Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] Data race and stale state usage for `allocated_pbn` between asynchronous commit tail and synchronous atomic check. --
commit 60c9feaab7e5f91e4bacf61ee648e76157adea65 Author: Xiao Lu <[email protected]> drm/dp_mst: Track allocated_pbn from ALLOCATE_PAYLOAD down-reply This commit stores the allocated_pbn returned by the ALLOCATE_PAYLOAD down-reply sideband message and uses it to perform bandwidth limit checks based on the larger of the requested and allocated PBN. > diff --git a/drivers/gpu/drm/display/drm_dp_mst_topology.c > b/drivers/gpu/drm/display/drm_dp_mst_topology.c > index 7ce9e212770ad..b0e4cf8458eda 100644 > --- a/drivers/gpu/drm/display/drm_dp_mst_topology.c > +++ b/drivers/gpu/drm/display/drm_dp_mst_topology.c [ ... ] > @@ -3166,10 +3167,14 @@ static int drm_dp_payload_send_msg(struct > drm_dp_mst_topology_mgr *mgr, > */ > ret = drm_dp_mst_wait_tx_reply(mstb, txmsg); > if (ret > 0) { > - if (txmsg->reply.reply_type == DP_SIDEBAND_REPLY_NAK) > + if (txmsg->reply.reply_type == DP_SIDEBAND_REPLY_NAK) { > ret = -EINVAL; > - else > + } else { > + if (payload) > + payload->allocated_pbn = > + > txmsg->reply.u.allocate_payload.allocated_pbn; > ret = 0; > + } > } > kfree(txmsg); > fail_put: [Severity: High] Does modifying payload->allocated_pbn asynchronously here in drm_dp_payload_send_msg() during the commit tail introduce a data race? When a non-blocking atomic commit updates the topology, its asynchronous commit tail waits for the ALLOCATE_PAYLOAD reply. If a subsequent atomic commit ioctl triggers the atomic check phase concurrently, that check phase duplicates the MST state without locks, which could capture a partially updated or zeroed allocated_pbn. [ ... ] > @@ -4670,6 +4678,7 @@ void drm_dp_mst_atomic_wait_for_dependencies(struct > drm_atomic_commit *state) > new_payload = > drm_atomic_get_mst_payload_state(new_mst_state, > > old_payload->port); > new_payload->vc_start_slot = old_payload->vc_start_slot; > + new_payload->allocated_pbn = old_payload->allocated_pbn; > new_payload->payload_allocation_status = > > old_payload->payload_allocation_status; > } [Severity: High] Since drm_dp_mst_atomic_wait_for_dependencies() is executed in the commit tail after dependencies resolve, does copying allocated_pbn here mean the check phase of the current commit has already executed? If so, the check phase would have already run using the un-updated, stale value of allocated_pbn from the previous commit state before this synchronization occurs. [ ... ] > @@ -5352,7 +5361,7 @@ drm_dp_mst_atomic_check_port_bw_limit(struct > drm_dp_mst_port *port, > return -EINVAL; > } > > - pbn_used = payload->pbn; > + pbn_used = max(payload->pbn, payload->allocated_pbn); > } else { > pbn_used = drm_dp_mst_atomic_check_mstb_bw_limit(port->mstb, > state, [Severity: High] Does reading payload->allocated_pbn here in drm_dp_mst_atomic_check_port_bw_limit() risk bypassing bandwidth validation if the previous hardware reply hasn't arrived yet? If unprivileged userspace (like a Wayland compositor) submits rapid back-to-back atomic commits, the check phase here could evaluate max(payload->pbn, payload->allocated_pbn) using a stale or zeroed value. This might allow configurations that exceed link bandwidth because intermediate branch constraints would be silently ignored. -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=1
