[PATCH net-next 0/6] net/mlx5: Move notifiers outside the devlink lock

[PATCH net-next 0/6] net/mlx5: Move notifiers outside the devlink lock

[Tariq Toukan]

Hi,

This series by Cosmin moves blocking notifier registration in the mlx5
driver outside the devlink lock during probe.

This is mostly a no-op refactoring that consists of multiple pieces.
It is necessary because upcoming code will introduce a potential locking
cycle between the devlink lock and the blocking notifier head mutexes,
so these notifiers must move out of the devlink-locked critical section.

Regards,
Tariq


Cosmin Ratiu (6):
  net/mlx5: Initialize events outside devlink lock
  net/mlx5: Move the esw mode notifier chain outside the devlink lock
  net/mlx5: Move the vhca event notifier outside of the devlink lock
  net/mlx5: Move the SF HW table notifier outside the devlink lock
  net/mlx5: Move the SF table notifiers outside the devlink lock
  net/mlx5: Move SF dev table notifier registration outside the PF
    devlink lock

 .../net/ethernet/mellanox/mlx5/core/eswitch.c | 13 +--
 .../net/ethernet/mellanox/mlx5/core/eswitch.h |  7 +-
 .../net/ethernet/mellanox/mlx5/core/main.c    | 70 +++++++++++----
 .../ethernet/mellanox/mlx5/core/sf/dev/dev.c  | 48 ++++++----
 .../ethernet/mellanox/mlx5/core/sf/dev/dev.h  | 11 +++
 .../ethernet/mellanox/mlx5/core/sf/devlink.c  | 90 ++++++++++++-------
 .../ethernet/mellanox/mlx5/core/sf/hw_table.c | 61 ++++++++-----
 .../net/ethernet/mellanox/mlx5/core/sf/sf.h   | 20 ++++-
 .../mellanox/mlx5/core/sf/vhca_event.c        | 69 ++++++--------
 .../mellanox/mlx5/core/sf/vhca_event.h        |  5 ++
 include/linux/mlx5/driver.h                   | 10 ++-
 11 files changed, 253 insertions(+), 151 deletions(-)


base-commit: c9dfb92de0738eb7fe6a591ad1642333793e8b6e
-- 
2.31.1

[PATCH net-next 1/6] net/mlx5: Initialize events outside devlink lock

[Tariq Toukan]

From: Cosmin Ratiu <cratiu@nvidia.com>

Move event init/cleanup outside of mlx5_init_one() / mlx5_uninit_one()
and into the mlx5_mdev_init() / mlx5_mdev_uninit() functions.

By doing this, we avoid the events being reinitialized on devlink reload
and, more importantly, the events->sw_nh notifier chain becomes
available earlier in the init procedure, which will be used in
subsequent patches. This makes sense because the events struct is pure
software, independent of any HW details.

Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/main.c    | 34 +++++++++++++------
 1 file changed, 24 insertions(+), 10 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index c904696cbc3a..612fc4de9d3c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1010,16 +1010,10 @@ static int mlx5_init_once(struct mlx5_core_dev *dev)
 		goto err_irq_cleanup;
 	}
 
-	err = mlx5_events_init(dev);
-	if (err) {
-		mlx5_core_err(dev, "failed to initialize events\n");
-		goto err_eq_cleanup;
-	}
-
 	err = mlx5_fw_reset_init(dev);
 	if (err) {
 		mlx5_core_err(dev, "failed to initialize fw reset events\n");
-		goto err_events_cleanup;
+		goto err_eq_cleanup;
 	}
 
 	mlx5_cq_debugfs_init(dev);
@@ -1121,8 +1115,6 @@ static int mlx5_init_once(struct mlx5_core_dev *dev)
 	mlx5_cleanup_reserved_gids(dev);
 	mlx5_cq_debugfs_cleanup(dev);
 	mlx5_fw_reset_cleanup(dev);
-err_events_cleanup:
-	mlx5_events_cleanup(dev);
 err_eq_cleanup:
 	mlx5_eq_table_cleanup(dev);
 err_irq_cleanup:
@@ -1155,7 +1147,6 @@ static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
 	mlx5_cleanup_reserved_gids(dev);
 	mlx5_cq_debugfs_cleanup(dev);
 	mlx5_fw_reset_cleanup(dev);
-	mlx5_events_cleanup(dev);
 	mlx5_eq_table_cleanup(dev);
 	mlx5_irq_table_cleanup(dev);
 	mlx5_devcom_unregister_device(dev->priv.devc);
@@ -1833,6 +1824,24 @@ static int vhca_id_show(struct seq_file *file, void *priv)
 
 DEFINE_SHOW_ATTRIBUTE(vhca_id);
 
+static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
+{
+	int err;
+
+	err = mlx5_events_init(dev);
+	if (err) {
+		mlx5_core_err(dev, "failed to initialize events\n");
+		return err;
+	}
+
+	return 0;
+}
+
+static void mlx5_notifiers_cleanup(struct mlx5_core_dev *dev)
+{
+	mlx5_events_cleanup(dev);
+}
+
 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
 {
 	struct mlx5_priv *priv = &dev->priv;
@@ -1888,6 +1897,10 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
 	if (err)
 		goto err_hca_caps;
 
+	err = mlx5_notifiers_init(dev);
+	if (err)
+		goto err_hca_caps;
+
 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
 	 * unique id per function which uses mlx5_core.
 	 * Those values are supplied to FW as part of the init HCA command to
@@ -1930,6 +1943,7 @@ void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
 	if (priv->sw_vhca_id > 0)
 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
 
+	mlx5_notifiers_cleanup(dev);
 	mlx5_hca_caps_free(dev);
 	mlx5_adev_cleanup(dev);
 	mlx5_pagealloc_cleanup(dev);
-- 
2.31.1

[PATCH net-next 2/6] net/mlx5: Move the esw mode notifier chain outside the devlink lock

[Tariq Toukan]

From: Cosmin Ratiu <cratiu@nvidia.com>

The esw mode change notifier chain is initialized/cleaned up in
mlx5_init_one() / mlx5_uninit_one() with the devlink lock held.

Move the notifier head from the eswitch struct into mlx5_priv directly,
and initialize it outside the critical section. This will allow notifier
registration to happen earlier in the init procedure in subsequent
patches.

Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/eswitch.c   | 13 +++++++------
 drivers/net/ethernet/mellanox/mlx5/core/eswitch.h   |  7 ++++---
 drivers/net/ethernet/mellanox/mlx5/core/main.c      |  2 ++
 .../net/ethernet/mellanox/mlx5/core/sf/devlink.c    |  6 +++---
 include/linux/mlx5/driver.h                         |  1 +
 5 files changed, 17 insertions(+), 12 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c b/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
index 25af8bd7f077..3adf2b1cd26a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
@@ -1474,7 +1474,7 @@ static void mlx5_esw_mode_change_notify(struct mlx5_eswitch *esw, u16 mode)
 
 	info.new_mode = mode;
 
-	blocking_notifier_call_chain(&esw->n_head, 0, &info);
+	blocking_notifier_call_chain(&esw->dev->priv.esw_n_head, 0, &info);
 }
 
 static int mlx5_esw_egress_acls_init(struct mlx5_core_dev *dev)
@@ -2050,7 +2050,6 @@ int mlx5_eswitch_init(struct mlx5_core_dev *dev)
 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
 	else
 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
-	BLOCKING_INIT_NOTIFIER_HEAD(&esw->n_head);
 
 	esw_info(dev,
 		 "Total vports %d, per vport: max uc(%d) max mc(%d)\n",
@@ -2379,14 +2378,16 @@ bool mlx5_esw_multipath_prereq(struct mlx5_core_dev *dev0,
 		dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS);
 }
 
-int mlx5_esw_event_notifier_register(struct mlx5_eswitch *esw, struct notifier_block *nb)
+int mlx5_esw_event_notifier_register(struct mlx5_core_dev *dev,
+				     struct notifier_block *nb)
 {
-	return blocking_notifier_chain_register(&esw->n_head, nb);
+	return blocking_notifier_chain_register(&dev->priv.esw_n_head, nb);
 }
 
-void mlx5_esw_event_notifier_unregister(struct mlx5_eswitch *esw, struct notifier_block *nb)
+void mlx5_esw_event_notifier_unregister(struct mlx5_core_dev *dev,
+					struct notifier_block *nb)
 {
-	blocking_notifier_chain_unregister(&esw->n_head, nb);
+	blocking_notifier_chain_unregister(&dev->priv.esw_n_head, nb);
 }
 
 /**
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eswitch.h b/drivers/net/ethernet/mellanox/mlx5/core/eswitch.h
index beaec450a734..ad1073f7b79f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eswitch.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eswitch.h
@@ -403,7 +403,6 @@ struct mlx5_eswitch {
 	struct {
 		u32             large_group_num;
 	}  params;
-	struct blocking_notifier_head n_head;
 	struct xarray paired;
 	struct mlx5_devcom_comp_dev *devcom;
 	u16 enabled_ipsec_vf_count;
@@ -864,8 +863,10 @@ struct mlx5_esw_event_info {
 	u16 new_mode;
 };
 
-int mlx5_esw_event_notifier_register(struct mlx5_eswitch *esw, struct notifier_block *n);
-void mlx5_esw_event_notifier_unregister(struct mlx5_eswitch *esw, struct notifier_block *n);
+int mlx5_esw_event_notifier_register(struct mlx5_core_dev *dev,
+				     struct notifier_block *n);
+void mlx5_esw_event_notifier_unregister(struct mlx5_core_dev *dev,
+					struct notifier_block *n);
 
 bool mlx5_esw_hold(struct mlx5_core_dev *dev);
 void mlx5_esw_release(struct mlx5_core_dev *dev);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 612fc4de9d3c..05f16f3e9c4f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1834,6 +1834,8 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 		return err;
 	}
 
+	BLOCKING_INIT_NOTIFIER_HEAD(&dev->priv.esw_n_head);
+
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c
index 3304f25cc805..2ece4983d33f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c
@@ -481,7 +481,7 @@ int mlx5_sf_table_init(struct mlx5_core_dev *dev)
 	xa_init(&table->function_ids);
 	dev->priv.sf_table = table;
 	table->esw_nb.notifier_call = mlx5_sf_esw_event;
-	err = mlx5_esw_event_notifier_register(dev->priv.eswitch, &table->esw_nb);
+	err = mlx5_esw_event_notifier_register(dev, &table->esw_nb);
 	if (err)
 		goto reg_err;
 
@@ -496,7 +496,7 @@ int mlx5_sf_table_init(struct mlx5_core_dev *dev)
 	return 0;
 
 vhca_err:
-	mlx5_esw_event_notifier_unregister(dev->priv.eswitch, &table->esw_nb);
+	mlx5_esw_event_notifier_unregister(dev, &table->esw_nb);
 reg_err:
 	mutex_destroy(&table->sf_state_lock);
 	kfree(table);
@@ -513,7 +513,7 @@ void mlx5_sf_table_cleanup(struct mlx5_core_dev *dev)
 
 	mlx5_blocking_notifier_unregister(dev, &table->mdev_nb);
 	mlx5_vhca_event_notifier_unregister(table->dev, &table->vhca_nb);
-	mlx5_esw_event_notifier_unregister(dev->priv.eswitch, &table->esw_nb);
+	mlx5_esw_event_notifier_unregister(dev, &table->esw_nb);
 	mutex_destroy(&table->sf_state_lock);
 	WARN_ON(!xa_empty(&table->function_ids));
 	kfree(table);
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 046396269ccf..27eeadbcff50 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -599,6 +599,7 @@ struct mlx5_priv {
 
 	struct mlx5_flow_steering *steering;
 	struct mlx5_mpfs        *mpfs;
+	struct blocking_notifier_head esw_n_head;
 	struct mlx5_eswitch     *eswitch;
 	struct mlx5_core_sriov	sriov;
 	struct mlx5_lag		*lag;
-- 
2.31.1

[PATCH net-next 3/6] net/mlx5: Move the vhca event notifier outside of the devlink lock

[Tariq Toukan]

From: Cosmin Ratiu <cratiu@nvidia.com>

The vhca event notifier consists of an atomic notifier for vhca state
changes (used for SF events), multiple workqueues and a blocking
notifier chain for delivering the vhca state change events for further
processing.

This patch moves the vhca notifier head outside of mlx5_init_one() /
mlx5_uninit_one() and into the mlx5_mdev_init() / mlx5_mdev_uninit()
functions.

This allows called notifiers to grab the PF devlink lock which was
previously impossible because it would create a circular lock
dependency.

mlx5_vhca_event_stop() is now called earlier in the cleanup phase and
flushes the workqueues to ensure that after the call, there are no
pending events. This simplifies the cleanup flow for vhca event
consumers.

Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/main.c    |  3 +-
 .../ethernet/mellanox/mlx5/core/sf/dev/dev.c  |  1 -
 .../ethernet/mellanox/mlx5/core/sf/hw_table.c |  1 +
 .../mellanox/mlx5/core/sf/vhca_event.c        | 69 +++++++------------
 .../mellanox/mlx5/core/sf/vhca_event.h        |  5 ++
 include/linux/mlx5/driver.h                   |  4 +-
 6 files changed, 35 insertions(+), 48 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 05f16f3e9c4f..097ba7ab90a4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1438,12 +1438,12 @@ static void mlx5_unload(struct mlx5_core_dev *dev)
 {
 	mlx5_eswitch_disable(dev->priv.eswitch);
 	mlx5_devlink_traps_unregister(priv_to_devlink(dev));
+	mlx5_vhca_event_stop(dev);
 	mlx5_sf_dev_table_destroy(dev);
 	mlx5_sriov_detach(dev);
 	mlx5_lag_remove_mdev(dev);
 	mlx5_ec_cleanup(dev);
 	mlx5_sf_hw_table_destroy(dev);
-	mlx5_vhca_event_stop(dev);
 	mlx5_fs_core_cleanup(dev);
 	mlx5_fpga_device_stop(dev);
 	mlx5_rsc_dump_cleanup(dev);
@@ -1835,6 +1835,7 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 	}
 
 	BLOCKING_INIT_NOTIFIER_HEAD(&dev->priv.esw_n_head);
+	mlx5_vhca_state_notifier_init(dev);
 
 	return 0;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
index 99219ea52c4b..a68a8ee24dce 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
@@ -381,7 +381,6 @@ void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev)
 
 	mlx5_sf_dev_destroy_active_works(table);
 	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
-	mlx5_vhca_event_work_queues_flush(dev);
 
