Re: ETHTOOL_GSET IOCTL on GigE links

Re: ETHTOOL_GSET IOCTL on GigE links

[Jeff Garzik]

Jeff Haran wrote:
> With 10/100 Mbps links it wasn't such an issue since the devices tend to
> support the same forced speeds and duplexities as they are capable of
> negotiating, but with GigE links that's not always the case, at least
> not according to what I've read. For instance, the following doc from
> Sun http://www.sun.com/blueprints/0704/817-7526.pdf says that IEEE
> 802.3ab says you can't force 1000Base-T over copper media (see page 4),
> whereas some other physical media allow GigE to run without
> autonegotiation (there's apparently this "serdes" interface that allows
> it, for instance).
> 
> Seems like there should be another field named something like
> supported_forced to indicate what can be forced on the interface. Either
> that or some more SUPPORTED_* bits to indicate supported forced modes.


The 'supported' field has nothing at all to do with auto-negotiation.

The driver should list all possibilities in that field, even if some are 
ONLY supported via 'forced' selection.

	Jeff


P.S.  netdev@linux.kernel.org is a far more active list, and is where 
the network stack/driver developers appear.

RE: ETHTOOL_GSET IOCTL on GigE links

[Jeff Haran]

> -----Original Message-----
> From: Jeff Garzik [mailto:jeff@garzik.org] 
> Sent: Friday, June 01, 2007 3:40 PM
> To: Jeff Haran
> Cc: linux-net@vger.kernel.org; netdev
> Subject: Re: ETHTOOL_GSET IOCTL on GigE links
> 
> Jeff Haran wrote:
> > With 10/100 Mbps links it wasn't such an issue since the 
> devices tend to
> > support the same forced speeds and duplexities as they are 
> capable of
> > negotiating, but with GigE links that's not always the 
> case, at least
> > not according to what I've read. For instance, the 
> following doc from
> > Sun http://www.sun.com/blueprints/0704/817-7526.pdf says that IEEE
> > 802.3ab says you can't force 1000Base-T over copper media 
> (see page 4),
> > whereas some other physical media allow GigE to run without
> > autonegotiation (there's apparently this "serdes" interface 
> that allows
> > it, for instance).
> > 
> > Seems like there should be another field named something like
> > supported_forced to indicate what can be forced on the 
> interface. Either
> > that or some more SUPPORTED_* bits to indicate supported 
> forced modes.
> 
> 
> The 'supported' field has nothing at all to do with auto-negotiation.
> 
> The driver should list all possibilities in that field, even 
> if some are 
> ONLY supported via 'forced' selection.
> 
> 	Jeff
> 

OK, but my question remains. In the case where a device supports one set
of speeds via autonegotiation and another set via forcing, how does one
tell which speeds can be forced and which can be autonegotiated?

Thanks,

Jeff Haran

Re: ETHTOOL_GSET IOCTL on GigE links

[David Miller]

From: "Jeff Haran" <jharan@Brocade.COM>
Date: Fri, 1 Jun 2007 15:51:18 -0700

> OK, but my question remains. In the case where a device supports one set
> of speeds via autonegotiation and another set via forcing, how does one
> tell which speeds can be forced and which can be autonegotiated?

Unfortunately, as you have noted, this information is not
given at the moment.

I got bit by this the other week when I wanted to try forcing
gigabit on a tg3 and the driver wouldn't allow it because it
only allows forcing link speeds at 10/100 on copper links.

Re: ETHTOOL_GSET IOCTL on GigE links

[Jeff Garzik]

Jeff Haran wrote:
> OK, but my question remains. In the case where a device supports one set
> of speeds via autonegotiation and another set via forcing, how does one
> tell which speeds can be forced and which can be autonegotiated?

The interface does not currently support such enumeration.

You can certainly attempt forcing a speed, and see what happens.  The 
driver will either (a) work, (b) refuse and give you an error message, 
or (c) present you with a situation that requires filing a driver bug 
report :)

	Jeff

RE: ETHTOOL_GSET IOCTL on GigE links

[Jeff Haran]

> -----Original Message-----
> From: Jeff Garzik [mailto:jeff@garzik.org] 
> Sent: Friday, June 01, 2007 4:05 PM
> To: Jeff Haran
> Cc: linux-net@vger.kernel.org; netdev
> Subject: Re: ETHTOOL_GSET IOCTL on GigE links
> 
> Jeff Haran wrote:
> > OK, but my question remains. In the case where a device 
> supports one set
> > of speeds via autonegotiation and another set via forcing, 
> how does one
> > tell which speeds can be forced and which can be autonegotiated?
> 
> The interface does not currently support such enumeration.
> 
> You can certainly attempt forcing a speed, and see what happens.  The 
> driver will either (a) work, (b) refuse and give you an error 
> message, 
> or (c) present you with a situation that requires filing a driver bug 
> report :)
> 
> 	Jeff

Are there any plans in the works to modify the interface to provide this
information?

Thanks,

Jeff Haran