Commit 93f82444 authored by Leon Romanovsky's avatar Leon Romanovsky
Browse files

RDMA/mlx5: Convert mlx5_ib to use auxiliary bus 



The conversion to auxiliary bus solves long standing issue with
existing mlx5_ib<->mlx5_core coupling. It required to have both
modules in initramfs if one of them needed for the boot.

Signed-off-by: default avatarLeon Romanovsky <leonro@nvidia.com>
parent 912cebf4

drivers/infiniband/hw/mlx5/ib_rep.c

+58 −18
Original line number Diff line number Diff line
@@ -33,6 +33,7 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep) 
	const struct mlx5_ib_profile *profile;

	struct mlx5_ib_dev *ibdev;

	int vport_index;

	int ret;



	if (rep->vport == MLX5_VPORT_UPLINK)

		profile = &raw_eth_profile;
@@ -46,8 +47,8 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep) 
	ibdev->port = kcalloc(num_ports, sizeof(*ibdev->port),

			      GFP_KERNEL);

	if (!ibdev->port) {

		ib_dealloc_device(&ibdev->ib_dev);

		return -ENOMEM;

		ret = -ENOMEM;

		goto fail_port;

	}



	ibdev->is_rep = true;
@@ -58,12 +59,19 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep) 
	ibdev->mdev = dev;

	ibdev->num_ports = num_ports;



	if (!__mlx5_ib_add(ibdev, profile))

		return -EINVAL;

	ret = __mlx5_ib_add(ibdev, profile);

	if (ret)

		goto fail_add;



	rep->rep_data[REP_IB].priv = ibdev;



	return 0;



fail_add:

	kfree(ibdev->port);

fail_port:

	ib_dealloc_device(&ibdev->ib_dev);

	return ret;

}



static void
@@ -94,20 +102,6 @@ static const struct mlx5_eswitch_rep_ops rep_ops = { 
	.get_proto_dev = mlx5_ib_vport_get_proto_dev,

};



void mlx5_ib_register_vport_reps(struct mlx5_core_dev *mdev)

{

	struct mlx5_eswitch *esw = mdev->priv.eswitch;



	mlx5_eswitch_register_vport_reps(esw, &rep_ops, REP_IB);

}



void mlx5_ib_unregister_vport_reps(struct mlx5_core_dev *mdev)

{

	struct mlx5_eswitch *esw = mdev->priv.eswitch;



	mlx5_eswitch_unregister_vport_reps(esw, REP_IB);

}



u8 mlx5_ib_eswitch_mode(struct mlx5_eswitch *esw)

{

	return mlx5_eswitch_mode(esw);
@@ -154,3 +148,49 @@ struct mlx5_flow_handle *create_flow_rule_vport_sq(struct mlx5_ib_dev *dev, 
	return mlx5_eswitch_add_send_to_vport_rule(esw, rep->vport,

						   sq->base.mqp.qpn);

}



static int mlx5r_rep_probe(struct auxiliary_device *adev,

			   const struct auxiliary_device_id *id)

{

	struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);

	struct mlx5_core_dev *mdev = idev->mdev;

	struct mlx5_eswitch *esw;



	esw = mdev->priv.eswitch;

	mlx5_eswitch_register_vport_reps(esw, &rep_ops, REP_IB);

	return 0;

}



static void mlx5r_rep_remove(struct auxiliary_device *adev)

{

	struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);

	struct mlx5_core_dev *mdev = idev->mdev;

	struct mlx5_eswitch *esw;



	esw = mdev->priv.eswitch;

	mlx5_eswitch_unregister_vport_reps(esw, REP_IB);

}



static const struct auxiliary_device_id mlx5r_rep_id_table[] = {

	{ .name = MLX5_ADEV_NAME ".rdma-rep", },

	{},

};



MODULE_DEVICE_TABLE(auxiliary, mlx5r_rep_id_table);



static struct auxiliary_driver mlx5r_rep_driver = {

	.name = "rep",

	.probe = mlx5r_rep_probe,

	.remove = mlx5r_rep_remove,

	.id_table = mlx5r_rep_id_table,

};



int mlx5r_rep_init(void)

{

	return auxiliary_driver_register(&mlx5r_rep_driver);

}



void mlx5r_rep_cleanup(void)

{

	auxiliary_driver_unregister(&mlx5r_rep_driver);

