> Steven Stallion wrote:
>>> Steven Stallion writes:
>>>
>>>>> Right, see
>>>>> http://gdamore.blogspot.com/2008/08/why-you-dont-want-to-force-link-modes.html
>>>>> for my most recent rant on this particular topic.
>>>>>
>>>> I'm suddenly reminded of hours spent in the lab years ago trying to
>>>> get
>>>> a
>>>> PIX and ProCurve to play nice...
>>>>
>>>> It really sounds like this behavior could be abstracted out and placed
>>>> into a capability. Essentially all you would really need is a couple
>>>> of
>>>> utility functions to simplify exposing a set of adv_ props, and to
>>>> read
>>>> (in order of 802.3 precedence) a set of config modes.
>>>>
>>> I'm not sure what "capability" means here, but I do agree with you
>>> that having an analysis tool that reads the various stable
>>> configuration and status bits and produces a human-ready report (along
>>> with a diagnosis) would be extremely useful -- and not terribly hard
>>> to write.
>>>
>>> This is something that's been discussed more than once ...
>>>
>>
>> A Nemo/GLDv3 capability (i.e. MAC_CAPAB_ADV or somesuch).
>>
>> Essentially a driver which wants to use this cap would register a couple
>> of callbacks so Nemo could request what the device will advertise and
>> provide a utility function or two for getting any en_ modes which have
>> been set by the user.
>>
>> As far as stable status bits, this pretty well limits the cap to devices
>> which support MII/GMII/XGMII xcvrs.
>>
>> Right now, I am also working on adding 802.3 MII/GMII support as a
>> capability to Nemo; this just seemed to be another cap that could be
>> added
>> (pretty simply actually) which would life a little easier for driver
>> dev's.
>>
>
> Actually, the current approach, where MII bits are expressed directly,
> is pretty simple for most devices.  The reason for this is that most
> devices (at least modern ones) actually expose MII registers directly to
> the driver, and so there is a one-to-one correspondence between what
> gets requested and what gets set.  Trying to abstract this further would
> actually make *more* work for such drivers.
>
> Now there are *some* devices (some dnet variants using SYM are good
> examples... mxfe has the same problem) where the MII registers are *not*
> exposed, but abstracted somehow.  This is mostly older devices, designed
> before 802.3u was ratified, and they have more work to do.  I'd rather
> leave the work for those devices, and keep the more modern ones simple.

Fair enough. ;)

I think the real thrust of what I was suggesting was to provide a utility
function to return a list of configured modes to help during negotiation.

Steve

_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to