> On (06/17/07 20:51), Cathy Zhou wrote:
>   
>> I was actually discussing this (the initialization of the linkprop) with 
>> Meem the other day, and I am thinking of to call the dladm_init_prop() in 
>> the rcm framework (specifically, in network_rcm when it discovers a new 
>> physical link.
>>     
>
> That would be a good thing for Brussels, but what is the definition
> of "link discovery"? Is that defined as hitting mac_start()? 
> Or  the attach function?
>   
The network_rcm should be able to "discover" a new physical link when 
the link is attached. Today, that logic is in the ip_notify_event() 
function of ip_rcm.c today, and I am going to move it to network_rcm.
>   
>> I don't know whether it is already late to initialize some linkprops when 
>> network_rcm discovers the new links though.
>>     
>
> Well, in theory, if the driver property implementation  has been written
> flexibly, it should not be too late. We are doing this part for some 
> properties like jumbo frames. 
>
> But, in general, if the driver has been written based on the assumption
> that all prop definitions are available at attach, then that driver
> cannot leverage from then network_rcm -> dladm_init_linkprop enhancement.
> We will still have the door call invocation from attach to do the
> dladm_init_linkprop so that things are no worse than the update_drv method
> for such drivers.
>   
Understood.

- Cathy

Reply via email to