Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <6a9ad9b1-a6f2-c380-77a9-f10317f7054f@linux.intel.com>
Date:   Wed, 24 Aug 2022 09:42:22 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Firefox/91.0 Thunderbird/91.11.0
Subject: Re: [PATCH 0/3] soundwire: allocate device_number with IDA
Content-Language: en-US
To:     Vinod Koul <vkoul@kernel.org>,
        Bard Liao <yung-chuan.liao@linux.intel.com>
Cc:     alsa-devel@alsa-project.org, linux-kernel@vger.kernel.org,
        bard.liao@intel.com
References: <20220823045004.2670658-1-yung-chuan.liao@linux.intel.com>
 <YwUHp3ivSktx6/Lo@matsya>
From:   Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
In-Reply-To: <YwUHp3ivSktx6/Lo@matsya>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi Vinod,

>> The device_number is currently allocated in the scope of each bus and does
>> not need to be unique at the system level.
>> This leads e.g. on Dell devices with three or four Device1 on different
>> bus segments. 
> 
> And how is that an issue that should be fixed?

1. this makes things simpler for debug. You can look at the logs and see
what the device is without having to look at the link. It's been much
easier to track complicated issues when each device is tracked with a
unique system level number.

2. Intel hardware has a requirement that the device number be unique for
all links managed by  a given controller. This patch prepares that
transition. I cannot disclose more details at the moment, you will have
to trust me on this one.

>> To make the device_number unique at the system level, and
> 
> Why should we do that...
>> unified with the HDaudio/iDISP SDI values, this series allocates the
>> dev_number with an IDA restricted between 4 and 11 (inclusive).
> 
> Does this not place an artificial restriction that a system can have
> only 12 devices if we have unique device

It does indeed, but it simplifies debug and it allows future hardware to
be supported.

In practice, the most we've seen on Intel platforms is 4 devices at the
system level. Even if the SoundWire spec does allow for 11 devices per
link, it's way over-engineered due to capacitive load and signal
integrity issues. We've seen an absolute maximum of 4 devices on a
single link in test rigs, and that was never deployed. Two devices per
link were also only used in prototypes which never hit the market.

At any rate, this in an opt-in solution, not a requirement for non-Intel
platforms at all. Others can keep using as many devices as they wish,
within the bounds allowed by the standard.

Note that the device_number creates a de-facto priority, and it would be
perfectly acceptable for some platforms to tweak the current first-come
first-serve allocation to improve interrupt response time on each link,
etc. The standard says nothing about how the device numbers should be
allocated, only that they be unique on each link. In other words, you
should expect additional changes to the existing way of allocating
device numbers.

> Perhaps a better way would be to use dus:dev combination while dealing
> with a device always..??

That is not compatible with Intel hardware requirements. This wasn't a
decision where Bard and I were consulted, it is what it is.

Does this clarify things?
Thanks
-Pierre