On Fri, 18 Jun 2004, David Brownell wrote:
> Alan Stern wrote:
> > Greg:
> >
> > This patch fixes another bug in the dummy_hcd driver. When a gadget
> > driver unregisters (simulating a device disconnect), dummy_hcd stops the
> > timer that it uses for processing outstanding URBs. Unfortunately this
> > means that those URBs will never be given back. The patch lets the timer
> > continue running; when it expires all the remaining URBs will fail so the
> > timer won't be reinitialized.
>
> Wouldn't it be better to just force the timer to run
> right away? And fail all URBs and usb_requests right
> at that point. The timer needs to stop running at
> clearly defined points, notably "before rmmod" ...
I thought about this and no, it wouldn't be better, for a couple of
reasons:
Real host controllers don't fail URBs as soon as the device
disconnects. They continue to transmit, and get protocol errors
when there's no reply. The dummy_hcd simulation should work
the same way.
Just because the gadget driver has unregistered doesn't mean
the timer can't keep running. It's only when the corresponding
usb_device has been freed or dummy_hcd is exiting that we need
the timer to stop.
As you note, rmmod presents a potential problem. Obviously we don't want
the timer to fire when the code is being unloaded! But in this regard
dummy_hcd is no different from any other host controller driver:
The timer doesn't run when there are no outstanding URBs.
If there are outstanding URBs then the usb_device must exist,
since the hcd glue layer guarantees that an URB in flight
owns a reference to the usb_device.
The usb_device and the root hub device both own a reference
to the usb_bus.
The HCD's bus-release callback won't get called until after all
references to the bus are dropped. This can be after the
unregister_bus call returns (although I would like to change
usbcore so that it can't be).
If an HCD is unloaded from memory before its bus-release
routine runs, there's a problem. That's true for any HCD.
As you can see, any issues are generic and ought to be solved in the core.
They aren't specific to dummy_hcd.
Alan Stern
-------------------------------------------------------
This SF.Net email is sponsored by The 2004 JavaOne(SM) Conference
Learn from the experts at JavaOne(SM), Sun's Worldwide Java Developer
Conference, June 28 - July 1 at the Moscone Center in San Francisco, CA
REGISTER AND SAVE! http://java.sun.com/javaone/sf Priority Code NWMGYKND
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
