[PATCH net-next 0/4] nfp: devlink port implementation

[PATCH net-next 0/4] nfp: devlink port implementation

[Jakub Kicinski]

Hi!

This series adds basic devlink support.  The operations we can perfrom
are port show and port split/unsplit.

Jakub Kicinski (3):
  nfp: register ports as devlink ports
  nfp: calculate total port lanes for split
  nfp: support port splitting via devlink

Simon Horman (1):
  nfp: add devlink support

 drivers/net/ethernet/netronome/nfp/Makefile        |   1 +
 drivers/net/ethernet/netronome/nfp/nfp_devlink.c   | 185 +++++++++++++++++++++
 drivers/net/ethernet/netronome/nfp/nfp_main.c      |  30 +++-
 drivers/net/ethernet/netronome/nfp/nfp_main.h      |   5 +
 drivers/net/ethernet/netronome/nfp/nfp_net_main.c  |  57 ++++++-
 drivers/net/ethernet/netronome/nfp/nfp_port.c      |  19 +++
 drivers/net/ethernet/netronome/nfp/nfp_port.h      |  12 ++
 .../net/ethernet/netronome/nfp/nfpcore/nfp_nsp.h   |   3 +
 .../ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c   |  11 +-
 9 files changed, 307 insertions(+), 16 deletions(-)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_devlink.c

-- 
2.11.0

[PATCH net-next 1/4] nfp: add devlink support

[Jakub Kicinski]

From: Simon Horman <simon.horman@netronome.com>

Add initial devlink support.  This patch simply switches allocation
of per-adapter structure to devlink's priv and register devlink
with empty ops table.  See following patches for implementation
of particular ops.

Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/Makefile      |  1 +
 drivers/net/ethernet/netronome/nfp/nfp_devlink.c | 39 ++++++++++++++++++++++++
 drivers/net/ethernet/netronome/nfp/nfp_main.c    | 22 ++++++++++---
 drivers/net/ethernet/netronome/nfp/nfp_main.h    |  3 ++
 4 files changed, 61 insertions(+), 4 deletions(-)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfp_devlink.c

diff --git a/drivers/net/ethernet/netronome/nfp/Makefile b/drivers/net/ethernet/netronome/nfp/Makefile
index e8333283ada6..95f6b97b5d71 100644
--- a/drivers/net/ethernet/netronome/nfp/Makefile
+++ b/drivers/net/ethernet/netronome/nfp/Makefile
@@ -15,6 +15,7 @@ nfp-objs := \
 	    nfpcore/nfp_rtsym.o \
 	    nfpcore/nfp_target.o \
 	    nfp_app.o \
+	    nfp_devlink.o \
 	    nfp_main.o \
 	    nfp_net_common.o \
 	    nfp_net_ethtool.o \
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_devlink.c b/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
new file mode 100644
index 000000000000..a8a52b3ff42b
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
@@ -0,0 +1,39 @@
+/*
+ * Copyright (C) 2017 Netronome Systems, Inc.
+ *
+ * This software is dual licensed under the GNU General License Version 2,
+ * June 1991 as shown in the file COPYING in the top-level directory of this
+ * source tree or the BSD 2-Clause License provided below.  You have the
+ * option to license this software under the complete terms of either license.
+ *
+ * The BSD 2-Clause License:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      1. Redistributions of source code must retain the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer.
+ *
+ *      2. Redistributions in binary form must reproduce the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer in the documentation and/or other materials
+ *         provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <net/devlink.h>
+
+#include "nfp_main.h"
+
+const struct devlink_ops nfp_devlink_ops = {
+};
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_main.c b/drivers/net/ethernet/netronome/nfp/nfp_main.c
index bb586ce1ea06..f22c8f083ab0 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_main.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_main.c
@@ -44,6 +44,7 @@
 #include <linux/pci.h>
 #include <linux/firmware.h>
 #include <linux/vermagic.h>
+#include <net/devlink.h>
 
