On 29/11/2022 15:44, Pierre-Louis Bossart wrote:
Don't hold sdw_dev_lock while calling the peripheral driver probe() and remove() callbacks.
Holding sdw_dev_lock around the probe() and remove() calls causes a theoretical mutex inversion which lockdep will assert on. The peripheral driver probe will probably register a soundcard, which will take ALSA and ASoC locks. During
It's extremely unlikely that a peripheral driver would register a sound card, this is what machine drivers do.
Which leads me to the question: is this a real problem?
Yes, try turning on lockdep checking and you will get an assert. During probe the existing code takes sdw_dev_lock and then calls the codec driver probe, so you will get a mutex sequence like:
sdw_dev_lock -> controls_rw_sem -> pcm_mutex
but in normal operation the ALSA/ASoC code will take its mutexes first and call runtime_resume which then takes the sdw_dev_lock, so you get
pcm_mutex -> sdw_dev_lock
and lockdep will assert on that opposite ordering. The full assert is at the end of this email.
Humm, you lost me with the reference to runtime_resume. I don't fully understand how it's possible to invoke pm_runtime during probe. pm_runtime should only enabled during the codec update_status() which can only be done once the probe completes.
I am fine with the changes that you are suggesting, the introduction of the sdw_dev_lock was probably too conservative and it'd be fine to only protect what is required.
However we do have lockdep enabled
I wonder whether this is because the Cirrus devices use full DP prepare, so there will be a DP prepare interrupt during the attempt to prepare the dailink. The lockdep assert was when sdw_update_slave_status() tried to take sdw_dev_lock.
If the Realtek codecs only use Soundwire interrupts for jack detect you probably won't see a sdw_dev_lock inside a pcm_mutex.