Nouveau - Oct 2018 - [PATCH v2 4/4] drm/nouveau: Use atomic VCPI helpers for MST

This patchset does some cleaning up of the atomic VCPI helpers for MST,
and converts nouveau over to using them. I would have included amdgpu in
this patch as well, but at the moment moving them over to the atomic
helpers is nontrivial.

Cc: Daniel Vetter <daniel at ffwll.ch>

Lyude Paul (4):
  drm/dp_mst: Add some atomic state iterator macros
  drm/dp_mst: Start tracking per-port VCPI allocations
  drm/dp_mst: Check payload count in drm_dp_mst_atomic_check()
  drm/nouveau: Use atomic VCPI helpers for MST

 drivers/gpu/drm/drm_dp_mst_topology.c   | 196 ++++++++++++++++++++----
 drivers/gpu/drm/i915/intel_display.c    |   8 +
 drivers/gpu/drm/i915/intel_dp_mst.c     |  31 ++--
 drivers/gpu/drm/nouveau/dispnv50/disp.c |  54 ++++++-
 include/drm/drm_dp_mst_helper.h         |  88 ++++++++++-
 5 files changed, 322 insertions(+), 55 deletions(-)

-- 
2.17.2

Signed-off-by: Lyude Paul <lyude at redhat.com>
Cc: Daniel Vetter <daniel at ffwll.ch>
---
 include/drm/drm_dp_mst_helper.h | 77 +++++++++++++++++++++++++++++++++
 1 file changed, 77 insertions(+)

diff --git a/include/drm/drm_dp_mst_helper.h b/include/drm/drm_dp_mst_helper.h
index 59f005b419cf..3faceb66f5cb 100644
--- a/include/drm/drm_dp_mst_helper.h
+++ b/include/drm/drm_dp_mst_helper.h
@@ -628,4 +628,81 @@ int drm_dp_atomic_release_vcpi_slots(struct
drm_atomic_state *state,
 int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
 				 struct drm_dp_mst_port *port, bool power_up);
 
+extern const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs;
+
+static inline bool
+__drm_dp_mst_state_iter_get(struct drm_atomic_state *state,
+			    struct drm_dp_mst_topology_mgr **mgr,
+			    struct drm_dp_mst_topology_state **old_state,
+			    struct drm_dp_mst_topology_state **new_state,
+			    int i)
+{
+	struct __drm_private_objs_state *objs_state = &state->private_objs[i];
+
+	if (objs_state->ptr->funcs != &drm_dp_mst_topology_state_funcs)
+		return false;
+
+	*mgr = to_dp_mst_topology_mgr(objs_state->ptr);
+	if (old_state)
+		*old_state = to_dp_mst_topology_state(objs_state->old_state);
+	if (new_state)
+		*new_state = to_dp_mst_topology_state(objs_state->new_state);
+
+	return true;
+}
+
+/**
+ * for_each_oldnew_mst_mgr_in_state - iterate over all DP MST topology
+ * managers in an atomic update
+ * @__state: &struct drm_atomic_state pointer
+ * @mgr: &struct drm_dp_mst_topology_mgr iteration cursor
+ * @old_state: &struct drm_dp_mst_topology_state iteration cursor for the
old
+ * state
+ * @new_state: &struct drm_dp_mst_topology_state iteration cursor for the
new
+ * state
+ * @__i: int iteration cursor, for macro-internal use
+ *
+ * This iterates over all DRM DP MST topology managers in an atomic update,
+ * tracking both old and new state. This is useful in places where the state
+ * delta needs to be considered, for example in atomic check functions.
+ */
+#define for_each_oldnew_mst_mgr_in_state(__state, mgr, old_state, new_state,
__i) \
+	for ((__i) = 0; (__i) < (__state)->num_private_objs; (__i)++) \
+		for_each_if(__drm_dp_mst_state_iter_get((__state), &(mgr),
&(old_state), &(new_state), (__i)))
+
+/**
+ * for_each_old_mst_mgr_in_state - iterate over all DP MST topology managers
+ * in an atomic update
+ * @__state: &struct drm_atomic_state pointer
+ * @mgr: &struct drm_dp_mst_topology_mgr iteration cursor
+ * @old_state: &struct drm_dp_mst_topology_state iteration cursor for the
old
+ * state
+ * @__i: int iteration cursor, for macro-internal use
+ *
+ * This iterates over all DRM DP MST topology managers in an atomic update,
+ * tracking only the old state. This is useful in disable functions, where we
+ * need the old state the hardware is still in.
+ */
+#define for_each_old_mst_mgr_in_state(__state, mgr, old_state, __i) \
+	for ((__i) = 0; (__i) < (__state)->num_private_objs; (__i)++) \
+		for_each_if(__drm_dp_mst_state_iter_get((__state), &(mgr),
&(old_state), NULL, (__i)))
+
+/**
+ * for_each_new_mst_mgr_in_state - iterate over all DP MST topology managers
+ * in an atomic update
+ * @__state: &struct drm_atomic_state pointer
+ * @mgr: &struct drm_dp_mst_topology_mgr iteration cursor
+ * @new_state: &struct drm_dp_mst_topology_state iteration cursor for the
new
+ * state
+ * @__i: int iteration cursor, for macro-internal use
+ *
+ * This iterates over all DRM DP MST topology managers in an atomic update,
+ * tracking only the new state. This is useful in enable functions, where we
+ * need the new state the hardware should be in when the atomic commit
+ * operation has completed.
+ */
+#define for_each_new_mst_mgr_in_state(__state, mgr, new_state, __i) \
+	for ((__i) = 0; (__i) < (__state)->num_private_objs; (__i)++) \
+		for_each_if(__drm_dp_mst_state_iter_get((__state), &(mgr), NULL,
&(new_state), (__i)))
+
 #endif
-- 
2.17.2

There has been a TODO waiting for quite a long time in
drm_dp_mst_topology.c:

	/* We cannot rely on port->vcpi.num_slots to update
	 * topology_state->avail_slots as the port may not exist if the parent
	 * branch device was unplugged. This should be fixed by tracking
	 * per-port slot allocation in drm_dp_mst_topology_state instead of
	 * depending on the caller to tell us how many slots to release.
	 */

That's not the only reason we should fix this: forcing the driver to
track the VCPI allocations throughout a state's atomic check is
error prone, because it means that extra care has to be taken with the
order that drm_dp_atomic_find_vcpi_slots() and
drm_dp_atomic_release_vcpi_slots() are called in in order to ensure
idempotency. Currently the only driver actually using these helpers,
i915, doesn't even do this correctly: multiple ->best_encoder() checks
with i915's current implementation would not be idempotent and would
over-allocate VCPI slots, something I learned trying to implement
fallback retraining in MST.

So: simplify this whole mess, and teach drm_dp_atomic_find_vcpi_slots()
and drm_dp_atomic_release_vcpi_slots() to track the VCPI allocations for
each port. This allows us to ensure idempotency without having to rely
on the driver as much. Additionally: the driver doesn't need to do any
kind of VCPI slot tracking anymore if it doesn't need it for it's own
internal state.

Additionally; this adds a new drm_dp_mst_atomic_check() helper which
must be used by atomic drivers to perform validity checks for the new
VCPI allocations incurred by a state.

Also: update the documentation and make it more obvious that these
/must/ be called by /all/ atomic drivers supporting MST.

