From: Frederic Weisbecker on
On Tue, Apr 13, 2010 at 08:01:42PM -0700, Joe Perches wrote:
> On Tue, 2010-04-13 at 22:44 -0400, Eric Paris wrote:
> > On Tue, 2010-04-13 at 18:33 -0700, Joe Perches wrote:
> > > On Tue, 2010-04-13 at 21:13 -0400, Eric Paris wrote:
> > > > Patch ef0658f3de484bf9b173639cd47544584e01efa5 changed the precision field
> > > > from and int to an s8. Problem is that we have code which uses a much larger
> > > > precision in the kernel. An example would in the audit code where we have:
> > > >
> > > > vsnprintf(...,..., " msg='%.1024s'", (char *)data);
> > > >
> > > > which causes precision to be too large and end up truncating to nothing.
> > > > Raising the size of the precision fixes the audit system issue. It also does
> > > > not affect the alignment of the struct according to pahole and is still
> > > > approprietely packed.
> > >
> > > I don't see how it could be appropriately packed.
> >
> > I was just saying there was no padding inside the struct, although you
> > are right about it now being longer than 64.
>
> Which is bad.
>
> > But what does __attribute__((packed)) buy us?
>
> It could force the size to be 64 bits on more platforms.
>
> > I'll gladly resend with u8 type and s16 precision if that's the best
> > solution.
>
> Reordering struct members to keep width and precision
> together seems appropriate. The attribute may not be.
>
> struct printf_spec {
> u8 type;
> u8 flags; /* flags to number() */
> u8 base;
> u8 qualifier;
> s16 field_width; /* width of output field */
> s16 precision; /* # of digits/chars */
> };


Yeah, we should avoid the attribute. Packing struct should be
done for pretty special cases, not to fix padding holes, because
the hole problem would be turned into an alignment access unefficiency.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/