[PATCH net-next 00/11] nfp: update TX path to enable repr offloads

[PATCH net-next 00/11] nfp: update TX path to enable repr offloads

[Jakub Kicinski]

Hi!

This set starts with three micro optimizations to the TX path.
The improvement is measurable, but below 1% of CPU utilization.

Patches 4 - 9 add basic TX offloads to representor devices, like
checksum offload or TSO, and remove the unnecessary TX lock and
Qdisc (our representors are software constructs on top of the PF).

The last 2 patches add more info to error messages - id of command
which failed and exact location of incorrect TLVs, very useful for
debugging.


Jakub Kicinski (11):
  nfp: copy only the relevant part of the TX descriptor for frags
  nfp: move temporary variables in nfp_net_tx_complete()
  nfp: move queue variable init
  nfp: correct descriptor offsets in presence of metadata
  nfp: avoid oversized TSO headers with metadata prepend
  nfp: run representor TX locklessly
  nfp: run don't require Qdiscs on representor netdevs
  nfp: add locking around representor changes
  nfp: add offloads on representors
  nfp: add offset to all TLV parsing errors
  nfp: report more info when reconfiguration fails

 drivers/net/ethernet/netronome/nfp/abm/main.c |  4 +
 drivers/net/ethernet/netronome/nfp/nfp_app.c  | 42 +++++++++
 drivers/net/ethernet/netronome/nfp/nfp_net.h  |  3 +
 .../ethernet/netronome/nfp/nfp_net_common.c   | 81 ++++++++++-------
 .../net/ethernet/netronome/nfp/nfp_net_ctrl.c | 21 +++--
 .../net/ethernet/netronome/nfp/nfp_net_ctrl.h |  7 ++
 .../net/ethernet/netronome/nfp/nfp_net_repr.c | 90 +++++++++++++++++++
 .../net/ethernet/netronome/nfp/nfp_net_repr.h |  2 +
 8 files changed, 213 insertions(+), 37 deletions(-)

-- 
2.17.1

[PATCH net-next 01/11] nfp: copy only the relevant part of the TX descriptor for frags

[Jakub Kicinski]

Chained descriptors for fragments need to duplicate all the descriptor
fields of the skb head, so we copy the descriptor and then modify the
relevant fields.  This is wasteful, because the top half of the descriptor
will get overwritten entirely while the bottom half is not modified at all.
Copy only the bottom half.  This saves us 0.3% of CPU in a GSO test.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net.h        |  1 +
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 12 +++++++-----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index bb3dbd74583b..5a9a6178cf26 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -158,6 +158,7 @@ struct nfp_net_tx_desc {
 			__le16 data_len; /* Length of frame + meta data */
 		} __packed;
 		__le32 vals[4];
+		__le64 vals8[2];
 	};
 };
 
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 9aa6265bf4de..b54c6e481229 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -786,11 +786,11 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 {
 	struct nfp_net *nn = netdev_priv(netdev);
 	const struct skb_frag_struct *frag;
-	struct nfp_net_tx_desc *txd, txdg;
 	int f, nr_frags, wr_idx, md_bytes;
 	struct nfp_net_tx_ring *tx_ring;
 	struct nfp_net_r_vector *r_vec;
 	struct nfp_net_tx_buf *txbuf;
+	struct nfp_net_tx_desc *txd;
 	struct netdev_queue *nd_q;
 	struct nfp_net_dp *dp;
 	dma_addr_t dma_addr;
@@ -860,8 +860,10 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 
 	/* Gather DMA */
 	if (nr_frags > 0) {
+		__le64 second_half;
+
 		/* all descs must match except for in addr, length and eop */
-		txdg = *txd;
+		second_half = txd->vals8[1];
 
 		for (f = 0; f < nr_frags; f++) {
 			frag = &skb_shinfo(skb)->frags[f];
@@ -878,11 +880,11 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 			tx_ring->txbufs[wr_idx].fidx = f;
 
 			txd = &tx_ring->txds[wr_idx];
-			*txd = txdg;
 			txd->dma_len = cpu_to_le16(fsize);
 			nfp_desc_set_dma_addr(txd, dma_addr);
-			txd->offset_eop |=
-				(f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
+			txd->offset_eop = md_bytes |
+				((f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0);
+			txd->vals8[1] = second_half;
 		}
 
 		u64_stats_update_begin(&r_vec->tx_sync);
-- 
2.17.1

[PATCH net-next 02/11] nfp: move temporary variables in nfp_net_tx_complete()

