Re: 2.6.39-rc2 regression: X201s fails to resume b77dcf8460ae57d4eb9fd3633eb4f97b8fb20716

["Gustavo F. Padovan" <padovan@profusion.mobi>]

* Thomas Gleixner <tglx@linutronix.de> [2011-04-12 00:19:32 +0200]:

> On Tue, 12 Apr 2011, Thomas Gleixner wrote:
> > On Mon, 11 Apr 2011, Marcel Holtmann wrote:
> > 
> > > Hi Thomas,
> > > 
> > > > > > > Can the bluetooth folks please have a look at that ASAP? The obvious
> > > > > > > fast fix for Linus tree is to revert the second hunk for now, but this
> > > > > > > needs to be fixed proper.
> > > > > > 
> > > > > > Who will submit this patch? I'd rather have your name on it so that
> > > > > > people come complain at you...
> > > > > 
> > > > > I took a shot at it and just sent a patch (also attached for convenience) 
> > > > > that should solve the problem.
> > > > 
> > > > Aaarg. No. That patch reverts both hunks.
> > > > 
> > > > --- a/net/bluetooth/hci_core.c
> > > > +++ b/net/bluetooth/hci_core.c
> > > > @@ -586,9 +586,6 @@ static int hci_dev_do_close(struct hci_dev *hdev)
> > > >  	hci_req_cancel(hdev, ENODEV);
> > > >  	hci_req_lock(hdev);
> > > >  
> > > > -	/* Stop timer, it might be running */
> > > > -	del_timer_sync(&hdev->cmd_timer);
> > > > -
> > > >  	if (!test_and_clear_bit(HCI_UP, &hdev->flags)) {
> > > >  		hci_req_unlock(hdev);
> > > >  		return 0;
> > > > 
> > > > As I said before you need that first hunk to stay for the case where
> > > > there is no device up and you return via the !HCI_UP check. You just
> > > > moved back to the state before as the stupid timer is active for
> > > > whatever reason even when HCI_UP is not set.
> > > 
> > > if I read this right then we have the case that we arm this timer for no
> > > real reason. A device in !HCI_UP should have nothing running. Certainly
> > > not the cmd_timer since it will never process any commands.
> > > 
> > > According to Gustavo, the problem is really in the hci_reset logic were
> > > we arm the timer even when shutting down the device.
> > 
> > The reason why the original patch was sent is, that the timer was
> > running when the thing went out via the !HCI_UP path, which caused the
> > whole thing to explode in the first place. I had no time to figure out
> > why, but moving the del_timer_sync above the
> > if (!test_and_clear_bit(HCI_UP, &hdev->flags)) solved it.
> 
> Oops. Hit send too fast.
> 
> Then it broke the resume on Keith machine and reverting just the hunk
> which disarms the timer in the 
> 
>         if (hdev->sent_cmd) {
> 
> path made both scenarios working. So there are two problems:
> 
>      1) Why do we need the del_timer_sync() above the !HCI_UP check

That is still a mysterious to me, the real bug the hiding here. I'm trying to
track this down but no luck yet.

> 
>      2) Why gets the timer rearmed after that

It is armed at each HCI command we send. In the close path we send out an HCI
RESET command that rearms it.

-- 
Gustavo F. Padovan
http://profusion.mobi