 #include "nfpcore/nfp.h"
 #include "nfpcore/nfp_cpp.h"
@@ -315,6 +316,7 @@ static void nfp_fw_unload(struct nfp_pf *pf)
 static int nfp_pci_probe(struct pci_dev *pdev,
 			 const struct pci_device_id *pci_id)
 {
+	struct devlink *devlink;
 	struct nfp_pf *pf;
 	int err;
 
@@ -335,11 +337,12 @@ static int nfp_pci_probe(struct pci_dev *pdev,
 		goto err_pci_disable;
 	}
 
-	pf = kzalloc(sizeof(*pf), GFP_KERNEL);
-	if (!pf) {
+	devlink = devlink_alloc(&nfp_devlink_ops, sizeof(*pf));
+	if (!devlink) {
 		err = -ENOMEM;
 		goto err_rel_regions;
 	}
+	pf = devlink_priv(devlink);
 	INIT_LIST_HEAD(&pf->vnics);
 	INIT_LIST_HEAD(&pf->ports);
 	pci_set_drvdata(pdev, pf);
@@ -370,8 +373,14 @@ static int nfp_pci_probe(struct pci_dev *pdev,
 	if (err)
 		goto err_fw_unload;
 
+	err = devlink_register(devlink, &pdev->dev);
+	if (err)
+		goto err_kill_ndevs;
+
 	return 0;
 
+err_kill_ndevs:
+	nfp_net_pci_remove(pf);
 err_fw_unload:
 	if (pf->fw_loaded)
 		nfp_fw_unload(pf);
@@ -380,7 +389,7 @@ static int nfp_pci_probe(struct pci_dev *pdev,
 	nfp_cpp_free(pf->cpp);
 err_disable_msix:
 	pci_set_drvdata(pdev, NULL);
-	kfree(pf);
+	devlink_free(devlink);
 err_rel_regions:
 	pci_release_regions(pdev);
 err_pci_disable:
@@ -392,6 +401,11 @@ static int nfp_pci_probe(struct pci_dev *pdev,
 static void nfp_pci_remove(struct pci_dev *pdev)
 {
 	struct nfp_pf *pf = pci_get_drvdata(pdev);
+	struct devlink *devlink;
+
+	devlink = priv_to_devlink(pf);
+
+	devlink_unregister(devlink);
 
 	nfp_net_pci_remove(pf);
 
@@ -404,7 +418,7 @@ static void nfp_pci_remove(struct pci_dev *pdev)
 	nfp_cpp_free(pf->cpp);
 
 	kfree(pf->eth_tbl);
-	kfree(pf);
+	devlink_free(devlink);
 	pci_release_regions(pdev);
 	pci_disable_device(pdev);
 }
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_main.h b/drivers/net/ethernet/netronome/nfp/nfp_main.h
index 991c4cba0bbf..526db8029dea 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_main.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_main.h
@@ -47,6 +47,7 @@
 #include <linux/workqueue.h>
 
 struct dentry;
+struct devlink_ops;
 struct pci_dev;
 
 struct nfp_cpp;
@@ -107,6 +108,8 @@ struct nfp_pf {
 
 extern struct pci_driver nfp_netvf_pci_driver;
 
+extern const struct devlink_ops nfp_devlink_ops;
+
 int nfp_net_pci_probe(struct nfp_pf *pf);
 void nfp_net_pci_remove(struct nfp_pf *pf);
 
-- 
2.11.0

[PATCH net-next 2/4] nfp: register ports as devlink ports

[Jakub Kicinski]

Extend nfp_port to contain devlink_port structures.  Register the
ports to allow users inspecting device ports.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_devlink.c  | 49 +++++++++++++++++++++++
 drivers/net/ethernet/netronome/nfp/nfp_main.c     |  8 ++++
 drivers/net/ethernet/netronome/nfp/nfp_main.h     |  2 +
 drivers/net/ethernet/netronome/nfp/nfp_net_main.c | 34 ++++++++++++++++
 drivers/net/ethernet/netronome/nfp/nfp_port.h     |  8 ++++
 5 files changed, 101 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_devlink.c b/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
index a8a52b3ff42b..b65c50c0d211 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
@@ -31,9 +31,58 @@
  * SOFTWARE.
  */
 
