Merge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2025-05-30 (ice, idpf)

For ice:
Michal resolves XDP issues related to Tx scheduler configuration with
large number of Tx queues.

Additional information:
https://lore.kernel.org/intel-wired-lan/20250513105529.241745-1-michal.kubiak@intel.com/

For idpf:
Brian Vazquez updates netif_subqueue_maybe_stop() condition check to
prevent possible races.

Emil shuts down virtchannel mailbox during reset to reduce timeout
delays as it's unavailable during that time.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
  idpf: avoid mailbox timeout delays during reset
  idpf: fix a race in txq wakeup
  ice: fix rebuilding the Tx scheduler tree for large queue counts
  ice: create new Tx scheduler nodes for new queues only
  ice: fix Tx scheduler error handling in XDP callback
====================

Link: https://patch.msgid.link/20250530211221.2170484-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/intel/ice/ice_main.c

@@ -2740,6 +2740,27 @@ void ice_map_xdp_rings(struct ice_vsi *vsi)
 	}
 }
 
+/**
+ * ice_unmap_xdp_rings - Unmap XDP rings from interrupt vectors
+ * @vsi: the VSI with XDP rings being unmapped
+ */
+static void ice_unmap_xdp_rings(struct ice_vsi *vsi)
+{
+	int v_idx;
+
+	ice_for_each_q_vector(vsi, v_idx) {
+		struct ice_q_vector *q_vector = vsi->q_vectors[v_idx];
+		struct ice_tx_ring *ring;
+
+		ice_for_each_tx_ring(ring, q_vector->tx)
+			if (!ring->tx_buf || !ice_ring_is_xdp(ring))
+				break;
+
+		/* restore the value of last node prior to XDP setup */
+		q_vector->tx.tx_ring = ring;
+	}
+}
+
 /**
  * ice_prepare_xdp_rings - Allocate, configure and setup Tx rings for XDP
  * @vsi: VSI to bring up Tx rings used by XDP
@@ -2803,7 +2824,7 @@ int ice_prepare_xdp_rings(struct ice_vsi *vsi, struct bpf_prog *prog,
 	if (status) {
 		dev_err(dev, "Failed VSI LAN queue config for XDP, error: %d\n",
 			status);
-		goto clear_xdp_rings;
+		goto unmap_xdp_rings;
 	}
 
 	/* assign the prog only when it's not already present on VSI;
@@ -2819,6 +2840,8 @@ int ice_prepare_xdp_rings(struct ice_vsi *vsi, struct bpf_prog *prog,
 		ice_vsi_assign_bpf_prog(vsi, prog);
 
 	return 0;
+unmap_xdp_rings:
+	ice_unmap_xdp_rings(vsi);
 clear_xdp_rings:
 	ice_for_each_xdp_txq(vsi, i)
 		if (vsi->xdp_rings[i]) {
@@ -2835,6 +2858,8 @@ err_map_xdp:
 	mutex_unlock(&pf->avail_q_mutex);
 
 	devm_kfree(dev, vsi->xdp_rings);
+	vsi->xdp_rings = NULL;
+
 	return -ENOMEM;
 }
 
@@ -2850,7 +2875,7 @@ int ice_destroy_xdp_rings(struct ice_vsi *vsi, enum ice_xdp_cfg cfg_type)
 {
 	u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 };
 	struct ice_pf *pf = vsi->back;
-	int i, v_idx;
+	int i;
 
 	/* q_vectors are freed in reset path so there's no point in detaching
 	 * rings
@@ -2858,17 +2883,7 @@ int ice_destroy_xdp_rings(struct ice_vsi *vsi, enum ice_xdp_cfg cfg_type)
 	if (cfg_type == ICE_XDP_CFG_PART)
 		goto free_qmap;
 
