RDMA RX RoCE Steering Support
Subsystems (5)
cover RDMA RX RoCE Steering Support
From: Leon Romanovsky <[email protected]>
Hi,
This series from Mark extends mlx5 with RDMA_RX RoCE flow steering support for DEVX and QP objects.
Thanks
Mark Zhang (3):
net/mlx5: Add per-namespace flow table default miss action support net/mlx5: Create bypass and loopback flow steering namespaces for RDMA
RX
RDMA/mlx5: RDMA_RX flow type support for user applications
drivers/infiniband/hw/mlx5/flow.c | 13 +- drivers/infiniband/hw/mlx5/main.c | 7 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 1 + .../net/ethernet/mellanox/mlx5/core/fs_cmd.c | 4 +- .../net/ethernet/mellanox/mlx5/core/fs_core.c | 120 ++++++++++++------ .../net/ethernet/mellanox/mlx5/core/fs_core.h | 3 +- .../net/ethernet/mellanox/mlx5/core/rdma.c | 2 +- include/linux/mlx5/fs.h | 1 + include/uapi/rdma/mlx5_user_ioctl_verbs.h | 1 + 9 files changed, 107 insertions(+), 45 deletions(-)
--
2.20.1
PATCH 1/3 net/mlx5: Add per-namespace flow table default miss action support
From: Mark Zhang <[email protected]>
Currently all the namespaces under the same steering domain share the same default table miss action, however in some situations (e.g., RDMA RX) different actions are required. This patch adds a per-namespace default table miss action instead of using the miss action of the steering domain.
--- .../net/ethernet/mellanox/mlx5/core/fs_cmd.c | 4 +- .../net/ethernet/mellanox/mlx5/core/fs_core.c | 73 +++++++++++-------- .../net/ethernet/mellanox/mlx5/core/fs_core.h | 3 +- 3 files changed, 47 insertions(+), 33 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c +2 -2
@@ -182,7 +182,7 @@ static int mlx5_cmd_create_flow_table(struct mlx5_flow_root_namespace *ns, }else {MLX5_SET (create_flow_table_in, in, flow_table_context.table_miss_action, - ns->def_miss_action); + ft->def_miss_action); }break ;
@@ -262,7 +262,7 @@ static int mlx5_cmd_modify_flow_table(struct mlx5_flow_root_namespace *ns, }else {MLX5_SET (modify_flow_table_in, in, flow_table_context.table_miss_action, - ns->def_miss_action); + ft->def_miss_action); } }
drivers/net/ethernet/mellanox/mlx5/core/fs_core.c +43 -30
130 lines not on this page · raw message
drivers/net/ethernet/mellanox/mlx5/core/fs_core.h +2 -1
@@ -145,6 +145,7 @@ struct mlx5_flow_table {struct list_head fwd_rules;u32 flags;struct rhltable fgs_hash; +enum mlx5_flow_table_miss_action def_miss_action; };struct mlx5_ft_underlay_qp {
@@ -191,6 +192,7 @@ struct fs_prio {struct mlx5_flow_namespace {/* parent == NULL => root ns */ struct fs_node node; +enum mlx5_flow_table_miss_action def_miss_action; };struct mlx5_flow_group_mask {
@@ -219,7 +221,6 @@ struct mlx5_flow_root_namespace {struct mutex chain_lock;struct list_head underlay_qpns;const struct mlx5_flow_cmds *cmds; -enum mlx5_flow_table_miss_action def_miss_action; };int mlx5_init_fc_stats (struct mlx5_core_dev *dev);
Signature · 1 line
2.20.1
PATCH 3/3 RDMA/mlx5: RDMA_RX flow type support for user applications
In reply to Leon RomanovskyFrom: Mark Zhang <[email protected]>
Currently user applications can only steer TCP/IP(NIC RX/RX) traffic. This patch adds RDMA_RX as a new flow type to allow the user to insert steering rules to control RDMA traffic.
Two destinations are supported(but not set at the same time): devx flow table object and QP.