 	/* Now that event handler is not running, it is safe to destroy
 	 * the sf device without race.
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c
index 1f613320fe07..a14b1aa5fb5a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c
@@ -389,6 +389,7 @@ void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev)
 		return;
 
 	mlx5_vhca_event_notifier_unregister(dev, &table->vhca_nb);
+
 	/* Dealloc SFs whose firmware event has been missed. */
 	mlx5_sf_hw_table_dealloc_all(table);
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.c
index cda01ba441ae..b04cf6cf8956 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.c
@@ -9,15 +9,9 @@
 #define CREATE_TRACE_POINTS
 #include "diag/vhca_tracepoint.h"
 
-struct mlx5_vhca_state_notifier {
-	struct mlx5_core_dev *dev;
-	struct mlx5_nb nb;
-	struct blocking_notifier_head n_head;
-};
-
 struct mlx5_vhca_event_work {
 	struct work_struct work;
-	struct mlx5_vhca_state_notifier *notifier;
+	struct mlx5_core_dev *dev;
 	struct mlx5_vhca_state_event event;
 };
 
@@ -95,16 +89,14 @@ mlx5_vhca_event_notify(struct mlx5_core_dev *dev, struct mlx5_vhca_state_event *
 	mlx5_vhca_event_arm(dev, event->function_id);
 	trace_mlx5_sf_vhca_event(dev, event);
 
-	blocking_notifier_call_chain(&dev->priv.vhca_state_notifier->n_head, 0, event);
+	blocking_notifier_call_chain(&dev->priv.vhca_state_n_head, 0, event);
 }
 
 static void mlx5_vhca_state_work_handler(struct work_struct *_work)
 {
 	struct mlx5_vhca_event_work *work = container_of(_work, struct mlx5_vhca_event_work, work);
-	struct mlx5_vhca_state_notifier *notifier = work->notifier;
-	struct mlx5_core_dev *dev = notifier->dev;
 
-	mlx5_vhca_event_notify(dev, &work->event);
+	mlx5_vhca_event_notify(work->dev, &work->event);
 	kfree(work);
 }
 
@@ -116,8 +108,8 @@ void mlx5_vhca_events_work_enqueue(struct mlx5_core_dev *dev, int idx, struct wo
 static int
 mlx5_vhca_state_change_notifier(struct notifier_block *nb, unsigned long type, void *data)
 {
-	struct mlx5_vhca_state_notifier *notifier =
-				mlx5_nb_cof(nb, struct mlx5_vhca_state_notifier, nb);
+	struct mlx5_core_dev *dev = mlx5_nb_cof(nb, struct mlx5_core_dev,
+						priv.vhca_state_nb);
 	struct mlx5_vhca_event_work *work;
 	struct mlx5_eqe *eqe = data;
 	int wq_idx;
@@ -126,10 +118,10 @@ mlx5_vhca_state_change_notifier(struct notifier_block *nb, unsigned long type, v
 	if (!work)
 		return NOTIFY_DONE;
 	INIT_WORK(&work->work, &mlx5_vhca_state_work_handler);
-	work->notifier = notifier;
+	work->dev = dev;
 	work->event.function_id = be16_to_cpu(eqe->data.vhca_state.function_id);
 	wq_idx = work->event.function_id % MLX5_DEV_MAX_WQS;
-	mlx5_vhca_events_work_enqueue(notifier->dev, wq_idx, &work->work);
+	mlx5_vhca_events_work_enqueue(dev, wq_idx, &work->work);
 	return NOTIFY_OK;
 }
 
@@ -145,9 +137,15 @@ void mlx5_vhca_state_cap_handle(struct mlx5_core_dev *dev, void *set_hca_cap)
 	MLX5_SET(cmd_hca_cap, set_hca_cap, event_on_vhca_state_teardown_request, 1);
 }
 
+void mlx5_vhca_state_notifier_init(struct mlx5_core_dev *dev)
+{
+	BLOCKING_INIT_NOTIFIER_HEAD(&dev->priv.vhca_state_n_head);
+	MLX5_NB_INIT(&dev->priv.vhca_state_nb, mlx5_vhca_state_change_notifier,
+		     VHCA_STATE_CHANGE);
+}
+
 int mlx5_vhca_event_init(struct mlx5_core_dev *dev)
 {
-	struct mlx5_vhca_state_notifier *notifier;
 	char wq_name[MLX5_CMD_WQ_MAX_NAME];
 	struct mlx5_vhca_events *events;
 	int err, i;
@@ -160,7 +158,6 @@ int mlx5_vhca_event_init(struct mlx5_core_dev *dev)
 		return -ENOMEM;
 
 	events->dev = dev;
-	dev->priv.vhca_events = events;
 	for (i = 0; i < MLX5_DEV_MAX_WQS; i++) {
 		snprintf(wq_name, MLX5_CMD_WQ_MAX_NAME, "mlx5_vhca_event%d", i);
 		events->handler[i].wq = create_singlethread_workqueue(wq_name);
@@ -169,20 +166,10 @@ int mlx5_vhca_event_init(struct mlx5_core_dev *dev)
 			goto err_create_wq;
 		}
 	}
+	dev->priv.vhca_events = events;
 
-	notifier = kzalloc(sizeof(*notifier), GFP_KERNEL);
-	if (!notifier) {
-		err = -ENOMEM;
-		goto err_notifier;
-	}
-
-	dev->priv.vhca_state_notifier = notifier;
-	notifier->dev = dev;
-	BLOCKING_INIT_NOTIFIER_HEAD(&notifier->n_head);
-	MLX5_NB_INIT(&notifier->nb, mlx5_vhca_state_change_notifier, VHCA_STATE_CHANGE);
 	return 0;
 
-err_notifier:
 err_create_wq:
 	for (--i; i >= 0; i--)
 		destroy_workqueue(events->handler[i].wq);
@@ -211,8 +198,6 @@ void mlx5_vhca_event_cleanup(struct mlx5_core_dev *dev)
 	if (!mlx5_vhca_event_supported(dev))
 		return;
 
-	kfree(dev->priv.vhca_state_notifier);
-	dev->priv.vhca_state_notifier = NULL;
 	vhca_events = dev->priv.vhca_events;
 	for (i = 0; i < MLX5_DEV_MAX_WQS; i++)
 		destroy_workqueue(vhca_events->handler[i].wq);
@@ -221,34 +206,30 @@ void mlx5_vhca_event_cleanup(struct mlx5_core_dev *dev)
 
 void mlx5_vhca_event_start(struct mlx5_core_dev *dev)
 {
-	struct mlx5_vhca_state_notifier *notifier;
-
-	if (!dev->priv.vhca_state_notifier)
+	if (!mlx5_vhca_event_supported(dev))
 		return;
 
-	notifier = dev->priv.vhca_state_notifier;
-	mlx5_eq_notifier_register(dev, &notifier->nb);
+	mlx5_eq_notifier_register(dev, &dev->priv.vhca_state_nb);
 }
 
 void mlx5_vhca_event_stop(struct mlx5_core_dev *dev)
 {
-	struct mlx5_vhca_state_notifier *notifier;
-
-	if (!dev->priv.vhca_state_notifier)
+	if (!mlx5_vhca_event_supported(dev))
 		return;
 
