Re: [linux-pm] Re: The evilness of struct usb_device->auto_pm

On Friday, 28 September 2007 16:33, Alan Stern wrote:
> On Fri, 28 Sep 2007, Oliver Neukum wrote:
> 
> > > > > What exactly would you like to do?
> > > > 
> > > > Put the stuff into struct device and move the autosuspend code into
> > > > driver core.
> 
> Moving some of the data fields might be acceptable.  Moving the code is 
> premature at best.  You must always bear in mind that the autosuspend 
> model used by USB is not suitable for every type of device.  Each 
> subsystem needs to implement its own form of runtime PM; the core 
> should contain only the lowest-denominator common parts.

Well, that's what I've been always thinking.

> > > Well, I thought that it would be nicer if drivers handled autosuspend in a
> > > transparent (from the point of view of the core) way.
> > 
> > That is possible only for drivers which are leaves in the device tree.
> 
> Not true.  For example, the USB hub driver handles autosuspend 
> perfectly well.  The main issue is what to do when the autosuspend 
> notifications need to cross a subsystem boundary.
> 
> > > Now, it seems to follow from what you're saying that this is not possible in
> > > general, because of USB mass storage devices that need some special handling.
> > 
> > No, they need the ordinary handling. All a device's children must be suspended
> > if a device is suspended.
> > 
> > > I'm not familiar with the USB code, so can you please tell me how the USB
> > > autosuspend is handled at present and what exactly the problem with that is?
> > 
> > The drivers for interfaces manage a count. If all a device's interfaces' count
> > goes to zero, the device is suspended. The drivers are notified through the
> > normal suspend() method. But children of these devices are not notified.
> 
> That's because the notifications go up the device tree, not down.  When 
> a child is suspended, the parent is notified.  Then if the driver sees 
> that all the children are suspended, the parent can be suspended as 
> well.
> 
> Oliver has been trying to subvert this model by making usb-storage 
> responsible for suspending the SCSI disk and CD/DVD drivers, which are 
> located beneath it in the device tree.  I've been trying to convince 
> him that the proper way to handle things is to let the SCSI drivers 
> decide for themselves when their devices can be suspended, and then 
> have them inform usb-storage.

Yes, I think that the SCSI layer should decide.

I guess that the SCSI layer is not really autosuspend-aware, is it?

Greetings,
Rafael

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel