From: Ursula Braun [mailto:ubr...@linux.vnet.ibm.com]
> Sent: 20 August 2015 12:44
> On Thu, 2015-08-20 at 10:46 +0000, David Laight wrote:
> > From: Ursula Braun
> > > Sent: 19 August 2015 09:21
> > > In little endian cases, the macro htons unfolds to __swab16 which
> > > provides special case for constants. In big endian cases,
> > > __constant_htons and htons expand directly to the same expression.
> > > So, replace __constant_htons with htons with the goal of getting
> > > rid of the definition of __constant_htons completely.
> > ...
> > > diff --git a/drivers/s390/net/qeth_l3_main.c
> > > b/drivers/s390/net/qeth_l3_main.c
> > > index 70eb2f6..ecfe622 100644
> > > --- a/drivers/s390/net/qeth_l3_main.c
> > > +++ b/drivers/s390/net/qeth_l3_main.c
> > > @@ -1887,13 +1887,13 @@ static inline int qeth_l3_rebuild_skb(struct
> > > qeth_card *card,
> > > case QETH_CAST_MULTICAST:
> > > switch (prot) {
> > > #ifdef CONFIG_QETH_IPV6
> > > - case __constant_htons(ETH_P_IPV6):
> > > + case htons(ETH_P_IPV6):
> >
> > I didn't think htons() was 'constant enough' to be used as a case label.
> >
> > Using byteswapped constants in a case statement can change it from being
> > implemented as a jump table to a branch tree.
> > This might be more expensive than byteswapping the value (even on systems
> > that don't have cheap byteswap instructions).
> >
> > David
> >
> For big endian systems both __constant_htons(x) and htons(x) are
> resolved to ((__force __be16)(__u16)(x)). Thus I do not see a reason to
> reject the patch proposal from Vaishali Thakkar.
Look at a little-endian one (eg amd64).
I think you'll find a C ?: expression that uses __builtin_constant() to
select between an expression the compiler can evaluate and a call to an
inline function that uses some appropriate asm.
The latter isn't a compile-time constant so can't be used as a case
label or as an initialiser.
David