-	ice_for_each_q_vector(vsi, v_idx) {
-		struct ice_q_vector *q_vector = vsi->q_vectors[v_idx];
-		struct ice_tx_ring *ring;
-
-		ice_for_each_tx_ring(ring, q_vector->tx)
-			if (!ring->tx_buf || !ice_ring_is_xdp(ring))
-				break;
-
-		/* restore the value of last node prior to XDP setup */
-		q_vector->tx.tx_ring = ring;
-	}
+	ice_unmap_xdp_rings(vsi);
 
 free_qmap:
 	mutex_lock(&pf->avail_q_mutex);
@@ -3013,11 +3028,14 @@ ice_xdp_setup_prog(struct ice_vsi *vsi, struct bpf_prog *prog,
 		xdp_ring_err = ice_vsi_determine_xdp_res(vsi);
 		if (xdp_ring_err) {
 			NL_SET_ERR_MSG_MOD(extack, "Not enough Tx resources for XDP");
+			goto resume_if;
 		} else {
 			xdp_ring_err = ice_prepare_xdp_rings(vsi, prog,
 							     ICE_XDP_CFG_FULL);
-			if (xdp_ring_err)
+			if (xdp_ring_err) {
 				NL_SET_ERR_MSG_MOD(extack, "Setting up XDP Tx resources failed");
+				goto resume_if;
+			}
 		}
 		xdp_features_set_redirect_target(vsi->netdev, true);
 		/* reallocate Rx queues that are used for zero-copy */
@@ -3035,6 +3053,7 @@ ice_xdp_setup_prog(struct ice_vsi *vsi, struct bpf_prog *prog,
 			NL_SET_ERR_MSG_MOD(extack, "Freeing XDP Rx resources failed");
 	}
 
+resume_if:
 	if (if_running)
 		ret = ice_up(vsi);
 

drivers/net/ethernet/intel/ice/ice_sched.c

@@ -84,6 +84,27 @@ ice_sched_find_node_by_teid(struct ice_sched_node *start_node, u32 teid)
 	return NULL;
 }
 
+/**
+ * ice_sched_find_next_vsi_node - find the next node for a given VSI
+ * @vsi_node: VSI support node to start search with
+ *
+ * Return: Next VSI support node, or NULL.
+ *
+ * The function returns a pointer to the next node from the VSI layer
+ * assigned to the given VSI, or NULL if there is no such a node.
+ */
+static struct ice_sched_node *
+ice_sched_find_next_vsi_node(struct ice_sched_node *vsi_node)
+{
+	unsigned int vsi_handle = vsi_node->vsi_handle;
+
+	while ((vsi_node = vsi_node->sibling) != NULL)
+		if (vsi_node->vsi_handle == vsi_handle)
+			break;
+
+	return vsi_node;
+}
+
 /**
  * ice_aqc_send_sched_elem_cmd - send scheduling elements cmd
  * @hw: pointer to the HW struct
@@ -1084,8 +1105,10 @@ ice_sched_add_nodes_to_layer(struct ice_port_info *pi,
 		if (parent->num_children < max_child_nodes) {
 			new_num_nodes = max_child_nodes - parent->num_children;
 		} else {
-			/* This parent is full, try the next sibling */
-			parent = parent->sibling;
+			/* This parent is full,
+			 * try the next available sibling.
+			 */
+			parent = ice_sched_find_next_vsi_node(parent);
 			/* Don't modify the first node TEID memory if the
 			 * first node was added already in the above call.
 			 * Instead send some temp memory for all other
@@ -1528,12 +1551,23 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
 	/* get the first queue group node from VSI sub-tree */
 	qgrp_node = ice_sched_get_first_node(pi, vsi_node, qgrp_layer);
 	while (qgrp_node) {
+		struct ice_sched_node *next_vsi_node;
+
 		/* make sure the qgroup node is part of the VSI subtree */
 		if (ice_sched_find_node_in_subtree(pi->hw, vsi_node, qgrp_node))
 			if (qgrp_node->num_children < max_children &&
 			    qgrp_node->owner == owner)
 				break;
 		qgrp_node = qgrp_node->sibling;
+		if (qgrp_node)
+			continue;
+
+		next_vsi_node = ice_sched_find_next_vsi_node(vsi_node);
+		if (!next_vsi_node)
+			break;
+
+		vsi_node = next_vsi_node;
+		qgrp_node = ice_sched_get_first_node(pi, vsi_node, qgrp_layer);
 	}
 