+#include <linux/rtnetlink.h>
 #include <net/devlink.h>
 
+#include "nfpcore/nfp_nsp.h"
+#include "nfp_app.h"
 #include "nfp_main.h"
+#include "nfp_port.h"
+
+static int
+nfp_devlink_fill_eth_port(struct nfp_port *port,
+			  struct nfp_eth_table_port *copy)
+{
+	struct nfp_eth_table_port *eth_port;
+
+	eth_port = __nfp_port_get_eth_port(port);
+	if (!eth_port)
+		return -EINVAL;
+
+	memcpy(copy, eth_port, sizeof(*eth_port));
+
+	return 0;
+}
 
 const struct devlink_ops nfp_devlink_ops = {
 };
+
+int nfp_devlink_port_register(struct nfp_app *app, struct nfp_port *port)
+{
+	struct nfp_eth_table_port eth_port;
+	struct devlink *devlink;
+	int ret;
+
+	rtnl_lock();
+	ret = nfp_devlink_fill_eth_port(port, &eth_port);
+	rtnl_unlock();
+	if (ret)
+		return ret;
+
+	devlink_port_type_eth_set(&port->dl_port, port->netdev);
+	if (eth_port.is_split)
+		devlink_port_split_set(&port->dl_port, eth_port.label_port);
+
+	devlink = priv_to_devlink(app->pf);
+
+	return devlink_port_register(devlink, &port->dl_port, port->eth_id);
+}
+
+void nfp_devlink_port_unregister(struct nfp_port *port)
+{
+	/* Due to unpleasant lock ordering we may see the port go away before
+	 * we have fully probed.
+	 */
+	if (port->dl_port.registered)
+		devlink_port_unregister(&port->dl_port);
+}
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_main.c b/drivers/net/ethernet/netronome/nfp/nfp_main.c
index f22c8f083ab0..ea7c122092e7 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_main.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_main.c
@@ -377,8 +377,14 @@ static int nfp_pci_probe(struct pci_dev *pdev,
 	if (err)
 		goto err_kill_ndevs;
 
+	err = nfp_net_devlink_register(pf);
+	if (err)
+		goto err_devlink_unregiser;
+
 	return 0;
 
+err_devlink_unregiser:
+	devlink_unregister(devlink);
 err_kill_ndevs:
 	nfp_net_pci_remove(pf);
 err_fw_unload:
@@ -405,6 +411,8 @@ static void nfp_pci_remove(struct pci_dev *pdev)
 
 	devlink = priv_to_devlink(pf);
 
+	nfp_net_devlink_unregister(pf);
+
 	devlink_unregister(devlink);
 
 	nfp_net_pci_remove(pf);
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_main.h b/drivers/net/ethernet/netronome/nfp/nfp_main.h
index 526db8029dea..7fdb9036f666 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_main.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_main.h
@@ -112,5 +112,7 @@ extern const struct devlink_ops nfp_devlink_ops;
 
 int nfp_net_pci_probe(struct nfp_pf *pf);
 void nfp_net_pci_remove(struct nfp_pf *pf);
+int nfp_net_devlink_register(struct nfp_pf *pf);
+void nfp_net_devlink_unregister(struct nfp_pf *pf);
 
 #endif /* NFP_MAIN_H */
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_main.c b/drivers/net/ethernet/netronome/nfp/nfp_net_main.c
index dd1118c7e1a4..bf10c0a316f3 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_main.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_main.c
@@ -584,6 +584,8 @@ static void nfp_net_refresh_vnics(struct work_struct *work)
 		if (!nn->port || nn->port->type != NFP_PORT_INVALID)
 			continue;
 