}

drivers/infiniband/hw/mlx5/ib_rep.h

+4 −4
Original line number Diff line number Diff line
@@ -18,8 +18,8 @@ struct mlx5_ib_dev *mlx5_ib_get_rep_ibdev(struct mlx5_eswitch *esw, 
struct mlx5_ib_dev *mlx5_ib_get_uplink_ibdev(struct mlx5_eswitch *esw);

struct mlx5_eswitch_rep *mlx5_ib_vport_rep(struct mlx5_eswitch *esw,

					   u16 vport_num);

void mlx5_ib_register_vport_reps(struct mlx5_core_dev *mdev);

void mlx5_ib_unregister_vport_reps(struct mlx5_core_dev *mdev);

int mlx5r_rep_init(void);

void mlx5r_rep_cleanup(void);

struct mlx5_flow_handle *create_flow_rule_vport_sq(struct mlx5_ib_dev *dev,

						   struct mlx5_ib_sq *sq,

						   u16 port);
@@ -51,8 +51,8 @@ struct mlx5_eswitch_rep *mlx5_ib_vport_rep(struct mlx5_eswitch *esw, 
	return NULL;

}



static inline void mlx5_ib_register_vport_reps(struct mlx5_core_dev *mdev) {}

static inline void mlx5_ib_unregister_vport_reps(struct mlx5_core_dev *mdev) {}

static inline int mlx5r_rep_init(void) { return 0; }

static inline void mlx5r_rep_cleanup(void) {}

static inline

struct mlx5_flow_handle *create_flow_rule_vport_sq(struct mlx5_ib_dev *dev,

						   struct mlx5_ib_sq *sq,

drivers/infiniband/hw/mlx5/main.c

+97 −56
Original line number Diff line number Diff line
@@ -4593,7 +4593,7 @@ void __mlx5_ib_remove(struct mlx5_ib_dev *dev, 
	ib_dealloc_device(&dev->ib_dev);

}



void *__mlx5_ib_add(struct mlx5_ib_dev *dev,

int __mlx5_ib_add(struct mlx5_ib_dev *dev,

		  const struct mlx5_ib_profile *profile)

{

	int err;
@@ -4610,13 +4610,16 @@ void *__mlx5_ib_add(struct mlx5_ib_dev *dev, 
	}



	dev->ib_active = true;



	return dev;

	return 0;



err_out:

	__mlx5_ib_remove(dev, profile, i);



	return NULL;

	/* Clean up stages which were initialized */

	while (i) {

		i--;

		if (profile->stage[i].cleanup)

			profile->stage[i].cleanup(dev);

	}

	return -ENOMEM;

}



static const struct mlx5_ib_profile pf_profile = {
@@ -4739,8 +4742,11 @@ const struct mlx5_ib_profile raw_eth_profile = { 
		     NULL),

};



static void *mlx5_ib_add_slave_port(struct mlx5_core_dev *mdev)

static int mlx5r_mp_probe(struct auxiliary_device *adev,

			  const struct auxiliary_device_id *id)

{

	struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);

	struct mlx5_core_dev *mdev = idev->mdev;

	struct mlx5_ib_multiport_info *mpi;

	struct mlx5_ib_dev *dev;

	bool bound = false;
@@ -4748,15 +4754,14 @@ static void *mlx5_ib_add_slave_port(struct mlx5_core_dev *mdev) 


	mpi = kzalloc(sizeof(*mpi), GFP_KERNEL);

	if (!mpi)

		return NULL;

		return -ENOMEM;



	mpi->mdev = mdev;



	err = mlx5_query_nic_vport_system_image_guid(mdev,

						     &mpi->sys_image_guid);

	if (err) {

		kfree(mpi);

		return NULL;

		return err;

	}



	mutex_lock(&mlx5_ib_multiport_mutex);
@@ -4777,40 +4782,46 @@ static void *mlx5_ib_add_slave_port(struct mlx5_core_dev *mdev) 
	}

	mutex_unlock(&mlx5_ib_multiport_mutex);



	return mpi;

	dev_set_drvdata(&adev->dev, mpi);

	return 0;

}



static void *mlx5_ib_add(struct mlx5_core_dev *mdev)

static void mlx5r_mp_remove(struct auxiliary_device *adev)

{

	struct mlx5_ib_multiport_info *mpi;



	mpi = dev_get_drvdata(&adev->dev);

	mutex_lock(&mlx5_ib_multiport_mutex);

	if (mpi->ibdev)

		mlx5_ib_unbind_slave_port(mpi->ibdev, mpi);

	list_del(&mpi->list);

	mutex_unlock(&mlx5_ib_multiport_mutex);

	kfree(mpi);

}



static int mlx5r_probe(struct auxiliary_device *adev,

		       const struct auxiliary_device_id *id)

