[dpdk-dev] [PATCH v2] ixgbe: fix multi-process support

[Bruce Richardson]

When using multiple processes, the TX function used in all processes
should be the same, otherwise the secondary processes cannot transmit
more than tx-ring-size - 1 packets.
To achieve this, we extract out the code to select the ixgbe TX function
to be used into a separate function inside the ixgbe driver, and call
that from a secondary process when it is attaching to an
already-configured NIC.

Testing with symmetric MP app shows that we are able to RX and TX from
both primary and secondary processes once this patch is applied.

Signed-off-by: Bruce Richardson <bruce.richardson at intel.com>

----
V2 Changes:
* Moved check for primary/secondary process into set_tx_function instead
  of ixgbe_txq_vec_setup, which reduces scope of diff.
* Added/cleaned up some code comments for this change
---
 lib/librte_pmd_ixgbe/ixgbe_ethdev.c | 12 +++++--
 lib/librte_pmd_ixgbe/ixgbe_rxtx.c   | 62 ++++++++++++++++++++++---------------
 lib/librte_pmd_ixgbe/ixgbe_rxtx.h   |  7 +++++
 3 files changed, 54 insertions(+), 27 deletions(-)

diff --git a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
index 937fc3c..c19852b 100644
--- a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
+++ b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
@@ -68,6 +68,7 @@
 #include "ixgbe/ixgbe_common.h"
 #include "ixgbe_ethdev.h"
 #include "ixgbe_bypass.h"
+#include "ixgbe_rxtx.h"
 
 /*
  * High threshold controlling when to start sending XOFF frames. Must be at
@@ -741,10 +742,17 @@ eth_ixgbe_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
 	eth_dev->rx_pkt_burst = &ixgbe_recv_pkts;
 	eth_dev->tx_pkt_burst = &ixgbe_xmit_pkts;
 
-	/* for secondary processes, we don't initialise any further as primary
+	/*
+	 * for secondary processes, we don't initialise any further as primary
 	 * has already done this work. Only check we don't need a different
-	 * RX function */
+	 * RX and TX function
+	 */
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
+		struct igb_tx_queue *txq;
+		/* TX queue function in primary, set by last queue initialized */
+		txq = eth_dev->data->tx_queues[eth_dev->data->nb_tx_queues-1];
+		set_tx_function(eth_dev, txq);
+
 		if (eth_dev->data->scattered_rx)
 			eth_dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
 		return 0;
diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
index 5c36bff..f1d1b25 100644
--- a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
+++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
@@ -1771,6 +1771,41 @@ static struct ixgbe_txq_ops def_txq_ops = {
 	.reset = ixgbe_reset_tx_queue,
 };
 
+/* takes an ethdev and a queue and sets up the tx function to be used based on
+ * the queue parameters. Used in tx_queue_setup by primary process and then
+ * in dev_init by secondary process when attaching to an existing ethdev
+ */
+void
+set_tx_function(struct rte_eth_dev* dev, struct igb_tx_queue* txq)
+{
+	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
+	if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS)
+			&& (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
+		PMD_INIT_LOG(INFO, "Using simple tx code path");
+#ifdef RTE_IXGBE_INC_VECTOR
+		if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
+				(rte_eal_process_type() != RTE_PROC_PRIMARY ||
+					ixgbe_txq_vec_setup(txq) == 0)) {
+			PMD_INIT_LOG(INFO, "Vector tx enabled.");
+			dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
+		}
+		else
+#endif
+		dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
+	} else {
+		PMD_INIT_LOG(INFO, "Using full-featured tx code path");
+		PMD_INIT_LOG(INFO,
+				" - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
+				(long unsigned )txq->txq_flags,
+				(long unsigned)IXGBE_SIMPLE_FLAGS);
+		PMD_INIT_LOG(INFO,
+				" - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
+				(long unsigned )txq->tx_rs_thresh,
+				(long unsigned)RTE_PMD_IXGBE_TX_MAX_BURST);
+		dev->tx_pkt_burst = ixgbe_xmit_pkts;
+	}
+}
+
 int
 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
 			 uint16_t queue_idx,
@@ -1933,31 +1968,8 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
 	PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
 		     txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
 
-	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
-	if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS) &&
-	    (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
-		PMD_INIT_LOG(INFO, "Using simple tx code path");
-#ifdef RTE_IXGBE_INC_VECTOR
-		if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
-		    ixgbe_txq_vec_setup(txq) == 0) {
-			PMD_INIT_LOG(INFO, "Vector tx enabled.");
-			dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
-		}
-		else
-#endif
-			dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
-	} else {
-		PMD_INIT_LOG(INFO, "Using full-featured tx code path");
-		PMD_INIT_LOG(INFO, " - txq_flags = %lx "
-			     "[IXGBE_SIMPLE_FLAGS=%lx]",
-			     (long unsigned)txq->txq_flags,
-			     (long unsigned)IXGBE_SIMPLE_FLAGS);
-		PMD_INIT_LOG(INFO, " - tx_rs_thresh = %lu "
-			     "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
-			     (long unsigned)txq->tx_rs_thresh,
-			     (long unsigned)RTE_PMD_IXGBE_TX_MAX_BURST);
-		dev->tx_pkt_burst = ixgbe_xmit_pkts;
-	}
+	/* set up vector or scalar TX function as appropriate */
+	set_tx_function(dev, txq);
 
 	txq->ops->reset(txq);
 
diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx.h b/lib/librte_pmd_ixgbe/ixgbe_rxtx.h
index 13099af..873656d 100644
--- a/lib/librte_pmd_ixgbe/ixgbe_rxtx.h
+++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx.h
@@ -248,6 +248,13 @@ struct ixgbe_txq_ops {
 			 IXGBE_ADVTXD_DCMD_DEXT |\
 			 IXGBE_ADVTXD_DCMD_EOP)
 
+
+/* takes an ethdev and a queue and sets up the tx function to be used based on
+ * the queue parameters. Used in tx_queue_setup by primary process and then
+ * in dev_init by secondary process when attaching to an existing ethdev
+ */
+void set_tx_function(struct rte_eth_dev* dev, struct igb_tx_queue* txq);
+
 #ifdef RTE_IXGBE_INC_VECTOR
 uint16_t ixgbe_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 		uint16_t nb_pkts);
-- 
1.9.3