 	/* Select the best queue group */
@@ -1604,16 +1638,16 @@ ice_sched_get_agg_node(struct ice_port_info *pi, struct ice_sched_node *tc_node,
 /**
  * ice_sched_calc_vsi_child_nodes - calculate number of VSI child nodes
  * @hw: pointer to the HW struct
- * @num_qs: number of queues
+ * @num_new_qs: number of new queues that will be added to the tree
  * @num_nodes: num nodes array
  *
  * This function calculates the number of VSI child nodes based on the
  * number of queues.
  */
 static void
-ice_sched_calc_vsi_child_nodes(struct ice_hw *hw, u16 num_qs, u16 *num_nodes)
+ice_sched_calc_vsi_child_nodes(struct ice_hw *hw, u16 num_new_qs, u16 *num_nodes)
 {
-	u16 num = num_qs;
+	u16 num = num_new_qs;
 	u8 i, qgl, vsil;
 
 	qgl = ice_sched_get_qgrp_layer(hw);
@@ -1779,7 +1813,11 @@ ice_sched_add_vsi_support_nodes(struct ice_port_info *pi, u16 vsi_handle,
 		if (!parent)
 			return -EIO;
 
-		if (i == vsil)
+		/* Do not modify the VSI handle for already existing VSI nodes,
+		 * (if no new VSI node was added to the tree).
+		 * Assign the VSI handle only to newly added VSI nodes.
+		 */
+		if (i == vsil && num_added)
 			parent->vsi_handle = vsi_handle;
 	}
 
@@ -1812,6 +1850,41 @@ ice_sched_add_vsi_to_topo(struct ice_port_info *pi, u16 vsi_handle, u8 tc)
 					       num_nodes);
 }
 
+/**
+ * ice_sched_recalc_vsi_support_nodes - recalculate VSI support nodes count
+ * @hw: pointer to the HW struct
+ * @vsi_node: pointer to the leftmost VSI node that needs to be extended
+ * @new_numqs: new number of queues that has to be handled by the VSI
+ * @new_num_nodes: pointer to nodes count table to modify the VSI layer entry
+ *
+ * This function recalculates the number of supported nodes that need to
+ * be added after adding more Tx queues for a given VSI.
+ * The number of new VSI support nodes that shall be added will be saved
+ * to the @new_num_nodes table for the VSI layer.
+ */
+static void
+ice_sched_recalc_vsi_support_nodes(struct ice_hw *hw,
+				   struct ice_sched_node *vsi_node,
+				   unsigned int new_numqs, u16 *new_num_nodes)
+{
+	u32 vsi_nodes_cnt = 1;
+	u32 max_queue_cnt = 1;
+	u32 qgl, vsil;
+
+	qgl = ice_sched_get_qgrp_layer(hw);
+	vsil = ice_sched_get_vsi_layer(hw);
+
+	for (u32 i = vsil; i <= qgl; i++)
+		max_queue_cnt *= hw->max_children[i];
+
+	while ((vsi_node = ice_sched_find_next_vsi_node(vsi_node)) != NULL)
+		vsi_nodes_cnt++;
+
+	if (new_numqs > (max_queue_cnt * vsi_nodes_cnt))
+		new_num_nodes[vsil] = DIV_ROUND_UP(new_numqs, max_queue_cnt) -
+				      vsi_nodes_cnt;
+}
+
 /**
  * ice_sched_update_vsi_child_nodes - update VSI child nodes
  * @pi: port information structure
@@ -1863,15 +1936,25 @@ ice_sched_update_vsi_child_nodes(struct ice_port_info *pi, u16 vsi_handle,
 			return status;
 	}
 
-	if (new_numqs)
-		ice_sched_calc_vsi_child_nodes(hw, new_numqs, new_num_nodes);
-	/* Keep the max number of queue configuration all the time. Update the
-	 * tree only if number of queues > previous number of queues. This may
+	ice_sched_recalc_vsi_support_nodes(hw, vsi_node,
+					   new_numqs, new_num_nodes);
+	ice_sched_calc_vsi_child_nodes(hw, new_numqs - prev_numqs,
+				       new_num_nodes);
+
+	/* Never decrease the number of queues in the tree. Update the tree
+	 * only if number of queues > previous number of queues. This may
 	 * leave some extra nodes in the tree if number of queues < previous
 	 * number but that wouldn't harm anything. Removing those extra nodes
 	 * may complicate the code if those nodes are part of SRL or
 	 * individually rate limited.
+	 * Also, add the required VSI support nodes if the existing ones cannot
+	 * handle the requested new number of queues.
 	 */
+	status = ice_sched_add_vsi_support_nodes(pi, vsi_handle, tc_node,
+						 new_num_nodes);
+	if (status)
+		return status;
+
 	status = ice_sched_add_vsi_child_nodes(pi, vsi_handle, tc_node,
 					       new_num_nodes, owner);
 	if (status)
@@ -2012,6 +2095,58 @@ static bool ice_sched_is_leaf_node_present(struct ice_sched_node *node)
 	return (node->info.data.elem_type == ICE_AQC_ELEM_TYPE_LEAF);
 }
 
+/**
+ * ice_sched_rm_vsi_subtree - remove all nodes assigned to a given VSI
+ * @pi: port information structure
+ * @vsi_node: pointer to the leftmost node of the VSI to be removed
+ * @owner: LAN or RDMA
+ * @tc: TC number
+ *
+ * Return: Zero in case of success, or -EBUSY if the VSI has leaf nodes in TC.
+ *
+ * This function removes all the VSI support nodes associated with a given VSI
+ * and its LAN or RDMA children nodes from the scheduler tree.
+ */
+static int
+ice_sched_rm_vsi_subtree(struct ice_port_info *pi,
+			 struct ice_sched_node *vsi_node, u8 owner, u8 tc)
+{
+	u16 vsi_handle = vsi_node->vsi_handle;
+	bool all_vsi_nodes_removed = true;
+	int j = 0;
+
+	while (vsi_node) {
+		struct ice_sched_node *next_vsi_node;
+
+		if (ice_sched_is_leaf_node_present(vsi_node)) {
+			ice_debug(pi->hw, ICE_DBG_SCHED, "VSI has leaf nodes in TC %d\n", tc);
+			return -EBUSY;
+		}
+		while (j < vsi_node->num_children) {
+			if (vsi_node->children[j]->owner == owner)
+				ice_free_sched_node(pi, vsi_node->children[j]);
+			else
+				j++;
+		}
+
+		next_vsi_node = ice_sched_find_next_vsi_node(vsi_node);
+
+		/* remove the VSI if it has no children */
+		if (!vsi_node->num_children)
+			ice_free_sched_node(pi, vsi_node);
+		else
+			all_vsi_nodes_removed = false;
+
+		vsi_node = next_vsi_node;
+	}
+
+	/* clean up aggregator related VSI info if any */
+	if (all_vsi_nodes_removed)
+		ice_sched_rm_agg_vsi_info(pi, vsi_handle);
+
+	return 0;
+}
+
 /**
  * ice_sched_rm_vsi_cfg - remove the VSI and its children nodes
  * @pi: port information structure
@@ -2038,7 +2173,6 @@ ice_sched_rm_vsi_cfg(struct ice_port_info *pi, u16 vsi_handle, u8 owner)
 
 	ice_for_each_traffic_class(i) {
 		struct ice_sched_node *vsi_node, *tc_node;
-		u8 j = 0;
 
 		tc_node = ice_sched_get_tc_node(pi, i);
 		if (!tc_node)
@@ -2048,31 +2182,12 @@ ice_sched_rm_vsi_cfg(struct ice_port_info *pi, u16 vsi_handle, u8 owner)
 		if (!vsi_node)
 			continue;
 
-		if (ice_sched_is_leaf_node_present(vsi_node)) {
-			ice_debug(pi->hw, ICE_DBG_SCHED, "VSI has leaf nodes in TC %d\n", i);
-			status = -EBUSY;
+		status = ice_sched_rm_vsi_subtree(pi, vsi_node, owner, i);
+		if (status)
 			goto exit_sched_rm_vsi_cfg;
-		}
-		while (j < vsi_node->num_children) {
-			if (vsi_node->children[j]->owner == owner) {
-				ice_free_sched_node(pi, vsi_node->children[j]);
 
