[PATCH net-next 0/3] dpaa2-eth: add bql support

[PATCH net-next 0/3] dpaa2-eth: add bql support

[Ioana Ciocoi Radulescu]

The first two patches make minor tweaks to the driver to
simplify bql implementation. The third patch adds the actual
bql support.

Ioana Radulescu (3):
  dpaa2-eth: Update callback signature
  dpaa2-eth: Don't use multiple queues per channel
  dpaa2-eth: bql support

 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c | 71 ++++++++++++++++--------
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h |  8 ++-
 2 files changed, 54 insertions(+), 25 deletions(-)

-- 
2.7.4

[PATCH net-next 2/3] dpaa2-eth: Don't use multiple queues per channel

[Ioana Ciocoi Radulescu]

The DPNI object on which we build a network interface has a
certain number of {Rx, Tx, Tx confirmation} frame queues as
resources. The default hardware setup offers one queue of each
type, as well as one DPCON channel, for each core available
in the system.

There are however cases where the number of queues is greater
than the number of cores or channels. Until now, we configured
and used all the frame queues associated with a DPNI, even if it
meant assigning multiple queues of one type to the same channel.

Update the driver to only use a number of queues equal to the
number of channels, ensuring each channel will contain exactly
one Rx and one Tx confirmation queue.

>From the user viewpoint, this change is completely transparent.
Performance wise there is no impact in most scenarios. In case
the number of queues is larger than and not a multiple of the
number of channels, Rx hash distribution offers now better load
balancing between cores, which can have a positive impact on
overall system performance.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
---
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c | 2 +-
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
index 048414a..be31287 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
@@ -1601,7 +1601,7 @@ static int setup_dpio(struct dpaa2_eth_priv *priv)
 		/* Stop if we already have enough channels to accommodate all
 		 * RX and TX conf queues
 		 */
-		if (priv->num_channels == dpaa2_eth_queue_count(priv))
+		if (priv->num_channels == priv->dpni_attrs.num_queues)
 			break;
 	}
 
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
index 287b6853..3af706a 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
@@ -433,9 +433,10 @@ static inline unsigned int dpaa2_eth_rx_head_room(struct dpaa2_eth_priv *priv)
 	       DPAA2_ETH_RX_HWA_SIZE;
 }
 
+/* We have exactly one {Rx, Tx conf} queue per channel */
 static int dpaa2_eth_queue_count(struct dpaa2_eth_priv *priv)
 {
-	return priv->dpni_attrs.num_queues;
+	return priv->num_channels;
 }
 
 int dpaa2_eth_set_hash(struct net_device *net_dev, u64 flags);
-- 
2.7.4

[PATCH net-next 1/3] dpaa2-eth: Update callback signature

[Ioana Ciocoi Radulescu]

Change the frame consume callback signature:
* the entire FQ structure is passed to the callback instead
of just the queue index
* the NAPI structure can be easily obtained from the channel
it is associated to, so we don't need to pass it explicitly

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
---
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c | 12 +++++-------
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h |  3 +--
 2 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
index bdfb13b..048414a 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
@@ -203,8 +203,7 @@ static struct sk_buff *build_frag_skb(struct dpaa2_eth_priv *priv,
 static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
 			 struct dpaa2_eth_channel *ch,
 			 const struct dpaa2_fd *fd,
-			 struct napi_struct *napi,
-			 u16 queue_id)
+			 struct dpaa2_eth_fq *fq)
 {
 	dma_addr_t addr = dpaa2_fd_get_addr(fd);
 	u8 fd_format = dpaa2_fd_get_format(fd);
@@ -267,12 +266,12 @@ static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
 	}
 
 	skb->protocol = eth_type_trans(skb, priv->net_dev);
-	skb_record_rx_queue(skb, queue_id);
+	skb_record_rx_queue(skb, fq->flowid);
 
 	percpu_stats->rx_packets++;
 	percpu_stats->rx_bytes += dpaa2_fd_get_len(fd);
 