+		nfp_devlink_port_unregister(nn->port);
+
 		nfp_net_debugfs_dir_clean(&nn->debugfs_dir);
 		nfp_net_clean(nn);
 
@@ -774,3 +776,35 @@ void nfp_net_pci_remove(struct nfp_pf *pf)
 
 	cancel_work_sync(&pf->port_refresh_work);
 }
+
+int nfp_net_devlink_register(struct nfp_pf *pf)
+{
+	struct nfp_port *port;
+	int err;
+
+	mutex_lock(&pf->lock);
+	list_for_each_entry(port, &pf->ports, port_list) {
+		err = nfp_devlink_port_register(pf->app, port);
+		if (err)
+			goto err_prev_ports_unreg;
+	}
+	mutex_unlock(&pf->lock);
+
+	return 0;
+
+err_prev_ports_unreg:
+	list_for_each_entry_continue_reverse(port, &pf->ports, port_list)
+		nfp_devlink_port_unregister(port);
+	mutex_unlock(&pf->lock);
+	return err;
+}
+
+void nfp_net_devlink_unregister(struct nfp_pf *pf)
+{
+	struct nfp_port *port;
+
+	mutex_lock(&pf->lock);
+	list_for_each_entry(port, &pf->ports, port_list)
+		nfp_devlink_port_unregister(port);
+	mutex_unlock(&pf->lock);
+}
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_port.h b/drivers/net/ethernet/netronome/nfp/nfp_port.h
index 641617de41cc..36a540b514be 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_port.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_port.h
@@ -34,6 +34,8 @@
 #ifndef _NFP_PORT_H_
 #define _NFP_PORT_H_
 
+#include <net/devlink.h>
+
 struct net_device;
 struct nfp_app;
 struct nfp_port;
@@ -66,6 +68,7 @@ enum nfp_port_flags {
  * @type:	what port type does the entity represent
  * @flags:	port flags
  * @app:	backpointer to the app structure
+ * @dl_port:	devlink port structure
  * @eth_id:	for %NFP_PORT_PHYS_PORT port ID in NFP enumeration scheme
  * @eth_port:	for %NFP_PORT_PHYS_PORT translated ETH Table port entry
  * @port_list:	entry on pf's list of ports
@@ -78,6 +81,8 @@ struct nfp_port {
 
 	struct nfp_app *app;
 
+	struct devlink_port dl_port;
+
 	unsigned int eth_id;
 	struct nfp_eth_table_port *eth_port;
 
@@ -99,4 +104,7 @@ void nfp_port_free(struct nfp_port *port);
 int nfp_net_refresh_eth_port(struct nfp_port *port);
 void nfp_net_refresh_port_table(struct nfp_port *port);
 
+int nfp_devlink_port_register(struct nfp_app *app, struct nfp_port *port);
+void nfp_devlink_port_unregister(struct nfp_port *port);
+
 #endif
-- 
2.11.0

[PATCH net-next 3/4] nfp: calculate total port lanes for split

[Jakub Kicinski]

For port splitting we will need to know the total number of lanes
in a port.  Calculate that based on eth_table information.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.h     |  3 +++
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c | 11 ++++++-----
 2 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.h b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.h
index 36b21e4dc56d..84a1d20adae1 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.h
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.h
@@ -96,6 +96,7 @@ enum nfp_eth_aneg {
  * @override_changed: is media reconfig pending?
  *
  * @port_type:	one of %PORT_* defines for ethtool
+ * @port_lanes:	total number of lanes on the port (sum of lanes of all subports)
  * @is_split:	is interface part of a split port
  */
 struct nfp_eth_table {
@@ -127,6 +128,8 @@ struct nfp_eth_table {
 		/* Computed fields */
 		u8 port_type;
 
