Re: [RFC PATCH 3/5] i2c: core: add function to request an alias

From: Luca Ceresoli <luca@lucaceresoli.net>
To: Wolfram Sang <wsa@the-dreams.de>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>,
	Wolfram Sang <wsa+renesas@sang-engineering.com>,
	linux-i2c@vger.kernel.org, linux-renesas-soc@vger.kernel.org,
	Kieran Bingham <kieran@ksquared.org.uk>,
	Jacopo Mondi <jacopo@jmondi.org>,
	Vladimir Zapolskiy <vz@mleia.com>
Subject: Re: [RFC PATCH 3/5] i2c: core: add function to request an alias
Date: Thu, 2 Jan 2020 23:27:57 +0100	[thread overview]
Message-ID: <cc2a10ab-9f05-2c61-3a37-0e5e0184e379@lucaceresoli.net> (raw)
In-Reply-To: <20200102211327.GB1030@kunai>

Hi Wolfram,

On 02/01/20 22:13, Wolfram Sang wrote:
> Hi Luca,
> 
>>> This looks quite inefficient, especially if the beginning of the range
>>> is populated with devices. Furthermore, I think there's a high risk of
>>> false negatives, as acquiring a free address and reprogramming the
>>> client to make use of it are separate operations.
>>
>> Right. Applying the alias could raise other errors, thus one would need
>> i2c_new_alias_device() to keep the alias locked until programming it has
>> either failed or has been successfully programmed.
> 
> Please see my reply to Laurent, I don't think it is racy. But please
> elaborate if you think I am wrong.

Uhm, you are right here, it's not racy. Sorry, I had read the code
quickly and didn't notice the i2c_new_dummy_device() call.

So this means if i2c_new_alias_device() succeeds but the caller later
fails while applying the alias, then it has to call
i2c_unregister_device() to free the alias. Correct?

>>> What happened to the idea of reporting busy address ranges in the
>>> firmware (DT, ACPI, ...) ?
>>
>> Indeed that's how I remember it as well, and I'm a bit suspicious about
>> sending out probe messages that might have side effects (even if the
>> false negative issue mentioned by Laurent were solved). You know, I've
>> been taught to "expect the worse" :) so I'd like to better understand
>> what are the strong reasons in favor of probing, as well as the
>> potential side effects.
> 
> As I said to Laurent, too, I think the risk that a bus is not fully
> described is higher than a device which does not respond to a read_byte.
> In both cases, we would wrongly use an address in use.

OK, I'm still uncomfortable with sending unexpected transactions to the
dark outer space, but this is more a feeling than based on facts, and
you know more than me, so I guess I can live with that.

> Also, all the best for you in 2020!

Thanks. Best wishes to you too for the new year!

-- 
Luca