-	notifier = dev->priv.vhca_state_notifier;
-	mlx5_eq_notifier_unregister(dev, &notifier->nb);
+	mlx5_eq_notifier_unregister(dev, &dev->priv.vhca_state_nb);
+
+	/* Flush workqueues of all pending events. */
+	mlx5_vhca_event_work_queues_flush(dev);
 }
 
 int mlx5_vhca_event_notifier_register(struct mlx5_core_dev *dev, struct notifier_block *nb)
 {
-	if (!dev->priv.vhca_state_notifier)
-		return -EOPNOTSUPP;
-	return blocking_notifier_chain_register(&dev->priv.vhca_state_notifier->n_head, nb);
+	return blocking_notifier_chain_register(&dev->priv.vhca_state_n_head,
+						nb);
 }
 
 void mlx5_vhca_event_notifier_unregister(struct mlx5_core_dev *dev, struct notifier_block *nb)
 {
-	blocking_notifier_chain_unregister(&dev->priv.vhca_state_notifier->n_head, nb);
+	blocking_notifier_chain_unregister(&dev->priv.vhca_state_n_head, nb);
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.h b/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.h
index 1725ba64f8af..52790423874c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/vhca_event.h
@@ -18,6 +18,7 @@ static inline bool mlx5_vhca_event_supported(const struct mlx5_core_dev *dev)
 }
 
 void mlx5_vhca_state_cap_handle(struct mlx5_core_dev *dev, void *set_hca_cap);
+void mlx5_vhca_state_notifier_init(struct mlx5_core_dev *dev);
 int mlx5_vhca_event_init(struct mlx5_core_dev *dev);
 void mlx5_vhca_event_cleanup(struct mlx5_core_dev *dev);
 void mlx5_vhca_event_start(struct mlx5_core_dev *dev);
@@ -37,6 +38,10 @@ static inline void mlx5_vhca_state_cap_handle(struct mlx5_core_dev *dev, void *s
 {
 }
 
+static inline void mlx5_vhca_state_notifier_init(struct mlx5_core_dev *dev)
+{
+}
+
 static inline int mlx5_vhca_event_init(struct mlx5_core_dev *dev)
 {
 	return 0;
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 27eeadbcff50..57368db0a5b7 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -488,7 +488,6 @@ struct mlx5_devcom_dev;
 struct mlx5_fw_reset;
 struct mlx5_eq_table;
 struct mlx5_irq_table;
-struct mlx5_vhca_state_notifier;
 struct mlx5_sf_dev_table;
 struct mlx5_sf_hw_table;
 struct mlx5_sf_table;
@@ -615,7 +614,8 @@ struct mlx5_priv {
 	struct mlx5_bfreg_data		bfregs;
 	struct mlx5_sq_bfreg bfreg;
 #ifdef CONFIG_MLX5_SF
-	struct mlx5_vhca_state_notifier *vhca_state_notifier;
+	struct mlx5_nb vhca_state_nb;
+	struct blocking_notifier_head vhca_state_n_head;
 	struct mlx5_sf_dev_table *sf_dev_table;
 	struct mlx5_core_dev *parent_mdev;
 #endif
-- 
2.31.1

[PATCH net-next 4/6] net/mlx5: Move the SF HW table notifier outside the devlink lock

[Tariq Toukan]

From: Cosmin Ratiu <cratiu@nvidia.com>

Move the SF HW table notifier registration/unregistration outside of
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.

This is only done for non-SFs, since SFs do not have a SF HW table
themselves.

Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/main.c    | 17 ++---
 .../ethernet/mellanox/mlx5/core/sf/hw_table.c | 62 +++++++++++--------
 .../net/ethernet/mellanox/mlx5/core/sf/sf.h   |  9 ++-
 include/linux/mlx5/driver.h                   |  1 +
 4 files changed, 54 insertions(+), 35 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 097ba7ab90a4..843ee452239f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1377,12 +1377,6 @@ static int mlx5_load(struct mlx5_core_dev *dev)
 
 	mlx5_vhca_event_start(dev);
 
-	err = mlx5_sf_hw_table_create(dev);
-	if (err) {
-		mlx5_core_err(dev, "sf table create failed %d\n", err);
-		goto err_vhca;
-	}
-
 	err = mlx5_ec_init(dev);
 	if (err) {
 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
@@ -1411,8 +1405,6 @@ static int mlx5_load(struct mlx5_core_dev *dev)
 	mlx5_lag_remove_mdev(dev);
 	mlx5_ec_cleanup(dev);
 err_ec:
-	mlx5_sf_hw_table_destroy(dev);
-err_vhca:
 	mlx5_vhca_event_stop(dev);
 err_set_hca:
 	mlx5_fs_core_cleanup(dev);
@@ -1837,11 +1829,20 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 	BLOCKING_INIT_NOTIFIER_HEAD(&dev->priv.esw_n_head);
 	mlx5_vhca_state_notifier_init(dev);
 
+	err = mlx5_sf_hw_notifier_init(dev);
+	if (err)
+		goto err_sf_hw_notifier;
+
 	return 0;
+
+err_sf_hw_notifier:
+	mlx5_events_cleanup(dev);
+	return err;
 }
 
 static void mlx5_notifiers_cleanup(struct mlx5_core_dev *dev)
 {
+	mlx5_sf_hw_notifier_cleanup(dev);
 	mlx5_events_cleanup(dev);
 }
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c
index a14b1aa5fb5a..bd968f3b3855 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/hw_table.c
@@ -30,9 +30,7 @@ enum mlx5_sf_hwc_index {
 };
 
 struct mlx5_sf_hw_table {
-	struct mlx5_core_dev *dev;
 	struct mutex table_lock; /* Serializes sf deletion and vhca state change handler. */
-	struct notifier_block vhca_nb;
 	struct mlx5_sf_hwc_table hwc[MLX5_SF_HWC_MAX];
 };
 
@@ -71,14 +69,16 @@ mlx5_sf_table_fn_to_hwc(struct mlx5_sf_hw_table *table, u16 fn_id)
 	return NULL;
 }
 
-static int mlx5_sf_hw_table_id_alloc(struct mlx5_sf_hw_table *table, u32 controller,
+static int mlx5_sf_hw_table_id_alloc(struct mlx5_core_dev *dev,
+				     struct mlx5_sf_hw_table *table,
+				     u32 controller,
 				     u32 usr_sfnum)
 {
 	struct mlx5_sf_hwc_table *hwc;
 	int free_idx = -1;
 	int i;
 
-	hwc = mlx5_sf_controller_to_hwc(table->dev, controller);
+	hwc = mlx5_sf_controller_to_hwc(dev, controller);
 	if (!hwc->sfs)
 		return -ENOSPC;
 
@@ -100,11 +100,13 @@ static int mlx5_sf_hw_table_id_alloc(struct mlx5_sf_hw_table *table, u32 control
 	return free_idx;
 }
 
-static void mlx5_sf_hw_table_id_free(struct mlx5_sf_hw_table *table, u32 controller, int id)
+static void mlx5_sf_hw_table_id_free(struct mlx5_core_dev *dev,
+				     struct mlx5_sf_hw_table *table,
+				     u32 controller, int id)
 {
 	struct mlx5_sf_hwc_table *hwc;
 