Changes since v1:
 - Don't use the now-removed ->atomic_check() for private objects hook,
   just give drivers a function to call themselves

Signed-off-by: Lyude Paul <lyude at redhat.com>
Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
---
 drivers/gpu/drm/drm_dp_mst_topology.c | 190 +++++++++++++++++++++-----
 drivers/gpu/drm/i915/intel_display.c  |   8 ++
 drivers/gpu/drm/i915/intel_dp_mst.c   |  31 +++--
 include/drm/drm_dp_mst_helper.h       |  11 +-
 4 files changed, 192 insertions(+), 48 deletions(-)

diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
b/drivers/gpu/drm/drm_dp_mst_topology.c
index 8c3cfac437f4..dcfab7536914 100644
--- a/drivers/gpu/drm/drm_dp_mst_topology.c
+++ b/drivers/gpu/drm/drm_dp_mst_topology.c
@@ -2614,21 +2614,33 @@ static int drm_dp_init_vcpi(struct
drm_dp_mst_topology_mgr *mgr,
 }
 
 /**
- * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to the state
+ * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to the state
  * @state: global atomic state
  * @mgr: MST topology manager for the port
  * @port: port to find vcpi slots for
  * @pbn: bandwidth required for the mode in PBN
  *
+ * Allocates VCPI slots to @port, replacing any previous VCPI allocations it
+ * may have had. Any atomic drivers which support MST must call this function
+ * in their atomic_check() handlers to change the current VCPI allocation for
+ * the new state. After the ->atomic_check() hooks of the driver and all
other
+ * mode objects in the state have been called, DRM will check the final VCPI
+ * allocations to ensure that they will fit into the available bandwidth on
+ * the topology.
+ *
+ * See also: drm_dp_atomic_release_vcpi_slots()
+ *
  * RETURNS:
- * Total slots in the atomic state assigned for this port or error
+ * Total slots in the atomic state assigned for this port, or a negative error
+ * code if the port no longer exists
  */
 int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
 				  struct drm_dp_mst_topology_mgr *mgr,
 				  struct drm_dp_mst_port *port, int pbn)
 {
 	struct drm_dp_mst_topology_state *topology_state;
-	int req_slots;
+	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
+	int prev_slots, req_slots, ret;
 
 	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
 	if (IS_ERR(topology_state))
@@ -2637,20 +2649,41 @@ int drm_dp_atomic_find_vcpi_slots(struct
drm_atomic_state *state,
 	port = drm_dp_get_validated_port_ref(mgr, port);
 	if (port == NULL)
 		return -EINVAL;
-	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
-	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
-			req_slots, topology_state->avail_slots);
 
-	if (req_slots > topology_state->avail_slots) {
-		drm_dp_put_port(port);
-		return -ENOSPC;
+	/* Find the current allocation for this port, if any */
+	list_for_each_entry(pos, &topology_state->vcpis, next) {
+		if (pos->port == port) {
+			vcpi = pos;
+			prev_slots = vcpi->vcpi;
+			break;
+		}
 	}
+	if (!vcpi)
+		prev_slots = 0;
+
+	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
+
+	DRM_DEBUG_KMS("[CONNECTOR:%d:%s] [MST PORT:%p] vcpi %d -> %d\n",
+		      port->connector->base.id, port->connector->name, port,
+		      prev_slots, req_slots);
+
+	/* Add the new allocation to the state */
+	if (!vcpi) {
+		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
+		if (!vcpi) {
+			ret = -ENOMEM;
+			goto out;
+		}
 
-	topology_state->avail_slots -= req_slots;
-	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
+		vcpi->port = port;
+		list_add(&vcpi->next, &topology_state->vcpis);
+	}
+	vcpi->vcpi = req_slots;
 
+	ret = req_slots;
+out:
 	drm_dp_put_port(port);
-	return req_slots;
+	return ret;
 }
 EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
 
@@ -2658,32 +2691,46 @@ EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
  * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi slots
  * @state: global atomic state
  * @mgr: MST topology manager for the port
- * @slots: number of vcpi slots to release
+ *
+ * Releases any VCPI slots that have been allocated to a port in the atomic
+ * state. Any atomic drivers which support MST must call this function in
+ * their connector's atomic_check() handler when the connector will no
longer
+ * have VCPI allocated (e.g. because it's CRTC was removed).
+ *
+ * It is OK to call this even if @port has been removed from the system, in
+ * which case it will just amount to a no-op.
+ *
+ * See also: drm_dp_atomic_find_vcpi_slots()
  *
  * RETURNS:
- * 0 if @slots were added back to
&drm_dp_mst_topology_state->avail_slots or
- * negative error code
+ * 0 if all slots for this port were added back to
+ * &drm_dp_mst_topology_state->avail_slots or negative error code
  */
 int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
 				     struct drm_dp_mst_topology_mgr *mgr,
-				     int slots)
+				     struct drm_dp_mst_port *port)
 {
 	struct drm_dp_mst_topology_state *topology_state;
+	struct drm_dp_vcpi_allocation *tmp, *pos;
 
 	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
 	if (IS_ERR(topology_state))
 		return PTR_ERR(topology_state);
 
-	/* We cannot rely on port->vcpi.num_slots to update
-	 * topology_state->avail_slots as the port may not exist if the parent
-	 * branch device was unplugged. This should be fixed by tracking
-	 * per-port slot allocation in drm_dp_mst_topology_state instead of
-	 * depending on the caller to tell us how many slots to release.
-	 */
-	topology_state->avail_slots += slots;
-	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
-			slots, topology_state->avail_slots);
+	list_for_each_entry_safe(pos, tmp, &topology_state->vcpis, next) {
+		if (pos->port == port) {
+			list_del(&pos->next);
+			DRM_DEBUG_KMS("[MST PORT:%p] vcpi %d -> 0\n",
+				      port, pos->vcpi);
 
+			kfree(pos);
+			return 0;
+		}
+	}
+
+	/* If no allocation was found, all that means is that the port was
+	 * destroyed since the last atomic commit. That's OK!
+	 */
 	return 0;
 }
 EXPORT_SYMBOL(drm_dp_atomic_release_vcpi_slots);
@@ -3112,15 +3159,50 @@ static void drm_dp_destroy_connector_work(struct
work_struct *work)
 static struct drm_private_state *
 drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
 {
-	struct drm_dp_mst_topology_state *state;
+	struct drm_dp_mst_topology_state *state, *old_state +	
to_dp_mst_topology_state(obj->state);
+	struct drm_dp_mst_topology_mgr *mgr = old_state->mgr;
+	struct drm_dp_mst_port *port;
+	struct drm_dp_vcpi_allocation *pos, *vcpi;
 
-	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
+	state = kzalloc(sizeof(*state), GFP_KERNEL);
 	if (!state)
 		return NULL;
 
