Re: [PATCH][ATM] use rtnl_{lock,unlock} during device operations (take 2)

[Werner Almesberger <wa@almesberger.net>]

chas williams wrote:
> but parts of the control/config plane still arent synchronus.
> how about atm_async_release()?

That's just "tag and go away". If I understood Dave right, he
wants the device to actually disappear at that point (well, at
what would be the equivalent point for devices).

Of course, "tag and go away" is relatively easy, and you can
even make it partially look as if you'd destroy things
immediately, e.g. by putting the old "ATM device" layer under
a network device layer. When that dreaded async call comes,
you drop the network device layer part, leave the ATM device
intact, and start working towards getting rid of it too. (E.g.
by failing all system calls related to it, much like what
happens when atmsigd dies.)

> actually i believe its up to the driver author to make sure that
> a vcc isnt used after the close completes.

When a driver's "close" function returns, the driver must have
released all externally visible resources (e.g. VPI/VCI) that
belong to the VCC, and ideally all invisible resources (e.g.
buffers). There must be no more calls to "push".

In return, the stack must not make any other calls for that
VCC after invoking "close".

Hmm, and I should have written all this in the device driver
interface document :-)

>  but very few (if any) try to do this.  

Looks like trouble ...

- Werner

-- 
  _________________________________________________________________________
 / Werner Almesberger, Buenos Aires, Argentina         wa@almesberger.net /
/_http://www.almesberger.net/____________________________________________/