Skip to contentSkip to conversation

Add ICM header-modify-pattern RDMA API

PATCH In review Leon Romanovsky 2 participants 7 messages Jun 7, 2022 – Jun 14, 2022

Subsystems (5)
Networking Network drivers InfiniBand Mellanox mlx5 core +1 more subsystems
lore
Share & tools
lore

Leon Romanovsky

author
#0 lore raw

cover Add ICM header-modify-pattern RDMA API

From: Leon Romanovsky <[email protected]>

SW steering manipulates packet's header using "modifying header" actions. Many of these actions do the same operation, but use different data each time. Currently we create and keep every one of these actions, which use expensive and limited resources.

Now we introduce a new mechanism - pattern and argument, which splits a modifying action into two parts:
1. action pattern: contains the operations to be applied on packet's header, mainly set/add/copy of fields in the packet
2. action data/argument: contains the data to be used by each operation in the pattern.

This way we reuse same patterns with different arguments to create new modifying actions, and since many actions share the same operations, we end up creating a small number of patterns that we keep in a dedicated cache.

These modify header patterns are implemented as new type of ICM memory, so the following kernel patch series add the support for this new ICM type.

Thanks

Yevgeny Kliteynik (3):
net/mlx5: Introduce header-modify-pattern ICM properties net/mlx5: Manage ICM of type modify-header pattern RDMA/mlx5: Support handling of modify-header pattern ICM area

 drivers/infiniband/hw/mlx5/dm.c               | 53 ++++++++++++-------
 drivers/infiniband/hw/mlx5/mr.c               |  1 +
 .../net/ethernet/mellanox/mlx5/core/lib/dm.c  | 42 +++++++++++++++
 include/linux/mlx5/driver.h                   |  1 +
 include/linux/mlx5/mlx5_ifc.h                 |  7 ++-
 include/uapi/rdma/mlx5_user_ioctl_verbs.h     |  1 +
 6 files changed, 85 insertions(+), 20 deletions(-)
Signature · 1 line
2.36.1
author
#1 lore raw

PATCH 1/3 net/mlx5: Introduce header-modify-pattern ICM properties

From: Yevgeny Kliteynik <[email protected]>

Added new fields for device memory capabilities, in order to support creation of ICM memory for modify header patterns.

Signed-off-by Erez Shitrit <[email protected]>
Signed-off-by Yevgeny Kliteynik <[email protected]>
Signed-off-by Leon Romanovsky <[email protected]>
---
 include/linux/mlx5/mlx5_ifc.h | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
include/linux/mlx5/mlx5_ifc.h +5 -2
index 006836ae7e43..4e41b3164dc8 100644
@@ -1086,11 +1086,14 @@ struct mlx5_ifc_device_mem_cap_bits {
 	u8         log_sw_icm_alloc_granularity[0x6];
 	u8         log_steering_sw_icm_size[0x8];
 
-	u8         reserved_at_120[0x20];
+	u8         reserved_at_120[0x18];
+	u8         log_header_modify_pattern_sw_icm_size[0x8];
 
 	u8         header_modify_sw_icm_start_address[0x40];
 
-	u8         reserved_at_180[0x80];
+	u8         reserved_at_180[0x40];
+
+	u8         header_modify_pattern_sw_icm_start_address[0x40];
 
 	u8         memic_operations[0x20];
Signature · 1 line
2.36.1
Acked-by on v1 1/3
#2 lore raw

Re: [PATCH mlx5-next 1/3] net/mlx5: Introduce header-modify-pattern ICM properties

Quoted 8 lines · Leon Romanovsky

From: Yevgeny Kliteynik <[email protected]>

Added new fields for device memory capabilities, in order to support creation of ICM memory for modify header patterns.

Signed-off-by Erez Shitrit <[email protected]>
Signed-off-by Yevgeny Kliteynik <[email protected]>
Signed-off-by Leon Romanovsky <[email protected]>
Acked-by Saeed Mahameed <[email protected]>
author
#3 lore raw

PATCH 2/3 net/mlx5: Manage ICM of type modify-header pattern

In reply to Leon Romanovsky

From: Yevgeny Kliteynik <[email protected]>

Added support for managing new type of ICM for devices that support sw_owner_v2.

