I believe I have found a mutex leak in drm_dp_get_mst_branch_device() in linux/drivers/gpu/drm/drm_dp_mst_topology.c in kernels 4.2-rc1 through the current version, 4.3-rc5, where the function can take an error exit without doing mutex_unlock(&mgr_lock).  I have attached a proposed patch.

Unfortunately, since I made the patch, I have not been able to trigger the error branch that the patch fixes, but I did find the bug by observing that error branch executing and printing an error message that I believe is unique to that branch ("*ERROR* failed to lookup MSTB"):

[   36.944273] [drm:gen8_irq_handler [i915]] *ERROR* The master control interrupt lied (SDE)!
[   74.576811] [drm:drm_dp_get_mst_branch_device [drm_kms_helper]] *ERROR* failed to lookup MSTB with lct 2, rad 10
[   74.587704] BUG: workqueue leaked lock or atomic: kworker/u8:0/0x00000000/6
[   74.587704]      last function: i915_digport_work_func [i915]
[   74.655752] [drm:gen8_irq_handler [i915]] *ERROR* The master control interrupt lied (SDE)!

I think this patch is also arguably an infinitesimal improvement in that it depends less on the optimizer, by reducing two references to port->mstb to one.

Anyhow, if the patch looks good, I would be delighted to if the appropriate person (Dave?) would integrate it for submission upstream.

Thanks in advance for any feedback on this.