[Jakub Kicinski]

Move temporary variables in scope of the loop in nfp_net_tx_complete(),
and add a temp for txbuf software structure.  This saves us 0.2% of CPU.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 .../ethernet/netronome/nfp/nfp_net_common.c   | 31 ++++++++++---------
 1 file changed, 17 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index b54c6e481229..78c651f9f774 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -940,14 +940,10 @@ static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
 {
 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
-	const struct skb_frag_struct *frag;
 	struct netdev_queue *nd_q;
 	u32 done_pkts = 0, done_bytes = 0;
-	struct sk_buff *skb;
-	int todo, nr_frags;
 	u32 qcp_rd_p;
-	int fidx;
-	int idx;
+	int todo;
 
 	if (tx_ring->wr_p == tx_ring->rd_p)
 		return;
@@ -961,26 +957,33 @@ static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
 
 	while (todo--) {
+		const struct skb_frag_struct *frag;
+		struct nfp_net_tx_buf *tx_buf;
+		struct sk_buff *skb;
+		int fidx, nr_frags;
+		int idx;
+
 		idx = D_IDX(tx_ring, tx_ring->rd_p++);
+		tx_buf = &tx_ring->txbufs[idx];
 
-		skb = tx_ring->txbufs[idx].skb;
+		skb = tx_buf->skb;
 		if (!skb)
 			continue;
 
 		nr_frags = skb_shinfo(skb)->nr_frags;
-		fidx = tx_ring->txbufs[idx].fidx;
+		fidx = tx_buf->fidx;
 
 		if (fidx == -1) {
 			/* unmap head */
-			dma_unmap_single(dp->dev, tx_ring->txbufs[idx].dma_addr,
+			dma_unmap_single(dp->dev, tx_buf->dma_addr,
 					 skb_headlen(skb), DMA_TO_DEVICE);
 
-			done_pkts += tx_ring->txbufs[idx].pkt_cnt;
-			done_bytes += tx_ring->txbufs[idx].real_len;
+			done_pkts += tx_buf->pkt_cnt;
+			done_bytes += tx_buf->real_len;
 		} else {
 			/* unmap fragment */
 			frag = &skb_shinfo(skb)->frags[fidx];
-			dma_unmap_page(dp->dev, tx_ring->txbufs[idx].dma_addr,
+			dma_unmap_page(dp->dev, tx_buf->dma_addr,
 				       skb_frag_size(frag), DMA_TO_DEVICE);
 		}
 
@@ -988,9 +991,9 @@ static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
 		if (fidx == nr_frags - 1)
 			napi_consume_skb(skb, budget);
 
-		tx_ring->txbufs[idx].dma_addr = 0;
-		tx_ring->txbufs[idx].skb = NULL;
-		tx_ring->txbufs[idx].fidx = -2;
+		tx_buf->dma_addr = 0;
+		tx_buf->skb = NULL;
+		tx_buf->fidx = -2;
 	}
 
 	tx_ring->qcp_rd_p = qcp_rd_p;
-- 
2.17.1

[PATCH net-next 03/11] nfp: move queue variable init

[Jakub Kicinski]

nd_q is only used at the very end of nfp_net_tx(), there is no need
to initialize it early.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 78c651f9f774..0302a38149b6 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -801,13 +801,13 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 	qidx = skb_get_queue_mapping(skb);
 	tx_ring = &dp->tx_rings[qidx];
 	r_vec = tx_ring->r_vec;
-	nd_q = netdev_get_tx_queue(dp->netdev, qidx);
 
 	nr_frags = skb_shinfo(skb)->nr_frags;
 
