[PATCH v2 0/2] Fix MST VCPI lookup and modeset-lock splat

[Suraj Kandpal <suraj.kandpal@intel.com>]

intel_conn_to_vcpi() is called from the HDCP enable path in commit_tail
to look
up the VCPI of the MST stream that HDCP is being enabled on. Today it
reads the
topology state directly from mgr->base.state and takes mgr->base.lock
with the
in-flight state->base.acquire_ctx.

This breaks in two ways on MST + HDCP, especially when an
ALLOCATE_PAYLOAD
transaction is being driven on the mgr just before HDCP enable runs:

- mgr->base.state's payload list is being mutated in place by the MST
  payload code, so drm_atomic_get_mst_payload_state() can fail to find
  the entry for this connector's port. That trips drm_WARN_ON(!payload)
  and HDCP ends up programmed with VCPI 0.

- By the time we reach commit_tail/encoder enable,
  state->base.acquire_ctx
  is no longer valid for acquiring new modeset locks. Taking
  mgr->base.lock here produces a modeset-lock splat.

This series fixes both.

Suraj Kandpal (2):
  drm/i915/hdcp: Use new MST topology state in intel_conn_to_vcpi()
  drm/i915/hdcp: Drop mgr->base.lock acquisition in intel_conn_to_vcpi()

 drivers/gpu/drm/i915/display/intel_hdcp.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

-- 
2.34.1