 	__drm_atomic_helper_private_obj_duplicate_state(obj, &state->base);
 
+	state->mgr = mgr;
+	INIT_LIST_HEAD(&state->vcpis);
+
+	/* Copy over the VCPI allocations for ports that still exist */
+	list_for_each_entry(pos, &old_state->vcpis, next) {
+		port = drm_dp_get_validated_port_ref(mgr, pos->port);
+		if (!port) {
+			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone, vcpi %d -> 0\n",
+					 pos->port, pos->vcpi);
+			continue;
+		}
+
+		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
+		if (!vcpi) {
+			drm_dp_put_port(port);
+			goto fail_alloc;
+		}
+
+		vcpi->port = port;
+		vcpi->vcpi = pos->vcpi;
+		list_add(&vcpi->next, &state->vcpis);
+		drm_dp_put_port(port);
+	}
+
 	return &state->base;
+
+fail_alloc:
+	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next)
+		kfree(pos);
+	kfree(state);
+
+	return NULL;
 }
 
 static void drm_dp_mst_destroy_state(struct drm_private_obj *obj,
@@ -3128,14 +3210,60 @@ static void drm_dp_mst_destroy_state(struct
drm_private_obj *obj,
 {
 	struct drm_dp_mst_topology_state *mst_state  	
to_dp_mst_topology_state(state);
+	struct drm_dp_vcpi_allocation *pos, *tmp;
+
+	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis, next)
+		kfree(pos);
 
 	kfree(mst_state);
 }
 
-static const struct drm_private_state_funcs mst_state_funcs = {
+/**
+ * drm_dp_mst_atomic_check - Check that the new state of an MST topology in an
+ * atomic update is valid
+ * @state: Pointer to the new &struct drm_dp_mst_topology_state
+ *
+ * Checks the given topology state for an atomic update to ensure that it's
+ * valid. This includes checking whether there's enough bandwidth to
support
+ * the new VCPI allocations in the atomic update.
+ *
+ * Any atomic drivers supporting DP MST must make sure to call this after
+ * checking the rest of their state in their ->atomic_check() callback.
+ *
+ * Returns:
+ *
+ * 0 if the new state is valid, negative error code otherwise.
+ */
+int drm_dp_mst_atomic_check(struct drm_dp_mst_topology_state *state)
+{
+	struct drm_dp_mst_topology_mgr *mgr = state->mgr;
+	struct drm_dp_vcpi_allocation *pos;
+	int avail_slots = 63;
+
+	list_for_each_entry(pos, &state->vcpis, next) {
+		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi slots\n",
+				 pos->port, pos->vcpi);
+
+		avail_slots -= pos->vcpi;
+		if (avail_slots < 0) {
+			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough vcpi slots in state %p
(avail=%d)\n",
+					 pos->port, state,
+					 avail_slots + pos->vcpi);
+			return -ENOSPC;
+		}
+	}
+	DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p vcpi avail=%d used=%d\n",
+			 mgr, state, avail_slots, 63 - avail_slots);
+
+	return 0;
+}
+EXPORT_SYMBOL(drm_dp_mst_atomic_check);
+
+const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs = {
 	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
 	.atomic_destroy_state = drm_dp_mst_destroy_state,
 };
+EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
 
 /**
  * drm_atomic_get_mst_topology_state: get MST topology state
@@ -3213,13 +3341,11 @@ int drm_dp_mst_topology_mgr_init(struct
drm_dp_mst_topology_mgr *mgr,
 		return -ENOMEM;
 
 	mst_state->mgr = mgr;
-
-	/* max. time slots - one slot for MTP header */
-	mst_state->avail_slots = 63;
+	INIT_LIST_HEAD(&mst_state->vcpis);
 
 	drm_atomic_private_obj_init(&mgr->base,
 				    &mst_state->base,
-				    &mst_state_funcs);
+				    &drm_dp_mst_topology_state_funcs);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/i915/intel_display.c
b/drivers/gpu/drm/i915/intel_display.c
index fe045abb6472..f66af1465686 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -12484,6 +12484,8 @@ static int intel_atomic_check(struct drm_device *dev,
 	struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
 	struct drm_crtc *crtc;
 	struct drm_crtc_state *old_crtc_state, *crtc_state;
+	struct drm_dp_mst_topology_mgr *mgr;
+	struct drm_dp_mst_topology_state *mst_state;
 	int ret, i;
 	bool any_ms = false;
 
@@ -12534,6 +12536,12 @@ static int intel_atomic_check(struct drm_device *dev,
 				       "[modeset]" : "[fastset]");
 	}
 
+	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
+		ret = drm_dp_mst_atomic_check(mst_state);
+		if (ret)
+			return ret;
+	}
+
 	if (any_ms) {
 		ret = intel_modeset_checks(state);
 
diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
b/drivers/gpu/drm/i915/intel_dp_mst.c
index 8b71d64ebd9d..aaf904738b78 100644
--- a/drivers/gpu/drm/i915/intel_dp_mst.c
+++ b/drivers/gpu/drm/i915/intel_dp_mst.c
@@ -114,28 +114,31 @@ static int intel_dp_mst_atomic_check(struct drm_connector
*connector,
 	struct drm_connector_state *old_conn_state;
 	struct drm_crtc *old_crtc;
 	struct drm_crtc_state *crtc_state;
-	int slots, ret = 0;
+	struct intel_connector *intel_connector +		to_intel_connector(connector);
+	struct drm_dp_mst_topology_mgr *mgr +	
&intel_connector->mst_port->mst_mgr;
+	struct drm_dp_mst_port *port = intel_connector->port;
+	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;
+	/* Free VCPI, since compute_config() won't be run */
+	if (!new_conn_state->crtc) {
+		crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
 
-		old_encoder = old_conn_state->best_encoder;
-		mgr = &enc_to_mst(old_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
-			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
+		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, port);
+		if (ret) {
+			DRM_DEBUG_KMS("failed releasing vcpi slots: %d\n",
+				      ret);
+			return ret;
+		}
+		to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
 	}
+
 	return ret;
 }
 
diff --git a/include/drm/drm_dp_mst_helper.h b/include/drm/drm_dp_mst_helper.h
index 3faceb66f5cb..0aa7d3658013 100644
--- a/include/drm/drm_dp_mst_helper.h
+++ b/include/drm/drm_dp_mst_helper.h
@@ -406,9 +406,15 @@ struct drm_dp_payload {
 
 #define to_dp_mst_topology_state(x) container_of(x, struct
drm_dp_mst_topology_state, base)
 
+struct drm_dp_vcpi_allocation {
+	struct drm_dp_mst_port *port;
+	int vcpi;
+	struct list_head next;
+};
+
 struct drm_dp_mst_topology_state {
 	struct drm_private_state base;
-	int avail_slots;
+	struct list_head vcpis;
 	struct drm_dp_mst_topology_mgr *mgr;
 };
 
@@ -624,9 +630,10 @@ int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state
*state,
 				  struct drm_dp_mst_port *port, int pbn);
 int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
 				     struct drm_dp_mst_topology_mgr *mgr,
-				     int slots);
+				     struct drm_dp_mst_port *port);
 int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
 				 struct drm_dp_mst_port *port, bool power_up);
+int drm_dp_mst_atomic_check(struct drm_dp_mst_topology_state *state);
 
 extern const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs;
 
-- 
2.17.2

It occurred to me that we never actually check this! So let's start
doing that.

Signed-off-by: Lyude Paul <lyude at redhat.com>
Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
---
 drivers/gpu/drm/drm_dp_mst_topology.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
b/drivers/gpu/drm/drm_dp_mst_topology.c
index dcfab7536914..8bb03700e199 100644
--- a/drivers/gpu/drm/drm_dp_mst_topology.c
+++ b/drivers/gpu/drm/drm_dp_mst_topology.c
@@ -3238,7 +3238,7 @@ int drm_dp_mst_atomic_check(struct
drm_dp_mst_topology_state *state)
 {
 	struct drm_dp_mst_topology_mgr *mgr = state->mgr;
 	struct drm_dp_vcpi_allocation *pos;
-	int avail_slots = 63;
+	int avail_slots = 63, payload_count = 0;
 
 	list_for_each_entry(pos, &state->vcpis, next) {
 		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi slots\n",
@@ -3251,6 +3251,12 @@ int drm_dp_mst_atomic_check(struct
drm_dp_mst_topology_state *state)
 					 avail_slots + pos->vcpi);
 			return -ENOSPC;
 		}
+
+		if (++payload_count > mgr->max_payloads) {
+			DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p has too many payloads
(max=%d)\n",
+					 mgr, state, mgr->max_payloads);
+			return -EINVAL;
+		}
 	}
 	DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p vcpi avail=%d used=%d\n",
 			 mgr, state, avail_slots, 63 - avail_slots);
-- 
2.17.2

Currently, nouveau uses the yolo method of setting up MST displays: it
uses the old VCPI helpers (drm_dp_find_vcpi_slots()) for computing the
display configuration. These helpers don't take care to make sure they
take a reference to the mstb port that they're checking, and
additionally don't actually check whether or not the topology still has
enough bandwidth to provide the VCPI tokens required.

So, drop usage of the old helpers and move entirely over to the atomic
helpers.

Signed-off-by: Lyude Paul <lyude at redhat.com>
Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
---
 drivers/gpu/drm/nouveau/dispnv50/disp.c | 54 +++++++++++++++++++++----
 1 file changed, 47 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/dispnv50/disp.c
b/drivers/gpu/drm/nouveau/dispnv50/disp.c
index 6cbbae3f438b..37503319e5b1 100644
--- a/drivers/gpu/drm/nouveau/dispnv50/disp.c
+++ b/drivers/gpu/drm/nouveau/dispnv50/disp.c
@@ -747,16 +747,22 @@ nv50_msto_atomic_check(struct drm_encoder *encoder,
 		       struct drm_crtc_state *crtc_state,
 		       struct drm_connector_state *conn_state)
 {
-	struct nv50_mstc *mstc = nv50_mstc(conn_state->connector);
+	struct drm_atomic_state *state = crtc_state->state;
+	struct drm_connector *connector = conn_state->connector;
+	struct nv50_mstc *mstc = nv50_mstc(connector);
 	struct nv50_mstm *mstm = mstc->mstm;
-	int bpp = conn_state->connector->display_info.bpc * 3;
+	int bpp = connector->display_info.bpc * 3;
 	int slots;
 
-	mstc->pbn = drm_dp_calc_pbn_mode(crtc_state->adjusted_mode.clock, bpp);
-
-	slots = drm_dp_find_vcpi_slots(&mstm->mgr, mstc->pbn);
-	if (slots < 0)
-		return slots;
+	mstc->pbn = drm_dp_calc_pbn_mode(crtc_state->adjusted_mode.clock,
+					 bpp);
+	/* Zombies don't need VCPI */
+	if (!drm_connector_is_unregistered(connector)) {
+		slots = drm_dp_atomic_find_vcpi_slots(state, &mstm->mgr,
+						      mstc->port, mstc->pbn);
+		if (slots < 0)
+			return slots;
+	}
 
 	return nv50_outp_atomic_check_view(encoder, crtc_state, conn_state,
 					   mstc->native);
@@ -920,12 +926,38 @@ nv50_mstc_get_modes(struct drm_connector *connector)
 	return ret;
 }
 
+static int
+nv50_mstc_atomic_check(struct drm_connector *connector,
+		       struct drm_connector_state *new_conn_state)
+{
+	struct drm_atomic_state *state = new_conn_state->state;
+	struct nv50_mstc *mstc = nv50_mstc(connector);
+	struct drm_dp_mst_topology_mgr *mgr = &mstc->mstm->mgr;
+	struct drm_connector_state *old_conn_state;
+	struct drm_crtc *old_crtc;
+
+	old_conn_state = drm_atomic_get_old_connector_state(state, connector);
+	old_crtc = old_conn_state->crtc;
+
+	/* We only need to release VCPI here if we're going from having a CRTC
+	 * on this connector, to not having one
+	 */
+	if (!old_crtc || new_conn_state->crtc)
+		return 0;
+
+	/* Release the previous VCPI allocation since the encoder's
+	 * atomic_check() won't be called
+	 */
+	return drm_dp_atomic_release_vcpi_slots(state, mgr, mstc->port);
+}
+
 static const struct drm_connector_helper_funcs
 nv50_mstc_help = {
 	.get_modes = nv50_mstc_get_modes,
 	.mode_valid = nv50_mstc_mode_valid,
 	.best_encoder = nv50_mstc_best_encoder,
 	.atomic_best_encoder = nv50_mstc_atomic_best_encoder,
+	.atomic_check = nv50_mstc_atomic_check,
 };
 
 static enum drm_connector_status
@@ -2084,6 +2116,8 @@ nv50_disp_atomic_check(struct drm_device *dev, struct
drm_atomic_state *state)
 	struct drm_connector *connector;
 	struct drm_crtc_state *new_crtc_state;
 	struct drm_crtc *crtc;
+	struct drm_dp_mst_topology_mgr *mgr;
+	struct drm_dp_mst_topology_state *mst_state;
 	int ret, i;
 
 	/* We need to handle colour management on a per-plane basis. */
@@ -2109,6 +2143,12 @@ nv50_disp_atomic_check(struct drm_device *dev, struct
drm_atomic_state *state)
 			return ret;
 	}
 
+	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
+		ret = drm_dp_mst_atomic_check(mst_state);
+		if (ret)
+			return ret;
+	}
+
 	return 0;
 }
 
-- 
2.17.2

On Fri, Oct 26, 2018 at 04:35:46PM -0400, Lyude Paul
wrote:
> Signed-off-by: Lyude Paul <lyude at redhat.com>
> Cc: Daniel Vetter <daniel at ffwll.ch>
> ---
>  include/drm/drm_dp_mst_helper.h | 77 +++++++++++++++++++++++++++++++++
>  1 file changed, 77 insertions(+)
> 
> diff --git a/include/drm/drm_dp_mst_helper.h
b/include/drm/drm_dp_mst_helper.h
> index 59f005b419cf..3faceb66f5cb 100644
> --- a/include/drm/drm_dp_mst_helper.h
> +++ b/include/drm/drm_dp_mst_helper.h
> @@ -628,4 +628,81 @@ int drm_dp_atomic_release_vcpi_slots(struct
drm_atomic_state *state,
>  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
>  				 struct drm_dp_mst_port *port, bool power_up);
>  
> +extern const struct drm_private_state_funcs
drm_dp_mst_topology_state_funcs;
> +
> +static inline bool
> +__drm_dp_mst_state_iter_get(struct drm_atomic_state *state,
> +			    struct drm_dp_mst_topology_mgr **mgr,
> +			    struct drm_dp_mst_topology_state **old_state,
> +			    struct drm_dp_mst_topology_state **new_state,
> +			    int i)
> +{
> +	struct __drm_private_objs_state *objs_state =
&state->private_objs[i];
> +
> +	if (objs_state->ptr->funcs != &drm_dp_mst_topology_state_funcs)
> +		return false;
> +
> +	*mgr = to_dp_mst_topology_mgr(objs_state->ptr);
> +	if (old_state)
> +		*old_state = to_dp_mst_topology_state(objs_state->old_state);
> +	if (new_state)
> +		*new_state = to_dp_mst_topology_state(objs_state->new_state);
> +
> +	return true;
> +}
> +
> +/**
> + * for_each_oldnew_mst_mgr_in_state - iterate over all DP MST topology
> + * managers in an atomic update
> + * @__state: &struct drm_atomic_state pointer
> + * @mgr: &struct drm_dp_mst_topology_mgr iteration cursor
> + * @old_state: &struct drm_dp_mst_topology_state iteration cursor for
the old
> + * state
> + * @new_state: &struct drm_dp_mst_topology_state iteration cursor for
the new
> + * state
> + * @__i: int iteration cursor, for macro-internal use
> + *
> + * This iterates over all DRM DP MST topology managers in an atomic
update,
> + * tracking both old and new state. This is useful in places where the
state
> + * delta needs to be considered, for example in atomic check functions.
> + */
> +#define for_each_oldnew_mst_mgr_in_state(__state, mgr, old_state,
new_state, __i) \
> +	for ((__i) = 0; (__i) < (__state)->num_private_objs; (__i)++) \
> +		for_each_if(__drm_dp_mst_state_iter_get((__state), &(mgr),
&(old_state), &(new_state), (__i)))
I just realized that all our iterator macros do not compose at all. Whic
is rather disappointing :-/

list_for_each solves that with a list_for_each_member, which you can then
easily use to create more specialized macros. I wonder whether we should
have a master macros, which also takes the obj type (we have very uniform
naming, so a few well placed ## should make this work neatly), and then
you could just #define more specialized iterators without having to retype
them (or the need for dummy iterator markers and tons of casting).

