From mboxrd@z Thu Jan  1 00:00:00 1970
From: Yoshinori Sato <ysato@users.sourceforge.jp>
Date: Tue, 05 Aug 2008 19:21:31 +0000
Subject: Re: [PATCH] sh_eth Add SH7619 support
Message-Id: <87tzdzmfno.wl%ysato@users.sourceforge.jp>
List-Id: <linux-sh.vger.kernel.org>
References: <878wvhtynu.wl%ysato@users.sourceforge.jp>
In-Reply-To: <878wvhtynu.wl%ysato@users.sourceforge.jp>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: linux-sh@vger.kernel.org

At Mon, 4 Aug 2008 04:23:35 +0900,
Paul Mundt wrote:
> 
> On Sat, Aug 02, 2008 at 11:34:21AM -0400, Yoshinori Sato wrote:
> > At Sat, 2 Aug 2008 17:30:22 +1000,
> > Simon Horman wrote:
> > > On Fri, Aug 01, 2008 at 11:32:28AM -0400, Yoshinori Sato wrote:
> > > > > > +/* CPU endian from/to EDMAC endian */
> > > > > > +#if !defined(CONFIG_CPU_SUBTYPE_SH7619)
> > > > > > +#define cpu_to_edmac(x) cpu_to_le32(x)
> > > > > > +#define edmac_to_cpu(x) le32_to_cpu(x)
> > > > > > +#else
> > > > > > +#define cpu_to_edmac(x) cpu_to_be32(x)
> > > > > > +#define edmac_to_cpu(x) be32_to_cpu(x)
> > > > > > +#endif
> > > > > > +
> > > > > >  #endif
> > > > > I think that this depends on an endian not CPU dependence.
> > > > 
> > > > I agree.
> > > > But I'm no idea.
> > > > Are you have good idea?
> > > 
> > > Iwamatsu-san,
> > > 
> > > what is the difference between CONFIG_CPU_LITTLE_ENDIAN and
> > > __LITTLE_ENDIAN__ as used in drivers/net/sh_eth.h ?
> > 
> > EDMAC endian is no depend CPU endian.
> > 
> The cleaner thing to do might just be to introduce a config symbol for
> specifying whether the EDMAC is in big or little endian mode, depending
> on how it's wired up in the CPU. Also, you should use a consistent ifdef
> style. In some cases you only check if 7619 is defined and do your corner
> cases there, in others (like this one) you check if it is not defined and
> push its corner cases further down. The latter of which is very error
> prone.
> 
> If the EDMAC endianness can be changed within a given subtype (or hooked
> up externally), then we also need to making the conversion run-time
> configurable, and we can toss the endian hint in the platform data.

I think it's best solution.

example

struct sh_eth_drvadata drvdata {
       .phy = 1,
       .edmac_endian = EDMAC_LITTLE, /* or EDMAC_BIG */
};

static struct platform_device eth_device = {
 :
        .dev = {
                .platform_data = &drvdata,
        },
 :
};

It's OK?

-- 
Yoshinori Sato
<ysato@users.sourceforge.jp>