+		unsigned int port_lanes;
+
 		bool is_split;
 	} ports[0];
 };
diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c
index 639438d8313a..b0f8785c064f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c
@@ -186,17 +186,19 @@ nfp_eth_port_translate(struct nfp_nsp *nsp, const union eth_table_entry *src,
 }
 
 static void
-nfp_eth_mark_split_ports(struct nfp_cpp *cpp, struct nfp_eth_table *table)
+nfp_eth_calc_port_geometry(struct nfp_cpp *cpp, struct nfp_eth_table *table)
 {
 	unsigned int i, j;
 
 	for (i = 0; i < table->count; i++)
 		for (j = 0; j < table->count; j++) {
-			if (i == j)
-				continue;
 			if (table->ports[i].label_port !=
 			    table->ports[j].label_port)
 				continue;
+			table->ports[i].port_lanes += table->ports[j].lanes;
+
+			if (i == j)
+				continue;
 			if (table->ports[i].label_subport ==
 			    table->ports[j].label_subport)
 				nfp_warn(cpp,
@@ -205,7 +207,6 @@ nfp_eth_mark_split_ports(struct nfp_cpp *cpp, struct nfp_eth_table *table)
 					 table->ports[i].label_subport);
 
 			table->ports[i].is_split = true;
-			break;
 		}
 }
 
@@ -289,7 +290,7 @@ __nfp_eth_read_ports(struct nfp_cpp *cpp, struct nfp_nsp *nsp)
 			nfp_eth_port_translate(nsp, &entries[i], i,
 					       &table->ports[j++]);
 
-	nfp_eth_mark_split_ports(cpp, table);
+	nfp_eth_calc_port_geometry(cpp, table);
 	for (i = 0; i < table->count; i++)
 		nfp_eth_calc_port_type(cpp, &table->ports[i]);
 
-- 
2.11.0

[PATCH net-next 4/4] nfp: support port splitting via devlink

[Jakub Kicinski]

Add support for configuring port split with devlink.  Add devlink
callbacks to validate requested config and call NSP helpers.
Getting the right nfp_port structure can be done with simple iteration.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_devlink.c  | 97 +++++++++++++++++++++++
 drivers/net/ethernet/netronome/nfp/nfp_net_main.c | 23 ++++--
 drivers/net/ethernet/netronome/nfp/nfp_port.c     | 19 +++++
 drivers/net/ethernet/netronome/nfp/nfp_port.h     |  4 +
 4 files changed, 136 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_devlink.c b/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
index b65c50c0d211..f001e3d60bd1 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_devlink.c
@@ -54,7 +54,104 @@ nfp_devlink_fill_eth_port(struct nfp_port *port,
 	return 0;
 }
 
