RE: [PATCH v10 13/15] platform/x86: wmi: create userspace interface for drivers

[Mario.Limonciello]

> -----Original Message-----
> From: Christoph Hellwig [mailto:hch@xxxxxx]
> Sent: Friday, October 20, 2017 9:49 AM
> To: Limonciello, Mario <Mario_Limonciello@xxxxxxxx>
> Cc: greg@xxxxxxxxx; dvhart@xxxxxxxxxxxxx; andy.shevchenko@xxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; platform-driver-x86@xxxxxxxxxxxxxxx; luto@xxxxxxxxxx;
> quasisec@xxxxxxxxxx; pali.rohar@xxxxxxxxx; rjw@xxxxxxxxxxxxx;
> mjg59@xxxxxxxxxx; hch@xxxxxx; gnomes@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [PATCH v10 13/15] platform/x86: wmi: create userspace interface for
> drivers
> 
> On Fri, Oct 20, 2017 at 01:54:36PM +0000, Mario.Limonciello@xxxxxxxx wrote:
> > > -----Original Message-----
> > > From: Greg KH [mailto:greg@xxxxxxxxx]
> > > Sent: Friday, October 20, 2017 8:22 AM
> > > To: Limonciello, Mario <Mario_Limonciello@xxxxxxxx>
> > > Cc: dvhart@xxxxxxxxxxxxx; Andy Shevchenko <andy.shevchenko@xxxxxxxxx>;
> > > LKML <linux-kernel@xxxxxxxxxxxxxxx>; platform-driver-x86@xxxxxxxxxxxxxxx;
> Andy
> > > Lutomirski <luto@xxxxxxxxxx>; quasisec@xxxxxxxxxx; pali.rohar@xxxxxxxxx;
> > > rjw@xxxxxxxxxxxxx; mjg59@xxxxxxxxxx; hch@xxxxxx; Alan Cox
> > > <gnomes@xxxxxxxxxxxxxxxxxxx>
> > > Subject: Re: [PATCH v10 13/15] platform/x86: wmi: create userspace interface
> for
> > > drivers
> > >
> > > On Thu, Oct 19, 2017 at 12:50:16PM -0500, Mario Limonciello wrote:
> > > > +	wblock = container_of(wdev, struct wmi_block, dev);
> > > > +	if (!wblock)
> > > > +		return -ENODEV;
> > >
> > > How can container_of() ever return NULL?  If so, you have a very odd
> > > memory layout...
> > >
> >
> > I'm assuming this is from set_required_buffer_size right?
> >
> > The symbol is exported out for other drivers to use.  It's possible for another
> > driver to allocate a wmi_device structure that's not part of a wblock.
> 
> container_of can never return NULL, it does arithmetics on a pointer
> based on the type it is embedded into.
> 
> You better don't register a wmi_device that's not part of the block
> with your driver.  Which others drivers are those, btw?

No drivers do this today, it's obviously not a good idea.
I was just saying it's hypothetical.

I see that the other methods exported (wmi_evaluate_method and such) to drivers 
require that it's part of a wblock, so this seems like a reasonable expectation
from other drivers.  I'll remove this invalid check.

> > This one it's possible that a driver isn't bound to a device, and when
> > that happens wdriver is NULL.

 
> See above, no it can't.  Maybe wblock->dev.dev.driver can be NULL,
> but in that case you must not call container_of on it.

> container_of() is just pointer math.  If you pass in NULL, you will get
> a non-NULL value (incremented or decremented).  If you pass in a very
> tiny number, you might get NULL, but that's still really wrong.
> 
> In other words, these tests will _NEVER_ fail.  Go ahead, try it :)

I was seeing failures (with NULL) when I tested with some drivers unbound, 
but I now understand my check is definitely wrong.  I'll adjust the check and make 
sure it's valid.

Thank you both for your feedback here.