 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
+		nd_q = netdev_get_tx_queue(dp->netdev, qidx);
 		netif_tx_stop_queue(nd_q);
 		nfp_net_tx_xmit_more_flush(tx_ring);
 		u64_stats_update_begin(&r_vec->tx_sync);
@@ -894,6 +894,8 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 
 	skb_tx_timestamp(skb);
 
+	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
+
 	tx_ring->wr_p += nr_frags + 1;
 	if (nfp_net_tx_ring_should_stop(tx_ring))
 		nfp_net_tx_ring_stop(nd_q, tx_ring);
-- 
2.17.1

[PATCH net-next 04/11] nfp: correct descriptor offsets in presence of metadata

[Jakub Kicinski]

The TSO-related offsets in the descriptor should not include
the length of the prepended metadata.  Adjust them.  Note that
this could not have caused issues in the past as we don't
support TSO with metadata prepend as of this patch.

Signed-off-by: Michael Rapson <michael.rapson@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 .../ethernet/netronome/nfp/nfp_net_common.c   | 20 +++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 0302a38149b6..9cc2c204baa3 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -647,27 +647,29 @@ static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
  * @txbuf: Pointer to driver soft TX descriptor
  * @txd: Pointer to HW TX descriptor
  * @skb: Pointer to SKB
+ * @md_bytes: Prepend length
  *
  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
  * Return error on packet header greater than maximum supported LSO header size.
  */
 static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
 			   struct nfp_net_tx_buf *txbuf,
-			   struct nfp_net_tx_desc *txd, struct sk_buff *skb)
+			   struct nfp_net_tx_desc *txd, struct sk_buff *skb,
+			   u32 md_bytes)
 {
-	u32 hdrlen;
+	u32 l3_offset, l4_offset, hdrlen;
 	u16 mss;
 
 	if (!skb_is_gso(skb))
 		return;
 
 	if (!skb->encapsulation) {
-		txd->l3_offset = skb_network_offset(skb);
-		txd->l4_offset = skb_transport_offset(skb);
+		l3_offset = skb_network_offset(skb);
+		l4_offset = skb_transport_offset(skb);
 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
 	} else {
-		txd->l3_offset = skb_inner_network_offset(skb);
-		txd->l4_offset = skb_inner_transport_offset(skb);
+		l3_offset = skb_inner_network_offset(skb);
+		l4_offset = skb_inner_transport_offset(skb);
 		hdrlen = skb_inner_transport_header(skb) - skb->data +
 			inner_tcp_hdrlen(skb);
 	}
@@ -676,7 +678,9 @@ static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
 
 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
-	txd->lso_hdrlen = hdrlen;
+	txd->l3_offset = l3_offset - md_bytes;
+	txd->l4_offset = l4_offset - md_bytes;
+	txd->lso_hdrlen = hdrlen - md_bytes;
 	txd->mss = cpu_to_le16(mss);
 	txd->flags |= PCIE_DESC_TX_LSO;
 
@@ -851,7 +855,7 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 	txd->lso_hdrlen = 0;
 
 	/* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
-	nfp_net_tx_tso(r_vec, txbuf, txd, skb);
+	nfp_net_tx_tso(r_vec, txbuf, txd, skb, md_bytes);
 	nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
 	if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
 		txd->flags |= PCIE_DESC_TX_VLAN;
-- 
2.17.1

[PATCH net-next 05/11] nfp: avoid oversized TSO headers with metadata prepend

[Jakub Kicinski]

In preparation for TSO over representors make sure the port id
prepend will always fit in the frame.  The current max header
length is 255, which is ample, so assume worst case scenario
of 8 byte prepend and save ourselves the conditionals.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 9cc2c204baa3..3cb7dceca2d9 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -3284,7 +3284,10 @@ nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
 		hdrlen = skb_inner_transport_header(skb) - skb->data +
 			inner_tcp_hdrlen(skb);
 
-		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
+		/* Assume worst case scenario of having longest possible
+		 * metadata prepend - 8B
+		 */
+		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8))
 			features &= ~NETIF_F_GSO_MASK;
 	}
 
-- 
2.17.1

[PATCH net-next 06/11] nfp: run representor TX locklessly

[Jakub Kicinski]

Our representors are software devices built on top of the PF
vNIC, the only state they have are per-cpu stats, so make
the TX run locklessly.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_repr.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
index c09b893c30dd..769fb5210aaf 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
@@ -299,6 +299,8 @@ int nfp_repr_init(struct nfp_app *app, struct net_device *netdev,
 
 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
 
+	netdev->features |= NETIF_F_LLTX;
+
 	if (nfp_app_has_tc(app)) {
 		netdev->features |= NETIF_F_HW_TC;
 		netdev->hw_features |= NETIF_F_HW_TC;
-- 
2.17.1

[PATCH net-next 07/11] nfp: run don't require Qdiscs on representor netdevs

[Jakub Kicinski]

Our representors are software devices built on top of the PF
vNIC, the queuing should only happen at the vNIC netdevice.
Allow representors to run qdisc-less.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_repr.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
index 769fb5210aaf..b9904f6b41f8 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
@@ -299,6 +299,7 @@ int nfp_repr_init(struct nfp_app *app, struct net_device *netdev,
 
 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
 
+	netdev->priv_flags |= IFF_NO_QUEUE;
 	netdev->features |= NETIF_F_LLTX;
 
 	if (nfp_app_has_tc(app)) {
-- 
2.17.1

[PATCH net-next 08/11] nfp: add locking around representor changes

[Jakub Kicinski]

Up until now we never needed to keep a networking locks around
representors accesses, we only accessed them when device was
reconfigured (under nfp pf->lock) or on fast path (under RCU).
Now we want to be able to iterate over all representors during
notifications, so make sure representor assignment is done
under RTNL lock.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/abm/main.c     | 4 ++++
 drivers/net/ethernet/netronome/nfp/nfp_app.c      | 2 ++
 drivers/net/ethernet/netronome/nfp/nfp_net_repr.c | 2 ++
 3 files changed, 8 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/abm/main.c b/drivers/net/ethernet/netronome/nfp/abm/main.c
index 7a4d55f794c2..80e79447b644 100644
--- a/drivers/net/ethernet/netronome/nfp/abm/main.c
+++ b/drivers/net/ethernet/netronome/nfp/abm/main.c
@@ -126,7 +126,9 @@ nfp_abm_spawn_repr(struct nfp_app *app, struct nfp_abm_link *alink,
 
 	reprs = nfp_reprs_get_locked(app, rtype);
 	WARN(nfp_repr_get_locked(app, reprs, alink->id), "duplicate repr");
+	rtnl_lock();
 	rcu_assign_pointer(reprs->reprs[alink->id], netdev);
+	rtnl_unlock();
 
 	nfp_info(app->cpp, "%s Port %d Representor(%s) created\n",
 		 ptype == NFP_PORT_PF_PORT ? "PCIe" : "Phys",
@@ -152,7 +154,9 @@ nfp_abm_kill_repr(struct nfp_app *app, struct nfp_abm_link *alink,
 	netdev = nfp_repr_get_locked(app, reprs, alink->id);
 	if (!netdev)
 		return;
+	rtnl_lock();
 	rcu_assign_pointer(reprs->reprs[alink->id], NULL);
+	rtnl_unlock();
 	synchronize_rcu();
 	/* Cast to make sure nfp_repr_clean_and_free() takes a nfp_repr */
 	nfp_repr_clean_and_free((struct nfp_repr *)netdev_priv(netdev));
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_app.c b/drivers/net/ethernet/netronome/nfp/nfp_app.c
index 4a1b8f79e731..117553914342 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_app.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_app.c
@@ -131,7 +131,9 @@ nfp_app_reprs_set(struct nfp_app *app, enum nfp_repr_type type,
 	struct nfp_reprs *old;
 
 	old = nfp_reprs_get_locked(app, type);
+	rtnl_lock();
 	rcu_assign_pointer(app->reprs[type], reprs);
+	rtnl_unlock();
 
 	return old;
 }
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
index b9904f6b41f8..0a1d761c6692 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
@@ -445,7 +445,9 @@ int nfp_reprs_resync_phys_ports(struct nfp_app *app)
 			continue;
 
 		nfp_app_repr_preclean(app, netdev);
+		rtnl_lock();
 		rcu_assign_pointer(reprs->reprs[i], NULL);
+		rtnl_unlock();
 		synchronize_rcu();
 		nfp_repr_clean(repr);
 	}
-- 
2.17.1

[PATCH net-next 09/11] nfp: add offloads on representors

[Jakub Kicinski]

FW/HW can generally support the standard networking offloads
on representors without any trouble.  Add the ability for FW
to advertise which features should be available on representors.

Because representors are muxed on top of the vNIC we need to listen
on feature changes of their lower devices, and update their features
appropriately.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_app.c  | 40 +++++++++
 .../net/ethernet/netronome/nfp/nfp_net_ctrl.c |  9 ++
 .../net/ethernet/netronome/nfp/nfp_net_ctrl.h |  7 ++
 .../net/ethernet/netronome/nfp/nfp_net_repr.c | 85 +++++++++++++++++++
 .../net/ethernet/netronome/nfp/nfp_net_repr.h |  2 +
 5 files changed, 143 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_app.c b/drivers/net/ethernet/netronome/nfp/nfp_app.c
index 117553914342..3a973282b2bb 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_app.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_app.c
@@ -138,6 +138,38 @@ nfp_app_reprs_set(struct nfp_app *app, enum nfp_repr_type type,
 	return old;
 }
 
+static void
+nfp_app_netdev_feat_change(struct nfp_app *app, struct net_device *netdev)
+{
+	struct nfp_net *nn;
+	unsigned int type;
+
+	if (!nfp_netdev_is_nfp_net(netdev))
+		return;
+	nn = netdev_priv(netdev);
+	if (nn->app != app)
+		return;
+
+	for (type = 0; type < __NFP_REPR_TYPE_MAX; type++) {
+		struct nfp_reprs *reprs;
+		unsigned int i;
+
+		reprs = rtnl_dereference(app->reprs[type]);
+		if (!reprs)
+			continue;
+
+		for (i = 0; i < reprs->num_reprs; i++) {
+			struct net_device *repr;
+
+			repr = rtnl_dereference(reprs->reprs[i]);
+			if (!repr)
+				continue;
+
+			nfp_repr_transfer_features(repr, netdev);
+		}
+	}
+}
+
 static int
 nfp_app_netdev_event(struct notifier_block *nb, unsigned long event, void *ptr)
 {
@@ -147,6 +179,14 @@ nfp_app_netdev_event(struct notifier_block *nb, unsigned long event, void *ptr)
 	netdev = netdev_notifier_info_to_dev(ptr);
 	app = container_of(nb, struct nfp_app, netdev_nb);
 
+	/* Handle events common code is interested in */
+	switch (event) {
+	case NETDEV_FEAT_CHANGE:
+		nfp_app_netdev_feat_change(app, netdev);
+		break;
+	}
+
+	/* Call offload specific handlers */
 	if (app->type->netdev_event)
 		return app->type->netdev_event(app, netdev, event, ptr);
 	return NOTIFY_DONE;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c
index f2aaef976c7d..d26ea32df8d9 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c
@@ -90,6 +90,15 @@ int nfp_net_tlv_caps_parse(struct device *dev, u8 __iomem *ctrl_mem,
 				 FIELD_GET(NFP_NET_CFG_TLV_HEADER_TYPE, hdr),
 				 offset, length);
 			break;
+		case NFP_NET_CFG_TLV_TYPE_REPR_CAP:
+			if (length < 4) {
+				dev_err(dev, "REPR CAP TLV short %dB < 4B\n",
+					length);
+				return -EINVAL;
+			}
+
+			caps->repr_cap = readl(data);
+			break;
 		default:
 			if (!FIELD_GET(NFP_NET_CFG_TLV_HEADER_REQUIRED, hdr))
 				break;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
index ccec69944de5..166d7f71442e 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.h
@@ -466,6 +466,10 @@
  * Variable, experimental IDs.  IDs designated for internal development and
  * experiments before a stable TLV ID has been allocated to a feature.  Should
  * never be present in production firmware.
+ *
+ * %NFP_NET_CFG_TLV_TYPE_REPR_CAP:
+ * Single word, equivalent of %NFP_NET_CFG_CAP for representors, features which
+ * can be used on representors.
  */
 #define NFP_NET_CFG_TLV_TYPE_UNKNOWN		0
 #define NFP_NET_CFG_TLV_TYPE_RESERVED		1
@@ -474,6 +478,7 @@
 #define NFP_NET_CFG_TLV_TYPE_MBOX		4
 #define NFP_NET_CFG_TLV_TYPE_EXPERIMENTAL0	5
 #define NFP_NET_CFG_TLV_TYPE_EXPERIMENTAL1	6
+#define NFP_NET_CFG_TLV_TYPE_REPR_CAP		7
 