+static int
+nfp_devlink_fill_eth_port_from_id(struct nfp_pf *pf, unsigned int port_index,
+				  struct nfp_eth_table_port *copy)
+{
+	struct nfp_port *port;
+
+	port = nfp_port_from_id(pf, NFP_PORT_PHYS_PORT, port_index);
+
+	return nfp_devlink_fill_eth_port(port, copy);
+}
+
+static int
+nfp_devlink_set_lanes(struct nfp_pf *pf, unsigned int idx, unsigned int lanes)
+{
+	struct nfp_nsp *nsp;
+	int ret;
+
+	nsp = nfp_eth_config_start(pf->cpp, idx);
+	if (IS_ERR(nsp))
+		return PTR_ERR(nsp);
+
+	ret = __nfp_eth_set_split(nsp, lanes);
+	if (ret) {
+		nfp_eth_config_cleanup_end(nsp);
+		return ret;
+	}
+
+	ret = nfp_eth_config_commit_end(nsp);
+	if (ret < 0)
+		return ret;
+	if (ret) /* no change */
+		return 0;
+
+	return nfp_net_refresh_port_table_sync(pf);
+}
+
+static int
+nfp_devlink_port_split(struct devlink *devlink, unsigned int port_index,
+		       unsigned int count)
+{
+	struct nfp_pf *pf = devlink_priv(devlink);
+	struct nfp_eth_table_port eth_port;
+	int ret;
+
+	if (count < 2)
+		return -EINVAL;
+
+	mutex_lock(&pf->lock);
+
+	rtnl_lock();
+	ret = nfp_devlink_fill_eth_port_from_id(pf, port_index, &eth_port);
+	rtnl_unlock();
+	if (ret)
+		goto out;
+
+	if (eth_port.is_split || eth_port.port_lanes % count) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = nfp_devlink_set_lanes(pf, eth_port.index,
+				    eth_port.port_lanes / count);
+out:
+	mutex_unlock(&pf->lock);
+
+	return ret;
+}
+
+static int
+nfp_devlink_port_unsplit(struct devlink *devlink, unsigned int port_index)
+{
+	struct nfp_pf *pf = devlink_priv(devlink);
+	struct nfp_eth_table_port eth_port;
+	int ret;
+
+	mutex_lock(&pf->lock);
+
+	rtnl_lock();
+	ret = nfp_devlink_fill_eth_port_from_id(pf, port_index, &eth_port);
+	rtnl_unlock();
+	if (ret)
+		goto out;
+
+	if (!eth_port.is_split) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = nfp_devlink_set_lanes(pf, eth_port.index, eth_port.port_lanes);
+out:
+	mutex_unlock(&pf->lock);
+
+	return ret;
+}
+
 const struct devlink_ops nfp_devlink_ops = {
+	.port_split		= nfp_devlink_port_split,
+	.port_unsplit		= nfp_devlink_port_unsplit,
 };
 
 int nfp_devlink_port_register(struct nfp_app *app, struct nfp_port *port)
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_main.c b/drivers/net/ethernet/netronome/nfp/nfp_net_main.c
index bf10c0a316f3..a50acca25cfa 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_main.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_main.c
@@ -43,6 +43,7 @@
 #include <linux/etherdevice.h>
 #include <linux/kernel.h>
 #include <linux/init.h>
+#include <linux/lockdep.h>
 #include <linux/pci.h>
 #include <linux/pci_regs.h>
 #include <linux/msi.h>
@@ -543,19 +544,17 @@ nfp_net_eth_port_update(struct nfp_cpp *cpp, struct nfp_port *port,
 	return 0;
 }
 
-static void nfp_net_refresh_vnics(struct work_struct *work)
+int nfp_net_refresh_port_table_sync(struct nfp_pf *pf)
 {
-	struct nfp_pf *pf = container_of(work, struct nfp_pf,
-					 port_refresh_work);
 	struct nfp_eth_table *eth_table;
 	struct nfp_net *nn, *next;
 	struct nfp_port *port;
 
-	mutex_lock(&pf->lock);
+	lockdep_assert_held(&pf->lock);
 
 	/* Check for nfp_net_pci_remove() racing against us */
 	if (list_empty(&pf->vnics))
-		goto out;
+		return 0;
 
 	/* Update state of all ports */
 	rtnl_lock();
@@ -569,7 +568,7 @@ static void nfp_net_refresh_vnics(struct work_struct *work)
 				set_bit(NFP_PORT_CHANGED, &port->flags);
 		rtnl_unlock();
 		nfp_err(pf->cpp, "Error refreshing port config!\n");
-		goto out;
+		return -EIO;
 	}
 
 	list_for_each_entry(port, &pf->ports, port_list)
@@ -594,7 +593,17 @@ static void nfp_net_refresh_vnics(struct work_struct *work)
 
 	if (list_empty(&pf->vnics))
 		nfp_net_pci_remove_finish(pf);
-out:
+
+	return 0;
+}
+
+static void nfp_net_refresh_vnics(struct work_struct *work)
+{
+	struct nfp_pf *pf = container_of(work, struct nfp_pf,
+					 port_refresh_work);
+
+	mutex_lock(&pf->lock);
+	nfp_net_refresh_port_table_sync(pf);
 	mutex_unlock(&pf->lock);
 }
 
