Hi Laurent, Sorry for the very late reply, and thanks for the suggestion.
I agree that pm_runtime_resume_and_get() is preferable here. I also noticed that cdns_dsi_bridge_atomic_post_disable() unconditionally accesses the controller registers and releases the PHY and runtime PM reference. Since the pre-enable callback returns void, the post-disable callback may still be invoked after a failed runtime resume. I'll update the patch to use pm_runtime_resume_and_get() and track whether the runtime PM reference was successfully acquired, so that post-disable skips cleanup when pre-enable fails. Thanks, Wentao
