Re: [PATCH 1/2] byteorder: add load/store_{endian} API

[Harvey Harrison <harvey.harrison@gmail.com>]

On Mon, 2009-03-02 at 17:51 -0800, Linus Torvalds wrote:
> 
> On Mon, 2 Mar 2009, Harvey Harrison wrote:
> > 
> > OK, static inline it is then. Would you be opposed to an API like:
> > 
> > get_le16
> > put_le16
> > 
> > to match with 
> >
> > get_unaligned_le16
> > put_unaligned_le16
> > 
> > And make the existing unaligned helpers typesafe?
> 
> That sounds much better to me. That said, I'm also wondering what the 
> upside is of this all?
> 

1) Recognize that some drivers/subsystems want this, or already
implement it privately

2) [micro-optimization] Allow arches that do provide load/store swapped
instructions to be used more often.

3) make it more likely that people will actually use the unaligned
helpers rather than open-coding the byteswapping, allowing arches that
have no alignment constraints to just do regular loads if possible.

Disadvantages:

1) existing users of the get_unaligned bits may/will produce sparse
warnings on the flag day

2) the existing argument ordering of put_unaligned is opposite what
is usually expected in such a helper

3) [nitpicky] get/put almost always means reference taking/releasing,
load/store is the usual verb used for such an API.

Harvey