-	napi_gro_receive(napi, skb);
+	napi_gro_receive(&ch->napi, skb);
 
 	return;
 
@@ -312,7 +311,7 @@ static int consume_frames(struct dpaa2_eth_channel *ch,
 		fd = dpaa2_dq_fd(dq);
 		fq = (struct dpaa2_eth_fq *)(uintptr_t)dpaa2_dq_fqd_ctx(dq);
 
-		fq->consume(priv, ch, fd, &ch->napi, fq->flowid);
+		fq->consume(priv, ch, fd, fq);
 		cleaned++;
 	} while (!is_last);
 
@@ -661,8 +660,7 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
 			      struct dpaa2_eth_channel *ch __always_unused,
 			      const struct dpaa2_fd *fd,
-			      struct napi_struct *napi __always_unused,
-			      u16 queue_id __always_unused)
+			      struct dpaa2_eth_fq *fq __always_unused)
 {
 	struct rtnl_link_stats64 *percpu_stats;
 	struct dpaa2_eth_drv_stats *percpu_extras;
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
index 452a8e9..287b6853 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
@@ -277,8 +277,7 @@ struct dpaa2_eth_fq {
 	void (*consume)(struct dpaa2_eth_priv *priv,
 			struct dpaa2_eth_channel *ch,
 			const struct dpaa2_fd *fd,
-			struct napi_struct *napi,
-			u16 queue_id);
+			struct dpaa2_eth_fq *fq);
 	struct dpaa2_eth_fq_stats stats;
 };
 
-- 
2.7.4

[PATCH net-next 3/3] dpaa2-eth: bql support

[Ioana Ciocoi Radulescu]

Add support for byte queue limit.

On NAPI poll, we save the total number of Tx confirmed frames/bytes
and register them with bql at the end of the poll function.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
---
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c | 59 ++++++++++++++++++------
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h |  2 +
 2 files changed, 46 insertions(+), 15 deletions(-)

diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
index be31287..640967a 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
@@ -288,7 +288,7 @@ static void dpaa2_eth_rx(struct dpaa2_eth_priv *priv,
  * Observance of NAPI budget is not our concern, leaving that to the caller.
  */
 static int consume_frames(struct dpaa2_eth_channel *ch,
-			  enum dpaa2_eth_fq_type *type)
+			  struct dpaa2_eth_fq **src)
 {
 	struct dpaa2_eth_priv *priv = ch->priv;
 	struct dpaa2_eth_fq *fq = NULL;
@@ -322,10 +322,10 @@ static int consume_frames(struct dpaa2_eth_channel *ch,
 	ch->stats.frames += cleaned;
 
 	/* A dequeue operation only pulls frames from a single queue
-	 * into the store. Return the frame queue type as an out param.
+	 * into the store. Return the frame queue as an out param.
 	 */
-	if (type)
-		*type = fq->type;
+	if (src)
+		*src = fq;
 
 	return cleaned;
 }
@@ -570,8 +570,10 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 	struct rtnl_link_stats64 *percpu_stats;
 	struct dpaa2_eth_drv_stats *percpu_extras;
 	struct dpaa2_eth_fq *fq;
+	struct netdev_queue *nq;
 	u16 queue_mapping;
 	unsigned int needed_headroom;
+	u32 fd_len;
 	int err, i;
 
