[PATCH 2/3] soundwire: qcom: add auto enumeration support

[Pierre-Louis Bossart]

>>> +        if (!val1 && !val2)
>>> +            break;
>>> +
>>> +        addr = buf2[1] | (buf2[0] << 8) | (buf1[3] << 16) |
>>> +            ((u64)buf1[2] << 24) | ((u64)buf1[1] << 32) |
>>> +            ((u64)buf1[0] << 40);
>>> +
>>> +        sdw_extract_slave_id(bus, addr, &id);
>>> +        /* Now compare with entries */
>>> +        list_for_each_entry_safe(slave, _s, &bus->slaves, node) {
>>> +            if (sdw_compare_devid(slave, id) == 0) {
>>> +                u32 status = qcom_swrm_get_n_device_status(ctrl, i);
>>> +                if (status == SDW_SLAVE_ATTACHED) {
>>> +                    slave->dev_num = i;
>>> +                    mutex_lock(&bus->bus_lock);
>>> +                    set_bit(i, bus->assigned);
>>> +                    mutex_unlock(&bus->bus_lock);
>>> +
>>> +                }
>>
>> And that part is strange as well. The bus->assigned bit should be set 
>> even if the Slave is not in the list provided by platform firmware. 
>> It's really tracking the state of the hardware, and it should not be 
>> influenced by what software knows to manage.
> 
> Am not 100% sure If I understand the concern here, but In normal (non 
> auto enum) cases this bit is set by the bus code while its doing 
> enumeration to assign a dev number from the assigned bitmap!
> 
> However in this case where auto enumeration happens it makes sense to 
> set this here with matching dev number!
> 
> AFAIU from code, each bit in this bitmap corresponds to slave dev number!

Yes, but the point was "why do you compare with information coming from 
platform firmware"? if the hardware reports the presence of devices on 
the link, why not use the information as is?

You recently added code that helps us deal with devices that are not 
listed in DT or ACPI tables, so why would we filter in this specific loop?

>>> +    complete(&ctrl->enumeration);
>>
>> you have init_completion() and complete() in this patch, but no 
>> wait_for_completion(), so that should be added in a later patch, no?
> 
> make sense, will move that to other patch!

Actually on this one comment that I missed last time is that you are 
using a completion only for the resume() case, and I think it should 
also be used for the regular probe() case, no?