On (06/02/08 14:25), Artem Kachitchkine wrote:
> One important thing I forgot to mention. The in-kernel property cache is 
> technically no longer need: we always propagate properties from the 
> ioctl straight into mc_setprop. I see two reasons to retain the cache:
> 
> - Observability/diagnostics. Using ::dls_prop to list loaded properties 
> is so much easier than digging through each driver's soft state structure.
> 
> - Potentially useful for future projects (e.g. Crossbow).

I also have an instinct that it should be kept, so that the mac
layer can know the property setting without having to consult the
driver at all times. Also, wasn't this needed for autopush/zone
temporary settings?

I looked at your latest webrev, it looks good. Thanks for keeping
up with this so patiently!

We should document the lessons learned over the last few weeks
somewhere though, so that others don't stumble on the same constraints
that we did. Although we could put this in the design doc, I think it's
more likely to be useful as comments in the code, perhaps in the
vicinity of dls_devnet_set/dls_devnet_prop_load_task to explain why we
need to asynchronously do the property loading, and how it all fits
together. 

--Sowmini


_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to