 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
@@ -643,8 +645,12 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 		/* Clean up everything, including freeing the skb */
 		free_tx_fd(priv, &fd);
 	} else {
+		fd_len = dpaa2_fd_get_len(&fd);
 		percpu_stats->tx_packets++;
-		percpu_stats->tx_bytes += dpaa2_fd_get_len(&fd);
+		percpu_stats->tx_bytes += fd_len;
+
+		nq = netdev_get_tx_queue(net_dev, queue_mapping);
+		netdev_tx_sent_queue(nq, fd_len);
 	}
 
 	return NETDEV_TX_OK;
@@ -660,10 +666,11 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
 			      struct dpaa2_eth_channel *ch __always_unused,
 			      const struct dpaa2_fd *fd,
-			      struct dpaa2_eth_fq *fq __always_unused)
+			      struct dpaa2_eth_fq *fq)
 {
 	struct rtnl_link_stats64 *percpu_stats;
 	struct dpaa2_eth_drv_stats *percpu_extras;
+	u32 fd_len = dpaa2_fd_get_len(fd);
 	u32 fd_errors;
 
 	/* Tracing point */
@@ -671,7 +678,10 @@ static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
 
 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
 	percpu_extras->tx_conf_frames++;
-	percpu_extras->tx_conf_bytes += dpaa2_fd_get_len(fd);
+	percpu_extras->tx_conf_bytes += fd_len;
+
+	fq->dq_frames++;
+	fq->dq_bytes += fd_len;
 
 	/* Check frame errors in the FD field */
 	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
@@ -932,8 +942,9 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 	struct dpaa2_eth_channel *ch;
 	struct dpaa2_eth_priv *priv;
 	int rx_cleaned = 0, txconf_cleaned = 0;
-	enum dpaa2_eth_fq_type type = 0;
-	int store_cleaned;
+	struct dpaa2_eth_fq *fq, *txc_fq = NULL;
+	struct netdev_queue *nq;
+	int store_cleaned, work_done;
 	int err;
 
 	ch = container_of(napi, struct dpaa2_eth_channel, napi);
@@ -947,18 +958,25 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 		/* Refill pool if appropriate */
 		refill_pool(priv, ch, priv->bpid);
 
-		store_cleaned = consume_frames(ch, &type);
-		if (type == DPAA2_RX_FQ)
+		store_cleaned = consume_frames(ch, &fq);
+		if (!store_cleaned)
+			break;
+		if (fq->type == DPAA2_RX_FQ) {
 			rx_cleaned += store_cleaned;
-		else
+		} else {
 			txconf_cleaned += store_cleaned;
+			/* We have a single Tx conf FQ on this channel */
+			txc_fq = fq;
+		}
 
 		/* If we either consumed the whole NAPI budget with Rx frames
 		 * or we reached the Tx confirmations threshold, we're done.
 		 */
 		if (rx_cleaned >= budget ||
-		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI)
-			return budget;
+		    txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
+			work_done = budget;
+			goto out;
+		}
 	} while (store_cleaned);
 
 	/* We didn't consume the entire budget, so finish napi and
@@ -972,7 +990,18 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 	WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
 		  ch->nctx.desired_cpu);
 
-	return max(rx_cleaned, 1);
+	work_done = max(rx_cleaned, 1);
+
+out:
+	if (txc_fq) {
+		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
+		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
+					  txc_fq->dq_bytes);
+		txc_fq->dq_frames = 0;
+		txc_fq->dq_bytes = 0;
+	}
+
+	return work_done;
 }
 
 static void enable_ch_napi(struct dpaa2_eth_priv *priv)
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
index 3af706a..16545e9 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
@@ -271,6 +271,8 @@ struct dpaa2_eth_fq {
 	u32 tx_qdbin;
 	u16 flowid;
 	int target_cpu;
+	u32 dq_frames;
+	u32 dq_bytes;
 	struct dpaa2_eth_channel *channel;
 	enum dpaa2_eth_fq_type type;
 
-- 
2.7.4

Re: [PATCH net-next 0/3] dpaa2-eth: add bql support

[David Miller]

From: Ioana Ciocoi Radulescu <ruxandra.radulescu@nxp.com>
Date: Wed, 14 Nov 2018 11:48:34 +0000

> The first two patches make minor tweaks to the driver to
> simplify bql implementation. The third patch adds the actual
> bql support.

Series applied, thanks!