual
> > "link is down" and "link is up" messages show up unless netconsole
> > logging is enabled upon boot using (e.g.):
> > 
> > 	netconsole=@172.17.1.65/,514@172.17.1.64/00:12:3f:85:17:52
> > 
> > In that case the "link is up" message no longer makes it to the virtual
> > console/remote machine and the machine seems to lock up. alt-sysrq-p
> > doesn't show anything on the VC (caps- and numlock are dead too) but
> > alt-sysrq-b works provided it is not preceded by too many unprinted kernel
> > messages (so it seems). The link does go up according to the link state
> > LED and there is some activity (incoming packets only I suppose).
> > 
> > unplugging and plugging again doesn't help.
> > 
> > During boot the kernel says:
> > 
> > 	kernel: netconsole: device eth0 not up yet, forcing it
> > 	kernel: b44: eth0: Link is up at 100 Mbps, full duplex.
> > 	kernel: b44: eth0: Flow control is off for TX and off for RX.
> > 	kernel: netconsole: network logging started
> 
> I think this is not a b44 but a netconsole problem.
> I saw this behaviour with netconsole and sungem in the past, too.
> I simply decided to not remove the cable while using netconsole :P
> But that's not really a solution, hehe.

If I'm not mistaken, this is an issue with the netpoll stack not being
able to handle printk output if the netpoll output device is down.
I forget the details...anyone?

John
-- 
John W. Linville
linville@tuxdriver.com