Re: [PATCH] drivers: staging: lustre: lustre: include: add "__attribute__((packed))" for the related union

[Dan Carpenter <dan.carpenter@oracle.com>]

On Mon, Jan 20, 2014 at 12:37:57PM +0000, James Hogan wrote:
> On 20/01/14 12:30, Dan Carpenter wrote:
> > Ah.  From so metag is a new arch and not a compiler like the changelog
> > says.
> > 
> > On Mon, Jan 20, 2014 at 11:56:47AM +0000, James Hogan wrote:
> >> struct a {
> >> 	struct b {
> >> 		unsigned int x;
> >> 		unsigned short y;
> >> 	} x;
> >> 	unsigned short y;
> >> } __packed;
> > 
> > This is not the code we are discussing.  It should look like:
> > 
> > struct a {
> > 	union {
> > 		short x;
> > 		short y;
> > 	}
> > 	short z;
> > };
> > 
> > Any normal person would assume that sizeof(struct a) would be 4 but
> > apparently on metag it is 8.  That totally defeats the point of using
> > a union in the first place.  It's easy enough to add a __packed to the
> > lustre declaration but I expect this to cause an endless stream of bugs.
> > 
> > It it is really stupid.
> 
> I agree completely (and did request this be changed when I first found
> out about it, but since it's an ABI issue it was really too late).
> That's why I'm not actively pushing for every case to be fixed unless
> it's in generic code that actually affects metag.
> 

It would be easy enough to make the compiler complain about any union
which would normally have size which is not a multiple of 4.

Warning: union will be padded with 2 bytes unless __attribute__((packed)).

Otherwise you will be fighting this for ever.

regards,
dan carpenter