Re: [PATCH] staging: wlan-ng: Replace zero-length arrays with DECLARE_FLEX_ARRAY() helper

[Deepak R Varma]

On Thu, Nov 17, 2022 at 07:03:21PM +0100, Greg Kroah-Hartman wrote:
> On Thu, Nov 17, 2022 at 06:50:55PM +0530, Deepak R Varma wrote:
> > On Thu, Nov 17, 2022 at 01:54:49PM +0100, Greg Kroah-Hartman wrote:
> > > On Thu, Nov 17, 2022 at 03:48:45PM +0530, Deepak R Varma wrote:
> > > >
> > > >  struct hfa384x_pdr_refdac_measurements {
> > > > -	u16 value[0];
> > > > +	DECLARE_FLEX_ARRAY(u16, value);
> > > >  } __packed;
> > >
> > > Why?  This structure is never used anywhere, right?  So why is this
> > > needed to be changed and not just removed entirely?  Same for the other
> > > structures in this patch.
> >
> > Hello Greg,
> > I am unable to confirm that these structures are truly not needed in the absence
> > if a real device based testing. I could only validate that using the compile
> > build and driver loading.
>
> Think this through, if no one is actually using this structure, and it
> is of 0 size, then do you think it is being used?

Hello Greg,
I did not find any memory allocation for these zero length array structures.
Also, the union or its enclosing structure do not appear to access the members.
Hence I am leaning towards concluding that these zero length array structures do
not appear to be necessary.

There are a few other structures that are part of the same union, however, they
too do not appear to be used for accessing the memory assigned to the union [or
its enclosing structure]. I think most of the members of these unions can be
replaced by one max size structure of this union [e.g. struct
hfa384x_pdr_mkk_measurements].

Could you please comment if I am reading the code right?

For your quick reference, the zero length structure declaration are online 963
whereas the union is on line number 1080 of the file drivers/staging/wlan-ng/hfa384x.h


Thank you,
./drv


>
> > This change that I am proposing in the interim would enable the compiler to
> > protect the structure from addition of a new member below the zero length array.
>
> Why would you want to add a new member below this?  That's not what is
> happening here at all.

I came across this one old commit where such an accident happened. This is from
a recent LWN article:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e48f129c2f20

I understand the C99 now protects from such an attempt at the compile time
itself.

Thank you,
./drv


>
> Please think this through a bit more.
>
> good luck!
>
> greg k-h
>