Re: [RFC PATCH 1/2] tee: generic TEE subsystem

[Russell King - ARM Linux <linux@arm.linux.org.uk>]

On Sat, Apr 18, 2015 at 10:37:16PM +0200, Greg Kroah-Hartman wrote:
> On Sat, Apr 18, 2015 at 08:02:24PM +0100, Russell King - ARM Linux wrote:
> > On Sat, Apr 18, 2015 at 08:47:13PM +0200, Greg Kroah-Hartman wrote:
> > > On Sat, Apr 18, 2015 at 10:04:20AM +0100, Russell King - ARM Linux wrote:
> > > > On Sat, Apr 18, 2015 at 10:57:12AM +0200, Greg Kroah-Hartman wrote:
> > > > > On Fri, Apr 17, 2015 at 09:50:56AM +0200, Jens Wiklander wrote:
> > > > > > +struct tee_device {
> > > > > > +	char name[TEE_MAX_DEV_NAME_LEN];
> > > > > > +	const struct tee_desc *desc;
> > > > > > +	struct device *dev;
> > > > > 
> > > > > No, please embed the device in your structure, don't have a pointer to
> > > > > it.
> > > > 
> > > > Greg, "dev" here is not a locally allocated device, but the parent device.
> > > > It's actually the same as struct tee_device.miscdev.parent, which could be
> > > > used instead and this member deleted.
> > > 
> > > A miscdev doesn't need to have a "parent", it's just there to provide a
> > > character device node to userspace, not to represent a "device that you
> > > can do things with in the heirachy".
> > > 
> > > If you really want that, then use a real 'struct device' as should be
> > > done here.  Have just a pointer to a misc device, that is meant to be
> > > dynamic.
> > 
> > Let's rewind.
> > 
> > You are saying that "struct device *dev;" should be "struct device dev;"
> 
> Yes.
> 
> > I'm saying that you are mis-interpreting in your review what _that_ is.
> 
> Probably, I really have no idea what it is anymore.  What it _should_ be
> is the thing that controls the lifecycle of the structure.  Do not use a
> miscdevice for that, it will not work, as the TPM developers found out
> the hard way.

I _really_ don't understand what you're going on about.

The "struct device *dev" is a pointer to the struct device corresponding
to the _device_ which is being probed and the tee device is being
registered for - in the case of the submitted code, that is the
struct device embedded in the platform device.

This is a /really/ standard thing to do in drivers - saving a pointer
to the struct device which the driver is responsible for.

So why should this pointer become a struct device itself?

Greg, I think you have performed a disservice by poorly reviewing the
driver, and giving _incorrect_ comments.  Please can you have another
look at both patches together and provide a better review.  Thanks.


Second point _against_ embedding a struct device here - a struct device
is exposed to userspace.  Why expose this to userspace - we have other
ways to manage the lifetime of data structures, such as krefs, which
are not exposed to userspace.  What's wrong with using a kref to
control the lifetime of this structure?

-- 
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.