Re: handling failure in suspend() during autosuspend

On Fri, 5 Oct 2007, Oliver Neukum wrote:

> Hi,
> 
> this code tries to handle suspend() returning errors (usb_suspend_both):
> 
> 	/* If the suspend failed, resume interfaces that did get suspended */
> 	if (status != 0) {
> 		while (--i >= 0) {
> 			intf = udev->actconfig->interface[i];
> 			usb_resume_interface(intf, 0);
> 		}
> 
> 		/* Try another autosuspend when the interfaces aren't busy */
> 		if (udev->auto_pm)
> 			autosuspend_check(udev);
> 
> It uses autosuspend_check() to resubmit the work if the problem isn't
> permanent. It in turn schedules the retry only if last_busy
> is recent enough. A failure to set it if you refuse to autosuspend will
> kill autosuspend on the devices that use exclusively last_busy.

Right.  If a driver refuses to autosuspend, it should update last_busy.  
In most cases the refusal will be because an I/O event just happened, 
which means last_busy should already be (almost) equal to the current 
time.

> However, even if a driver uses it, it still is a race condition.
> The clean solution would be to split autosuspend_check(), with
> check really only checking, and a new function to schedule another
> attempt.

Do you really want to schedule another attempt at this point?  What 
if the drivers's suspend method incremented the pm_usage_count?  I 
don't see anything wrong with doing both a check and a reschedule.

As far as the race is concerned...  In practice it shouldn't matter
unless autosuspend_delay is set to 0.  I suppose autosuspend_check()  
could take an extra "reschedule" flag.  If the flag was set, the
routine would schedule another autosuspend attempt (with a 1-second
delay) if the idle-time period had already expired.

Alan Stern


-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel