[PATCH 3/3] drm/dp_mst: Rewrite and fix bandwidth limit checks
Lyude Paul
lyude at redhat.com
Wed Mar 4 22:36:13 UTC 2020
Sigh, this is mostly my fault for not giving commit cd82d82cbc04
("drm/dp_mst: Add branch bandwidth validation to MST atomic check")
enough scrutiny during review. The way we're checking bandwidth
limitations here is mostly wrong.
First things first, we need to follow the locking conventions for MST.
Whenever traversing downwards (upwards is always safe) in the topology,
we need to hold &mgr->lock to prevent the topology from changing under
us. We don't currently do that when performing bandwidth limit checks.
Next we need to figure out the actual PBN limit for the primary MSTB.
Here we actually want to use the highest available_pbn value we can find
on each level of the topology, then make sure that the combined sum of
allocated PBN on each port of the branch device doesn't exceed that
amount. Currently, we just loop through each level of the topology and
use the last non-zero PBN we find.
Once we've done that, we then want to traverse down each branch device
we find in the topology with at least one downstream port that has PBN
allocated in our atomic state, and repeat the whole process on each
level of the topology as we travel down. While doing this, we need to
take care to avoid attempting to traverse down end devices. We don't
currently do this, although I'm not actually sure whether or not this
broke anything before.
Since there's a bit too many issues here to try to fix one by one, and
the drm_dp_mst_atomic_check_bw_limit() code is not entirely clear on all
of these pain points anyway, let's just take the easy way out and
rewrite the whole function. Additionally, we also add a kernel warning
if we find that any ports we performed bandwidth limit checks on didn't
actually have available_pbn populated - as this is always a bug in the
MST helpers.
This should fix regressions seen on nouveau, i915 and amdgpu where we
erroneously reject atomic states that should fit within bandwidth
limitations.
Signed-off-by: Lyude Paul <lyude at redhat.com>
Fixes: cd82d82cbc04 ("drm/dp_mst: Add branch bandwidth validation to MST atomic check")
Cc: Mikita Lipski <mikita.lipski at amd.com>
Cc: Alex Deucher <alexander.deucher at amd.com>
Cc: Sean Paul <seanpaul at google.com>
Cc: Hans de Goede <hdegoede at redhat.com>
---
drivers/gpu/drm/drm_dp_mst_topology.c | 101 ++++++++++++++++++++------
1 file changed, 78 insertions(+), 23 deletions(-)
diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c b/drivers/gpu/drm/drm_dp_mst_topology.c
index 7b0ff0cff954..87dc7c92d339 100644
--- a/drivers/gpu/drm/drm_dp_mst_topology.c
+++ b/drivers/gpu/drm/drm_dp_mst_topology.c
@@ -4853,41 +4853,90 @@ static bool drm_dp_mst_port_downstream_of_branch(struct drm_dp_mst_port *port,
return false;
}
-static inline
-int drm_dp_mst_atomic_check_bw_limit(struct drm_dp_mst_branch *branch,
- struct drm_dp_mst_topology_state *mst_state)
+static int
+drm_dp_mst_atomic_check_bw_limit(struct drm_dp_mst_branch *branch,
+ struct drm_dp_mst_topology_state *mst_state)
{
struct drm_dp_mst_port *port;
struct drm_dp_vcpi_allocation *vcpi;
- int pbn_limit = 0, pbn_used = 0;
+ int pbn_limit = 0, pbn_used = 0, ret;
- list_for_each_entry(port, &branch->ports, next) {
- if (port->mstb)
- if (drm_dp_mst_atomic_check_bw_limit(port->mstb, mst_state))
- return -ENOSPC;
+ if (branch->port_parent)
+ DRM_DEBUG_ATOMIC("[MSTB:%p] [MST PORT:%p] checking [MSTB:%p]\n",
+ branch->port_parent->parent,
+ branch->port_parent, branch);
+ else
+ DRM_DEBUG_ATOMIC("Checking [MSTB:%p]\n", branch);
- if (port->available_pbn > 0)
+ list_for_each_entry(port, &branch->ports, next) {
+ /* Since each port shares a link, the highest PBN we find
+ * should be assumed to be the limit for this branch device
+ */
+ if (pbn_limit < port->available_pbn)
pbn_limit = port->available_pbn;
- }
- DRM_DEBUG_ATOMIC("[MST BRANCH:%p] branch has %d PBN available\n",
- branch, pbn_limit);
- list_for_each_entry(vcpi, &mst_state->vcpis, next) {
- if (!vcpi->pbn)
+ if (port->pdt == DP_PEER_DEVICE_NONE)
continue;
- if (drm_dp_mst_port_downstream_of_branch(vcpi->port, branch))
- pbn_used += vcpi->pbn;
+ if (drm_dp_mst_is_end_device(port->pdt, port->mcs)) {
+ list_for_each_entry(vcpi, &mst_state->vcpis, next) {
+ if (vcpi->port != port)
+ continue;
+ if (!vcpi->pbn)
+ break;
+
+ /* This should never happen, as it means we
+ * tried to set a mode before querying the
+ * available_pbn
+ */
+ if (WARN_ON(!port->available_pbn))
+ return -EINVAL;
+
+ if (vcpi->pbn > port->available_pbn) {
+ DRM_DEBUG_ATOMIC("[MSTB:%p] [MST PORT:%p] %d exceeds available PBN of %d\n",
+ branch, port,
+ vcpi->pbn,
+ port->available_pbn);
+ return -ENOSPC;
+ }
+
+ DRM_DEBUG_ATOMIC("[MSTB:%p] [MST PORT:%p] using %d PBN\n",
+ branch, port, vcpi->pbn);
+ pbn_used += vcpi->pbn;
+ break;
+ }
+ } else {
+ list_for_each_entry(vcpi, &mst_state->vcpis, next) {
+ if (!vcpi->pbn ||
+ !drm_dp_mst_port_downstream_of_branch(vcpi->port,
+ port->mstb))
+ continue;
+
+ ret = drm_dp_mst_atomic_check_bw_limit(port->mstb,
+ mst_state);
+ if (ret < 0)
+ return ret;
+
+ pbn_used += ret;
+ break;
+ }
+ }
}
- DRM_DEBUG_ATOMIC("[MST BRANCH:%p] branch used %d PBN\n",
- branch, pbn_used);
+ if (!pbn_used)
+ return 0;
+
+ DRM_DEBUG_ATOMIC("[MSTB:%p] has total available PBN of %d\n",
+ branch, pbn_limit);
if (pbn_used > pbn_limit) {
- DRM_DEBUG_ATOMIC("[MST BRANCH:%p] No available bandwidth\n",
- branch);
+ DRM_DEBUG_ATOMIC("[MSTB:%p] Not enough bandwidth (need: %d)\n",
+ branch, pbn_used);
return -ENOSPC;
}
- return 0;
+
+ DRM_DEBUG_ATOMIC("[MSTB:%p] using %d PBN\n", branch, pbn_used);
+
+ return pbn_used;
}
static inline int
@@ -5085,9 +5134,15 @@ int drm_dp_mst_atomic_check(struct drm_atomic_state *state)
ret = drm_dp_mst_atomic_check_vcpi_alloc_limit(mgr, mst_state);
if (ret)
break;
- ret = drm_dp_mst_atomic_check_bw_limit(mgr->mst_primary, mst_state);
- if (ret)
+
+ mutex_lock(&mgr->lock);
+ ret = drm_dp_mst_atomic_check_bw_limit(mgr->mst_primary,
+ mst_state);
+ mutex_unlock(&mgr->lock);
+ if (ret < 0)
break;
+ else
+ ret = 0;
}
return ret;
--
2.24.1
More information about the amd-gfx
mailing list