@@ -30,6 +30,38 @@
#include <drm/drm_crtc_helper.h>
#include <drm/drm_edid.h>
+static int
+intel_dp_mst_atomic_release_vcpi_slots(struct drm_atomic_state *state,
+ struct drm_crtc_state *crtc_state,
+ struct drm_connector_state *conn_state)
+{
+ struct drm_crtc_state *new_crtc_state;
+ struct intel_crtc_state *intel_crtc_state =
+ to_intel_crtc_state(crtc_state);
+ struct drm_encoder *encoder;
+ struct drm_dp_mst_topology_mgr *mgr;
+ int slots, ret;
+
+ slots = intel_crtc_state->dp_m_n.tu;
+ if (slots <= 0)
+ return 0;
+
+ encoder = conn_state->best_encoder;
+ mgr = &enc_to_mst(encoder)->primary->dp.mst_mgr;
+
+ ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
+ if (ret) {
+ DRM_DEBUG_KMS("failed releasing %d vcpi slots:%d\n",
+ slots, ret);
+ } else {
+ new_crtc_state = drm_atomic_get_crtc_state(state,
+ crtc_state->crtc);
+ to_intel_crtc_state(new_crtc_state)->dp_m_n.tu = 0;
+ }
+
+ return ret;
+}
+
static bool intel_dp_mst_compute_config(struct intel_encoder *encoder,
struct intel_crtc_state *pipe_config,
struct drm_connector_state *conn_state)
@@ -41,11 +73,15 @@ static bool intel_dp_mst_compute_config(struct intel_encoder *encoder,
struct intel_connector *connector =
to_intel_connector(conn_state->connector);
struct drm_atomic_state *state = pipe_config->base.state;
+ struct drm_connector_state *old_conn_state =
+ drm_atomic_get_old_connector_state(state, &connector->base);
+ struct drm_crtc *old_crtc;
struct intel_dp_mst_topology_state *mst_state;
int bpp;
int slots;
const struct drm_display_mode *adjusted_mode = &pipe_config->base.adjusted_mode;
int mst_pbn;
+ int ret;
bool reduce_m_n = drm_dp_has_quirk(&intel_dp->desc,
DP_DPCD_QUIRK_LIMITED_M_N);
@@ -78,6 +114,17 @@ static bool intel_dp_mst_compute_config(struct intel_encoder *encoder,
mst_pbn = drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock, bpp);
pipe_config->pbn = mst_pbn;
+ /* Free any VCPI allocations on this connector from the previous
+ * state */
+ old_crtc = old_conn_state->crtc;
+ if (old_crtc) {
+ ret = intel_dp_mst_atomic_release_vcpi_slots(
+ state, drm_atomic_get_old_crtc_state(state, old_crtc),
+ old_conn_state);
+ if (ret)
+ return ret;
+ }
+
slots = drm_dp_atomic_find_vcpi_slots(state, &intel_dp->mst_mgr,
connector->port, mst_pbn);
if (slots < 0) {
@@ -107,31 +154,20 @@ static int intel_dp_mst_atomic_check(struct drm_connector *connector,
{
struct drm_atomic_state *state = new_conn_state->state;
struct drm_connector_state *old_conn_state;
- struct drm_crtc *old_crtc;
- struct drm_crtc_state *crtc_state;
- int slots, ret = 0;
+ int ret = 0;
old_conn_state = drm_atomic_get_old_connector_state(state, connector);
- old_crtc = old_conn_state->crtc;
- if (!old_crtc)
- return ret;
- crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
- slots = to_intel_crtc_state(crtc_state)->dp_m_n.tu;
- if (drm_atomic_crtc_needs_modeset(crtc_state) && slots > 0) {
- struct drm_dp_mst_topology_mgr *mgr;
- struct drm_encoder *old_encoder;
-
- old_encoder = old_conn_state->best_encoder;
- mgr = &enc_to_mst(old_encoder)->primary->dp.mst_mgr;
+ /* Only free VCPI here if we're not going to be detaching the
+ * connector's current CRTC, since no config will be computed
+ */
+ if (new_conn_state->crtc || !old_conn_state->crtc)
+ return ret;
- ret = drm_dp_atomic_release_vcpi_slots(state, mgr, slots);
- if (ret)
- DRM_DEBUG_KMS("failed releasing %d vcpi slots:%d\n", slots, ret);
- else
- to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
- }
- return ret;
+ return intel_dp_mst_atomic_release_vcpi_slots(
+ state,
+ drm_atomic_get_new_crtc_state(state, old_conn_state->crtc),
+ old_conn_state);
}
static void intel_mst_disable_dp(struct intel_encoder *encoder,
The current way of handling changing VCPI allocations doesn't make a whole ton of sense. Since drm_atomic_helper_check_modeset() can be called multiple times which means that intel_dp_mst_atomic_check() can also be called multiple times. However, since we make the silly mistake of trying to free VCPI slots based off the /new/ atomic state instead of the old atomic state, we'll end up potentially double freeing the vcpi slots for the ports. This also has another unintended consequence that came back up while implementing MST fallback retraining: if a modeset is forced on a connector that's already part of the state, but it's atomic_check() has already been run once and doesn't get run again, we'll end up not freeing the VCPI allocations on the connector at all. The easiest way to solve this is to be clever and, with the exception of connectors that are being disabled and thus will never have compute_config() ran on them, move vcpi freeing out of the atomic check and into compute_config(). Cc: Manasi Navare <manasi.d.navare@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Lyude Paul <lyude@redhat.com> --- drivers/gpu/drm/i915/intel_dp_mst.c | 78 +++++++++++++++++++++++++++---------- 1 file changed, 57 insertions(+), 21 deletions(-)