Date: Mon, 29 Nov 2010 21:11:13 +0100 (CET)
From: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
To: Greg KH <greg@kroah.com>
cc: linux-kernel@vger.kernel.org
Subject: Re: [PATCH/RFC] core: add a function to safely try to get device
 driver owner
In-Reply-To: <20101129195649.GA32500@kroah.com>
Message-ID: <Pine.LNX.4.64.1011292107250.27962@axis700.grange>
References: <Pine.LNX.4.64.1011292043000.27962@axis700.grange>
 <20101129195649.GA32500@kroah.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 4505
Lines: 137

On Mon, 29 Nov 2010, Greg KH wrote:

> On Mon, Nov 29, 2010 at 08:43:28PM +0100, Guennadi Liakhovetski wrote:
> > When two drivers interoperate without an explicit dependency, it is often
> > required to prevent one of them from being unloaded safely by dereferencing
> > dev->driver->owner. This patch provides a generic function to do this in a
> > race-free way.
> > 
> > Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
> > ---
> > 
> > Not run-time tested in this form, but this is just a generalisation of the 
> > code in drivers/media/video/sh_mobile_ceu_camera.c::sh_mobile_ceu_probe(). 
> > If the idea is accepted in principle, I will replace that specific 
> > implementation with a call to this function, test... But I am not sure, if 
> > I'd be able to test it for races. If such testing is required on SMP, I'd 
> > have to write some test-case for it. Thoughts?
> > 
> >  drivers/base/dd.c      |   63 ++++++++++++++++++++++++++++++++++++++++++++++++
> >  include/linux/device.h |    1 +
> >  2 files changed, 64 insertions(+), 0 deletions(-)
> > 
> > diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> > index da57ee9..44c6672 100644
> > --- a/drivers/base/dd.c
> > +++ b/drivers/base/dd.c
> > @@ -17,10 +17,12 @@
> >   * This file is released under the GPLv2
> >   */
> >  
> > +#include <linux/completion.h>
> >  #include <linux/device.h>
> >  #include <linux/delay.h>
> >  #include <linux/module.h>
> >  #include <linux/kthread.h>
> > +#include <linux/notifier.h>
> >  #include <linux/wait.h>
> >  #include <linux/async.h>
> >  #include <linux/pm_runtime.h>
> > @@ -422,3 +424,64 @@ void dev_set_drvdata(struct device *dev, void *data)
> >  	dev->p->driver_data = data;
> >  }
> >  EXPORT_SYMBOL(dev_set_drvdata);
> > +
> > +struct bus_wait {
> > +	struct notifier_block	notifier;
> > +	struct completion	completion;
> > +	struct device		*dev;
> > +};
> > +
> > +static int bus_notify(struct notifier_block *nb,
> > +		      unsigned long action, void *data)
> > +{
> > +	struct device *dev = data;
> > +	struct bus_wait *wait = container_of(nb, struct bus_wait, notifier);
> > +
> > +	if (wait->dev != dev)
> > +		return NOTIFY_DONE;
> > +
> > +	switch (action) {
> > +	case BUS_NOTIFY_UNBOUND_DRIVER:
> > +		/* Protect from module unloading */
> > +		wait_for_completion(&wait->completion);
> > +		return NOTIFY_OK;
> > +	}
> > +	return NOTIFY_DONE;
> > +}
> > +
> > +int device_try_get_driver(struct device *dev)
> 
> Please create some kerneldoc information for this new function so that
> people know how to use it and what it is for.

Ok, will do.

> Also, do you want to provide a device_put_driver() function as well to
> decrement the owner count once the person who grabed the driver is done
> with it?

Well, once you've got the driver, you put it by just calling 
module_put(device->driver->owner), but you're right, I should add one for 
symmetry, and there I won't guard against the race, because if someone 
would be calling the put() without a successful get(), they deserve any 
Oops they get anyway;)

> 
> > +{
> > +	struct bus_wait wait = {
> > +		.completion = COMPLETION_INITIALIZER_ONSTACK(wait.completion),
> > +		.dev = dev,
> > +		.notifier.notifier_call = bus_notify,
> > +	};
> > +	struct bus_type	*bus;
> > +	int ret;
> > +
> > +	if (!dev || !dev->bus)
> > +		return 0;
> > +
> > +	bus = dev->bus;
> > +
> > +	if (bus_register_notifier(bus, &wait.notifier) < 0)
> > +		return 0;
> > +
> > +	/*
> > +	 * From this point the driver module will not unload, until we complete
> > +	 * the completion. In the worst case it is hanging in device release on
> > +	 * our completion. So, _now_ dereferencing the "owner" is safe.
> > +	 */
> > +	if (dev->driver && dev->driver->owner)
> > +		ret = try_module_get(dev->driver->owner);
> > +	else
> > +		/* Either no driver, or too late, or probing failed */
> > +		ret = 0;
> > +
> > +	/* Let notifier complete, if it has been blocked */
> > +	complete(&wait.completion);
> > +	bus_unregister_notifier(bus, &wait.notifier);
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL(device_try_get_driver);
> 
> EXPORT_SYMBOL_GPL() please.

Ok

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/