Anyway, that's a lament for another day&patch. This looks good.

Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> +
> +/**
> + * for_each_old_mst_mgr_in_state - iterate over all DP MST topology
managers
> + * in an atomic update
> + * @__state: &struct drm_atomic_state pointer
> + * @mgr: &struct drm_dp_mst_topology_mgr iteration cursor
> + * @old_state: &struct drm_dp_mst_topology_state iteration cursor for
the old
> + * state
> + * @__i: int iteration cursor, for macro-internal use
> + *
> + * This iterates over all DRM DP MST topology managers in an atomic
update,
> + * tracking only the old state. This is useful in disable functions, where
we
> + * need the old state the hardware is still in.
> + */
> +#define for_each_old_mst_mgr_in_state(__state, mgr, old_state, __i) \
> +	for ((__i) = 0; (__i) < (__state)->num_private_objs; (__i)++) \
> +		for_each_if(__drm_dp_mst_state_iter_get((__state), &(mgr),
&(old_state), NULL, (__i)))
> +
> +/**
> + * for_each_new_mst_mgr_in_state - iterate over all DP MST topology
managers
> + * in an atomic update
> + * @__state: &struct drm_atomic_state pointer
> + * @mgr: &struct drm_dp_mst_topology_mgr iteration cursor
> + * @new_state: &struct drm_dp_mst_topology_state iteration cursor for
the new
> + * state
> + * @__i: int iteration cursor, for macro-internal use
> + *
> + * This iterates over all DRM DP MST topology managers in an atomic
update,
> + * tracking only the new state. This is useful in enable functions, where
we
> + * need the new state the hardware should be in when the atomic commit
> + * operation has completed.
> + */
> +#define for_each_new_mst_mgr_in_state(__state, mgr, new_state, __i) \
> +	for ((__i) = 0; (__i) < (__state)->num_private_objs; (__i)++) \
> +		for_each_if(__drm_dp_mst_state_iter_get((__state), &(mgr), NULL,
&(new_state), (__i)))
> +
>  #endif
> -- 
> 2.17.2
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Fri, Oct 26, 2018 at 04:35:49PM -0400, Lyude Paul
wrote:
> Currently, nouveau uses the yolo method of setting up MST displays: it
> uses the old VCPI helpers (drm_dp_find_vcpi_slots()) for computing the
> display configuration. These helpers don't take care to make sure they
> take a reference to the mstb port that they're checking, and
> additionally don't actually check whether or not the topology still has
> enough bandwidth to provide the VCPI tokens required.
> 
> So, drop usage of the old helpers and move entirely over to the atomic
> helpers.
> 
> Signed-off-by: Lyude Paul <lyude at redhat.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> ---
>  drivers/gpu/drm/nouveau/dispnv50/disp.c | 54 +++++++++++++++++++++----
>  1 file changed, 47 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/gpu/drm/nouveau/dispnv50/disp.c
b/drivers/gpu/drm/nouveau/dispnv50/disp.c
> index 6cbbae3f438b..37503319e5b1 100644
> --- a/drivers/gpu/drm/nouveau/dispnv50/disp.c
> +++ b/drivers/gpu/drm/nouveau/dispnv50/disp.c
> @@ -747,16 +747,22 @@ nv50_msto_atomic_check(struct drm_encoder *encoder,
>  		       struct drm_crtc_state *crtc_state,
>  		       struct drm_connector_state *conn_state)
>  {
> -	struct nv50_mstc *mstc = nv50_mstc(conn_state->connector);
> +	struct drm_atomic_state *state = crtc_state->state;
> +	struct drm_connector *connector = conn_state->connector;
> +	struct nv50_mstc *mstc = nv50_mstc(connector);
>  	struct nv50_mstm *mstm = mstc->mstm;
> -	int bpp = conn_state->connector->display_info.bpc * 3;
> +	int bpp = connector->display_info.bpc * 3;
>  	int slots;
>  
> -	mstc->pbn = drm_dp_calc_pbn_mode(crtc_state->adjusted_mode.clock,
bpp);
> -
> -	slots = drm_dp_find_vcpi_slots(&mstm->mgr, mstc->pbn);
> -	if (slots < 0)
> -		return slots;
> +	mstc->pbn = drm_dp_calc_pbn_mode(crtc_state->adjusted_mode.clock,
> +					 bpp);
> +	/* Zombies don't need VCPI */
> +	if (!drm_connector_is_unregistered(connector)) {
> +		slots = drm_dp_atomic_find_vcpi_slots(state, &mstm->mgr,
> +						      mstc->port, mstc->pbn);
> +		if (slots < 0)
> +			return slots;
> +	}
>  
>  	return nv50_outp_atomic_check_view(encoder, crtc_state, conn_state,
>  					   mstc->native);
> @@ -920,12 +926,38 @@ nv50_mstc_get_modes(struct drm_connector *connector)
>  	return ret;
>  }
>  
> +static int
> +nv50_mstc_atomic_check(struct drm_connector *connector,
> +		       struct drm_connector_state *new_conn_state)
> +{
> +	struct drm_atomic_state *state = new_conn_state->state;
> +	struct nv50_mstc *mstc = nv50_mstc(connector);
> +	struct drm_dp_mst_topology_mgr *mgr = &mstc->mstm->mgr;
> +	struct drm_connector_state *old_conn_state;
> +	struct drm_crtc *old_crtc;
> +
> +	old_conn_state = drm_atomic_get_old_connector_state(state, connector);
> +	old_crtc = old_conn_state->crtc;
> +
> +	/* We only need to release VCPI here if we're going from having a
CRTC
> +	 * on this connector, to not having one
> +	 */
> +	if (!old_crtc || new_conn_state->crtc)
> +		return 0;
> +
> +	/* Release the previous VCPI allocation since the encoder's
> +	 * atomic_check() won't be called
> +	 */
> +	return drm_dp_atomic_release_vcpi_slots(state, mgr, mstc->port);
> +}
> +
>  static const struct drm_connector_helper_funcs
>  nv50_mstc_help = {
>  	.get_modes = nv50_mstc_get_modes,
>  	.mode_valid = nv50_mstc_mode_valid,
>  	.best_encoder = nv50_mstc_best_encoder,
>  	.atomic_best_encoder = nv50_mstc_atomic_best_encoder,
> +	.atomic_check = nv50_mstc_atomic_check,
>  };
>  
>  static enum drm_connector_status
> @@ -2084,6 +2116,8 @@ nv50_disp_atomic_check(struct drm_device *dev, struct
drm_atomic_state *state)
>  	struct drm_connector *connector;
>  	struct drm_crtc_state *new_crtc_state;
>  	struct drm_crtc *crtc;
> +	struct drm_dp_mst_topology_mgr *mgr;
> +	struct drm_dp_mst_topology_state *mst_state;
>  	int ret, i;
>  
>  	/* We need to handle colour management on a per-plane basis. */
> @@ -2109,6 +2143,12 @@ nv50_disp_atomic_check(struct drm_device *dev,
struct drm_atomic_state *state)
>  			return ret;
>  	}
>  
> +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> +		ret = drm_dp_mst_atomic_check(mst_state);
> +		if (ret)
> +			return ret;
> +	}
For even less code I think you could also push this loop into the helpers.
I can't come up with any scenario where a driver does not want to check
all of them at the same time in the atomic_check sequence.

