Re: drivers/block/ub.c

[Pete Zaitcev]

On Sun, 27 Jun 2004 11:19:38 -0400 (EDT)
Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:

> My favorite approach has always been:
> 
> 	put_be32(cmd->cdb + 2, block);
> 
> Unfortunately there is no such function or macro!  It's easy to define an
> inline function that would carry out the series of four single-byte
> assignments that originally started this discussion.  A more sophisticated
> implementation would expand to Andries' unspeakably ugly code on
> big-endian platforms that don't impose a large penalty for non-aligned
> 4-byte accesses.  I leave it up to others to decide which is best on
> little-endian platforms that can do unaligned accesses.
> 
> I think it would be great if some such utility routine were added to a
> standard header in the kernel, together with its siblings put_le32(),
> put_be16(), put_le16(), and the corresponding get_xxxx() functions.

This is very nice and would be a great help for Infiniband developers.
However, parts of SCSI commands are not defined in terms of C structures
or even 32 bit words with an endianness. They are streams of bytes, at
least historically. Please kindly refer to the WRITE(6) command for
the evidence. You'd need put_be20() to form that block address. :-)

I write these byte marshalling sequences since 1985 and I'm a little
used to them. I do not recall thinking twice about writing that element
of ub. It probably doesn't deserve all the tempest Oliver raised over it.

-- Pete
-
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/