From: Victor Raj <victor.raj@intel.com>
Date: Mon, 13 Jul 2020 13:53:12 -0700
Subject: ice: distribute Tx queues evenly
Patch-mainline: v5.9-rc1
Git-commit: 4043818c13ef5e610e6555b02d69dfd8d0a6b32d
References: jsc#SLE-12878

Distribute the Tx queues evenly across all queue groups. This will
help the queues to get more equal sharing among the queues when all
are in use.

In the previous algorithm, the next queue group node will be picked up
only after the previous one filled with max children.
For example: if VSI is configured with 9 queues, the first 8 queues
will be assigned to queue group 1 and the 9th queue will be assigned to
queue group 2.

The 2 queue groups split the bandwidth between them equally (50:50).
The first queue group node will share the 50% bandwidth with all of
its children (8 queues). And the second queue group node will share
the entire 50% bandwidth with its only children.

The new algorithm will fix this issue.

Signed-off-by: Victor Raj <victor.raj@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 drivers/net/ethernet/intel/ice/ice_sched.c |   55 ++++++++++++++++++++++++++---
 1 file changed, 51 insertions(+), 4 deletions(-)

--- a/drivers/net/ethernet/intel/ice/ice_sched.c
+++ b/drivers/net/ethernet/intel/ice/ice_sched.c
@@ -1276,6 +1276,53 @@ ice_sched_find_node_in_subtree(struct ic
 }
 
 /**
+ * ice_sched_get_free_qgrp - Scan all queue group siblings and find a free node
+ * @pi: port information structure
+ * @vsi_node: software VSI handle
+ * @qgrp_node: first queue group node identified for scanning
+ * @owner: LAN or RDMA
+ *
+ * This function retrieves a free LAN or RDMA queue group node by scanning
+ * qgrp_node and its siblings for the queue group with the fewest number
+ * of queues currently assigned.
+ */
+static struct ice_sched_node *
+ice_sched_get_free_qgrp(struct ice_port_info *pi,
+			struct ice_sched_node *vsi_node,
+			struct ice_sched_node *qgrp_node, u8 owner)
+{
+	struct ice_sched_node *min_qgrp;
+	u8 min_children;
+
+	if (!qgrp_node)
+		return qgrp_node;
+	min_children = qgrp_node->num_children;
+	if (!min_children)
+		return qgrp_node;
+	min_qgrp = qgrp_node;
+	/* scan all queue groups until find a node which has less than the
+	 * minimum number of children. This way all queue group nodes get
+	 * equal number of shares and active. The bandwidth will be equally
+	 * distributed across all queues.
+	 */
+	while (qgrp_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 < min_children &&
+			    qgrp_node->owner == owner) {
+				/* replace the new min queue group node */
+				min_qgrp = qgrp_node;
+				min_children = min_qgrp->num_children;
+				/* break if it has no children, */
+				if (!min_children)
+					break;
+			}
+		qgrp_node = qgrp_node->sibling;
+	}
+	return min_qgrp;
+}
+
+/**
  * ice_sched_get_free_qparent - Get a free LAN or RDMA queue group node
  * @pi: port information structure
  * @vsi_handle: software VSI handle
@@ -1288,7 +1335,7 @@ struct ice_sched_node *
 ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
 			   u8 owner)
 {
-	struct ice_sched_node *vsi_node, *qgrp_node = NULL;
+	struct ice_sched_node *vsi_node, *qgrp_node;
 	struct ice_vsi_ctx *vsi_ctx;
 	u16 max_children;
 	u8 qgrp_layer;
@@ -1302,7 +1349,7 @@ ice_sched_get_free_qparent(struct ice_po
 	vsi_node = vsi_ctx->sched.vsi_node[tc];
 	/* validate invalid VSI ID */
 	if (!vsi_node)
-		goto lan_q_exit;
+		return NULL;
 
 	/* get the first queue group node from VSI sub-tree */
 	qgrp_node = ice_sched_get_first_node(pi, vsi_node, qgrp_layer);
@@ -1315,8 +1362,8 @@ ice_sched_get_free_qparent(struct ice_po
 		qgrp_node = qgrp_node->sibling;
 	}
 
-lan_q_exit:
-	return qgrp_node;
+	/* Select the best queue group */
+	return ice_sched_get_free_qgrp(pi, vsi_node, qgrp_node, owner);
 }
 
 /**