mlxsw: PCI: Send EMAD traffic on a separate queue

Currently mlxsw distributes sent traffic among all the available send
queues. That includes control traffic as well as EMADs, which are used for
configuration of the device.

However because all the queues have the same traffic class of 3, they all
end up being directed to the same traffic class buffer. If the control
traffic in the buffer cannot be serviced quickly enough, the EMAD traffic
might be shut out, which causes transient failures, typically in FDB
maintenance, counter upkeep and other periodic work.

To address this issue, dedicate SDQ 0 to EMAD traffic, with TC 0.
Distribute the control traffic among the remaining queues, which are left
with their current TC 3.
Suggested-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/mellanox/mlxsw/pci.c

@@ -284,15 +284,18 @@ static dma_addr_t __mlxsw_pci_queue_page_get(struct mlxsw_pci_queue *q,
 static int mlxsw_pci_sdq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
 			      struct mlxsw_pci_queue *q)
 {
+	int tclass;
 	int i;
 	int err;
 
 	q->producer_counter = 0;
 	q->consumer_counter = 0;
+	tclass = q->num == MLXSW_PCI_SDQ_EMAD_INDEX ? MLXSW_PCI_SDQ_EMAD_TC :
+						      MLXSW_PCI_SDQ_CTL_TC;
 
 	/* Set CQ of same number of this SDQ. */
 	mlxsw_cmd_mbox_sw2hw_dq_cq_set(mbox, q->num);
-	mlxsw_cmd_mbox_sw2hw_dq_sdq_tclass_set(mbox, 3);
+	mlxsw_cmd_mbox_sw2hw_dq_sdq_tclass_set(mbox, tclass);
 	mlxsw_cmd_mbox_sw2hw_dq_log2_dq_sz_set(mbox, 3); /* 8 pages */
 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
@@ -963,6 +966,7 @@ static int mlxsw_pci_aqs_init(struct mlxsw_pci *mlxsw_pci, char *mbox)
 	eq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_eq_sz_get(mbox);
 
 	if (num_sdqs + num_rdqs > num_cqs ||
+	    num_sdqs < MLXSW_PCI_SDQS_MIN ||
 	    num_cqs > MLXSW_PCI_CQS_MAX || num_eqs != MLXSW_PCI_EQS_COUNT) {
 		dev_err(&pdev->dev, "Unsupported number of queues\n");
 		return -EINVAL;
@@ -1520,7 +1524,15 @@ static struct mlxsw_pci_queue *
 mlxsw_pci_sdq_pick(struct mlxsw_pci *mlxsw_pci,
 		   const struct mlxsw_tx_info *tx_info)
 {
-	u8 sdqn = tx_info->local_port % mlxsw_pci_sdq_count(mlxsw_pci);
+	u8 ctl_sdq_count = mlxsw_pci_sdq_count(mlxsw_pci) - 1;
+	u8 sdqn;
+
+	if (tx_info->is_emad) {
+		sdqn = MLXSW_PCI_SDQ_EMAD_INDEX;
+	} else {
+		BUILD_BUG_ON(MLXSW_PCI_SDQ_EMAD_INDEX != 0);
+		sdqn = 1 + (tx_info->local_port % ctl_sdq_count);
+	}
 
 	return mlxsw_pci_sdq_get(mlxsw_pci, sdqn);
 }

drivers/net/ethernet/mellanox/mlxsw/pci_hw.h

@@ -51,6 +51,11 @@
 #define MLXSW_PCI_EQ_ASYNC_NUM	0
 #define MLXSW_PCI_EQ_COMP_NUM	1
 
+#define MLXSW_PCI_SDQS_MIN	2 /* EMAD and control traffic */
+#define MLXSW_PCI_SDQ_EMAD_INDEX	0
+#define MLXSW_PCI_SDQ_EMAD_TC	0
+#define MLXSW_PCI_SDQ_CTL_TC	3
+
 #define MLXSW_PCI_AQ_PAGES	8
 #define MLXSW_PCI_AQ_SIZE	(MLXSW_PCI_PAGE_SIZE * MLXSW_PCI_AQ_PAGES)
 #define MLXSW_PCI_WQE_SIZE	32 /* 32 bytes per element */