Otherwise lgtm.
-Daniel
> +
>  	return 0;
>  }
>  
> -- 
> 2.17.2
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Fri, Oct 26, 2018 at 04:35:47PM -0400, Lyude Paul
wrote:
> There has been a TODO waiting for quite a long time in
> drm_dp_mst_topology.c:
> 
> 	/* We cannot rely on port->vcpi.num_slots to update
> 	 * topology_state->avail_slots as the port may not exist if the parent
> 	 * branch device was unplugged. This should be fixed by tracking
> 	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> 	 * depending on the caller to tell us how many slots to release.
> 	 */
> 
> That's not the only reason we should fix this: forcing the driver to
> track the VCPI allocations throughout a state's atomic check is
> error prone, because it means that extra care has to be taken with the
> order that drm_dp_atomic_find_vcpi_slots() and
> drm_dp_atomic_release_vcpi_slots() are called in in order to ensure
> idempotency. Currently the only driver actually using these helpers,
> i915, doesn't even do this correctly: multiple ->best_encoder()
checks
> with i915's current implementation would not be idempotent and would
> over-allocate VCPI slots, something I learned trying to implement
> fallback retraining in MST.
> 
> So: simplify this whole mess, and teach drm_dp_atomic_find_vcpi_slots()
> and drm_dp_atomic_release_vcpi_slots() to track the VCPI allocations for
> each port. This allows us to ensure idempotency without having to rely
> on the driver as much. Additionally: the driver doesn't need to do any
> kind of VCPI slot tracking anymore if it doesn't need it for it's
own
> internal state.
> 
> Additionally; this adds a new drm_dp_mst_atomic_check() helper which
> must be used by atomic drivers to perform validity checks for the new
> VCPI allocations incurred by a state.
> 
> Also: update the documentation and make it more obvious that these
> /must/ be called by /all/ atomic drivers supporting MST.
> 
> Changes since v1:
>  - Don't use the now-removed ->atomic_check() for private objects
hook,
>    just give drivers a function to call themselves
> 
> Signed-off-by: Lyude Paul <lyude at redhat.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> ---
>  drivers/gpu/drm/drm_dp_mst_topology.c | 190 +++++++++++++++++++++-----
>  drivers/gpu/drm/i915/intel_display.c  |   8 ++
>  drivers/gpu/drm/i915/intel_dp_mst.c   |  31 +++--
>  include/drm/drm_dp_mst_helper.h       |  11 +-
>  4 files changed, 192 insertions(+), 48 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
b/drivers/gpu/drm/drm_dp_mst_topology.c
> index 8c3cfac437f4..dcfab7536914 100644
> --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> @@ -2614,21 +2614,33 @@ static int drm_dp_init_vcpi(struct
drm_dp_mst_topology_mgr *mgr,
>  }
>  
>  /**
> - * drm_dp_atomic_find_vcpi_slots() - Find and add vcpi slots to the state
> + * drm_dp_atomic_find_vcpi_slots() - Find and add VCPI slots to the state
>   * @state: global atomic state
>   * @mgr: MST topology manager for the port
>   * @port: port to find vcpi slots for
>   * @pbn: bandwidth required for the mode in PBN
>   *
> + * Allocates VCPI slots to @port, replacing any previous VCPI allocations
it
> + * may have had. Any atomic drivers which support MST must call this
function
> + * in their atomic_check() handlers to change the current VCPI allocation
for
Maybe do a nice kerneldoc reference to the right atomic_check here.
> + * the new state. After the ->atomic_check() hooks of the driver and
all other
This will upset the kerneldoc parser I think.
> + * mode objects in the state have been called, DRM will check the final
VCPI
> + * allocations to ensure that they will fit into the available bandwidth
on
> + * the topology.
> + *
> + * See also: drm_dp_atomic_release_vcpi_slots()
Also need to reference drm_dp_mst_atomic_check() here and that drivers
must call it or nothing happens.
> + *
>   * RETURNS:
> - * Total slots in the atomic state assigned for this port or error
> + * Total slots in the atomic state assigned for this port, or a negative
error
> + * code if the port no longer exists
>   */
>  int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
>  				  struct drm_dp_mst_topology_mgr *mgr,
>  				  struct drm_dp_mst_port *port, int pbn)
>  {
>  	struct drm_dp_mst_topology_state *topology_state;
> -	int req_slots;
> +	struct drm_dp_vcpi_allocation *pos, *vcpi = NULL;
> +	int prev_slots, req_slots, ret;
>  
>  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
>  	if (IS_ERR(topology_state))
> @@ -2637,20 +2649,41 @@ int drm_dp_atomic_find_vcpi_slots(struct
drm_atomic_state *state,
>  	port = drm_dp_get_validated_port_ref(mgr, port);
>  	if (port == NULL)
>  		return -EINVAL;
> -	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> -	DRM_DEBUG_KMS("vcpi slots req=%d, avail=%d\n",
> -			req_slots, topology_state->avail_slots);
>  
> -	if (req_slots > topology_state->avail_slots) {
> -		drm_dp_put_port(port);
> -		return -ENOSPC;
> +	/* Find the current allocation for this port, if any */
