Hi, Jonathan, On 24.02.2025 14:06, Claudiu wrote: > From: Claudiu Beznea <claudiu.beznea.uj@xxxxxxxxxxxxxx> > > On all systems where the rzg2l_adc driver is used, the ADC clocks are part > of a PM domain. The code that implements the PM domains support is in > drivers/clk/renesas/rzg2l-cpg.c, the functions of interest for this commit > being rzg2l_cpg_attach_dev() and rzg2l_cpg_deattach_dev(). The PM > domains support is registered with GENPD_FLAG_PM_CLK which, according to > the documentation, instructs genpd to use the PM clk framework while > powering on/off attached devices. > > During probe, the ADC device is attached to the PM domain > controlling the ADC clocks. Similarly, during removal, the ADC device is > detached from the PM domain. > > The detachment call stack is as follows: > > device_driver_detach() -> > device_release_driver_internal() -> > __device_release_driver() -> > device_remove() -> > platform_remove() -> > dev_pm_domain_detach() > > During driver unbind, after the ADC device is detached from its PM domain, > the device_unbind_cleanup() function is called, which subsequently invokes > devres_release_all(). This function handles devres resource cleanup. > > If runtime PM is enabled via devm_pm_runtime_enable(), the cleanup process > triggers the action or reset function for disabling runtime PM. This > function is pm_runtime_disable_action(), which leads to the following call > stack of interest when called: > > pm_runtime_disable_action() -> > pm_runtime_dont_use_autosuspend() -> > __pm_runtime_use_autosuspend() -> > update_autosuspend() -> > rpm_idle() > > The rpm_idle() function attempts to runtime resume the ADC device. However, > at the point it is called, the ADC device is no longer part of the PM > domain (which manages the ADC clocks). Since the rzg2l_adc runtime PM > APIs directly modifies hardware registers, the > rzg2l_adc_pm_runtime_resume() function is invoked without the ADC clocks > being enabled. This is because the PM domain no longer resumes along with > the ADC device. As a result, this leads to system aborts. > > Open a devres group in the driver probe and release it in the driver > remove. This ensures the runtime PM is disabled (though the devres group) > after the rzg2l_adc_remove() finishes its execution avoiding the described > scenario. Can you please let me know if you consider this approach acceptable? Thank you, Claudiu