 struct device;
 
@@ -482,11 +487,13 @@ struct device;
  * @me_freq_mhz:	ME clock_freq (MHz)
  * @mbox_off:		vNIC mailbox area offset
  * @mbox_len:		vNIC mailbox area length
+ * @repr_cap:		capabilities for representors
  */
 struct nfp_net_tlv_caps {
 	u32 me_freq_mhz;
 	unsigned int mbox_off;
 	unsigned int mbox_len;
+	u32 repr_cap;
 };
 
 int nfp_net_tlv_caps_parse(struct device *dev, u8 __iomem *ctrl_mem,
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
index 0a1d761c6692..69d7aebda09b 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
@@ -11,6 +11,7 @@
 #include "nfpcore/nfp_nsp.h"
 #include "nfp_app.h"
 #include "nfp_main.h"
+#include "nfp_net.h"
 #include "nfp_net_ctrl.h"
 #include "nfp_net_repr.h"
 #include "nfp_net_sriov.h"
@@ -231,6 +232,27 @@ static int nfp_repr_open(struct net_device *netdev)
 	return err;
 }
 
+static netdev_features_t
+nfp_repr_fix_features(struct net_device *netdev, netdev_features_t features)
+{
+	struct nfp_repr *repr = netdev_priv(netdev);
+	netdev_features_t old_features = features;
+	netdev_features_t lower_features;
+	struct net_device *lower_dev;
+
+	lower_dev = repr->dst->u.port_info.lower_dev;
+
+	lower_features = lower_dev->features;
+	if (lower_features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
+		lower_features |= NETIF_F_HW_CSUM;
+
+	features = netdev_intersect_features(features, lower_features);
+	features |= old_features & (NETIF_F_SOFT_FEATURES | NETIF_F_HW_TC);
+	features |= NETIF_F_LLTX;
+
+	return features;
+}
+
 const struct net_device_ops nfp_repr_netdev_ops = {
 	.ndo_init		= nfp_app_ndo_init,
 	.ndo_uninit		= nfp_app_ndo_uninit,
@@ -248,10 +270,25 @@ const struct net_device_ops nfp_repr_netdev_ops = {
 	.ndo_set_vf_spoofchk	= nfp_app_set_vf_spoofchk,
 	.ndo_get_vf_config	= nfp_app_get_vf_config,
 	.ndo_set_vf_link_state	= nfp_app_set_vf_link_state,
+	.ndo_fix_features	= nfp_repr_fix_features,
 	.ndo_set_features	= nfp_port_set_features,
 	.ndo_set_mac_address    = eth_mac_addr,
 };
 
+void
+nfp_repr_transfer_features(struct net_device *netdev, struct net_device *lower)
+{
+	struct nfp_repr *repr = netdev_priv(netdev);
+
+	if (repr->dst->u.port_info.lower_dev != lower)
+		return;
+
+	netdev->gso_max_size = lower->gso_max_size;
+	netdev->gso_max_segs = lower->gso_max_segs;
+
+	netdev_update_features(netdev);
+}
+
 static void nfp_repr_clean(struct nfp_repr *repr)
 {
 	unregister_netdev(repr->netdev);
@@ -281,6 +318,8 @@ int nfp_repr_init(struct nfp_app *app, struct net_device *netdev,
 		  struct net_device *pf_netdev)
 {
 	struct nfp_repr *repr = netdev_priv(netdev);
+	struct nfp_net *nn = netdev_priv(pf_netdev);
+	u32 repr_cap = nn->tlv_caps.repr_cap;
 	int err;
 
 	nfp_repr_set_lockdep_class(netdev);
@@ -299,6 +338,52 @@ int nfp_repr_init(struct nfp_app *app, struct net_device *netdev,
 
 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
 
+	/* Set features the lower device can support with representors */
+	if (repr_cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
+		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
+
+	netdev->hw_features = NETIF_F_HIGHDMA;
+	if (repr_cap & NFP_NET_CFG_CTRL_RXCSUM_ANY)
+		netdev->hw_features |= NETIF_F_RXCSUM;
+	if (repr_cap & NFP_NET_CFG_CTRL_TXCSUM)
+		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
+	if (repr_cap & NFP_NET_CFG_CTRL_GATHER)
+		netdev->hw_features |= NETIF_F_SG;
+	if ((repr_cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
+	    repr_cap & NFP_NET_CFG_CTRL_LSO2)
+		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
+	if (repr_cap & NFP_NET_CFG_CTRL_RSS_ANY)
+		netdev->hw_features |= NETIF_F_RXHASH;
+	if (repr_cap & NFP_NET_CFG_CTRL_VXLAN) {
+		if (repr_cap & NFP_NET_CFG_CTRL_LSO)
+			netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
+	}
+	if (repr_cap & NFP_NET_CFG_CTRL_NVGRE) {
+		if (repr_cap & NFP_NET_CFG_CTRL_LSO)
+			netdev->hw_features |= NETIF_F_GSO_GRE;
+	}
+	if (repr_cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE))
+		netdev->hw_enc_features = netdev->hw_features;
+
+	netdev->vlan_features = netdev->hw_features;
+
+	if (repr_cap & NFP_NET_CFG_CTRL_RXVLAN)
+		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
+	if (repr_cap & NFP_NET_CFG_CTRL_TXVLAN) {
+		if (repr_cap & NFP_NET_CFG_CTRL_LSO2)
+			netdev_warn(netdev, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
+		else
+			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
+	}
+	if (repr_cap & NFP_NET_CFG_CTRL_CTAG_FILTER)
+		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
+
+	netdev->features = netdev->hw_features;
+
+	/* Advertise but disable TSO by default. */
+	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
+	netdev->gso_max_segs = NFP_NET_LSO_MAX_SEGS;
+
 	netdev->priv_flags |= IFF_NO_QUEUE;
 	netdev->features |= NETIF_F_LLTX;
 
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.h b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.h
index c412b94bfb97..e0f13dfe1f39 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.h
@@ -92,6 +92,8 @@ nfp_repr_get_locked(struct nfp_app *app, struct nfp_reprs *set,
 		    unsigned int id);
 
 void nfp_repr_inc_rx_stats(struct net_device *netdev, unsigned int len);
+void
+nfp_repr_transfer_features(struct net_device *netdev, struct net_device *lower);
 int nfp_repr_init(struct nfp_app *app, struct net_device *netdev,
 		  u32 cmsg_port_id, struct nfp_port *port,
 		  struct net_device *pf_netdev);
-- 
2.17.1

[PATCH net-next 10/11] nfp: add offset to all TLV parsing errors

[Jakub Kicinski]

When troubleshooting incorrect FW capabilities it's useful to know
where the faulty TLV is located.  Add offset to all errors messages.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 .../net/ethernet/netronome/nfp/nfp_net_ctrl.c    | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c
index d26ea32df8d9..6d5213b5bcb0 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_ctrl.c
@@ -41,8 +41,8 @@ int nfp_net_tlv_caps_parse(struct device *dev, u8 __iomem *ctrl_mem,
 		data += 4;
 
 		if (length % NFP_NET_CFG_TLV_LENGTH_INC) {
-			dev_err(dev, "TLV size not multiple of %u len:%u\n",
-				NFP_NET_CFG_TLV_LENGTH_INC, length);
+			dev_err(dev, "TLV size not multiple of %u offset:%u len:%u\n",
+				NFP_NET_CFG_TLV_LENGTH_INC, offset, length);
 			return -EINVAL;
 		}
 		if (data + length > end) {
@@ -61,14 +61,14 @@ int nfp_net_tlv_caps_parse(struct device *dev, u8 __iomem *ctrl_mem,
 			if (!length)
 				return 0;
 
