On Mon, 30 Dec 2013, Johannes Berg wrote:
> On Mon, 2013-12-30 at 20:58 +0100, Julia Lawall wrote:
> > > Is there any way we could catch (sparse, or some other script?) that
> > > struct reorganising won't break the condition needed ("within a
> > > structure that contains at least two more bytes")?
> >
> > What kind of reorganizing could happen? Do you mean that the programmer
> > might do at some time in the future, or something the compiler might do?
>
> I'm just thinking of a programmer, e.g. changing a struct like this:
>
> struct foo {
> u8 addr[ETH_ALEN];
> - u16 dummy;
> };
>
> for example.
That is easily resolved by:
struct foo {
u8 addr[ETH_ALEN];
u16 required_padding; /* do not remove upon pain of death */
};
Preferably with a comment nearby explaining just why that padding is so
important in the first place.
Unfortunately, we do have a lot of code with unexplained padding that one
should never remove, and it is often nowhere nearly as obvious as stray
elements with idiotic non-explanatory names in a structure.
One example is the via-rng hw_random driver, where the buffer used by the
"xstore" via-specific instriction *MUST* be longer than what the "xstore"
instruction is documented to require due to CPU errata (you ask it to write
from 1 to 8 bytes, and it may end up writing 16 bytes). The xstore buffer
is an array that was made long enough to avoid the errata, but that fact is
documented only in a git commit message.
--
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot
Henrique Holschuh
--
To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
