Re: [PATCH v2] device property: Get rid of union aliasing

[Lukas Wunner <lukas@wunner.de>]

On Mon, May 14, 2018 at 03:48:09PM +0300, Andy Shevchenko wrote:
> On Mon, 2018-05-14 at 14:18 +0200, Lukas Wunner wrote:
> > On Tue, May 08, 2018 at 04:15:47PM +0300, Andy Shevchenko wrote:
> > > --- a/drivers/firmware/efi/apple-properties.c
> > > +++ b/drivers/firmware/efi/apple-properties.c
> > > @@ -13,6 +13,9 @@
> > > + * FIXME: The approach is still based on union aliasing and should be
> > > + * replaced by a proper resource provider.
> > 
> > Why?  All Apple EFI properties are either boolean or u8 arrays.
> > You've correctly changed this file to always supply u8 arrays,
> > so I don't see where union aliasing is happening here?
> 
> Okay, for now I can see only Thunderbolt user of these properties (is it
> correct?) in upstream which uses u8 arrays indeed.

That is correct, thunderbolt.ko is so far the only user.


> Though the implementation is quite fragile in this sense, because it
> doesn't discourage people to use device_property_read_string() in case
> when it's indeed a string (I saw these kind of properties in the very
> dump you posted on your GH page).

Well if that is your concern then you need to prevent functions which
retrieve properties to use the wrong type.

E.g. to prevent retrieval of the u8 array as string, you'd have to
amend drivers/base/property.c:pset_prop_read_string_array() to
check the type of the property found and return -EINVAL if it's not
string.

Thanks,

Lukas