-	hwc = mlx5_sf_controller_to_hwc(table->dev, controller);
+	hwc = mlx5_sf_controller_to_hwc(dev, controller);
 	hwc->sfs[id].allocated = false;
 	hwc->sfs[id].pending_delete = false;
 }
@@ -120,7 +122,7 @@ int mlx5_sf_hw_table_sf_alloc(struct mlx5_core_dev *dev, u32 controller, u32 usr
 		return -EOPNOTSUPP;
 
 	mutex_lock(&table->table_lock);
-	sw_id = mlx5_sf_hw_table_id_alloc(table, controller, usr_sfnum);
+	sw_id = mlx5_sf_hw_table_id_alloc(dev, table, controller, usr_sfnum);
 	if (sw_id < 0) {
 		err = sw_id;
 		goto exist_err;
@@ -151,7 +153,7 @@ int mlx5_sf_hw_table_sf_alloc(struct mlx5_core_dev *dev, u32 controller, u32 usr
 vhca_err:
 	mlx5_cmd_dealloc_sf(dev, hw_fn_id);
 err:
-	mlx5_sf_hw_table_id_free(table, controller, sw_id);
+	mlx5_sf_hw_table_id_free(dev, table, controller, sw_id);
 exist_err:
 	mutex_unlock(&table->table_lock);
 	return err;
@@ -165,7 +167,7 @@ void mlx5_sf_hw_table_sf_free(struct mlx5_core_dev *dev, u32 controller, u16 id)
 	mutex_lock(&table->table_lock);
 	hw_fn_id = mlx5_sf_sw_to_hw_id(dev, controller, id);
 	mlx5_cmd_dealloc_sf(dev, hw_fn_id);
-	mlx5_sf_hw_table_id_free(table, controller, id);
+	mlx5_sf_hw_table_id_free(dev, table, controller, id);
 	mutex_unlock(&table->table_lock);
 }
 
@@ -216,10 +218,12 @@ static void mlx5_sf_hw_table_hwc_dealloc_all(struct mlx5_core_dev *dev,
 	}
 }
 
-static void mlx5_sf_hw_table_dealloc_all(struct mlx5_sf_hw_table *table)
+static void mlx5_sf_hw_table_dealloc_all(struct mlx5_core_dev *dev,
+					 struct mlx5_sf_hw_table *table)
 {
-	mlx5_sf_hw_table_hwc_dealloc_all(table->dev, &table->hwc[MLX5_SF_HWC_EXTERNAL]);
-	mlx5_sf_hw_table_hwc_dealloc_all(table->dev, &table->hwc[MLX5_SF_HWC_LOCAL]);
+	mlx5_sf_hw_table_hwc_dealloc_all(dev,
+					 &table->hwc[MLX5_SF_HWC_EXTERNAL]);
+	mlx5_sf_hw_table_hwc_dealloc_all(dev, &table->hwc[MLX5_SF_HWC_LOCAL]);
 }
 
 static int mlx5_sf_hw_table_hwc_init(struct mlx5_sf_hwc_table *hwc, u16 max_fn, u16 base_id)
@@ -301,7 +305,6 @@ int mlx5_sf_hw_table_init(struct mlx5_core_dev *dev)
 	}
 
 	mutex_init(&table->table_lock);
-	table->dev = dev;
 	dev->priv.sf_hw_table = table;
 
 	base_id = mlx5_sf_start_function_id(dev);
@@ -338,19 +341,22 @@ void mlx5_sf_hw_table_cleanup(struct mlx5_core_dev *dev)
 	mlx5_sf_hw_table_hwc_cleanup(&table->hwc[MLX5_SF_HWC_LOCAL]);
 	mutex_destroy(&table->table_lock);
 	kfree(table);
+	dev->priv.sf_hw_table = NULL;
 res_unregister:
 	mlx5_sf_hw_table_res_unregister(dev);
 }
 
 static int mlx5_sf_hw_vhca_event(struct notifier_block *nb, unsigned long opcode, void *data)
 {
-	struct mlx5_sf_hw_table *table = container_of(nb, struct mlx5_sf_hw_table, vhca_nb);
+	struct mlx5_core_dev *dev = container_of(nb, struct mlx5_core_dev,
+						 priv.sf_hw_table_vhca_nb);
+	struct mlx5_sf_hw_table *table = dev->priv.sf_hw_table;
 	const struct mlx5_vhca_state_event *event = data;
 	struct mlx5_sf_hwc_table *hwc;
 	struct mlx5_sf_hw *sf_hw;
 	u16 sw_id;
 
-	if (event->new_vhca_state != MLX5_VHCA_STATE_ALLOCATED)
+	if (!table || event->new_vhca_state != MLX5_VHCA_STATE_ALLOCATED)
 		return 0;
 
 	hwc = mlx5_sf_table_fn_to_hwc(table, event->function_id);
@@ -365,20 +371,28 @@ static int mlx5_sf_hw_vhca_event(struct notifier_block *nb, unsigned long opcode
 	 * Hence recycle the sf hardware id for reuse.
 	 */
 	if (sf_hw->allocated && sf_hw->pending_delete)
-		mlx5_sf_hw_table_hwc_sf_free(table->dev, hwc, sw_id);
+		mlx5_sf_hw_table_hwc_sf_free(dev, hwc, sw_id);
 	mutex_unlock(&table->table_lock);
 	return 0;
 }
 
-int mlx5_sf_hw_table_create(struct mlx5_core_dev *dev)
+int mlx5_sf_hw_notifier_init(struct mlx5_core_dev *dev)
 {
-	struct mlx5_sf_hw_table *table = dev->priv.sf_hw_table;
-
-	if (!table)
+	if (mlx5_core_is_sf(dev))
 		return 0;
 
-	table->vhca_nb.notifier_call = mlx5_sf_hw_vhca_event;
-	return mlx5_vhca_event_notifier_register(dev, &table->vhca_nb);
+	dev->priv.sf_hw_table_vhca_nb.notifier_call = mlx5_sf_hw_vhca_event;
+	return mlx5_vhca_event_notifier_register(dev,
+						 &dev->priv.sf_hw_table_vhca_nb);
+}
+
+void mlx5_sf_hw_notifier_cleanup(struct mlx5_core_dev *dev)
+{
+	if (mlx5_core_is_sf(dev))
+		return;
+
+	mlx5_vhca_event_notifier_unregister(dev,
+					    &dev->priv.sf_hw_table_vhca_nb);
 }
 
 void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev)
@@ -388,10 +402,8 @@ void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev)
 	if (!table)
 		return;
 
-	mlx5_vhca_event_notifier_unregister(dev, &table->vhca_nb);
-
 	/* Dealloc SFs whose firmware event has been missed. */
-	mlx5_sf_hw_table_dealloc_all(table);
+	mlx5_sf_hw_table_dealloc_all(dev, table);
 }
 
 bool mlx5_sf_hw_table_supported(const struct mlx5_core_dev *dev)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h b/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h
index 89559a37997a..3922dacffae8 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h
@@ -12,7 +12,8 @@
 int mlx5_sf_hw_table_init(struct mlx5_core_dev *dev);
 void mlx5_sf_hw_table_cleanup(struct mlx5_core_dev *dev);
 
-int mlx5_sf_hw_table_create(struct mlx5_core_dev *dev);
+int mlx5_sf_hw_notifier_init(struct mlx5_core_dev *dev);
+void mlx5_sf_hw_notifier_cleanup(struct mlx5_core_dev *dev);
 void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev);
 
 int mlx5_sf_table_init(struct mlx5_core_dev *dev);
@@ -44,11 +45,15 @@ static inline void mlx5_sf_hw_table_cleanup(struct mlx5_core_dev *dev)
 {
 }
 
-static inline int mlx5_sf_hw_table_create(struct mlx5_core_dev *dev)
+static inline int mlx5_sf_hw_notifier_init(struct mlx5_core_dev *dev)
 {
 	return 0;
 }
 
+static inline void mlx5_sf_hw_notifier_cleanup(struct mlx5_core_dev *dev)
+{
+}
+
 static inline void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev)
 {
 }
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 57368db0a5b7..a51bea44d57c 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -620,6 +620,7 @@ struct mlx5_priv {
 	struct mlx5_core_dev *parent_mdev;
 #endif
 #ifdef CONFIG_MLX5_SF_MANAGER
+	struct notifier_block sf_hw_table_vhca_nb;
 	struct mlx5_sf_hw_table *sf_hw_table;
 	struct mlx5_sf_table *sf_table;
 #endif
-- 
2.31.1

[PATCH net-next 5/6] net/mlx5: Move the SF table notifiers outside the devlink lock

[Tariq Toukan]

From: Cosmin Ratiu <cratiu@nvidia.com>

Move the SF table notifiers registration/unregistration outside of
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.