--- drivers/infiniband/hw/mlx5/flow.c | 13 +++++++++++-- drivers/infiniband/hw/mlx5/main.c | 7 +++++++ drivers/infiniband/hw/mlx5/mlx5_ib.h | 1 + include/uapi/rdma/mlx5_user_ioctl_verbs.h | 1 + 4 files changed, 20 insertions(+), 2 deletions(-)
drivers/infiniband/hw/mlx5/flow.c +11 -2
@@ -32,6 +32,9 @@ mlx5_ib_ft_type_to_namespace(enum mlx5_ib_uapi_flow_table_type table_type,case MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB : *namespace =MLX5_FLOW_NAMESPACE_FDB ;break ; +case MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX : + *namespace =MLX5_FLOW_NAMESPACE_RDMA_RX ; +break ;default :return -EINVAL ; }
@@ -101,6 +104,11 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(if (fs_matcher->ns_type ==MLX5_FLOW_NAMESPACE_FDB && !dest_devx)return -EINVAL ; +/* Allow only DEVX object or QP as dest when inserting to RDMA_RX */ +if ((fs_matcher->ns_type ==MLX5_FLOW_NAMESPACE_RDMA_RX ) && + ((!dest_devx && !dest_qp) || (dest_devx && dest_qp))) +return -EINVAL ; +if (dest_devx) { devx_obj =uverbs_attr_get_obj ( attrs,MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX );
@@ -112,8 +120,9 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(*/ if (!mlx5_ib_devx_is_flow_dest (devx_obj, &dest_id, &dest_type))return -EINVAL ; -/* Allow only flow table as dest when inserting to FDB */ -if (fs_matcher->ns_type ==MLX5_FLOW_NAMESPACE_FDB && +/* Allow only flow table as dest when inserting to FDB or RDMA_RX */ +if ((fs_matcher->ns_type ==MLX5_FLOW_NAMESPACE_FDB || + fs_matcher->ns_type ==MLX5_FLOW_NAMESPACE_RDMA_RX ) && dest_type !=MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE )return -EINVAL ; }else if (dest_qp) {
drivers/infiniband/hw/mlx5/main.c +7 -0
@@ -3963,6 +3963,11 @@ _get_flow_table(struct mlx5_ib_dev *dev, esw_encap) flags |= MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT; priority =FDB_BYPASS_PATH ; + }else if (fs_matcher->ns_type ==MLX5_FLOW_NAMESPACE_RDMA_RX ) { + max_table_size = +BIT (MLX5_CAP_FLOWTABLE_RDMA_RX (dev->mdev, + log_max_ft_size)); + priority = fs_matcher->priority; } max_table_size =min_t (int, max_table_size,MLX5_FS_MAX_ENTRIES );
@@ -3977,6 +3982,8 @@ _get_flow_table(struct mlx5_ib_dev *dev, prio = &dev->flow_db->egress_prios[priority]; else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_FDB) prio = &dev->flow_db->fdb; + else if (fs_matcher->ns_type == MLX5_FLOW_NAMESPACE_RDMA_RX) + prio = &dev->flow_db->rdma_rx[priority];if (!prio)return ERR_PTR (-EINVAL );
drivers/infiniband/hw/mlx5/mlx5_ib.h +1 -0
@@ -200,6 +200,7 @@ struct mlx5_ib_flow_db {struct mlx5_ib_flow_prio sniffer[MLX5_IB_NUM_SNIFFER_FTS ];struct mlx5_ib_flow_prio egress[MLX5_IB_NUM_EGRESS_FTS ];struct mlx5_ib_flow_prio fdb; +struct mlx5_ib_flow_prio rdma_rx[MLX5_IB_NUM_FLOW_FT ];struct mlx5_flow_table *lag_demux_ft;/* Protect flow steering bypass flow tables * when add/del flow rules.
include/uapi/rdma/mlx5_user_ioctl_verbs.h +1 -0
@@ -43,6 +43,7 @@ enum mlx5_ib_uapi_flow_table_type {MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_RX =0x0 ,MLX5_IB_UAPI_FLOW_TABLE_TYPE_NIC_TX =0x1 ,MLX5_IB_UAPI_FLOW_TABLE_TYPE_FDB =0x2 , + MLX5_IB_UAPI_FLOW_TABLE_TYPE_RDMA_RX = 0x3, }; enum mlx5_ib_uapi_flow_action_packet_reformat_type {
Signature · 1 line
2.20.1
PATCH 2/3 net/mlx5: Create bypass and loopback flow steering namespaces for RDMA RX
In reply to Leon RomanovskyFrom: Mark Zhang <[email protected]>
Use different namespaces for bypass and switchdev loopback because they have different priorities and default table miss action requirement:
1. bypass: with multiple priorities support, and MLX5_FLOW_TABLE_MISS_ACTION_DEF as the default table miss action;
2. switchdev loopback: with single priority support, and MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN as the default table miss action.
