On (03/27/08 09:05), Sebastien Roy wrote:
>
> Right, so given that the clamping to a minimum of 1500 by bge is a bug,  
> why would libdladm legitimize that be stating that the minimum for all  
> Ethernet links is 1500?  If there are different possible values per  
> driver, then perhaps something is missing to obtain those possible  
> values per-link?
>
> Right, I know that it's the convention from jumbo frame.  What I worry  
> about is 3rd party driver developers working on something outside of the  
> ON consolidation (so they don't have the option of recompiling  
> libdladm), perhaps "super jumbo frame".  They can't do that, since  
> libdladm artificially limits the MTU to 9000.  This doesn't seem like a  
> good thing to me.

Actually, the "Possible" part is just Advisory. In the example that you
cite (3rd party working on sjf) nothing would prohibit them from using
dladm- the output would just be out of sync with the behavior.

As for obtaining possible values for the link, I am starting to think
that this should be done with a separate property, similar to
the WL_DESIRED_RATES prop for wifi. However such a property would have
to be able to print ranges (e.g., "1500-9000" for bge) or discrete values
(as is done with WL_DESIRED_RATES), so the data passed in/out of the
property would need to be designed after some consideration. so I'm going
to defer this for later, and leave the current behavior for this unchanged.

> Given that the message itself doesn't allow one to correlate it with any  
> ioctl or command that issued it, I wouldn't mind if it wasn't printed at  
> all.

Ok.

> One minor nit on the updated webrev is that mactype_alloc() initializes  
> optional fields to appropriate values, so you don't need to update the  
> plugins that don't support ndd compatibility properties to set  
> mtr_mapping and mtr_mappingcount explicitly to NULL and 0.  They're both  
> allocated that way by default.

Ok, thanks for pointing this out. I was using mtr_stats as a hint
on the changes required, and noticed that mac_ib.c explicitly initializes 
this to null.

--Sowmini


Reply via email to