This is only done for non-SFs, since SFs do not have a SF table
themselves and thus don't need notifiers.

Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/main.c    |  7 ++
 .../ethernet/mellanox/mlx5/core/sf/devlink.c  | 90 ++++++++++++-------
 .../net/ethernet/mellanox/mlx5/core/sf/sf.h   | 11 +++
 include/linux/mlx5/driver.h                   |  3 +
 4 files changed, 78 insertions(+), 33 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 843ee452239f..0c3613ef39b1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1833,8 +1833,14 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 	if (err)
 		goto err_sf_hw_notifier;
 
+	err = mlx5_sf_notifiers_init(dev);
+	if (err)
+		goto err_sf_notifiers;
+
 	return 0;
 
+err_sf_notifiers:
+	mlx5_sf_hw_notifier_cleanup(dev);
 err_sf_hw_notifier:
 	mlx5_events_cleanup(dev);
 	return err;
@@ -1842,6 +1848,7 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 
 static void mlx5_notifiers_cleanup(struct mlx5_core_dev *dev)
 {
+	mlx5_sf_notifiers_cleanup(dev);
 	mlx5_sf_hw_notifier_cleanup(dev);
 	mlx5_events_cleanup(dev);
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c
index 2ece4983d33f..b82323b8449e 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/devlink.c
@@ -31,9 +31,6 @@ struct mlx5_sf_table {
 	struct mlx5_core_dev *dev; /* To refer from notifier context. */
 	struct xarray function_ids; /* function id based lookup. */
 	struct mutex sf_state_lock; /* Serializes sf state among user cmds & vhca event handler. */
-	struct notifier_block esw_nb;
-	struct notifier_block vhca_nb;
-	struct notifier_block mdev_nb;
 };
 
 static struct mlx5_sf *
@@ -391,11 +388,16 @@ static bool mlx5_sf_state_update_check(const struct mlx5_sf *sf, u8 new_state)
 
 static int mlx5_sf_vhca_event(struct notifier_block *nb, unsigned long opcode, void *data)
 {
-	struct mlx5_sf_table *table = container_of(nb, struct mlx5_sf_table, vhca_nb);
+	struct mlx5_core_dev *dev = container_of(nb, struct mlx5_core_dev,
+						 priv.sf_table_vhca_nb);
+	struct mlx5_sf_table *table = dev->priv.sf_table;
 	const struct mlx5_vhca_state_event *event = data;
 	bool update = false;
 	struct mlx5_sf *sf;
 
+	if (!table)
+		return 0;
+
 	mutex_lock(&table->sf_state_lock);
 	sf = mlx5_sf_lookup_by_function_id(table, event->function_id);
 	if (!sf)
@@ -407,7 +409,7 @@ static int mlx5_sf_vhca_event(struct notifier_block *nb, unsigned long opcode, v
 	update = mlx5_sf_state_update_check(sf, event->new_vhca_state);
 	if (update)
 		sf->hw_state = event->new_vhca_state;
-	trace_mlx5_sf_update_state(table->dev, sf->port_index, sf->controller,
+	trace_mlx5_sf_update_state(dev, sf->port_index, sf->controller,
 				   sf->hw_fn_id, sf->hw_state);
 unlock:
 	mutex_unlock(&table->sf_state_lock);
@@ -425,12 +427,16 @@ static void mlx5_sf_del_all(struct mlx5_sf_table *table)
 
 static int mlx5_sf_esw_event(struct notifier_block *nb, unsigned long event, void *data)
 {
-	struct mlx5_sf_table *table = container_of(nb, struct mlx5_sf_table, esw_nb);
+	struct mlx5_core_dev *dev = container_of(nb, struct mlx5_core_dev,
+						 priv.sf_table_esw_nb);
 	const struct mlx5_esw_event_info *mode = data;
 
+	if (!dev->priv.sf_table)
+		return 0;
+
 	switch (mode->new_mode) {
 	case MLX5_ESWITCH_LEGACY:
-		mlx5_sf_del_all(table);
+		mlx5_sf_del_all(dev->priv.sf_table);
 		break;
 	default:
 		break;
@@ -441,15 +447,16 @@ static int mlx5_sf_esw_event(struct notifier_block *nb, unsigned long event, voi
 
 static int mlx5_sf_mdev_event(struct notifier_block *nb, unsigned long event, void *data)
 {
-	struct mlx5_sf_table *table = container_of(nb, struct mlx5_sf_table, mdev_nb);
+	struct mlx5_core_dev *dev = container_of(nb, struct mlx5_core_dev,
+						 priv.sf_table_mdev_nb);
 	struct mlx5_sf_peer_devlink_event_ctx *event_ctx = data;
+	struct mlx5_sf_table *table = dev->priv.sf_table;
 	int ret = NOTIFY_DONE;
 	struct mlx5_sf *sf;
 
-	if (event != MLX5_DRIVER_EVENT_SF_PEER_DEVLINK)
+	if (!table || event != MLX5_DRIVER_EVENT_SF_PEER_DEVLINK)
 		return NOTIFY_DONE;
 
-
 	mutex_lock(&table->sf_state_lock);
 	sf = mlx5_sf_lookup_by_function_id(table, event_ctx->fn_id);
 	if (!sf)
@@ -464,10 +471,40 @@ static int mlx5_sf_mdev_event(struct notifier_block *nb, unsigned long event, vo
 	return ret;
 }
 
+int mlx5_sf_notifiers_init(struct mlx5_core_dev *dev)
+{
+	int err;
+
+	if (mlx5_core_is_sf(dev))
+		return 0;
+
+	dev->priv.sf_table_esw_nb.notifier_call = mlx5_sf_esw_event;
+	err = mlx5_esw_event_notifier_register(dev, &dev->priv.sf_table_esw_nb);
+	if (err)
+		return err;
+
+	dev->priv.sf_table_vhca_nb.notifier_call = mlx5_sf_vhca_event;
+	err = mlx5_vhca_event_notifier_register(dev,
+						&dev->priv.sf_table_vhca_nb);
+	if (err)
+		goto vhca_err;
+
+	dev->priv.sf_table_mdev_nb.notifier_call = mlx5_sf_mdev_event;
+	err = mlx5_blocking_notifier_register(dev, &dev->priv.sf_table_mdev_nb);
+	if (err)
+		goto mdev_err;
+
+	return 0;
+mdev_err:
+	mlx5_vhca_event_notifier_unregister(dev, &dev->priv.sf_table_vhca_nb);
+vhca_err:
+	mlx5_esw_event_notifier_unregister(dev, &dev->priv.sf_table_esw_nb);
+	return err;
+}
+
 int mlx5_sf_table_init(struct mlx5_core_dev *dev)
 {
 	struct mlx5_sf_table *table;
-	int err;
 
 	if (!mlx5_sf_table_supported(dev) || !mlx5_vhca_event_supported(dev))
 		return 0;
@@ -480,28 +517,18 @@ int mlx5_sf_table_init(struct mlx5_core_dev *dev)
 	table->dev = dev;
 	xa_init(&table->function_ids);
 	dev->priv.sf_table = table;
-	table->esw_nb.notifier_call = mlx5_sf_esw_event;
-	err = mlx5_esw_event_notifier_register(dev, &table->esw_nb);
-	if (err)
-		goto reg_err;
-
-	table->vhca_nb.notifier_call = mlx5_sf_vhca_event;
-	err = mlx5_vhca_event_notifier_register(table->dev, &table->vhca_nb);
-	if (err)
-		goto vhca_err;
-
-	table->mdev_nb.notifier_call = mlx5_sf_mdev_event;
-	mlx5_blocking_notifier_register(dev, &table->mdev_nb);
 
 	return 0;
+}
 
-vhca_err:
-	mlx5_esw_event_notifier_unregister(dev, &table->esw_nb);
-reg_err:
-	mutex_destroy(&table->sf_state_lock);
-	kfree(table);
-	dev->priv.sf_table = NULL;
-	return err;
+void mlx5_sf_notifiers_cleanup(struct mlx5_core_dev *dev)
+{
+	if (mlx5_core_is_sf(dev))
+		return;
+
+	mlx5_blocking_notifier_unregister(dev, &dev->priv.sf_table_mdev_nb);
+	mlx5_vhca_event_notifier_unregister(dev, &dev->priv.sf_table_vhca_nb);
+	mlx5_esw_event_notifier_unregister(dev, &dev->priv.sf_table_esw_nb);
 }
 
 void mlx5_sf_table_cleanup(struct mlx5_core_dev *dev)
@@ -511,9 +538,6 @@ void mlx5_sf_table_cleanup(struct mlx5_core_dev *dev)
 	if (!table)
 		return;
 
-	mlx5_blocking_notifier_unregister(dev, &table->mdev_nb);
-	mlx5_vhca_event_notifier_unregister(table->dev, &table->vhca_nb);
-	mlx5_esw_event_notifier_unregister(dev, &table->esw_nb);
 	mutex_destroy(&table->sf_state_lock);
 	WARN_ON(!xa_empty(&table->function_ids));
 	kfree(table);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h b/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h
index 3922dacffae8..d8a934a0e968 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/sf.h
@@ -16,7 +16,9 @@ int mlx5_sf_hw_notifier_init(struct mlx5_core_dev *dev);
 void mlx5_sf_hw_notifier_cleanup(struct mlx5_core_dev *dev);
 void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev);
 
+int mlx5_sf_notifiers_init(struct mlx5_core_dev *dev);
 int mlx5_sf_table_init(struct mlx5_core_dev *dev);
+void mlx5_sf_notifiers_cleanup(struct mlx5_core_dev *dev);
 void mlx5_sf_table_cleanup(struct mlx5_core_dev *dev);
 bool mlx5_sf_table_empty(const struct mlx5_core_dev *dev);
 
@@ -58,11 +60,20 @@ static inline void mlx5_sf_hw_table_destroy(struct mlx5_core_dev *dev)
 {
 }
 
+static inline int mlx5_sf_notifiers_init(struct mlx5_core_dev *dev)
+{
+	return 0;
+}
+
 static inline int mlx5_sf_table_init(struct mlx5_core_dev *dev)
 {
 	return 0;
 }
 
+static inline void mlx5_sf_notifiers_cleanup(struct mlx5_core_dev *dev)
+{
+}
+
 static inline void mlx5_sf_table_cleanup(struct mlx5_core_dev *dev)
 {
 }
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index a51bea44d57c..7dbef112deaf 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -622,6 +622,9 @@ struct mlx5_priv {
 #ifdef CONFIG_MLX5_SF_MANAGER
 	struct notifier_block sf_hw_table_vhca_nb;
 	struct mlx5_sf_hw_table *sf_hw_table;
+	struct notifier_block sf_table_esw_nb;
+	struct notifier_block sf_table_vhca_nb;
+	struct notifier_block sf_table_mdev_nb;
 	struct mlx5_sf_table *sf_table;
 #endif
 	struct blocking_notifier_head lag_nh;
-- 
2.31.1

[PATCH net-next 6/6] net/mlx5: Move SF dev table notifier registration outside the PF devlink lock

[Tariq Toukan]

From: Cosmin Ratiu <cratiu@nvidia.com>

This completes the previous patches by moving notifier registration for
SF dev tables outside the devlink locked critical section in
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.

This is only done for non-SFs, since SFs do not have a SF HW table
themselves.

After this patch, notifiers can grab the PF devlink lock (soon to be
necessary) without creating a locking cycle.

Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/main.c    |  7 +++
 .../ethernet/mellanox/mlx5/core/sf/dev/dev.c  | 47 ++++++++++++-------
 .../ethernet/mellanox/mlx5/core/sf/dev/dev.h  | 11 +++++
 include/linux/mlx5/driver.h                   |  1 +
 4 files changed, 49 insertions(+), 17 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 0c3613ef39b1..024339ce41f1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1837,8 +1837,14 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 	if (err)
 		goto err_sf_notifiers;
 
+	err = mlx5_sf_dev_notifier_init(dev);
+	if (err)
+		goto err_sf_dev_notifier;
+
 	return 0;
 
+err_sf_dev_notifier:
+	mlx5_sf_notifiers_cleanup(dev);
 err_sf_notifiers:
 	mlx5_sf_hw_notifier_cleanup(dev);
 err_sf_hw_notifier:
@@ -1848,6 +1854,7 @@ static int mlx5_notifiers_init(struct mlx5_core_dev *dev)
 
 static void mlx5_notifiers_cleanup(struct mlx5_core_dev *dev)
 {
+	mlx5_sf_dev_notifier_cleanup(dev);
 	mlx5_sf_notifiers_cleanup(dev);
 	mlx5_sf_hw_notifier_cleanup(dev);
 	mlx5_events_cleanup(dev);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
index a68a8ee24dce..f310bde3d11f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
@@ -16,7 +16,6 @@ struct mlx5_sf_dev_table {
 	struct xarray devices;
 	phys_addr_t base_address;
 	u64 sf_bar_length;
-	struct notifier_block nb;
 	struct workqueue_struct *active_wq;
 	struct work_struct work;
 	u8 stop_active_wq:1;
@@ -156,18 +155,23 @@ static void mlx5_sf_dev_del(struct mlx5_core_dev *dev, struct mlx5_sf_dev *sf_de
 static int
 mlx5_sf_dev_state_change_handler(struct notifier_block *nb, unsigned long event_code, void *data)
 {
-	struct mlx5_sf_dev_table *table = container_of(nb, struct mlx5_sf_dev_table, nb);
+	struct mlx5_core_dev *dev = container_of(nb, struct mlx5_core_dev,
+						 priv.sf_dev_nb);
+	struct mlx5_sf_dev_table *table = dev->priv.sf_dev_table;
 	const struct mlx5_vhca_state_event *event = data;
 	struct mlx5_sf_dev *sf_dev;
 	u16 max_functions;
 	u16 sf_index;
 	u16 base_id;
 
