Re: [PATCH 1/2] resource: Add device-managed request/release_resource()

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3035 bytes --]

On Tue, Jul 22, 2014 at 03:01:20PM -0400, Tejun Heo wrote:
> Hello,
> 
> On Tue, Jul 22, 2014 at 12:50:02PM -0600, Bjorn Helgaas wrote:
> > [+cc Tejun, LKML]
> > 
> > On Fri, Jul 11, 2014 at 09:08:24AM +0200, Thierry Reding wrote:
> > > From: Thierry Reding <treding@nvidia.com>
> > > 
> > > Provide device-managed implementations of the request_resource() and
> > > release_resource() functions. Upon failure to request a resource, the
> > > new devm_request_resource() function will output an error message for
> > > consistent error reporting.
> > > 
> > > Signed-off-by: Thierry Reding <treding@nvidia.com>
> > 
> > This seems OK to me, but I don't consider myself a devres maintainer.  I
> > added Tejun and LKML for any comment.  Minor nit below.
> 
> If there are gonna be users of the interface, sure.
> 
> > > +int devm_request_resource(struct device *dev, struct resource *root,
> > > +			  struct resource *new)
> > > +{
> > > +	struct resource *conflict, **ptr;
> > > +
> > > +	ptr = devres_alloc(devm_resource_release, sizeof(*ptr), GFP_KERNEL);
> > > +	if (!ptr)
> > > +		return -ENOMEM;
> > > +
> > > +	*ptr = new;
> > > +
> > > +	conflict = request_resource_conflict(root, new);
> > > +	if (!conflict) {
> > > +		devres_add(dev, ptr);
> > > +		return 0;
> > > +	}
> > > +
> > > +	dev_err(dev, "resource collision: %pR conflicts with %s %pR\n", new,
> > > +		conflict->name, conflict);
> > > +	devres_free(ptr);
> > > +	return -EBUSY;
> > 
> > Personally I would write this as:
> > 
> >   conflict = request_resource_conflict(...);
> >   if (conflict) {
> >     dev_err(...);
> >     devres_free(...);
> >     return -EBUSY;
> >   }
> > 
> >   devres_add(...);
> >   return 0;
> > 
> > so the straight-line path is the normal, non-error path and errors are
> > detected and dealt with in the "if" bodies.  Right now the "if" bodies
> > are a mix of error handling and normal path.  But that's just my personal
> > preference.
> 
> Agreed.
> 
> > > +static int devm_resource_match(struct device *dev, void *res, void *data)
> > > +{
> > > +	struct resource **ptr = res;
> > > +
> > > +	if (WARN_ON(!ptr || !*ptr))
> > > +		return 0;
> 
> How would !ptr or !*ptr possibly happen?  Wouldn't that be a bug
> already?

Honestly, I copied that from similar implementations. But checking the
code again, I don't think they can actually happen. The value returned
by devres_alloc() is a struct devres * with an added offset so that it
points at the payload immediately following the struct devres. So at
least !ptr can never happen.

!*ptr could happen since the devres code calls the match function on
every resource managed for the device and someone could've inserted a
NULL (or 0) value. But since we're not dereferencing *ptr this should
not be an issue. That said, having either res or data above be NULL
isn't something that devres was meant to deal with anyway, since it
relies on the pointers being unique for the matching.

Thierry

[-- Attachment #2: Type: application/pgp-signature, Size: 819 bytes --]