Joost Mulders wrote:
>> We want to leave ieee802.3 beneath the scene
>>
> I agree that drivers must maintain a "well defined" set of statistics.
> ieee802.3(5) is a good starting point for now but the set probable will
> need expansion in the future. For example, 10Gb is not in ieee802.3(5)
> and also the max size of jumbo frames -as mentioned as an example in
> brussels.pdf- is not in the ieee802.3(5) man page.
>
>
Yes, we define that in other parts of Brussels doc. Let's focus this
thread on pause
frame setting only. :-)
>> while providing user with a UI which makes more sense.
>> /* We use 0/1 to represent "off/on" respectively. */
>>
> Drivers use the 0/1 scheme because using other numbers is confusing and
> strings cannot be localized (see later post by Garret). I don't think
> dladm should limit itself to "on/off (1/0)" type of values. Why not use
> things like
> flowcontrol=send
> flowcontrol=accept
> flowcontrol=both (or symmetric)
> dladm in turn can do localization and translate the strings into
> ieee802.3(5) properties. dladm could also provide array type values like
> speed=10,100,1000
> duplex=half,full
> flowcontrol=send,accept
> I.e, I don't think dladm should limit itself to on/off values.
>
We don't limit dladm to on/off values, though there do exist some. I
think we could also use:
flowcontrol = {send, accept, all, disable}, while leaving symmetric or
asymmetric magic
off from user.
libdladm could parse the string.
> dladm could also introduce a property group concept whith groups for
> (for example) advertised-, link partner, capability and actual link
> properties and also (probably) a MAC private group.
>
>
>> UI: accept_pause = 0, send_pause = 0:
>> Protocol value: cap_asmpause = 0, cap_pause = 0, adv_cap_asmpause = 0,
>> adv_cap_pause = 0
>>
>
> I don't think dladm should modify the cap_ properties. I believe that
> the cap_ properties are a result of the capabilities of the PHY and MAC.
> The cap_ properties could be used to verify the values of the
> advertised capabilities though.
>
Yes, above UI -> Protocol value mapping does not indicate the r/w
properties at all, in fact.
> Regards, Joost
>