--- .../net/ethernet/mellanox/mlx5/core/fs_core.c | 49 +++++++++++++++---- .../net/ethernet/mellanox/mlx5/core/rdma.c | 2 +- include/linux/mlx5/fs.h | 1 + 3 files changed, 41 insertions(+), 11 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/fs_core.c +39 -10
@@ -182,6 +182,26 @@ static struct init_tree_node egress_root_fs = { } }; +#define RDMA_RX_BYPASS_PRIO 0 +#define RDMA_RX_KERNEL_PRIO 1 +static struct init_tree_node rdma_rx_root_fs = { + .type =FS_TYPE_NAMESPACE , + .ar_size =2 , + .children = (struct init_tree_node []) { + [RDMA_RX_BYPASS_PRIO ] = +ADD_PRIO (0 ,MLX5_BY_PASS_NUM_REGULAR_PRIOS ,0 , +FS_CHAINING_CAPS , +ADD_NS (MLX5_FLOW_TABLE_MISS_ACTION_DEF , +ADD_MULTIPLE_PRIO (MLX5_BY_PASS_NUM_REGULAR_PRIOS , +BY_PASS_PRIO_NUM_LEVELS ))), + [RDMA_RX_KERNEL_PRIO ] = +ADD_PRIO (0 ,MLX5_BY_PASS_NUM_REGULAR_PRIOS +1 ,0 , +FS_CHAINING_CAPS , +ADD_NS (MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN , +ADD_MULTIPLE_PRIO (1 ,1 ))), + } +}; +enum fs_i_lock_class {FS_LOCK_GRANDPARENT ,FS_LOCK_PARENT ,
@@ -2067,16 +2087,18 @@ struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,if (steering->sniffer_tx_root_ns)return &steering->sniffer_tx_root_ns->ns;return NULL ; -case MLX5_FLOW_NAMESPACE_RDMA_RX : -if (steering->rdma_rx_root_ns) -return &steering->rdma_rx_root_ns->ns; -return NULL ;default :break ; }if (type ==MLX5_FLOW_NAMESPACE_EGRESS ) { root_ns = steering->egress_root_ns; + }else if (type ==MLX5_FLOW_NAMESPACE_RDMA_RX ) { + root_ns = steering->rdma_rx_root_ns; + prio =RDMA_RX_BYPASS_PRIO ; + }else if (type ==MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL ) { + root_ns = steering->rdma_rx_root_ns; + prio =RDMA_RX_KERNEL_PRIO ; }else {/* Must be NIC RX */ root_ns = steering->root_ns; prio = type;
@@ -2507,18 +2529,25 @@ static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)static int init_rdma_rx_root_ns (struct mlx5_flow_steering *steering) { -struct fs_prio *prio; +int err; steering->rdma_rx_root_ns =create_root_ns (steering,FS_FT_RDMA_RX );if (!steering->rdma_rx_root_ns)return -ENOMEM ; - steering->rdma_rx_root_ns->ns.def_miss_action = -MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN ; + err =init_root_tree (steering, &rdma_rx_root_fs, + &steering->rdma_rx_root_ns->ns.node); +if (err) + goto out_err; -/* Create single prio */ - prio =fs_create_prio (&steering->rdma_rx_root_ns->ns,0 ,1 ); -return PTR_ERR_OR_ZERO (prio); +set_prio_attrs (steering->rdma_rx_root_ns); + +return 0 ; + +out_err: +cleanup_root_ns (steering->rdma_rx_root_ns); + steering->rdma_rx_root_ns =NULL ; +return err; }static int init_fdb_root_ns (struct mlx5_flow_steering *steering) {
drivers/net/ethernet/mellanox/mlx5/core/rdma.c +1 -1
@@ -51,7 +51,7 @@ static int mlx5_rdma_enable_roce_steering(struct mlx5_core_dev *dev)return -ENOMEM ; } - ns =mlx5_get_flow_namespace (dev,MLX5_FLOW_NAMESPACE_RDMA_RX ); + ns =mlx5_get_flow_namespace (dev,MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL );if (!ns) {mlx5_core_err (dev,"Failed to get RDMA RX namespace" ); err = -EOPNOTSUPP ;
include/linux/mlx5/fs.h +1 -0
@@ -75,6 +75,7 @@ enum mlx5_flow_namespace_type {MLX5_FLOW_NAMESPACE_SNIFFER_TX ,MLX5_FLOW_NAMESPACE_EGRESS ,MLX5_FLOW_NAMESPACE_RDMA_RX , + MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL, }; enum {
Signature · 1 line
2.20.1
Quoted 8 lines · Leon Romanovsky
From: Leon Romanovsky <[email protected]>
Hi,
This series from Mark extends mlx5 with
RDMA_RXRoCE flow steering support
for DEVX and QP objects.