> +	list_for_each_entry(pos, &topology_state->vcpis, next) {
> +		if (pos->port == port) {
> +			vcpi = pos;
> +			prev_slots = vcpi->vcpi;
> +			break;
> +		}
>  	}
> +	if (!vcpi)
> +		prev_slots = 0;
For robustness should we warn here, since drivers forgetting to release
vcpi slots is kinda a bug? Or do we need to have this to make life easier
for driver writers?
> +
> +	req_slots = DIV_ROUND_UP(pbn, mgr->pbn_div);
> +
> +	DRM_DEBUG_KMS("[CONNECTOR:%d:%s] [MST PORT:%p] vcpi %d ->
%d\n",
> +		      port->connector->base.id, port->connector->name, port,
> +		      prev_slots, req_slots);
> +
> +	/* Add the new allocation to the state */
> +	if (!vcpi) {
> +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
> +		if (!vcpi) {
> +			ret = -ENOMEM;
> +			goto out;
> +		}
>  
> -	topology_state->avail_slots -= req_slots;
> -	DRM_DEBUG_KMS("vcpi slots avail=%d",
topology_state->avail_slots);
> +		vcpi->port = port;
> +		list_add(&vcpi->next, &topology_state->vcpis);
> +	}
> +	vcpi->vcpi = req_slots;
>  
> +	ret = req_slots;
> +out:
>  	drm_dp_put_port(port);
> -	return req_slots;
> +	return ret;
>  }
>  EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
>  
> @@ -2658,32 +2691,46 @@ EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
>   * drm_dp_atomic_release_vcpi_slots() - Release allocated vcpi slots
>   * @state: global atomic state
>   * @mgr: MST topology manager for the port
> - * @slots: number of vcpi slots to release
> + *
> + * Releases any VCPI slots that have been allocated to a port in the
atomic
> + * state. Any atomic drivers which support MST must call this function in
> + * their connector's atomic_check() handler when the connector will no
longer
> + * have VCPI allocated (e.g. because it's CRTC was removed).
> + *
> + * It is OK to call this even if @port has been removed from the system,
in
> + * which case it will just amount to a no-op.
> + *
> + * See also: drm_dp_atomic_find_vcpi_slots()
Same comments as with the _find_vcpi_slots() function.
>   *
>   * RETURNS:
> - * 0 if @slots were added back to
&drm_dp_mst_topology_state->avail_slots or
> - * negative error code
> + * 0 if all slots for this port were added back to
> + * &drm_dp_mst_topology_state->avail_slots or negative error code
>   */
>  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
>  				     struct drm_dp_mst_topology_mgr *mgr,
> -				     int slots)
> +				     struct drm_dp_mst_port *port)
>  {
>  	struct drm_dp_mst_topology_state *topology_state;
> +	struct drm_dp_vcpi_allocation *tmp, *pos;
>  
>  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
>  	if (IS_ERR(topology_state))
>  		return PTR_ERR(topology_state);
>  
> -	/* We cannot rely on port->vcpi.num_slots to update
> -	 * topology_state->avail_slots as the port may not exist if the parent
> -	 * branch device was unplugged. This should be fixed by tracking
> -	 * per-port slot allocation in drm_dp_mst_topology_state instead of
> -	 * depending on the caller to tell us how many slots to release.
> -	 */
> -	topology_state->avail_slots += slots;
> -	DRM_DEBUG_KMS("vcpi slots released=%d, avail=%d\n",
> -			slots, topology_state->avail_slots);
> +	list_for_each_entry_safe(pos, tmp, &topology_state->vcpis, next) {
> +		if (pos->port == port) {
> +			list_del(&pos->next);
> +			DRM_DEBUG_KMS("[MST PORT:%p] vcpi %d -> 0\n",
> +				      port, pos->vcpi);
>  
> +			kfree(pos);
> +			return 0;
> +		}
> +	}
> +
> +	/* If no allocation was found, all that means is that the port was
> +	 * destroyed since the last atomic commit. That's OK!
> +	 */
>  	return 0;
>  }
>  EXPORT_SYMBOL(drm_dp_atomic_release_vcpi_slots);
> @@ -3112,15 +3159,50 @@ static void drm_dp_destroy_connector_work(struct
work_struct *work)
>  static struct drm_private_state *
>  drm_dp_mst_duplicate_state(struct drm_private_obj *obj)
>  {
> -	struct drm_dp_mst_topology_state *state;
> +	struct drm_dp_mst_topology_state *state, *old_state > +	
to_dp_mst_topology_state(obj->state);
> +	struct drm_dp_mst_topology_mgr *mgr = old_state->mgr;
> +	struct drm_dp_mst_port *port;
> +	struct drm_dp_vcpi_allocation *pos, *vcpi;
>  
> -	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
> +	state = kzalloc(sizeof(*state), GFP_KERNEL);
>  	if (!state)
>  		return NULL;
>  
>  	__drm_atomic_helper_private_obj_duplicate_state(obj,
&state->base);
>  
> +	state->mgr = mgr;
> +	INIT_LIST_HEAD(&state->vcpis);
> +
> +	/* Copy over the VCPI allocations for ports that still exist */
> +	list_for_each_entry(pos, &old_state->vcpis, next) {
> +		port = drm_dp_get_validated_port_ref(mgr, pos->port);
> +		if (!port) {
> +			DRM_DEBUG_ATOMIC("[MST PORT:%p] is gone, vcpi %d -> 0\n",
> +					 pos->port, pos->vcpi);
> +			continue;
> +		}
Hm, in general we have 0 validation in the duplicate_state functions, Just
dump copying, with minimal pointer/refcount updates.
> +
> +		vcpi = kzalloc(sizeof(*vcpi), GFP_KERNEL);
kmemdump + updating the list. Just in case we ever add more stuff here.
Probably needs an explicit memset to avoid list debugging.