-				/* reset the counter again since the num
-				 * children will be updated after node removal
-				 */
-				j = 0;
-			} else {
-				j++;
-			}
-		}
-		/* remove the VSI if it has no children */
-		if (!vsi_node->num_children) {
-			ice_free_sched_node(pi, vsi_node);
-			vsi_ctx->sched.vsi_node[i] = NULL;
+		vsi_ctx->sched.vsi_node[i] = NULL;
 
-			/* clean up aggregator related VSI info if any */
-			ice_sched_rm_agg_vsi_info(pi, vsi_handle);
-		}
 		if (owner == ICE_SCHED_NODE_OWNER_LAN)
 			vsi_ctx->sched.max_lanq[i] = 0;
 		else

drivers/net/ethernet/intel/idpf/idpf_lib.c

@@ -1801,11 +1801,19 @@ void idpf_vc_event_task(struct work_struct *work)
 	if (test_bit(IDPF_REMOVE_IN_PROG, adapter->flags))
 		return;
 
-	if (test_bit(IDPF_HR_FUNC_RESET, adapter->flags) ||
-	    test_bit(IDPF_HR_DRV_LOAD, adapter->flags)) {
-		set_bit(IDPF_HR_RESET_IN_PROG, adapter->flags);
-		idpf_init_hard_reset(adapter);
-	}
+	if (test_bit(IDPF_HR_FUNC_RESET, adapter->flags))
+		goto func_reset;
+
+	if (test_bit(IDPF_HR_DRV_LOAD, adapter->flags))
+		goto drv_load;
+
+	return;
+
+func_reset:
+	idpf_vc_xn_shutdown(adapter->vcxn_mngr);
+drv_load:
+	set_bit(IDPF_HR_RESET_IN_PROG, adapter->flags);
+	idpf_init_hard_reset(adapter);
 }
 
 /**

drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c

@@ -362,17 +362,18 @@ netdev_tx_t idpf_tx_singleq_frame(struct sk_buff *skb,
 {
 	struct idpf_tx_offload_params offload = { };
 	struct idpf_tx_buf *first;
+	int csum, tso, needed;
 	unsigned int count;
 	__be16 protocol;
-	int csum, tso;
 
 	count = idpf_tx_desc_count_required(tx_q, skb);
 	if (unlikely(!count))
 		return idpf_tx_drop_skb(tx_q, skb);
 
-	if (idpf_tx_maybe_stop_common(tx_q,
-				      count + IDPF_TX_DESCS_PER_CACHE_LINE +
-				      IDPF_TX_DESCS_FOR_CTX)) {
+	needed = count + IDPF_TX_DESCS_PER_CACHE_LINE + IDPF_TX_DESCS_FOR_CTX;
+	if (!netif_subqueue_maybe_stop(tx_q->netdev, tx_q->idx,
+				       IDPF_DESC_UNUSED(tx_q),
+				       needed, needed)) {
 		idpf_tx_buf_hw_update(tx_q, tx_q->next_to_use, false);
 
 		u64_stats_update_begin(&tx_q->stats_sync);

drivers/net/ethernet/intel/idpf/idpf_txrx.c

@@ -2184,6 +2184,19 @@ void idpf_tx_splitq_build_flow_desc(union idpf_tx_flex_desc *desc,
 	desc->flow.qw1.compl_tag = cpu_to_le16(params->compl_tag);
 }
 
+/* Global conditions to tell whether the txq (and related resources)
+ * has room to allow the use of "size" descriptors.
+ */
+static int idpf_txq_has_room(struct idpf_tx_queue *tx_q, u32 size)
+{
+	if (IDPF_DESC_UNUSED(tx_q) < size ||
+	    IDPF_TX_COMPLQ_PENDING(tx_q->txq_grp) >
+		IDPF_TX_COMPLQ_OVERFLOW_THRESH(tx_q->txq_grp->complq) ||
+	    IDPF_TX_BUF_RSV_LOW(tx_q))
+		return 0;
+	return 1;
+}
+
 /**
  * idpf_tx_maybe_stop_splitq - 1st level check for Tx splitq stop conditions
  * @tx_q: the queue to be checked
@@ -2194,29 +2207,11 @@ void idpf_tx_splitq_build_flow_desc(union idpf_tx_flex_desc *desc,
 static int idpf_tx_maybe_stop_splitq(struct idpf_tx_queue *tx_q,
 				     unsigned int descs_needed)
 {
-	if (idpf_tx_maybe_stop_common(tx_q, descs_needed))
-		goto out;
+	if (netif_subqueue_maybe_stop(tx_q->netdev, tx_q->idx,
+				      idpf_txq_has_room(tx_q, descs_needed),
+				      1, 1))
+		return 0;
 
