Re: [linux-pm] s2ram slow (radeon) / failing (usb)

["Bruno Prémont" <bonbons@linux-vserver.org>]

On Wed, 05 May 2010 Alan Stern <stern@rowland.harvard.edu> wrote:
> On Wed, 5 May 2010, Jiri Kosina wrote:
> > > > Ok, I've been digging some further...
> > > > 
> > > > The hid_device_probe properly returns -ENODEV, but:
> > > > 
> > > > Call trace:
> > > > [ 3228.866146]  [<ffffffffa01a00e6>] hid_device_probe+0xd6/0x1f0 [hid]
> > > >     return -ENODEV
> > > > [ 3228.874594]  [<ffffffff8130995a>] driver_probe_device+0xaa/0x1d0
> > > >     calls inlined really_probe from drivers/base/dd.c
> > > >     which ALLWAYS returns 0:
> > > >      dd.c:147 /*
> > > >           148  * Ignore errors returned by ->probe so that the next driver can try
> > > >           149  * its luck.
> > > >           150  */
> > > >           151 ret = 0;
> > > >      and has on line 139 (under same failure label):
> > > >               dev->driver = NULL;
> > > > [ 3228.882758]  [<ffffffff81309b20>] ? __device_attach+0x0/0x50
> > > > [ 3228.890555]  [<ffffffff81309b6b>] __device_attach+0x4b/0x50
> > > >      lets 0 bubble up
> > > > [ 3228.898272]  [<ffffffff81308d28>] bus_for_each_drv+0x68/0x90
> > > >      lets 0 bubble up
> > > > [ 3228.906080]  [<ffffffff81309c3b>] device_attach+0x8b/0xa0
> > > >      lets 0 bubble up
> > > > [ 3228.913603]  [<ffffffff81308b15>] bus_probe_device+0x25/0x40
> > > >      returns void and does WARN_ON(device_attach() < 0)
> > > > [ 3228.921356]  [<ffffffff81307166>] device_add+0x3d6/0x610
> > > >      returns 0 here as there was no local error
> > > > [ 3228.928772]  [<ffffffffa019fc53>] hid_add_device+0x183/0x1e0 [hid]
> > > > [ 3228.937098]  [<ffffffffa01b4a77>] usbhid_probe+0x287/0x420 [usbhid]
> > > > [ 3228.945535]  [<ffffffffa005006d>] usb_probe_interface+0x14d/0x230 [usbcore]
> > > > ...
> > > > 
> > > > So IMHO in hid_add_device() we should also check for hdev->dev.driver
> > > > when device_add() returns 0 and consider that one being NULL as a
> > > > (possible) error.
> 
> Note that it is perfectly normal for devices to be registered on a bus 
> without a driver.  Perhaps the usbhid core doesn't expect this, though, 
> or perhaps it doesn't make sense for HID devices.  Regardless, I don't 
> see how this could cause the problem.
>
> Earlier, Bruno said that the hang occurs in hid_cancel_delayed_stuff(), 
> presumably during one of its cancel_work_sync() calls, and presumably 
> because the workqueue has been frozen.  But as far as I can tell, 
> cancel_work_sync() should work just fine if the workqueue has been 
> frozen.  Maybe this should be investigated more closely.
> 
> Bruno, can you confirm that the hang occurs during one of those 
> cancel_work_sync() calls?

No, it's not one of the cancel_work_sync() that hangs but it's the
del_timer_sync() right before them that hangs!
(del_timer_sync() also hangs if I put it last, so the cancel_work_sync()
don't hang anything)

static void hid_cancel_delayed_stuff(struct usbhid_device *usbhid)
{
	del_timer_sync(&usbhid->io_retry); /* this one never returns */
	cancel_work_sync(&usbhid->restart_work);
	cancel_work_sync(&usbhid->reset_work);
}


Thanks,
Bruno