omap_remoteproc: Deadlock due to runtime PM

omap_remoteproc: Deadlock due to runtime PM

[Richard Weinberger]

Hi!

I'm seeing a reproducible deadlock in omap_remoteproc during system
reboot or shutdown while the remote processor is running.

The deadlock occurs as follows:
rproc_cdev_release() -> rproc_shutdown() -> mutex_lock_interruptible(&rproc->lock)
-> rproc_stop() -> omap_rproc_stop() -> pm_runtime_get_sync()

pm_runtime_get_sync() triggers omap_rproc_runtime_resume(),
which attempts to take rproc->lock again, leading to a deadlock.

In other words, rproc->lock is being used within the omap_remoteproc
runtime PM ops, which can conflict with other call paths that already
hold the same lock.

It was introduced with the following change:

commit 5f31b232c67434199558fd236e7644b432636b76
Author: Suman Anna <s-anna@ti.com>
Date:   Tue Mar 24 13:00:32 2020 +0200

    remoteproc/omap: Add support for runtime auto-suspend/resume

It seems the lock is taken in the PM ops likely because rproc->state 
is accessed, but perhaps this can be relaxed or refactored to avoid
the circular dependency.

Has anyone else encountered this issue,
or is there a known fix or approach to address it?

Thanks,
//richard