Re: drivers/media/v4l2-core/v4l2-ioctl.c:303:28: warning: taking address of packed member 'pixelformat' of class or structure 'v4l2_pix_format_mplane' may result in an unaligned pointer value

From: Nick Desaulniers
Date: Mon Jan 10 2022 - 16:23:02 EST


On Mon, Jan 10, 2022 at 12:50 PM Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
>
> On Mon, Jan 10, 2022 at 11:44:12AM -0800, Nick Desaulniers wrote:
> > On Sun, Jan 9, 2022 at 6:26 AM Andy Shevchenko
> > <andy.shevchenko@xxxxxxxxx> wrote:
> > >
> > > +Nick Desaulniers
> > >
> > > Any ideas on how to fix this?
> >
> > Not obviously.
>
> Thanks for looking into this!
>
> > This code probably will trigger a splat when run with
> > UBSan though.
> >
> > I'm curious if the structure needs to be packed for interfacing with
> > hardware or some ABI, or whether we could add an explicit alignment to
> > the member if that would be ok (which may add back some padding)?
>
> Looking into it I think removing packed attribute brings a regression
> immediately on 64-bit platforms since u32 member followed by u8.
>
> > Otherwise, I suspect to actually access this properly we may have
> > macros for performing underaligned loads and stores? I suspect you'd
> > read potentially unaligned data into an aligned copy, then do
> > operations on that, at which point printing the address of the copy is
> > legal, but perhaps useless.
>
> Current code does this:
>
> val = *fourcc & ~BIT(31);
> ...
> for (i = 0; i < sizeof(*fourcc); i++) {
> ...
> strcpy(p, *fourcc & BIT(31) ? " big-endian" : " little-endian");
> ...
> p = special_hex_number(p, output + sizeof(output) - 2, *fourcc, sizeof(u32));
>
> If unaligned access is not good it will crash on some architectures AFAIU.
>
> So, it should use get_unaligned() in the first place.

Regardless of your unaligned access patch
(https://lore.kernel.org/lkml/20220110205049.11696-1-andriy.shevchenko@xxxxxxxxxxxxxxx/),
taking the address of a packed member will still have a warning:
https://godbolt.org/z/8r1rcocPY

So it looks like packing a struct sets the alignment of members to 1,
ie. under-aligning them. You need to restore the alignment of the
members you plan to take the address of.

ie.

struct w {
char x, y, z;
} __attribute__((packed));

is equivalent in layout and alignment to:

struct w {
char __attribute__((aligned(1))) x;
char __attribute__((aligned(1))) y;
char __attribute__((aligned(1))) z;
} __attribute__((aligned(1)));

For `struct v4l2_pix_format_mplane` we probably want the trailing
char's to have alignment 1, rather than every member. We can either be
explicit with __aligned(1) on them, or __aligned(4) on the uint
members and retain __packed.

>
> > Perhaps:
> >
> > - __u32 pixelformat;
> > + __u32 pixelformat __aligned(4);
>
> This looks weird, however I can't immediately see any side effects of it.
> What if the address of the entire structure is unaligned, would we have a
> gap here? In any case I wouldn't go this way.
>
> > __u32 field;
> > __u32 colorspace;
> >
> > Perhaps we could tighten up this warning in clang; we don't have any
> > holes before this member, so I _wouldn't_ have assumed
> > __attribute__((packed)) would have caused the address of pixelformat
> > member of an instance of struct v4l2_pix_format_mplane to ever be
> > underaligned.
> >
> > > On Sun, Jan 9, 2022 at 3:53 PM kernel test robot <lkp@xxxxxxxxx> wrote:
> > > >
> > > > Hi Sakari,
> > > >
> > > > FYI, the error/warning still remains.
> > > >
> > > > tree: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> > > > head: d1587f7bfe9a0f97a75d42ac1489aeda551106bc
> > > > commit: e927e1e0f0dd3e353d5556503a71484008692c82 v4l: ioctl: Use %p4cc printk modifier to print FourCC codes
> > > > date: 11 months ago
> > > > config: mips-buildonly-randconfig-r002-20220107 (https://download.01.org/0day-ci/archive/20220108/202201081852.uTfBqS4b-lkp@xxxxxxxxx/config)
> > > > compiler: clang version 14.0.0 (https://github.com/llvm/llvm-project 32167bfe64a4c5dd4eb3f7a58e24f4cba76f5ac2)
> > > > reproduce (this is a W=1 build):
> > > > wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
> > > > chmod +x ~/bin/make.cross
> > > > # install mips cross compiling tool for clang build
> > > > # apt-get install binutils-mips-linux-gnu
> > > > # https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e927e1e0f0dd3e353d5556503a71484008692c82
> > > > git remote add linus https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> > > > git fetch --no-tags linus master
> > > > git checkout e927e1e0f0dd3e353d5556503a71484008692c82
> > > > # save the config file to linux build tree
> > > > mkdir build_dir
> > > > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=mips SHELL=/bin/bash drivers/hid/ drivers/media/v4l2-core/ fs/
> > > >
> > > > If you fix the issue, kindly add following tag as appropriate
> > > > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > > >
> > > > All warnings (new ones prefixed by >>):
> > > >
> > > > >> drivers/media/v4l2-core/v4l2-ioctl.c:303:28: warning: taking address of packed member 'pixelformat' of class or structure 'v4l2_pix_format_mplane' may result in an unaligned pointer value [-Waddress-of-packed-member]
> > > > mp->width, mp->height, &mp->pixelformat,
> > > > ^~~~~~~~~~~~~~~
> > > > include/linux/printk.h:385:26: note: expanded from macro 'pr_cont'
> > > > printk(KERN_CONT fmt, ##__VA_ARGS__)
> > > > ^~~~~~~~~~~
> > > > >> drivers/media/v4l2-core/v4l2-ioctl.c:347:37: warning: taking address of packed member 'pixelformat' of class or structure 'v4l2_sdr_format' may result in an unaligned pointer value [-Waddress-of-packed-member]
> > > > pr_cont(", pixelformat=%p4cc\n", &sdr->pixelformat);
> > > > ^~~~~~~~~~~~~~~~
> > > > include/linux/printk.h:385:26: note: expanded from macro 'pr_cont'
> > > > printk(KERN_CONT fmt, ##__VA_ARGS__)
> > > > ^~~~~~~~~~~
> > > > >> drivers/media/v4l2-core/v4l2-ioctl.c:353:5: warning: taking address of packed member 'dataformat' of class or structure 'v4l2_meta_format' may result in an unaligned pointer value [-Waddress-of-packed-member]
> > > > &meta->dataformat, meta->buffersize);
> > > > ^~~~~~~~~~~~~~~~
> > > > include/linux/printk.h:385:26: note: expanded from macro 'pr_cont'
> > > > printk(KERN_CONT fmt, ##__VA_ARGS__)
>
> --
> With Best Regards,
> Andy Shevchenko
>
>


--
Thanks,
~Nick Desaulniers