[gentoo-sparc] qfe (hme?) link bug ?

[gentoo-sparc] qfe (hme?) link bug ?

[Leif Sawyer]

[-- Attachment #1: Type: text/plain, Size: 900 bytes --]

I've been trying to figure out what the inconsistancy is in the
sparc ethernet model with relation to link up/down for hotplug events.

My qfe interface defaults at boot to "autoneg on", and 100/full 
as verified with ethtool.

if I unplug the cable, I should see a NETLINK event that the interface
is down, so that hotplug/netplug can take action.

This does not seem to happen.
What does seem to happen is that the interface resets itself into
10/half, and the link-detect stays active.

if I 'touch' the interface with ethtool in set mode, say, 
"ethtool -s eth2 autoneg on" -- which is the current setting anyway --
the interface finally shows link down, and I see kernel messages
from the driver stating this.

This would seem to be a bug in the re-negotiation code. Link-detect
should be 'no' until we've successfully renegotiated a carrier,
especially if we've suddenly changed speed/duplex.


[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 3115 bytes --]

Re: [gentoo-sparc] qfe (hme?) link bug ?

[Andrew Gaffney]

Leif Sawyer wrote:
> I've been trying to figure out what the inconsistancy is in the
> sparc ethernet model with relation to link up/down for hotplug events.
> 
> My qfe interface defaults at boot to "autoneg on", and 100/full 
> as verified with ethtool.
> 
> if I unplug the cable, I should see a NETLINK event that the interface
> is down, so that hotplug/netplug can take action.
> 
> This does not seem to happen.
> What does seem to happen is that the interface resets itself into
> 10/half, and the link-detect stays active.
> 
> if I 'touch' the interface with ethtool in set mode, say, 
> "ethtool -s eth2 autoneg on" -- which is the current setting anyway --
> the interface finally shows link down, and I see kernel messages
> from the driver stating this.
> 
> This would seem to be a bug in the re-negotiation code. Link-detect
> should be 'no' until we've successfully renegotiated a carrier,
> especially if we've suddenly changed speed/duplex.

This bug may not be specific to this driver. I saw something similar recently on 
some Dual Opteron servers here at work with Broadcom gigabit ethernet with the 
2006.1 LiveCD and kernel 2.6.17-gentoo-r5. When the machine booted up, I only 
had a cable plugged into eth1. While eth1 showed 100/full, eth0 showed 10/half 
until I did 'ifconfig eth0 up/down'.

I currently have 2.6.16-hardened-r11 running on those boxes, and one box still 
has nothing plugged into eth0, but mii-tool shows that it's down. It could be a 
difference in the bootup procedure or a bug that was introduced in 2.6.17. What 
kernel are you seeing this with?

-- 
Andrew Gaffney                            http://dev.gentoo.org/~agaffney/
Gentoo Linux Developer                                   Installer Project
Today's lesson in political correctness:      "Go asphyxiate on a phallus"
-- 
gentoo-sparc@gentoo.org mailing list

RE: [gentoo-sparc] qfe (hme?) link bug ?

[Leif Sawyer]

[-- Attachment #1: Type: text/plain, Size: 506 bytes --]

Andrew Gaffney writes:
> I currently have 2.6.16-hardened-r11 running on those boxes, 
> and one box still has nothing plugged into eth0, but mii-tool 
> shows that it's down. It could be a difference in the bootup 
> procedure or a bug that was introduced in 2.6.17. What kernel 
> are you seeing this with?

Latest  2.6.18-gentoo-1

also saw it in a number of previous kernels, but didn't pay it
too much mind until I started watching the console messages
trying to debug another issue with SIP and NAT.

[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 3115 bytes --]