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_port.c b/drivers/net/ethernet/netronome/nfp/nfp_port.c
index 3beed4167e2f..a17410ac01ab 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_port.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_port.c
@@ -31,6 +31,8 @@
  * SOFTWARE.
  */
 
+#include <linux/lockdep.h>
+
 #include "nfpcore/nfp_nsp.h"
 #include "nfp_app.h"
 #include "nfp_main.h"
@@ -48,6 +50,23 @@ struct nfp_port *nfp_port_from_netdev(struct net_device *netdev)
 	return nn->port;
 }
 
+struct nfp_port *
+nfp_port_from_id(struct nfp_pf *pf, enum nfp_port_type type, unsigned int id)
+{
+	struct nfp_port *port;
+
+	lockdep_assert_held(&pf->lock);
+
+	if (type != NFP_PORT_PHYS_PORT)
+		return NULL;
+
+	list_for_each_entry(port, &pf->ports, port_list)
+		if (port->eth_id == id)
+			return port;
+
+	return NULL;
+}
+
 struct nfp_eth_table_port *__nfp_port_get_eth_port(struct nfp_port *port)
 {
 	if (!port)
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_port.h b/drivers/net/ethernet/netronome/nfp/nfp_port.h
index 36a540b514be..4d1a9b3fed41 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_port.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_port.h
@@ -38,6 +38,7 @@
 
 struct net_device;
 struct nfp_app;
+struct nfp_pf;
 struct nfp_port;
 
 /**
@@ -90,6 +91,8 @@ struct nfp_port {
 };
 
 struct nfp_port *nfp_port_from_netdev(struct net_device *netdev);
+struct nfp_port *
+nfp_port_from_id(struct nfp_pf *pf, enum nfp_port_type type, unsigned int id);
 struct nfp_eth_table_port *__nfp_port_get_eth_port(struct nfp_port *port);
 struct nfp_eth_table_port *nfp_port_get_eth_port(struct nfp_port *port);
 
@@ -103,6 +106,7 @@ void nfp_port_free(struct nfp_port *port);
 
 int nfp_net_refresh_eth_port(struct nfp_port *port);
 void nfp_net_refresh_port_table(struct nfp_port *port);
+int nfp_net_refresh_port_table_sync(struct nfp_pf *pf);
 
 int nfp_devlink_port_register(struct nfp_app *app, struct nfp_port *port);
 void nfp_devlink_port_unregister(struct nfp_port *port);
-- 
2.11.0

Re: [PATCH net-next 2/4] nfp: register ports as devlink ports

[Jiri Pirko]

Tue, May 23, 2017 at 05:12:45PM CEST, jakub.kicinski@netronome.com wrote:
>Extend nfp_port to contain devlink_port structures.  Register the
>ports to allow users inspecting device ports.
>
>Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
>Reviewed-by: Simon Horman <simon.horman@netronome.com>
>---

[...]

>+void nfp_devlink_port_unregister(struct nfp_port *port)
>+{
>+	/* Due to unpleasant lock ordering we may see the port go away before
>+	 * we have fully probed.

Could you elaborate on this a bit more please?


>+	 */
>+	if (port->dl_port.registered)
>+		devlink_port_unregister(&port->dl_port);
>+}

Re: [PATCH net-next 2/4] nfp: register ports as devlink ports

[Jakub Kicinski]

On Wed, 24 May 2017 14:35:14 +0200, Jiri Pirko wrote:
> >+void nfp_devlink_port_unregister(struct nfp_port *port)
> >+{
> >+	/* Due to unpleasant lock ordering we may see the port go away before
> >+	 * we have fully probed.  
> 
> Could you elaborate on this a bit more please?

It's partially due to peculiarities of the management FW more than
kernel stuff.  Unfortunately some ethtool media config requires reboot
to be applied, so we print a friendly message to the logs and
unregister the associated netdevs.  Which means once netdevs get
registered ports may go away.

