From mboxrd@z Thu Jan  1 00:00:00 1970
Return-path: <linux-media-owner@vger.kernel.org>
Received: from zone0.gcu-squad.org ([212.85.147.21]:38944 "EHLO
	services.gcu-squad.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1754800Ab1KIKwP (ORCPT
	<rfc822;linux-media@vger.kernel.org>); Wed, 9 Nov 2011 05:52:15 -0500
Date: Wed, 9 Nov 2011 11:52:04 +0100
From: Jean Delvare <khali@linux-fr.org>
To: Antti Palosaari <crope@iki.fi>
Cc: Mauro Carvalho Chehab <mchehab@redhat.com>,
	linux-media <linux-media@vger.kernel.org>,
	Michael Krufky <mkrufky@kernellabs.com>
Subject: Re: [RFC 1/2] dvb-core: add generic helper function for I2C
 register
Message-ID: <20111109115204.401a8aa5@endymion.delvare>
In-Reply-To: <4EBA58E0.8080704@iki.fi>
References: <4EB9C13A.2060707@iki.fi>
	<4EBA4E3D.80105@redhat.com>
	<4EBA58E0.8080704@iki.fi>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-media-owner@vger.kernel.org
List-ID: <linux-media.vger.kernel.org>

On Wed, 09 Nov 2011 12:41:36 +0200, Antti Palosaari wrote:
> On 11/09/2011 11:56 AM, Mauro Carvalho Chehab wrote:
> > Due to the way I2C locks are bound, doing something like the above and something like:
> >
> >      struct i2c_msg msg[2] = {
> >          {
> >              .addr = i2c_cfg->addr,
> >              .flags = 0,
> >              .buf = buf,
> >          },
> >          {
> >              .addr = i2c_cfg->addr,
> >              .flags = 0,
> >              .buf = buf2,
> >          }
> >
> >      };
> >
> >      ret = i2c_transfer(i2c_cfg->adapter, msg, 2);
> >
> > Produces a different result. In the latter case, I2C core avoids having any other
> > transaction in the middle of the 2 messages.
> 
> In my understanding adding more messages than one means those should be 
> handled as one I2C transaction using REPEATED START.
> I see one big problem here, it is our adapters. I think again, for the 
> experience I have, most of our I2C-adapters can do only 3 different 
> types of I2C xfers;
> * I2C write
> * I2C write + I2C read (combined with REPEATED START)
> * I2C read (I suspect many adapters does not support that)
> That means, I2C REPEATED writes  are not possible.

Also, some adapters _or slaves_ won't support more than one repeated
start in a given transaction, so splitting block reads in continuous
chunks won't always work either. Which makes some sense if you think
about it: if both the slave and the controller supported larger blocks
then there would be no need to split the transfer into multiple
messages in the first place.

-- 
Jean Delvare