Re: [PATCH] UBIFS: endian handling fixes and annotations

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

On Mon, 2008-11-24 at 16:19 +0200, Adrian Hunter wrote:
> Sebastian Andrzej Siewior wrote:
> > * Artem Bityutskiy | 2008-11-21 19:19:24 [+0200]:
> > 
> >> index 9ee6508..3f1f16b 100644
> >> --- a/fs/ubifs/key.h
> >> +++ b/fs/ubifs/key.h
> >> @@ -345,7 +345,7 @@ static inline int key_type_flash(const struct ubifs_info *c, const void *k)
> >> {
> >> 	const union ubifs_key *key = k;
> >>
> >> -	return le32_to_cpu(key->u32[1]) >> UBIFS_S_KEY_BLOCK_BITS;
> >> +	return le32_to_cpu(key->j32[1]) >> UBIFS_S_KEY_BLOCK_BITS;
> > 
> > If you would change such references to something like
> > |return le32_to_cpup(&key->j32[1]) >> UBIFS_S_KEY_BLOCK_BITS;
> > then on powerpc
> > 
> >   text    data     bss     dec     hex filename
> > 155384    1284      24  156692   26414 ubifs-b4.ko
> > 155372    1284      24  156680   26408 ubifs-after.ko
> > 
> > because now it is possible to load the value as LE from memory instead
> > of loading it BE and swapping it afterwads.
> 
> Wouldn't that be true for every le32_to_cpu of an lvalue?  Shame you can't
> do:
> 
> is_lvalue(x) ? le32_to_cpup(&(x)) : le32_to_cpu(x)
> 

No, you wouldn't want to do the above if the lvalue was on the stack as most
of the time the extra code to setup a pointer to a stack variable ends up
being more expensive than just using cpu_to_le32.

Cheers,

Harvey