On (03/21/09 16:54), Steven Stallion wrote:
>
> [email protected] wrote:
> > On (03/21/09 14:58), Steven Stallion wrote:
> >> Would it be possible to add in some verbiage about the persistence
> >> lifecycle for driver developers?
> >
> > I'm not seeing why driver developers would have to ever consider
> > the persistence lifecycle for the IP interfaces (since these are IP
> > properties, not driver props). Can you clarify what you have in mind?
>
> Provided a driver is written using the Phase I interfaces,
> m_getprop/m_setprop may be called at any time (from an API perspective).
> AFAIK most devices do not support dynamic reconfiguration - meaning that
> most drivers tend to reset/reconfigure the device if m_setprop has been
> called after the interface has been started.
>
> I think its important to codify precisely when persisted properties are
> set on a given interface. Provided these are set after the plumb, and
> before m_start is called, this is fine. If these are done after the
> interface has started, this could be a serious drag on performance for
> those drivers which reset when a single property has been changed.
The driver callbacks would typically never be called, because
ipadm (as its name suggests) deals with TCP/IP properties, that should
be usually driver agnostic. Having said that, I do realize that there
may be a minority of properties that do impact the driver state itself.
If that happens, and a driver reset is genuinely required (as opposed
to a brute-force restart due to implementation details of the driver)
the GLD layer will take this into account and ensure that the {stop,
callback, start} sequence is handled transparently in the GLD layer.
>
> This is good. Is it possible to document this specifically? MAC drivers
> rely on the Brussels framework - detailing lifecycle is a good thing for
> those of us who arent intimately familiar with the project.
sure.
> It makes me a little nervous to rely on a given module always being
> available (or unloadable). It is certainly foreseeable to see an
> installation which does not make use of IP.
sure, but such an installation should not be using a property/feature
supplied by "ip"adm, right?
--Sowmini
_______________________________________________
networking-discuss mailing list
[email protected]