Enter devlink, I need the ability to grab the adapater lock in
split/unsplit callbacks to find the ports, which implies having to drop
that lock before I register devlink.  And only after I register devlink
can I register the ports.

I could do init without registering anything, drop the adapter lock,
register devlink, and then grab the adapter lock back and register
devlink ports and netdevs.  But there is another issue...

Since I look for ports on a list maintained in the adapter struct,
driver code doesn't care if devlink_port has been registered or not.
The moment devlink is registered, split/unsplit requests will be
accepted - potentially trying to unregister devlink_port before the
register could happen.

Further down the line, also, the eswitch mode setting is coming.  Which
means the moment I register devlink itself ports will get shuffled (due
to the plan of registering VFs as ports :)). 

I feel like registering devlink should be the last action of the
driver, really.  My plan was to keep that simple if() for now, and once
we get to extending devlink with SR-IOV stuff also add the ability to
pre-register ports.  Allow registering ports on not-yet-registered
devlink (probably put them on a private list within struct devlink).
This would make devlink_register() a single point when everything
devlink becomes visible, atomically, instead of devlink itself coming
first and then ports following.

Does that make sense?  Am I misreading the code (again :S)?

Re: [PATCH net-next 2/4] nfp: register ports as devlink ports

[Jiri Pirko]

Wed, May 24, 2017 at 09:24:13PM CEST, kubakici@wp.pl wrote:
>On Wed, 24 May 2017 14:35:14 +0200, Jiri Pirko wrote:
>> >+void nfp_devlink_port_unregister(struct nfp_port *port)
>> >+{
>> >+	/* Due to unpleasant lock ordering we may see the port go away before
>> >+	 * we have fully probed.  
>> 
>> Could you elaborate on this a bit more please?
>
>It's partially due to peculiarities of the management FW more than
>kernel stuff.  Unfortunately some ethtool media config requires reboot
>to be applied, so we print a friendly message to the logs and
>unregister the associated netdevs.  Which means once netdevs get
>registered ports may go away.
>
>Enter devlink, I need the ability to grab the adapater lock in
>split/unsplit callbacks to find the ports, which implies having to drop
>that lock before I register devlink.  And only after I register devlink
>can I register the ports.
>
>I could do init without registering anything, drop the adapter lock,
>register devlink, and then grab the adapter lock back and register
>devlink ports and netdevs.  But there is another issue...
>
>Since I look for ports on a list maintained in the adapter struct,
>driver code doesn't care if devlink_port has been registered or not.
>The moment devlink is registered, split/unsplit requests will be
>accepted - potentially trying to unregister devlink_port before the
>register could happen.
>
>Further down the line, also, the eswitch mode setting is coming.  Which
>means the moment I register devlink itself ports will get shuffled (due
>to the plan of registering VFs as ports :)). 
>
>I feel like registering devlink should be the last action of the
>driver, really.  My plan was to keep that simple if() for now, and once
>we get to extending devlink with SR-IOV stuff also add the ability to
>pre-register ports.  Allow registering ports on not-yet-registered
>devlink (probably put them on a private list within struct devlink).
>This would make devlink_register() a single point when everything
>devlink becomes visible, atomically, instead of devlink itself coming
>first and then ports following.
>
>Does that make sense?  Am I misreading the code (again :S)?

Well in mlxsw, we internally maintain a list of port and we assign a
pointer to the port struct only after all is initialized:

...
mlxsw_sp->ports[local_port] = mlxsw_sp_port;
err = register_netdev(dev);
...

Then in split, we check:
mlxsw_sp_port = mlxsw_sp->ports[local_port];
if (!mlxsw_sp_port) {
        dev_err(mlxsw_sp->bus_info->dev, "Port number \"%d\" does not exist\n",
                local_port);
        return -EINVAL;
}

I guess that you can do something similar. You should ensure that split
won't happen for half-initialized ports.