Hi, On Thu, Jan 05, 2023 at 01:42:40PM +0800, Linyu Yuan wrote: > When a PPM client unregisters with UCSI framework, connector specific work > queue is destroyed. However, a pending delayed work queued before may > still exist. Once the delay timer expires and the work is scheduled, > this can cause a system crash as the workqueue is destroyed already. When the workqueue is destroyed it's also flushed, no? So how could there be still something pending, delayed or not? Did you actually see this happening? > Fix this by moving all partner related delayed work to connector instance > and cancel all of them when ucsi_unregister() is called by PPM client. I would love to be able to cancel these works, though not because of driver removal - I'm more concerned about disconnections. The reason why each task is a unique work is because it allows the driver to add the same task to the queue as many times as needed, and that was needed inorder to recover from some firmware issues. If there's only a dedicated work per task like in your proposal, we can only schedule the task once at a time, and that may lead into other issues. But first things first. Is the problem that you are describing here a real problem that can actually happen? thanks, -- heikki