-	/* If there are too many outstanding completions expected on the
-	 * completion queue, stop the TX queue to give the device some time to
-	 * catch up
-	 */
-	if (unlikely(IDPF_TX_COMPLQ_PENDING(tx_q->txq_grp) >
-		     IDPF_TX_COMPLQ_OVERFLOW_THRESH(tx_q->txq_grp->complq)))
-		goto splitq_stop;
-
-	/* Also check for available book keeping buffers; if we are low, stop
-	 * the queue to wait for more completions
-	 */
-	if (unlikely(IDPF_TX_BUF_RSV_LOW(tx_q)))
-		goto splitq_stop;
-
-	return 0;
-
-splitq_stop:
-	netif_stop_subqueue(tx_q->netdev, tx_q->idx);
-
-out:
 	u64_stats_update_begin(&tx_q->stats_sync);
 	u64_stats_inc(&tx_q->q_stats.q_busy);
 	u64_stats_update_end(&tx_q->stats_sync);
@@ -2242,12 +2237,6 @@ void idpf_tx_buf_hw_update(struct idpf_tx_queue *tx_q, u32 val,
 	nq = netdev_get_tx_queue(tx_q->netdev, tx_q->idx);
 	tx_q->next_to_use = val;
 
-	if (idpf_tx_maybe_stop_common(tx_q, IDPF_TX_DESC_NEEDED)) {
-		u64_stats_update_begin(&tx_q->stats_sync);
-		u64_stats_inc(&tx_q->q_stats.q_busy);
-		u64_stats_update_end(&tx_q->stats_sync);
-	}
-
 	/* Force memory writes to complete before letting h/w
 	 * know there are new descriptors to fetch.  (Only
 	 * applicable for weak-ordered memory model archs,

drivers/net/ethernet/intel/idpf/idpf_txrx.h

@@ -1049,12 +1049,4 @@ bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_rx_queue *rxq,
 				      u16 cleaned_count);
 int idpf_tso(struct sk_buff *skb, struct idpf_tx_offload_params *off);
 
-static inline bool idpf_tx_maybe_stop_common(struct idpf_tx_queue *tx_q,
-					     u32 needed)
-{
-	return !netif_subqueue_maybe_stop(tx_q->netdev, tx_q->idx,
-					  IDPF_DESC_UNUSED(tx_q),
-					  needed, needed);
-}
-
 #endif /* !_IDPF_TXRX_H_ */

drivers/net/ethernet/intel/idpf/idpf_virtchnl.c

@@ -347,7 +347,7 @@ static void idpf_vc_xn_init(struct idpf_vc_xn_manager *vcxn_mngr)
  * All waiting threads will be woken-up and their transaction aborted. Further
  * operations on that object will fail.
  */
-static void idpf_vc_xn_shutdown(struct idpf_vc_xn_manager *vcxn_mngr)
+void idpf_vc_xn_shutdown(struct idpf_vc_xn_manager *vcxn_mngr)
 {
 	int i;
 

drivers/net/ethernet/intel/idpf/idpf_virtchnl.h

@@ -150,5 +150,6 @@ int idpf_send_get_stats_msg(struct idpf_vport *vport);
 int idpf_send_set_sriov_vfs_msg(struct idpf_adapter *adapter, u16 num_vfs);
 int idpf_send_get_set_rss_key_msg(struct idpf_vport *vport, bool get);
 int idpf_send_get_set_rss_lut_msg(struct idpf_vport *vport, bool get);
+void idpf_vc_xn_shutdown(struct idpf_vc_xn_manager *vcxn_mngr);
 
 #endif /* _IDPF_VIRTCHNL_H_ */