Re: [PATCH 2/4] soundwire: intel: skip suspend/resume/wake when link was not started

[Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>]

On 8/6/21 8:24 AM, Vinod Koul wrote:
> On 02-08-21, 08:59, Pierre-Louis Bossart wrote:
>>
>>
>>
>>>> On some HDaudio platforms, SoundWire devices are described in the
>>>> DSDT but never used. This patch adds a boolean status flag to skip all
>>>> suspend/resume/wake sequences for this configuration.
>>>
>>> Why are the sdw devices created in this case then? I would assume you
>>> are detecting this configuration and should skip device creation?
>>
>> The SDW Linux devices are created during the probe step, based on
>> information extracted from platform firmware. Since we see a matching
>> driver, there will be a probe and that driver also contains pm ops.
>>
>> We only know if the physical peripherals described in ACPI are real or
>> not during the startup() phase. After the bus reset, SoundWire
>> peripherals will report as ATTACHED as Device0 and the enumeration starts.
>>
>> So in these HDaudio cases, we create the Linux devices based on
>> incorrect ACPI information, but since we detect an HDaudio configuration
>> we never start the links and the suspend-resume fails.
>>
>> For a bit of historical context, the decision to handle devices in this
>> way was not the original proposal from Intel. In the initial patches,
>> the Linux devices were created when their matching physical peripheral
>> was showing signs of activity and attached after synchronizing. We
>> modified this behavior so that a device driver could use out-of-band
>> signaling to power-up a peripheral so that it could attach. That wasn't
>> a bad idea, but that also exposes a number of 'ghost devices' that are
>> not real. And unfortunately the Intel BIOS reference keeps using those
>> invalid _ADR values...
>>
>> Does this answer to the question?
> 
> Yes it does thanks, very helpful.
> 
> Can we add this to the changelog, am sure down the line people might
> forget why it was added.

yes, will do.