Matthew Dharm wrote:
> My previous observations are that URBs for disconnected devices are
> completed very quickly (i.e. <1 sec).  There has been some suggestion that
> there might be a case where this is not true... I have not seen it, and
> would like to know where it might exist.
> 
> And yes, this does apply to CBI devices.  I believe the interrupt
> completion gets called with a status to indicate the device is gone.
> Again, that's what I observed last time I tested this behavior....

Yes:  modulo bugs in the host controller driver.  If the
device neither transfers nor NAKs, there's a problem.

- Dave


> Matt
> 
> On Fri, Jul 05, 2002 at 07:22:01AM +0200, Manfred Spraul wrote:
> 
>>David Brownell wrote:
>>
>>>Matthew Dharm wrote:
>>>
>>>
>>>>Blocking until all URBs complete is (effectively) what we do now...
>>>>
>>>>But this is an important question.. do those URBs get 'completed' or 
>>>>not at
>>>>disconnection?  Better question: should they?
>>>
>>>
>>>Please explain your question a little better.  I just
>>>said they'll complete via unlink or transfer failure.
>>>What's the issue?
>>>
>>
>>The usb-storage driver submits an urb and waits for a reply from the 
>>hardware. If the user pulls out the cable, _disconnect() is called. The 
>>disconnect handler must wait until the currently processed request has 
>>finished.
>>
>>Right now it just waits on the request semaphore, without calling 
>>usb_unlink_urb().
>>Question: Is it guaranteed that the hcd will complete the urb within 
>>unnoticable time (< 1 sec or so) after the cable is pulled out.
>>
>>If the complete handler is not called immediately, then the user would 
>>notice a hang (20 seconds, until scsi midlayer timeout). The fix is 
>>obvious: call usb_unlink_urb() on the pending urb.
>>
>>--
>>      Manfred
> 
> 





-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Bringing you mounds of caffeinated joy.
http://thinkgeek.com/sf
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to