Signed-off-by Erez Shitrit <[email protected]>
Signed-off-by Yevgeny Kliteynik <[email protected]>
Signed-off-by Leon Romanovsky <[email protected]>
---
 .../net/ethernet/mellanox/mlx5/core/lib/dm.c  | 42 +++++++++++++++++++
 include/linux/mlx5/driver.h                   |  1 +
 2 files changed, 43 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/lib/dm.c +42 -0
index 3d5e57ff558c..7e02cbe8c3b9 100644
@@ -12,13 +12,16 @@ struct mlx5_dm {
 	spinlock_t lock;
 	unsigned long *steering_sw_icm_alloc_blocks;
 	unsigned long *header_modify_sw_icm_alloc_blocks;
+	unsigned long *header_modify_pattern_sw_icm_alloc_blocks;
 };
 
 struct mlx5_dm *mlx5_dm_create(struct mlx5_core_dev *dev)
 {
+	u64 header_modify_pattern_icm_blocks = 0;
 	u64 header_modify_icm_blocks = 0;
 	u64 steering_icm_blocks = 0;
 	struct mlx5_dm *dm;
+	bool support_v2;
 
 	if (!(MLX5_CAP_GEN_64(dev, general_obj_types) & MLX5_GENERAL_OBJ_TYPES_CAP_SW_ICM))
 		return NULL;

@@ -53,8 +56,27 @@ struct mlx5_dm *mlx5_dm_create(struct mlx5_core_dev *dev)
 			goto err_modify_hdr;
 	}
 
+	support_v2 = MLX5_CAP_FLOWTABLE_NIC_RX(dev, sw_owner_v2) &&
+		     MLX5_CAP_FLOWTABLE_NIC_TX(dev, sw_owner_v2) &&
+		     MLX5_CAP64_DEV_MEM(dev, header_modify_pattern_sw_icm_start_address);
+
+	if (support_v2) {
+		header_modify_pattern_icm_blocks =
+			BIT(MLX5_CAP_DEV_MEM(dev, log_header_modify_pattern_sw_icm_size) -
+			    MLX5_LOG_SW_ICM_BLOCK_SIZE(dev));
+
+		dm->header_modify_pattern_sw_icm_alloc_blocks =
+			kcalloc(BITS_TO_LONGS(header_modify_pattern_icm_blocks),
+				sizeof(unsigned long), GFP_KERNEL);
+		if (!dm->header_modify_pattern_sw_icm_alloc_blocks)
+			goto err_pattern;
+	}
+
 	return dm;
 
+err_pattern:
+	kfree(dm->header_modify_sw_icm_alloc_blocks);
+
 err_modify_hdr:
 	kfree(dm->steering_sw_icm_alloc_blocks);
 

@@ -86,6 +108,14 @@ void mlx5_dm_cleanup(struct mlx5_core_dev *dev)
 		kfree(dm->header_modify_sw_icm_alloc_blocks);
 	}
 
+	if (dm->header_modify_pattern_sw_icm_alloc_blocks) {
+		WARN_ON(!bitmap_empty(dm->header_modify_pattern_sw_icm_alloc_blocks,
+				      BIT(MLX5_CAP_DEV_MEM(dev,
+							   log_header_modify_pattern_sw_icm_size) -
+					  MLX5_LOG_SW_ICM_BLOCK_SIZE(dev))));
+		kfree(dm->header_modify_pattern_sw_icm_alloc_blocks);
+	}
+
 	kfree(dm);
 }
 

@@ -130,6 +160,13 @@ int mlx5_dm_sw_icm_alloc(struct mlx5_core_dev *dev, enum mlx5_sw_icm_type type,
 						log_header_modify_sw_icm_size);
 		block_map = dm->header_modify_sw_icm_alloc_blocks;
 		break;
+	case MLX5_SW_ICM_TYPE_HEADER_MODIFY_PATTERN:
+		icm_start_addr = MLX5_CAP64_DEV_MEM(dev,
+						    header_modify_pattern_sw_icm_start_address);
+		log_icm_size = MLX5_CAP_DEV_MEM(dev,
+						log_header_modify_pattern_sw_icm_size);
+		block_map = dm->header_modify_pattern_sw_icm_alloc_blocks;
+		break;
 	default:
 		return -EINVAL;
 	}