Quoted 8 lines
Thanks
Mark Zhang (3):
net/mlx5: Add per-namespace flow table default miss action support net/mlx5: Create bypass and loopback flow steering namespaces for
RDMA
RX
RDMA/mlx5:RDMA_RXflow type support for user applications
I have no objection to this series.
Signature · 3 lines
Doug Ledford <[email protected]> GPG KeyID: B826A3330E572FDD Fingerprint = AE6B 1BDA 122B 23B4 265B 1274 B826 A333 0E57 2FDD
Quoted 11 lines · Doug Ledford
Quoted 10 lines · Leon Romanovsky
From: Leon Romanovsky <[email protected]>
Hi,
This series from Mark extends mlx5 with
RDMA_RXRoCE flow steering support
for DEVX and QP objects.Thanks
Quoted 8 lines
Quoted 6 lines
Mark Zhang (3):
net/mlx5: Add per-namespace flow table default miss action support net/mlx5: Create bypass and loopback flow steering namespaces for
RDMA
RX
RDMA/mlx5:RDMA_RXflow type support for user applicationsI have no objection to this series.
Thanks, first two patches were applied to mlx5-next
e6806e9a63a7 net/mlx5: Create bypass and loopback flow steering namespaces for RDMA RX f66ad830b114 net/mlx5: Add per-namespace flow table default miss action support
Quoted 5 lines
--
Doug Ledford <[email protected]>GPG KeyID: B826A3330E572FDD Fingerprint = AE6B 1BDA 122B 23B4 265B 1274 B826 A333 0E57 2FDD
Quoted 21 lines · Leon Romanovsky
Quoted 20 lines · Doug Ledford
Quoted 18 lines · Leon Romanovsky
From: Leon Romanovsky <[email protected]>
Hi,
This series from Mark extends mlx5 with
RDMA_RXRoCE flow steering support
for DEVX and QP objects.Thanks
Mark Zhang (3):
net/mlx5: Add per-namespace flow table default miss action
support
net/mlx5: Create bypass and loopback flow steering namespaces
for
RDMA
RX
RDMA/mlx5:RDMA_RXflow type support for user applications
Quoted 8 lines
Quoted 1 line
I have no objection to this series.
Thanks, first two patches were applied to mlx5-next
e6806e9a63a7net/mlx5: Create bypass and loopback flow steering namespaces for RDMA RXf66ad830b114net/mlx5: Add per-namespace flow table default miss action support
mlx5-next merged into for-next, final patch applied, thanks.
Signature · 3 lines
Doug Ledford <[email protected]> GPG KeyID: B826A3330E572FDD Fingerprint = AE6B 1BDA 122B 23B4 265B 1274 B826 A333 0E57 2FDD
Quoted 24 lines · Doug Ledford
Quoted 23 lines · Leon Romanovsky
Quoted 21 lines · Doug Ledford
Quoted 10 lines · Leon Romanovsky
From: Leon Romanovsky <[email protected]>
Hi,
This series from Mark extends mlx5 with
RDMA_RXRoCE flow steering support
for DEVX and QP objects.Thanks
Quoted 8 lines
Mark Zhang (3):
net/mlx5: Add per-namespace flow table default miss action
support
net/mlx5: Create bypass and loopback flow steering namespaces
for
RDMA
RX
RDMA/mlx5:RDMA_RXflow type support for user applicationsI have no objection to this series.
Quoted 8 lines
Quoted 6 lines
Thanks, first two patches were applied to mlx5-next
e6806e9a63a7net/mlx5: Create bypass and loopback flow steering namespaces for RDMA RXf66ad830b114net/mlx5: Add per-namespace flow table default miss action supportmlx5-next merged into for-next, final patch applied, thanks.
Thanks
Quoted 5 lines
--
Doug Ledford <[email protected]>GPG KeyID: B826A3330E572FDD Fingerprint = AE6B 1BDA 122B 23B4 265B 1274 B826 A333 0E57 2FDD