powerpc/powernv: add hdat attribute to sysfs

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 628

You should pretty much never have a blank line in your variable declarations.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 632

It's traditional to call these "np" for "node pointer" (or less commonly "dn").

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 636

It's usually better to delay initialisation of variables until when they're needed, so in this case just prior to the loop that uses them. Also notice attr_count is only used prior to the use of n, so you could just use n for both. Though see below.

You should also group common types on one line when possible. And I and some other folks with good taste, like the longest variable lines first followed by the shorter ones, so in this case eg:

	struct bin_attribute *exported_attrs, *attr_tmp;
	struct kobject *opal_export_kobj;
	struct device_node *np;
	struct property *prop;
	int rc, attr_count, n;
	const __be64 *syms;
	unsigned int size;
	char **attr_name;

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 639

Now that this is a local it doesn't need such a verbose name, it could just be kobj.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 650

So here we count the properties of the node.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 655

And here we alloc space for that number - 2.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 658

But this loop does no checking vs attr_count. I know you have the check for "name" and "phandle" below, but if ever you didn't have those properties this loop would overflow the end of exported_attrs and corrupt something else on the heap.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 660

Here we allocate memory for the property name ...

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 665

But then here we don't free the attr_name we just kstrdup'ed.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 668

And same here, as well as we've now allocated an extra bin_attr in exported_attrs that we'll never use.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 675

The actual reading of the property can be done in a cleaner way using one of the helpers, eg:

	u64 vals[2];
	...
	if (of_property_read_u64_array(np, prop->name, &vals, 2))
		continue.

And if it succeeds then vals has the two values in CPU endian.

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 670

Notice here we use exported_attrs, ...

arch/powerpc/platforms/powernv/opal.c static void opal_export_symmap(void) · line 671

And here we use attr_name.

But we never use them again. We do need to keep the bin_attr alive, we gave it to sysfs_create_bin_file() after all, but we don't actually need an array of them. And attr_name is completely unnecessary once we've assigned the pointer to attr.name.

So the solution is to drop the allocation of exported_attrs and attr_name prior to the loop.

Then in the loop we do:

for_each_property_of_node(fw, prop)

Check for name or phandle, else continue
Read the property value using read_u64_array, else continue Allocate one attr else warn and continue
Strdup the prop->name into attr->name, else kfree(attr), warn, continue Intialise the attr
Call sysfs_create_bin_file(), if it fails, warn and kfree(attr)

cheers