@@ -203,6 +240,11 @@ int mlx5_dm_sw_icm_dealloc(struct mlx5_core_dev *dev, enum mlx5_sw_icm_type type
 		icm_start_addr = MLX5_CAP64_DEV_MEM(dev, header_modify_sw_icm_start_address);
 		block_map = dm->header_modify_sw_icm_alloc_blocks;
 		break;
+	case MLX5_SW_ICM_TYPE_HEADER_MODIFY_PATTERN:
+		icm_start_addr = MLX5_CAP64_DEV_MEM(dev,
+						    header_modify_pattern_sw_icm_start_address);
+		block_map = dm->header_modify_pattern_sw_icm_alloc_blocks;
+		break;
 	default:
 		return -EINVAL;
 	}
include/linux/mlx5/driver.h +1 -0
index aeaedb985c1f..220597c2f436 100644
@@ -676,6 +676,7 @@ struct mlx5e_resources {
 enum mlx5_sw_icm_type {
 	MLX5_SW_ICM_TYPE_STEERING,
 	MLX5_SW_ICM_TYPE_HEADER_MODIFY,
+	MLX5_SW_ICM_TYPE_HEADER_MODIFY_PATTERN,
 };
 
 #define MLX5_MAX_RESERVED_GIDS 8
Signature · 1 line
2.36.1
author
#5 lore raw

PATCH 3/3 RDMA/mlx5: Support handling of modify-header pattern ICM area

In reply to Leon Romanovsky

From: Yevgeny Kliteynik <[email protected]>

Add support for allocate/deallocate and registering MR of the new type of ICM area. Support exists only for devices that support sw_owner_v2.

Signed-off-by Yevgeny Kliteynik <[email protected]>
Signed-off-by Leon Romanovsky <[email protected]>
---
 drivers/infiniband/hw/mlx5/dm.c           | 53 +++++++++++++++--------
 drivers/infiniband/hw/mlx5/mr.c           |  1 +
 include/uapi/rdma/mlx5_user_ioctl_verbs.h |  1 +
 3 files changed, 37 insertions(+), 18 deletions(-)
drivers/infiniband/hw/mlx5/dm.c +35 -18
index 001d766cf291..3669c90b2dad 100644
@@ -336,9 +336,15 @@ static struct ib_dm *handle_alloc_dm_memic(struct ib_ucontext *ctx,
 
 static enum mlx5_sw_icm_type get_icm_type(int uapi_type)
 {
-	return uapi_type == MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM ?
-		       MLX5_SW_ICM_TYPE_STEERING :
-		       MLX5_SW_ICM_TYPE_HEADER_MODIFY;
+	switch (uapi_type) {
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
+		return MLX5_SW_ICM_TYPE_HEADER_MODIFY;
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_PATTERN_SW_ICM:
+		return MLX5_SW_ICM_TYPE_HEADER_MODIFY_PATTERN;
+	case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
+	default:
+		return MLX5_SW_ICM_TYPE_STEERING;
+	}
 }
 
 static struct ib_dm *handle_alloc_dm_sw_icm(struct ib_ucontext *ctx,

@@ -347,11 +353,32 @@ static struct ib_dm *handle_alloc_dm_sw_icm(struct ib_ucontext *ctx,
 					    int type)
 {
 	struct mlx5_core_dev *dev = to_mdev(ctx->device)->mdev;
-	enum mlx5_sw_icm_type icm_type = get_icm_type(type);
+	enum mlx5_sw_icm_type icm_type;
 	struct mlx5_ib_dm_icm *dm;
 	u64 act_size;
 	int err;
 
+	if (!capable(CAP_SYS_RAWIO) || !capable(CAP_NET_RAW))
+		return ERR_PTR(-EPERM);
+
+	switch (type) {
+	case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
+		if (!(MLX5_CAP_FLOWTABLE_NIC_RX(dev, sw_owner) ||
+		      MLX5_CAP_FLOWTABLE_NIC_TX(dev, sw_owner) ||
+		      MLX5_CAP_FLOWTABLE_NIC_RX(dev, sw_owner_v2) ||
+		      MLX5_CAP_FLOWTABLE_NIC_TX(dev, sw_owner_v2)))
+			return ERR_PTR(-EOPNOTSUPP);
+		break;
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_PATTERN_SW_ICM:
+		if (!MLX5_CAP_FLOWTABLE_NIC_RX(dev, sw_owner_v2) ||
+		    !MLX5_CAP_FLOWTABLE_NIC_TX(dev, sw_owner_v2))
+			return ERR_PTR(-EOPNOTSUPP);
+		break;
+	default:
+		return ERR_PTR(-EOPNOTSUPP);
+	}
+
 	dm = kzalloc(sizeof(*dm), GFP_KERNEL);
 	if (!dm)
 		return ERR_PTR(-ENOMEM);

@@ -359,19 +386,6 @@ static struct ib_dm *handle_alloc_dm_sw_icm(struct ib_ucontext *ctx,
 	dm->base.type = type;
 	dm->base.ibdm.device = ctx->device;
 
-	if (!capable(CAP_SYS_RAWIO) || !capable(CAP_NET_RAW)) {
-		err = -EPERM;
-		goto free;
-	}
-
-	if (!(MLX5_CAP_FLOWTABLE_NIC_RX(dev, sw_owner) ||
-	      MLX5_CAP_FLOWTABLE_NIC_TX(dev, sw_owner) ||
-	      MLX5_CAP_FLOWTABLE_NIC_RX(dev, sw_owner_v2) ||
-	      MLX5_CAP_FLOWTABLE_NIC_TX(dev, sw_owner_v2))) {
-		err = -EOPNOTSUPP;
-		goto free;
-	}
-
 	/* Allocation size must a multiple of the basic block size
 	 * and a power of 2.
 	 */

@@ -379,6 +393,8 @@ static struct ib_dm *handle_alloc_dm_sw_icm(struct ib_ucontext *ctx,
 	act_size = roundup_pow_of_two(act_size);
 
 	dm->base.size = act_size;
+	icm_type = get_icm_type(type);
+
 	err = mlx5_dm_sw_icm_alloc(dev, icm_type, act_size, attr->alignment,
 				   to_mucontext(ctx)->devx_uid,
 				   &dm->base.dev_addr, &dm->obj_id);

@@ -420,8 +436,8 @@ struct ib_dm *mlx5_ib_alloc_dm(struct ib_device *ibdev,
 	case MLX5_IB_UAPI_DM_TYPE_MEMIC:
 		return handle_alloc_dm_memic(context, attr, attrs);
 	case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
-		return handle_alloc_dm_sw_icm(context, attr, attrs, type);
 	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_PATTERN_SW_ICM:
 		return handle_alloc_dm_sw_icm(context, attr, attrs, type);
 	default:
 		return ERR_PTR(-EOPNOTSUPP);

@@ -474,6 +490,7 @@ static int mlx5_ib_dealloc_dm(struct ib_dm *ibdm,
 		return 0;
 	case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
 	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_PATTERN_SW_ICM:
 		return mlx5_dm_icm_dealloc(ctx, to_icm(ibdm));
 	default:
 		return -EOPNOTSUPP;
drivers/infiniband/hw/mlx5/mr.c +1 -0
index 90bb61ad8c64..dd8887bf83f0 100644
@@ -1132,6 +1132,7 @@ struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
 		break;
 	case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
 	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
+	case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_PATTERN_SW_ICM:
 		if (attr->access_flags & ~MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS)
 			return ERR_PTR(-EINVAL);
include/uapi/rdma/mlx5_user_ioctl_verbs.h +1 -0
index a21ca8ece8db..7af9e09ea556 100644
@@ -63,6 +63,7 @@ enum mlx5_ib_uapi_dm_type {
 	MLX5_IB_UAPI_DM_TYPE_MEMIC,
 	MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM,
 	MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM,
+	MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_PATTERN_SW_ICM,
 };
 
 enum mlx5_ib_uapi_devx_create_event_channel_flags {
Signature · 1 line
2.36.1
#6 lore raw
In reply to Leon Romanovsky
Quoted 7 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

SW steering manipulates packet's header using "modifying header" actions. Many of these actions do the same operation, but use different data each time. Currently we create and keep every one of these actions, which use expensive and limited resources.

Series applied to mlx5-next

Quoted 16 lines

Now we introduce a new mechanism - pattern and argument, which splits a modifying action into two parts:
1. action pattern: contains the operations to be applied on packet's header, mainly set/add/copy of fields in the packet
2. action data/argument: contains the data to be used by each operation in the pattern.

This way we reuse same patterns with different arguments to create new modifying actions, and since many actions share the same operations, we end up creating a small number of patterns that we keep in a dedicated cache.

These modify header patterns are implemented as new type of ICM memory, so the following kernel patch series add the support for this new ICM type.

Thanks