-	max_functions = mlx5_sf_max_functions(table->dev);
+	if (!table)
+		return 0;
+
+	max_functions = mlx5_sf_max_functions(dev);
 	if (!max_functions)
 		return 0;
 
-	base_id = mlx5_sf_start_function_id(table->dev);
+	base_id = mlx5_sf_start_function_id(dev);
 	if (event->function_id < base_id || event->function_id >= (base_id + max_functions))
 		return 0;
 
@@ -177,19 +181,19 @@ mlx5_sf_dev_state_change_handler(struct notifier_block *nb, unsigned long event_
 	case MLX5_VHCA_STATE_INVALID:
 	case MLX5_VHCA_STATE_ALLOCATED:
 		if (sf_dev)
-			mlx5_sf_dev_del(table->dev, sf_dev, sf_index);
+			mlx5_sf_dev_del(dev, sf_dev, sf_index);
 		break;
 	case MLX5_VHCA_STATE_TEARDOWN_REQUEST:
 		if (sf_dev)
-			mlx5_sf_dev_del(table->dev, sf_dev, sf_index);
+			mlx5_sf_dev_del(dev, sf_dev, sf_index);
 		else
-			mlx5_core_err(table->dev,
+			mlx5_core_err(dev,
 				      "SF DEV: teardown state for invalid dev index=%d sfnum=0x%x\n",
 				      sf_index, event->sw_function_id);
 		break;
 	case MLX5_VHCA_STATE_ACTIVE:
 		if (!sf_dev)
-			mlx5_sf_dev_add(table->dev, sf_index, event->function_id,
+			mlx5_sf_dev_add(dev, sf_index, event->function_id,
 					event->sw_function_id);
 		break;
 	default:
@@ -315,6 +319,15 @@ static void mlx5_sf_dev_destroy_active_works(struct mlx5_sf_dev_table *table)
 	}
 }
 
+int mlx5_sf_dev_notifier_init(struct mlx5_core_dev *dev)
+{
+	if (mlx5_core_is_sf(dev))
+		return 0;
+
+	dev->priv.sf_dev_nb.notifier_call = mlx5_sf_dev_state_change_handler;
+	return mlx5_vhca_event_notifier_register(dev, &dev->priv.sf_dev_nb);
+}
+
 void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 {
 	struct mlx5_sf_dev_table *table;
@@ -329,17 +342,12 @@ void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 		goto table_err;
 	}
 
-	table->nb.notifier_call = mlx5_sf_dev_state_change_handler;
 	table->dev = dev;
 	table->sf_bar_length = 1 << (MLX5_CAP_GEN(dev, log_min_sf_size) + 12);
 	table->base_address = pci_resource_start(dev->pdev, 2);
 	xa_init(&table->devices);
 	dev->priv.sf_dev_table = table;
 
-	err = mlx5_vhca_event_notifier_register(dev, &table->nb);
-	if (err)
-		goto vhca_err;
-
 	err = mlx5_sf_dev_create_active_works(table);
 	if (err)
 		goto add_active_err;
@@ -351,10 +359,8 @@ void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 
 arm_err:
 	mlx5_sf_dev_destroy_active_works(table);
-add_active_err:
-	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
 	mlx5_vhca_event_work_queues_flush(dev);
-vhca_err:
+add_active_err:
 	kfree(table);
 	dev->priv.sf_dev_table = NULL;
 table_err:
@@ -372,6 +378,14 @@ static void mlx5_sf_dev_destroy_all(struct mlx5_sf_dev_table *table)
 	}
 }
 
+void mlx5_sf_dev_notifier_cleanup(struct mlx5_core_dev *dev)
+{
+	if (mlx5_core_is_sf(dev))
+		return;
+
+	mlx5_vhca_event_notifier_unregister(dev, &dev->priv.sf_dev_nb);
+}
+
 void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev)
 {
 	struct mlx5_sf_dev_table *table = dev->priv.sf_dev_table;
@@ -380,7 +394,6 @@ void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev)
 		return;
 
 	mlx5_sf_dev_destroy_active_works(table);
-	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
 
 	/* Now that event handler is not running, it is safe to destroy
 	 * the sf device without race.
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.h b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.h
index b99131e95e37..3ab0449c770c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.h
@@ -25,7 +25,9 @@ struct mlx5_sf_peer_devlink_event_ctx {
 	int err;
 };
 
+int mlx5_sf_dev_notifier_init(struct mlx5_core_dev *dev);
 void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev);
+void mlx5_sf_dev_notifier_cleanup(struct mlx5_core_dev *dev);
 void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev);
 
 int mlx5_sf_driver_register(void);
@@ -35,10 +37,19 @@ bool mlx5_sf_dev_allocated(const struct mlx5_core_dev *dev);
 
 #else
 
+static inline int mlx5_sf_dev_notifier_init(struct mlx5_core_dev *dev)
+{
+	return 0;
+}
+
 static inline void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 {
 }
 
+static inline void mlx5_sf_dev_notifier_cleanup(struct mlx5_core_dev *dev)
+{
+}
+
 static inline void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev)
 {
 }
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 7dbef112deaf..6ff52bde1f40 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -616,6 +616,7 @@ struct mlx5_priv {
 #ifdef CONFIG_MLX5_SF
 	struct mlx5_nb vhca_state_nb;
 	struct blocking_notifier_head vhca_state_n_head;
+	struct notifier_block sf_dev_nb;
 	struct mlx5_sf_dev_table *sf_dev_table;
 	struct mlx5_core_dev *parent_mdev;
 #endif
-- 
2.31.1

Re: [PATCH net-next 0/6] net/mlx5: Move notifiers outside the devlink lock

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Sun, 16 Nov 2025 22:45:34 +0200 you wrote:
> Hi,
> 
> This series by Cosmin moves blocking notifier registration in the mlx5
> driver outside the devlink lock during probe.
> 
> This is mostly a no-op refactoring that consists of multiple pieces.
> It is necessary because upcoming code will introduce a potential locking
> cycle between the devlink lock and the blocking notifier head mutexes,
> so these notifiers must move out of the devlink-locked critical section.
> 
> [...]

Here is the summary with links:
  - [net-next,1/6] net/mlx5: Initialize events outside devlink lock
    https://git.kernel.org/netdev/net-next/c/b6b03097f982
  - [net-next,2/6] net/mlx5: Move the esw mode notifier chain outside the devlink lock
    https://git.kernel.org/netdev/net-next/c/3fee828789b1
  - [net-next,3/6] net/mlx5: Move the vhca event notifier outside of the devlink lock
    https://git.kernel.org/netdev/net-next/c/d3a356db853b
  - [net-next,4/6] net/mlx5: Move the SF HW table notifier outside the devlink lock
    https://git.kernel.org/netdev/net-next/c/e63c9c5f0a48
  - [net-next,5/6] net/mlx5: Move the SF table notifiers outside the devlink lock
    https://git.kernel.org/netdev/net-next/c/d4a0acbd94c2
  - [net-next,6/6] net/mlx5: Move SF dev table notifier registration outside the PF devlink lock
    https://git.kernel.org/netdev/net-next/c/64ad6470c882

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html