Re: byteorder helpers and void * (was: Re: [PATCH 01/21] lib: addbyteorder helpers for the aligned case)

[Harvey Harrison]

On Tue, 2008-06-24 at 14:54 +0200, Geert Uytterhoeven wrote:
> > 
> > Document the fact that void * passed in needs to be 16-bit aligned?
> 
> Why not let it just take a __le16 *? Because in many use-cases the pointer just
> points to an array of bytes?
> 
> For the unaligned case, e.g. get_unaligned_le16(), I can understand a bit the
> rationale about using void * (a typical use-case is accessing a little endian
> 16-bit value in the middle of an arrays of bytes).
> 
> However, a disadvantage is that you remove the ability of the compiler to check
> for using the wrong accessor in a (packed for the unaligned case) struct, e.g.
> 
>     struct {
> 	u8 pad;
> 	__le16 val;			/* 16-bit value */
>     } __attribute ((packed)) s;
> 
>     x = get_unaligned_le32(&s.val);	/* oops, 32-bit access */
> 

I'm starting to come around to the typechecking argument.  This would
also be a chance to fix the argument ordering in put_analigned_XXXX
that was noticed by others.  As there are already some existing users
in-tree, we could transition gradually by:

1) Introduce typed versions of get/put_unaligned_XXXX, that implies the
byteswap better:
u16 load_unaligned_le16(__le16 *)
void store_unaligned_le16(__le16 *, u16)

Then the aligned helpers could be:
le16_to_cpup -> aligned equivalent of load_unaligned_le16
store_le16(__le16 *, u16)

Implemented as (to allow constant folding)
#define store_le16(ptr, val)	(*(__le16 *)(ptr) = cpu_to_le16((u16)(val)))

> I noticed there's also a __get_unaligned_le(), which uses compile-time
> detection of the pointer time, to make sure the correct accessor is used.
> Do you intend this to be used by generic code? It's function name starts
> with double underscore, indicating otherwise.

It is not meant for generic use, it is just there as a helper for each
arch to wire up it's get_unaligned() macro depending on its endianness,
so each arch doesn't wire up its own version that may or may not have
the size checking.

Anything I missed?

Harvey

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