Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 31 May 2018 10:07:05 -0400 (EDT)
From:   Alan Stern <stern@rowland.harvard.edu>
To:     Johan Hovold <johan@kernel.org>
cc:     Roger Quadros <rogerq@ti.com>, <balbi@kernel.org>,
        <linux-usb@vger.kernel.org>, <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 2/2] usb: dwc3: of_simple: don't call pm_runtime_set_active()
In-Reply-To: <20180531132538.GH3259@localhost>
Message-ID: <Pine.LNX.4.44L0.1805311004260.1607-100000@iolanthe.rowland.org>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 31 May 2018, Johan Hovold wrote:

> > This breaks runtime pm as you now get a second round of clock enables
> > which are never balanced on runtime suspend (the clocks are first
> > enabled in dwc3_of_simple_clk_init() above and with your change again in
> > dwc3_of_simple_runtime_resume()).
> > 
> > On the other hand, we currently return from probe() with a positive RPM
> > count so perhaps the RPM callbacks can just be removed altogether (i.e.
> > unless some other entity drops that count at some point before
> > remove()).
> > 
> > >  	ret = of_platform_populate(np, NULL, NULL, dev);
> > >  	if (ret) {
> > >  		for (i = 0; i < simple->num_clocks; i++) {
> > > @@ -131,10 +134,6 @@ static int dwc3_of_simple_probe(struct platform_device *pdev)
> > >  		goto err_resetc_assert;
> > >  	}
> > >  
> > > -	pm_runtime_set_active(dev);
> > > -	pm_runtime_enable(dev);
> > > -	pm_runtime_get_sync(dev);
> > > -
> > >  	return 0;
> > >  
> > >  err_resetc_assert:
> > 
> > Also note that there's currently a use-after-free in remove(), where
> > pm_runtime_put_sync() is called after the clocks have been put.
> > Something like the below (untested) patch should fix it.
> 
> What about the use-after-free in remove? Shall I resubmit the fix below
> separately?
> 
> Thanks,
> Johan
> 
> > From 35c384c31010c344d403c26fc0a1dde0fd68ef4a Mon Sep 17 00:00:00 2001
> > From: Johan Hovold <johan@kernel.org>
> > Date: Mon, 28 May 2018 17:31:45 +0200
> > Subject: [PATCH] usb: dwc3: of-simple: fix use-after-free on remove
> > 
> > The clocks have already been explicitly disabled and put as part of
> > remove() so the runtime suspend callback must not be run when balancing
> > the runtime PM usage count before returning.
> > 
> > Fixes: 16adc674d0d6 ("usb: dwc3: add generic OF glue layer")
> > Signed-off-by: Johan Hovold <johan@kernel.org>
> > ---
> >  drivers/usb/dwc3/dwc3-of-simple.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/usb/dwc3/dwc3-of-simple.c b/drivers/usb/dwc3/dwc3-of-simple.c
> > index cb2ee96fd3e8..b9c869cd6585 100644
> > --- a/drivers/usb/dwc3/dwc3-of-simple.c
> > +++ b/drivers/usb/dwc3/dwc3-of-simple.c
> > @@ -165,8 +165,9 @@ static int dwc3_of_simple_remove(struct platform_device *pdev)
> >  
> >  	reset_control_put(simple->resets);
> >  
> > -	pm_runtime_put_sync(dev);
> > +	pm_runtime_put_noidle(dev);
> >  	pm_runtime_disable(dev);
> > +	pm_runtime_set_suspended(dev);
> >  
> >  	return 0;
> >  }

This is a little racy -- there might be a runtime-suspend callback
between the put_noidle and the disable.  (The put_noidle itself won't
cause a callback to happen, but something else could.)

It would be better to do the disable first and then the put_noidle.

Alan Stern