Am Montag, 19. April 2004 21:22 schrieb Alan Stern:
> On Mon, 19 Apr 2004, Oliver Neukum wrote:
> > This somehow doesn't add up. The data structure bound to is the interface
> > and the real addressing is done to a device/endpoint pair. In fact, why
> > do we have a pipe member anyway?
>
> The HCDs use it to determine the endpoint number, transfer direction, and
> transfer type (ctrl, bulk, int, or iso).
Specifying these parameters that way is bad, because it is illogical.
> > Yes, but we must not refcount interfaces. We cannot refcount _any_ data
> > structure being an internal representation of a physical USB device.
> > Probe() and disconnect() are entirely software events.
>
> I don't see why not. So long as the data structure is marked defunct so
> that it can't be used for communication with the device after a
> configuration change (for example), there's nothing wrong with leaving it
> in memory until its refcount goes to 0.
On which event would you zombify and reallocate?
The interface is a representation of a device's attributes. A binding to a
driver is just a convention at heart. You are making a layering violation.
> > The most elegant solution would seem to be refcounting URBs. We do that,
> > but an URB doesn't have an undead state. That would require us to add
> > an interface argument to usb_alloc_urb(). Then we could zombify URBs and
> > submissions to ep0 for no driver could be done by allocating URBs with
> > NULL as an interface parameter.
>
> Right now URBs aren't bound to anything when they are created. I've seen
> proposals that they be bound to endpoints, interfaces, and devices. Are
> there any possibilities left?
altsettings :-)
> Binding URBs to _something_ at or below the HCD level makes sense.
> Allocation could be handled by the HCD itself, with the necessary
Yes. That's a major performance boost. And will get rid of storage's deadlock
problems.
> hc-private data added automatically (but what about interference with the
> variable-length iso packet descriptors at the end?). However elegant it
Allocate the iso packet descriptors seperately.
> may be, though, I don't see any need for _inactive_ URBs to be identified
> with particular endpoints or interfaces.
1. It's a consequence of pushing allocation down the stack. They are specific
to the parameters of allocation now.
2. It's a simplification. You eliminate the need to associate on submit
and disassociate on the callback.
3. It's essential to let nuke_urbs() do all the work.
Regards
Oliver
-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
