Re: [PATCH 3/5] soundwire: qcom: wait for fifo to be empty before suspend

[Srinivas Kandagatla <srinivas.kandagatla@linaro.org>]

On 09/02/2023 16:33, Pierre-Louis Bossart wrote:
> 
> 
> On 2/9/23 09:52, Srinivas Kandagatla wrote:
>>
>>
>> On 09/02/2023 15:23, Pierre-Louis Bossart wrote:
>>>
>>>
>>> On 2/9/23 07:13, Srinivas Kandagatla wrote:
>>>> Wait for Fifo to be empty before going to suspend or before bank
>>>> switch happens. Just to make sure that all the reads/writes are done.
>>>
>>> For the suspend case that seems like a valid approach, but for bank
>>> switch don't we already have a bus->msg_lock mutex that will prevent the
>>> bank switch command from being sent before the other commands are
>>> handled?
>>
>> All read/writes are fifo based, so writes could be still pending.
> 
> I am not following. The bank switch happens with this function, where a
> mutex is taken.
> 
> int sdw_transfer(struct sdw_bus *bus, struct sdw_msg *msg)
> {
> 	int ret;
> 
> 	mutex_lock(&bus->msg_lock);
> 
> 	ret = sdw_transfer_unlocked(bus, msg);

Qualcomm controller uses fifo to read/write, so return after writing to 
fifo might not always indicate that write is completed.

Qcom Soundwire controller do not have any synchronous interrupt 
mechanism to indicate write complete.

--srini


> 
> 	mutex_unlock(&bus->msg_lock);
> 
> 	return ret;
> }
> 
> The transfer_unlocked is synchronous and waits for the command response
> to be available.
> 
> In other words, there's both a mutual exclusion and a synchronous
> behavior, so not sure how commands *before* the bank switch could be
> pending?