At least that's the style we use everywhere else.
> +		if (!vcpi) {
> +			drm_dp_put_port(port);
> +			goto fail_alloc;
> +		}
> +
> +		vcpi->port = port;
> +		vcpi->vcpi = pos->vcpi;
> +		list_add(&vcpi->next, &state->vcpis);
> +		drm_dp_put_port(port);
> +	}
> +
>  	return &state->base;
> +
> +fail_alloc:
> +	list_for_each_entry_safe(pos, vcpi, &state->vcpis, next)
> +		kfree(pos);
> +	kfree(state);
> +
> +	return NULL;
>  }
>  
>  static void drm_dp_mst_destroy_state(struct drm_private_obj *obj,
> @@ -3128,14 +3210,60 @@ static void drm_dp_mst_destroy_state(struct
drm_private_obj *obj,
>  {
>  	struct drm_dp_mst_topology_state *mst_state >  	
to_dp_mst_topology_state(state);
> +	struct drm_dp_vcpi_allocation *pos, *tmp;
> +
WARN_ON(!list_empty());

I think this not being empty on mgr teardown would be a driver bug. All
the ports/branch devices should be gone by now, and also all the modeset
state should have been torn down, using drm_atomic_helper_shutdown() or
friends.
> +	list_for_each_entry_safe(pos, tmp, &mst_state->vcpis, next)
> +		kfree(pos);
>  
>  	kfree(mst_state);
>  }
>  
> -static const struct drm_private_state_funcs mst_state_funcs = {
> +/**
> + * drm_dp_mst_atomic_check - Check that the new state of an MST topology
in an
> + * atomic update is valid
> + * @state: Pointer to the new &struct drm_dp_mst_topology_state
> + *
> + * Checks the given topology state for an atomic update to ensure that
it's
> + * valid. This includes checking whether there's enough bandwidth to
support
> + * the new VCPI allocations in the atomic update.
> + *
> + * Any atomic drivers supporting DP MST must make sure to call this after
> + * checking the rest of their state in their ->atomic_check() callback.
&drm_mode_config_funcs.atomic_check
> + *
> + * Returns:
> + *
> + * 0 if the new state is valid, negative error code otherwise.
> + */
> +int drm_dp_mst_atomic_check(struct drm_dp_mst_topology_state *state)
> +{
> +	struct drm_dp_mst_topology_mgr *mgr = state->mgr;
> +	struct drm_dp_vcpi_allocation *pos;
> +	int avail_slots = 63;
> +
> +	list_for_each_entry(pos, &state->vcpis, next) {
> +		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi slots\n",
> +				 pos->port, pos->vcpi);
> +
> +		avail_slots -= pos->vcpi;
> +		if (avail_slots < 0) {
> +			DRM_DEBUG_ATOMIC("[MST PORT:%p] not enough vcpi slots in state %p
(avail=%d)\n",
> +					 pos->port, state,
> +					 avail_slots + pos->vcpi);
> +			return -ENOSPC;
> +		}
> +	}
> +	DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p vcpi avail=%d
used=%d\n",
> +			 mgr, state, avail_slots, 63 - avail_slots);
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL(drm_dp_mst_atomic_check);
Commented on patch 4 already, but I think this would look a bit simpler if
we (also/instead, up to you really) expose a helper that checks all mgr in
a drm_atomic_state. No need to have the exact same loop in each driver.
> +
> +const struct drm_private_state_funcs drm_dp_mst_topology_state_funcs = {
>  	.atomic_duplicate_state = drm_dp_mst_duplicate_state,
>  	.atomic_destroy_state = drm_dp_mst_destroy_state,
>  };
> +EXPORT_SYMBOL(drm_dp_mst_topology_state_funcs);
>  
>  /**
>   * drm_atomic_get_mst_topology_state: get MST topology state
> @@ -3213,13 +3341,11 @@ int drm_dp_mst_topology_mgr_init(struct
drm_dp_mst_topology_mgr *mgr,
>  		return -ENOMEM;
>  
>  	mst_state->mgr = mgr;
> -
> -	/* max. time slots - one slot for MTP header */
> -	mst_state->avail_slots = 63;
> +	INIT_LIST_HEAD(&mst_state->vcpis);
>  
>  	drm_atomic_private_obj_init(&mgr->base,
>  				    &mst_state->base,
> -				    &mst_state_funcs);
> +				    &drm_dp_mst_topology_state_funcs);
>  
>  	return 0;
>  }
> diff --git a/drivers/gpu/drm/i915/intel_display.c
b/drivers/gpu/drm/i915/intel_display.c
> index fe045abb6472..f66af1465686 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -12484,6 +12484,8 @@ static int intel_atomic_check(struct drm_device
*dev,
>  	struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
>  	struct drm_crtc *crtc;
>  	struct drm_crtc_state *old_crtc_state, *crtc_state;
> +	struct drm_dp_mst_topology_mgr *mgr;
> +	struct drm_dp_mst_topology_state *mst_state;
>  	int ret, i;
>  	bool any_ms = false;
>  
> @@ -12534,6 +12536,12 @@ static int intel_atomic_check(struct drm_device
*dev,
>  				       "[modeset]" : "[fastset]");
>  	}
>  
> +	for_each_new_mst_mgr_in_state(state, mgr, mst_state, i) {
> +		ret = drm_dp_mst_atomic_check(mst_state);
> +		if (ret)
> +			return ret;
> +	}
> +
>  	if (any_ms) {
>  		ret = intel_modeset_checks(state);
>  
> diff --git a/drivers/gpu/drm/i915/intel_dp_mst.c
b/drivers/gpu/drm/i915/intel_dp_mst.c
> index 8b71d64ebd9d..aaf904738b78 100644
> --- a/drivers/gpu/drm/i915/intel_dp_mst.c
> +++ b/drivers/gpu/drm/i915/intel_dp_mst.c
> @@ -114,28 +114,31 @@ static int intel_dp_mst_atomic_check(struct
drm_connector *connector,
>  	struct drm_connector_state *old_conn_state;
>  	struct drm_crtc *old_crtc;
>  	struct drm_crtc_state *crtc_state;
> -	int slots, ret = 0;
> +	struct intel_connector *intel_connector > +	
to_intel_connector(connector);
> +	struct drm_dp_mst_topology_mgr *mgr > +	
&intel_connector->mst_port->mst_mgr;
> +	struct drm_dp_mst_port *port = intel_connector->port;
> +	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;
> +	/* Free VCPI, since compute_config() won't be run */
> +	if (!new_conn_state->crtc) {
> +		crtc_state = drm_atomic_get_new_crtc_state(state, old_crtc);
>  
> -		old_encoder = old_conn_state->best_encoder;
> -		mgr = &enc_to_mst(old_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
> -			to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
> +		ret = drm_dp_atomic_release_vcpi_slots(state, mgr, port);
> +		if (ret) {
> +			DRM_DEBUG_KMS("failed releasing vcpi slots: %d\n",
> +				      ret);
> +			return ret;
> +		}
> +		to_intel_crtc_state(crtc_state)->dp_m_n.tu = 0;
>  	}
> +
>  	return ret;
>  }
>  
> diff --git a/include/drm/drm_dp_mst_helper.h
b/include/drm/drm_dp_mst_helper.h
> index 3faceb66f5cb..0aa7d3658013 100644
> --- a/include/drm/drm_dp_mst_helper.h
> +++ b/include/drm/drm_dp_mst_helper.h
> @@ -406,9 +406,15 @@ struct drm_dp_payload {
>  
>  #define to_dp_mst_topology_state(x) container_of(x, struct
drm_dp_mst_topology_state, base)
>  
> +struct drm_dp_vcpi_allocation {
> +	struct drm_dp_mst_port *port;
> +	int vcpi;
> +	struct list_head next;
> +};
> +
>  struct drm_dp_mst_topology_state {
>  	struct drm_private_state base;
> -	int avail_slots;
> +	struct list_head vcpis;
>  	struct drm_dp_mst_topology_mgr *mgr;
>  };
>  
> @@ -624,9 +630,10 @@ int drm_dp_atomic_find_vcpi_slots(struct
drm_atomic_state *state,
>  				  struct drm_dp_mst_port *port, int pbn);
I think annotating the relase/find functions with __must_check would be
good.
>  int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
>  				     struct drm_dp_mst_topology_mgr *mgr,
> -				     int slots);
> +				     struct drm_dp_mst_port *port);
>  int drm_dp_send_power_updown_phy(struct drm_dp_mst_topology_mgr *mgr,
>  				 struct drm_dp_mst_port *port, bool power_up);
Same here.
> +int drm_dp_mst_atomic_check(struct drm_dp_mst_topology_state *state);
With the nits addressed:

Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>
>  
>  extern const struct drm_private_state_funcs
drm_dp_mst_topology_state_funcs;
>  
> -- 
> 2.17.2
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

On Fri, Oct 26, 2018 at 04:35:48PM -0400, Lyude Paul
wrote:
> It occurred to me that we never actually check this! So let's start
> doing that.
> 
> Signed-off-by: Lyude Paul <lyude at redhat.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>

One thought on testing all this: I think long term some unti tests, where
we have a fake driver doing fake mst branch/ports and a bunch of
allocations and then checking that it all works and validates would be
nice. Longer term project ofc, and maybe after Kunit has been merged ...
-Daniel
> ---
>  drivers/gpu/drm/drm_dp_mst_topology.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c
b/drivers/gpu/drm/drm_dp_mst_topology.c
> index dcfab7536914..8bb03700e199 100644
> --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> @@ -3238,7 +3238,7 @@ int drm_dp_mst_atomic_check(struct
drm_dp_mst_topology_state *state)
>  {
>  	struct drm_dp_mst_topology_mgr *mgr = state->mgr;
>  	struct drm_dp_vcpi_allocation *pos;
> -	int avail_slots = 63;
> +	int avail_slots = 63, payload_count = 0;
>  
>  	list_for_each_entry(pos, &state->vcpis, next) {
>  		DRM_DEBUG_ATOMIC("[MST PORT:%p] requires %d vcpi slots\n",
> @@ -3251,6 +3251,12 @@ int drm_dp_mst_atomic_check(struct
drm_dp_mst_topology_state *state)
>  					 avail_slots + pos->vcpi);
>  			return -ENOSPC;
>  		}
> +
> +		if (++payload_count > mgr->max_payloads) {
> +			DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p has too many payloads
(max=%d)\n",
> +					 mgr, state, mgr->max_payloads);
> +			return -EINVAL;
> +		}
>  	}
>  	DRM_DEBUG_ATOMIC("[MST MGR:%p] state %p vcpi avail=%d
used=%d\n",
>  			 mgr, state, avail_slots, 63 - avail_slots);
> -- 
> 2.17.2
> 
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch