-----Original Message----- From: Alex Deucher <alexdeucher@xxxxxxxxx> Sent: Monday, September 25, 2023 8:27 PM To: imre.deak@xxxxxxxxx Cc: Lyude Paul <lyude@xxxxxxxxxx>; Jani Nikula <jani.nikula@xxxxxxxxx>; Jeff Layton <jlayton@xxxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; dri-devel@xxxxxxxxxxxxxxxxxxxxx; Wayne Lin <Wayne.Lin@xxxxxxx>; Alex Deucher <alexander.deucher@xxxxxxx>; Ramya SR (QUIC) <quic_rsr@xxxxxxxxxxx> Subject: Re: [PATCH v1] drm/dp/mst: fix missing modeset unlock for MST port detect WARNING: This email originated from outside of Qualcomm. Please be wary of any links or attachments, and do not enable macros. On Fri, Sep 22, 2023 at 3:22 PM Imre Deak <imre.deak@xxxxxxxxx> wrote: > > On Fri, Sep 22, 2023 at 03:02:23PM -0400, Lyude Paul wrote: > > > > Oh! wow thank you for catching this: > > > > Reviewed-by: Lyude Paul <lyude@xxxxxxxxxx> > > > > I will go and push this to drm-misc-next in just a moment > > > > On Fri, 2023-09-15 at 10:24 +0530, Ramya SR wrote: > > > Modeset mutex unlock is missing in drm_dp_mst_detect_port function. > > > This will lead to deadlock if calling the function multiple times > > > in an atomic operation, for example, getting imultiple MST ports > > > status for a DP MST bonding scenario. > > > > > > Signed-off-by: Ramya SR <quic_rsr@xxxxxxxxxxx> > > > --- > > > drivers/gpu/drm/display/drm_dp_mst_topology.c | 4 +++- > > > 1 file changed, 3 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/gpu/drm/display/drm_dp_mst_topology.c > > > b/drivers/gpu/drm/display/drm_dp_mst_topology.c > > > index ed96cfc..d6512c4 100644 > > > --- a/drivers/gpu/drm/display/drm_dp_mst_topology.c > > > +++ b/drivers/gpu/drm/display/drm_dp_mst_topology.c > > > @@ -4154,7 +4154,7 @@ drm_dp_mst_detect_port(struct drm_connector > > > *connector, > > > > > > ret = drm_modeset_lock(&mgr->base.lock, ctx); > > > if (ret) > > > - goto out; > > > + goto fail; > > > > > > ret = connector_status_disconnected; > > > > > > @@ -4181,6 +4181,8 @@ drm_dp_mst_detect_port(struct drm_connector *connector, > > > break; > > > } > > > out: > > > + drm_modeset_unlock(&mgr->base.lock); > > Isn't this supposed to be unlocked only by > drm_helper_probe_detect_ctx() / drm_helper_probe_detect() ? Maybe adding a comment to that effect would be helpful for the future. Alex >this is different lock, above function mentioned is locking/unlocking the global connection_mutex, that is required always locked during the atomic >check/commit. Here we are talking about MST topology manager mutex "mgr->base.lock". >For normal non-bond MST, it's ok, the calling sequence for detecting MST connector status is > dp_mst_connector_detect ->mst->mst_fw_cbs->detect_port_ctx = dp_mst_detect_port ->drm_dp_mst_detect_port (where mgr->base.lock is locked) > Then for the bond MST case, to figure out if the sibling connectors are also connected, so that the bonding is possible, we need detect two or more connectors >in single dp_mst_connector_detect call >dp_mst_connector_detect ->mst->mst_fw_cbs->detect_port_ctx = dp_mst_detect_port ->drm_dp_mst_detect_port (where mgr->base.lock is locked) >dp_mst_find_sibling_connector ->mst->mst_fw_cbs->detect_port_ctx = dp_mst_detect_port ->drm_dp_mst_detect_port (blocked by mgr->base.lock) > > > > +fail: > > > drm_dp_mst_topology_put_port(port); > > > return ret; > > > } > > > > -- > > Cheers, > > Lyude Paul (she/her) > > Software Engineer at Red Hat > >