-			dev_err(dev, "END TLV should be empty, has len:%d\n",
-				length);
+			dev_err(dev, "END TLV should be empty, has offset:%u len:%d\n",
+				offset, length);
 			return -EINVAL;
 		case NFP_NET_CFG_TLV_TYPE_ME_FREQ:
 			if (length != 4) {
 				dev_err(dev,
-					"ME FREQ TLV should be 4B, is %dB\n",
-					length);
+					"ME FREQ TLV should be 4B, is %dB offset:%u\n",
+					length, offset);
 				return -EINVAL;
 			}
 
@@ -92,8 +92,8 @@ int nfp_net_tlv_caps_parse(struct device *dev, u8 __iomem *ctrl_mem,
 			break;
 		case NFP_NET_CFG_TLV_TYPE_REPR_CAP:
 			if (length < 4) {
-				dev_err(dev, "REPR CAP TLV short %dB < 4B\n",
-					length);
+				dev_err(dev, "REPR CAP TLV short %dB < 4B offset:%u\n",
+					length, offset);
 				return -EINVAL;
 			}
 
-- 
2.17.1

[PATCH net-next 11/11] nfp: report more info when reconfiguration fails

[Jakub Kicinski]

FW reconfiguration timeouts are a common indicator of FW trouble.
To make debugging easier print requested update and control word
when reconfiguration fails.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net.h        | 2 ++
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 9 +++++++--
 2 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index 5a9a6178cf26..be37c2d6151c 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -544,6 +544,7 @@ struct nfp_net_dp {
  * @reconfig_timer_active:  Timer for reading reconfiguration results is pending
  * @reconfig_sync_present:  Some thread is performing synchronous reconfig
  * @reconfig_timer:	Timer for async reading of reconfig results
+ * @reconfig_in_progress_update:	Update FW is processing now (debug only)
  * @link_up:            Is the link up?
  * @link_status_lock:	Protects @link_* and ensures atomicity with BAR reading
  * @rx_coalesce_usecs:      RX interrupt moderation usecs delay parameter
@@ -612,6 +613,7 @@ struct nfp_net {
 	bool reconfig_timer_active;
 	bool reconfig_sync_present;
 	struct timer_list reconfig_timer;
+	u32 reconfig_in_progress_update;
 
 	u32 rx_coalesce_usecs;
 	u32 rx_coalesce_max_frames;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 3cb7dceca2d9..e97636d2e6ee 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -101,6 +101,7 @@ static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
 	/* ensure update is written before pinging HW */
 	nn_pci_flush(nn);
 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
+	nn->reconfig_in_progress_update = update;
 }
 
 /* Pass 0 as update to run posted reconfigs. */
@@ -123,10 +124,14 @@ static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
 	if (reg == 0)
 		return true;
 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
-		nn_err(nn, "Reconfig error: 0x%08x\n", reg);
+		nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
+		       reg, nn->reconfig_in_progress_update,
+		       nn_readl(nn, NFP_NET_CFG_CTRL));
 		return true;
 	} else if (last_check) {
-		nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
+		nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
+		       reg, nn->reconfig_in_progress_update,
+		       nn_readl(nn, NFP_NET_CFG_CTRL));
 		return true;
 	}
 
-- 
2.17.1

Re: [PATCH net-next 00/11] nfp: update TX path to enable repr offloads

[David Miller]

From: Jakub Kicinski <jakub.kicinski@netronome.com>
Date: Tue, 27 Nov 2018 22:24:47 -0800

> This set starts with three micro optimizations to the TX path.
> The improvement is measurable, but below 1% of CPU utilization.
> 
> Patches 4 - 9 add basic TX offloads to representor devices, like
> checksum offload or TSO, and remove the unnecessary TX lock and
> Qdisc (our representors are software constructs on top of the PF).
> 
> The last 2 patches add more info to error messages - id of command
> which failed and exact location of incorrect TLVs, very useful for
> debugging.

Series applied, thanks Jakub.