[U-Boot] [PATCH v5] bitops: introduce BIT() definition

[Scott Wood]

On Wed, 2015-09-09 at 12:37 -0400, Tom Rini wrote:
> On Wed, Sep 09, 2015 at 11:22:25AM -0500, Scott Wood wrote:
> > On Tue, 2015-09-08 at 21:01 +0300, ivan.khoronzhuk wrote:
> > > Hi, Andreas
> > > 
> > > On 07.09.15 14:43, Andreas Bießmann wrote:
> > > > From: Heiko Schocher <hs at denx.de>
> > > > 
> > > > introduce BIT() definition, used in at91_udc gadget
> > > > driver.
> > > > 
> > > > Signed-off-by: Heiko Schocher <hs at denx.de>
> > > > [remove all other occurrences of BIT(x) definition]
> > > > Signed-off-by: Andreas Bießmann <andreas.devel at googlemail.com>
> > > > ---
> > > > Full buildman is running
> > > > 
> > > 
> > > ....
> > > 
> > > > 
> > > > +#define BIT(nr)            (1UL << (nr))
> > > 
> > > Why UL? Why not simply 1 << (nr)?
> > 
> > That would give the wrong result for nr == 31 if used as a 64-bit number, 
> > and 
> > would produce undefined behavior for nr >= 32 (though even with 1UL that 
> > would be undefined on 32-bit builds).
> > 
> > > What if I need set ULL bit on 32-bit system?
> > > Thanks for explanation.
> > 
> > Yes, ULL would be better.
> 
> That would be BIT_ULL(nr) ?  I want to assume that there was some care
> given upstream here.  It was about 2 years ago now the kernel added a
> specific BIT_ULL and family in addition to BIT(nr) from back in 2007.

A quick search didn't turn up much justification for keeping them separate 
(and it seems like using BIT where BIT_ULL is needed could be a source of 
difficult bugs), but sure, we don't want to encourage writing driver code 
that will break on Linux.

-Scott