Date: Tue, 12 Dec 2017 06:20:02 +0000
From: Al Viro <viro@ZenIV.linux.org.uk>
To: Jakub Kicinski <kubakici@wp.pl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [RFC][PATCH] new byteorder primitives - ..._{replace,get}_bits()
Message-ID: <20171212062002.GY21978@ZenIV.linux.org.uk>
References: <20171210045326.GO21978@ZenIV.linux.org.uk>
 <420a198d-61f8-81cf-646d-10446cb41def@synopsys.com>
 <20171211050520.GV21978@ZenIV.linux.org.uk>
 <20171211053803.GW21978@ZenIV.linux.org.uk>
 <20171211155422.GA12326@ZenIV.linux.org.uk>
 <20171211200224.23bc5df4@cakuba.netronome.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20171211200224.23bc5df4@cakuba.netronome.com>
User-Agent: Mutt/1.9.0 (2017-09-02)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1671
Lines: 36

On Mon, Dec 11, 2017 at 08:02:24PM -0800, Jakub Kicinski wrote:
> On Mon, 11 Dec 2017 15:54:22 +0000, Al Viro wrote:
> > Essentially, it gives helpers for work with bitfields in fixed-endian.
> > Suppose we have e.g. a little-endian 32bit value with fixed layout;
> > expressing that as a bitfield would go like
> > 	struct foo {
> > 		unsigned foo:4;		/* bits 0..3 */
> > 		unsigned :2;
> > 		unsigned bar:12;	/* bits 6..17 */
> > 		unsigned baz:14;	/* bits 18..31 */
> > 	}
> > Even for host-endian it doesn't work all that well - you end up with
> > ifdefs in structure definition and generated code stinks.  For fixed-endian
> > it gets really painful, and people tend to use explicit shift-and-mask
> > kind of macros for accessing the fields (and often enough get the
> > endianness conversions wrong, at that).  With these primitives
> > 
> > struct foo v		<=>	__le32 v
> > v.foo = i ? 1 : 2	<=>	v = le32_replace_bits(v, i ? 1 : 2, 0, 4)
> > f(4 + v.baz)		<=>	f(4 + le32_get_bits(v, 18, 14))
> 
> Looks very useful.  The [start bit, size] pair may not land itself
> too nicely to creating defines, though.  Which is why in
> include/linux/bitfield.h we tried to use a shifted mask and work
> backwards from that single value what the start and size are.  commit
> 3e9b3112ec74 ("add basic register-field manipulation macros") has the
> description.  Could a similar trick perhaps be applicable here?

Umm...  What's wrong with

#define FIELD_FOO 0,4
#define FIELD_BAR 6,12
#define FIELD_BAZ 18,14

A macro can bloody well expand to any sequence of tokens - le32_get_bits(v, FIELD_BAZ)
will become le32_get_bits(v, 18, 14) just fine.  What's the problem with that?