Some comments down below:
On Wed, 2024-06-26 at 16:48 +0800, Wayne Lin wrote:
[Why] During resume, observe that we receive CSN event before we start topology probing. Handling CSN at this moment based on uncertain topology is unnecessary.
[How] Add checking condition in drm_dp_mst_handle_up_req() to skip handling CSN if the topology is yet to be probed.
Cc: Lyude Paul lyude@redhat.com Cc: Harry Wentland hwentlan@amd.com Cc: Jani Nikula jani.nikula@intel.com Cc: Imre Deak imre.deak@intel.com Cc: Daniel Vetter daniel@ffwll.ch Cc: stable@vger.kernel.org Signed-off-by: Wayne Lin Wayne.Lin@amd.com
drivers/gpu/drm/display/drm_dp_mst_topology.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
diff --git a/drivers/gpu/drm/display/drm_dp_mst_topology.c b/drivers/gpu/drm/display/drm_dp_mst_topology.c index 68831f4e502a..fc2ceae61db2 100644 --- a/drivers/gpu/drm/display/drm_dp_mst_topology.c +++ b/drivers/gpu/drm/display/drm_dp_mst_topology.c @@ -4069,6 +4069,7 @@ static int drm_dp_mst_handle_up_req(struct drm_dp_mst_topology_mgr *mgr) if (up_req->msg.req_type == DP_CONNECTION_STATUS_NOTIFY) { const struct drm_dp_connection_status_notify *conn_stat = &up_req->msg.u.conn_stat;
bool handle_csn;drm_dbg_kms(mgr->dev, "Got CSN: pn: %d ldps:%d ddps: %d mcs: %d ip: %d pdt: %d\n", conn_stat->port_number, @@ -4077,6 +4078,16 @@ static int drm_dp_mst_handle_up_req(struct drm_dp_mst_topology_mgr *mgr) conn_stat->message_capability_status, conn_stat->input_port, conn_stat->peer_device_type);
mutex_lock(&mgr->probe_lock);handle_csn = mgr->mst_primary->link_address_sent;mutex_unlock(&mgr->probe_lock);if (!handle_csn) {drm_dbg_kms(mgr->dev, "Got CSN before finishtopology probing. Skip it.");
kfree(up_req);goto out;}
Hm. I think you're definitely on the right track here with not handling CSNs immediately after resume. My one question though is whether dropping the event entirely here is a good idea? In theory, we could receive a CSN at any time during the probe - including receiving a CSN for a connector that we've already probed in the initial post-resume process, which could result in us missing CSNs coming out of resume and still having an outdated topology layout.
I'm not totally sure about the solution I'm going to suggest but it seems like it would certainly be worth trying: what if we added a flag to drm_dp_mst_topology_mgr called something like "csn_during_resume" and simply set it to true in response to getting a CSN before we've finished reprobing? Then we at the end of the reprobe, we can simply restart the reprobing process if csn_during_resume gets set - which should still ensure we're up to date with reality.
} else if (up_req->msg.req_type == DP_RESOURCE_STATUS_NOTIFY) { const struct drm_dp_resource_status_notify *res_stat = &up_req->msg.u.resource_stat;