On 5/26/2010 1:14 PM, James Carlson wrote:
Willard Korfhage wrote:
For some reason, the Ethernet port on a server running build 134 suddenly
started auto-negotiating to 10 Mbps instead of 1000 Mbps. I tried to turn
auto-negotiation off so I could set the speed manually, but I am unable to do so
% pfexec dladm reset-linkprop -p adv_autoneg_cap bge0
% dladm show-linkprop -p adv_autoneg_cap bge0
LINK PROPERTY PERM VALUE DEFAULT POSSIBLE
bge0 adv_autoneg_cap rw 1 1 1,0
Should I be doing this a different way?
"reset-linkprop" means "set this property back to the standard default,
whatever that might be." So, it's doing what you asked (the default is
to enable autonegotation), but probably not what you wanted.
You would disable autonegotiation like this:
pfexec dladm set-linkprop -p adv_autoneg_cap=0 bge0
... but, in case you haven't been cautioned before, I'll do so now:
don't do this. Really. Don't. Disabling autonegotiation opens you up
to a world of extreme hurt.
In particular, you *must* nail all of the properties as needed on *both*
sides of the link. If you don't, then you'll see an obscure link failure.
The standards require that when one side negotiates, and the other does
not, then the one that tries to negotiate and fails has to fall back to
a minimum value -- like 10Mbps half-duplex. The error you're seeing is
most likely caused by having erroneously set the other side to "forced"
duplex or speed. That other end should be fixed, rather than disabling
autonegotiation locally.
Actually, I believe the standards *require* autonegotiation for
1000BaseTX. If you you're using gigabit networking, you need autoneg to
stay standards conformant.
What you *can* do is remove certain modes from what is negotiated, but
even that is a bad idea.
The days of needing to disable autoneg are long past us. The only sites
that still do this are those that were burned by crummy implementations
of 100BaseT way way back when and have not updated (perhaps they're
running 15 year old Cisco gear?), or those that simply do not know any
better.
-- Garrett
_______________________________________________
networking-discuss mailing list
[email protected]