{

	struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);

	struct mlx5_core_dev *mdev = idev->mdev;

	const struct mlx5_ib_profile *profile;

	int port_type_cap, num_ports, ret;

	enum rdma_link_layer ll;

	struct mlx5_ib_dev *dev;

	int port_type_cap;

	int num_ports;



	if (MLX5_ESWITCH_MANAGER(mdev) &&

	    mlx5_ib_eswitch_mode(mdev->priv.eswitch) == MLX5_ESWITCH_OFFLOADS) {

		if (!mlx5_core_mp_enabled(mdev))

			mlx5_ib_register_vport_reps(mdev);

		return mdev;

	}



	port_type_cap = MLX5_CAP_GEN(mdev, port_type);

	ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);



	if (mlx5_core_is_mp_slave(mdev) && ll == IB_LINK_LAYER_ETHERNET)

		return mlx5_ib_add_slave_port(mdev);



	num_ports = max(MLX5_CAP_GEN(mdev, num_ports),

			MLX5_CAP_GEN(mdev, num_vhca_ports));

	dev = ib_alloc_device(mlx5_ib_dev, ib_dev);

	if (!dev)

		return NULL;

		return -ENOMEM;

	dev->port = kcalloc(num_ports, sizeof(*dev->port),

			     GFP_KERNEL);

	if (!dev->port) {

		ib_dealloc_device(&dev->ib_dev);

		return NULL;

		return -ENOMEM;

	}



	dev->mdev = mdev;
@@ -4821,38 +4832,50 @@ static void *mlx5_ib_add(struct mlx5_core_dev *mdev) 
	else

		profile = &pf_profile;



	return __mlx5_ib_add(dev, profile);

	ret = __mlx5_ib_add(dev, profile);

	if (ret) {

		kfree(dev->port);

		ib_dealloc_device(&dev->ib_dev);

		return ret;

	}



	dev_set_drvdata(&adev->dev, dev);

	return 0;

}



static void mlx5_ib_remove(struct mlx5_core_dev *mdev, void *context)

static void mlx5r_remove(struct auxiliary_device *adev)

{

	struct mlx5_ib_multiport_info *mpi;

	struct mlx5_ib_dev *dev;



	if (MLX5_ESWITCH_MANAGER(mdev) && context == mdev) {

		mlx5_ib_unregister_vport_reps(mdev);

		return;

	dev = dev_get_drvdata(&adev->dev);

	__mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);

}



	if (mlx5_core_is_mp_slave(mdev)) {

		mpi = context;

		mutex_lock(&mlx5_ib_multiport_mutex);

		if (mpi->ibdev)

			mlx5_ib_unbind_slave_port(mpi->ibdev, mpi);

		list_del(&mpi->list);

		mutex_unlock(&mlx5_ib_multiport_mutex);

		kfree(mpi);

		return;

	}

static const struct auxiliary_device_id mlx5r_mp_id_table[] = {

	{ .name = MLX5_ADEV_NAME ".multiport", },

	{},

};



	dev = context;

	__mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);

}

static const struct auxiliary_device_id mlx5r_id_table[] = {

	{ .name = MLX5_ADEV_NAME ".rdma", },

	{},

};



static struct mlx5_interface mlx5_ib_interface = {

	.add            = mlx5_ib_add,

	.remove         = mlx5_ib_remove,

	.protocol	= MLX5_INTERFACE_PROTOCOL_IB,

MODULE_DEVICE_TABLE(auxiliary, mlx5r_mp_id_table);

MODULE_DEVICE_TABLE(auxiliary, mlx5r_id_table);



static struct auxiliary_driver mlx5r_mp_driver = {

	.name = "multiport",

	.probe = mlx5r_mp_probe,

	.remove = mlx5r_mp_remove,

	.id_table = mlx5r_mp_id_table,

};



static struct auxiliary_driver mlx5r_driver = {

	.name = "rdma",

	.probe = mlx5r_probe,

	.remove = mlx5r_remove,

	.id_table = mlx5r_id_table,

};



unsigned long mlx5_ib_get_xlt_emergency_page(void)
@@ -4868,7 +4891,7 @@ void mlx5_ib_put_xlt_emergency_page(void) 


static int __init mlx5_ib_init(void)

{

	int err;

	int ret;



	xlt_emergency_page = __get_free_page(GFP_KERNEL);

	if (!xlt_emergency_page)
@@ -4883,15 +4906,33 @@ static int __init mlx5_ib_init(void) 
	}



	mlx5_ib_odp_init();

	ret = mlx5r_rep_init();

	if (ret)

		goto rep_err;

	ret = auxiliary_driver_register(&mlx5r_mp_driver);

	if (ret)

		goto mp_err;

	ret = auxiliary_driver_register(&mlx5r_driver);

	if (ret)

		goto drv_err;

	return 0;



	err = mlx5_register_interface(&mlx5_ib_interface);



	return err;

drv_err:

	auxiliary_driver_unregister(&mlx5r_mp_driver);

mp_err:

	mlx5r_rep_cleanup();

rep_err:

	destroy_workqueue(mlx5_ib_event_wq);

	free_page((unsigned long)xlt_emergency_page);

	return ret;

}



static void __exit mlx5_ib_cleanup(void)

{

	mlx5_unregister_interface(&mlx5_ib_interface);

	auxiliary_driver_unregister(&mlx5r_driver);

	auxiliary_driver_unregister(&mlx5r_mp_driver);

	mlx5r_rep_cleanup();



	destroy_workqueue(mlx5_ib_event_wq);

	mutex_destroy(&xlt_emergency_page_mutex);

	free_page(xlt_emergency_page);

drivers/infiniband/hw/mlx5/mlx5_ib.h

+2 −2
Original line number Diff line number Diff line
@@ -1317,7 +1317,7 @@ extern const struct mmu_interval_notifier_ops mlx5_mn_ops; 
void __mlx5_ib_remove(struct mlx5_ib_dev *dev,

		      const struct mlx5_ib_profile *profile,

		      int stage);

void *__mlx5_ib_add(struct mlx5_ib_dev *dev,

int __mlx5_ib_add(struct mlx5_ib_dev *dev,

		  const struct mlx5_ib_profile *profile);



int mlx5_ib_get_vf_config(struct ib_device *device, int vf,

drivers/net/ethernet/mellanox/mlx5/core/dev.c

+66 −0
Original line number Diff line number Diff line
@@ -144,16 +144,82 @@ static bool is_vnet_supported(struct mlx5_core_dev *dev) 
	return true;

}



static bool is_ib_rep_supported(struct mlx5_core_dev *dev)

{

	if (!IS_ENABLED(CONFIG_MLX5_INFINIBAND))

		return false;



	if (dev->priv.flags & MLX5_PRIV_FLAGS_DISABLE_IB_ADEV)

		return false;



	if (!is_eth_rep_supported(dev))

		return false;



	if (!MLX5_ESWITCH_MANAGER(dev))

		return false;



	if (mlx5_eswitch_mode(dev->priv.eswitch) != MLX5_ESWITCH_OFFLOADS)

		return false;



	if (mlx5_core_mp_enabled(dev))

		return false;



	return true;

}



static bool is_mp_supported(struct mlx5_core_dev *dev)

{

	if (!IS_ENABLED(CONFIG_MLX5_INFINIBAND))

		return false;



	if (dev->priv.flags & MLX5_PRIV_FLAGS_DISABLE_IB_ADEV)

		return false;



	if (is_ib_rep_supported(dev))

		return false;



	if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)

		return false;



	if (!mlx5_core_is_mp_slave(dev))

		return false;



	return true;

}



static bool is_ib_supported(struct mlx5_core_dev *dev)

{

	if (!IS_ENABLED(CONFIG_MLX5_INFINIBAND))

		return false;



	if (dev->priv.flags & MLX5_PRIV_FLAGS_DISABLE_IB_ADEV)

		return false;



	if (is_ib_rep_supported(dev))

		return false;



	if (is_mp_supported(dev))

		return false;



	return true;

}



static const struct mlx5_adev_device {

	const char *suffix;

	bool (*is_supported)(struct mlx5_core_dev *dev);

} mlx5_adev_devices[] = {

	[MLX5_INTERFACE_PROTOCOL_VDPA] = { .suffix = "vnet",

					   .is_supported = &is_vnet_supported },

	[MLX5_INTERFACE_PROTOCOL_IB] = { .suffix = "rdma",

					 .is_supported = &is_ib_supported },

	[MLX5_INTERFACE_PROTOCOL_ETH] = { .suffix = "eth",

					  .is_supported = &is_eth_supported },

	[MLX5_INTERFACE_PROTOCOL_ETH_REP] = { .suffix = "eth-rep",

					   .is_supported = &is_eth_rep_supported },

	[MLX5_INTERFACE_PROTOCOL_IB_REP] = { .suffix = "rdma-rep",

					   .is_supported = &is_ib_rep_supported },

	[MLX5_INTERFACE_PROTOCOL_MPIB] = { .suffix = "multiport",

					   .is_supported = &is_mp_supported },

};



int mlx5_adev_idx_alloc(void)