* [PATCH for-next 1/6] IB/core: Add flow-steering support for IPoIB UD traffic
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
@ 2013-11-07 13:25 ` Or Gerlitz
2013-11-07 13:25 ` [PATCH for-next 2/6] IB/core: Add support for IB L2 device managed steering Or Gerlitz
` (6 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Or Gerlitz @ 2013-11-07 13:25 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, matanb-VPRAkNaXOzVWk0Htik3J/w,
Or Gerlitz
From: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
When creating the IPoIB UD QP, provide hint to the low level driver
that the QP should support flow-steering. This means that privileged
user space applications could steer TCP/IP IPoIB traffic from the
network stack, in a similar manner done with Ethernet RAW_PACKET QPs.
The hint is provided through new QP creation flag called NETIF_QP.
Signed-off-by: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/infiniband/ulp/ipoib/ipoib_verbs.c | 3 +++
include/rdma/ib_verbs.h | 1 +
2 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
index 049a997..c56d5d4 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
@@ -192,6 +192,9 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
if (priv->hca_caps & IB_DEVICE_BLOCK_MULTICAST_LOOPBACK)
init_attr.create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
+ if (priv->hca_caps & IB_DEVICE_MANAGED_FLOW_STEERING)
+ init_attr.create_flags |= IB_QP_CREATE_NETIF_QP;
+
if (dev->features & NETIF_F_SG)
init_attr.cap.max_send_sge = MAX_SKB_FRAGS + 1;
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index e393171..bb81c37 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -631,6 +631,7 @@ enum ib_qp_type {
enum ib_qp_create_flags {
IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
+ IB_QP_CREATE_NETIF_QP = 1 << 5,
/* reserve bits 26-31 for low level drivers' internal use */
IB_QP_CREATE_RESERVED_START = 1 << 26,
IB_QP_CREATE_RESERVED_END = 1 << 31,
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH for-next 2/6] IB/core: Add support for IB L2 device managed steering
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
2013-11-07 13:25 ` [PATCH for-next 1/6] IB/core: Add flow-steering support for IPoIB UD traffic Or Gerlitz
@ 2013-11-07 13:25 ` Or Gerlitz
2013-11-07 13:25 ` [PATCH for-next 3/6] net/mlx4_core: Add support for steer-able IB UD QPs Or Gerlitz
` (5 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Or Gerlitz @ 2013-11-07 13:25 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, matanb-VPRAkNaXOzVWk0Htik3J/w,
Or Gerlitz
From: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
This patch adds preliminary support for IB L2 device managed steering,
currently exposed only in the kernel.
This flow spec can be used by low level drivers that need to indicate
the link layer type when creating device managed flow rules.
Signed-off-by: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
include/rdma/ib_verbs.h | 16 +++++++++++++++-
1 files changed, 15 insertions(+), 1 deletions(-)
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index bb81c37..887e096 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -1098,13 +1098,14 @@ enum ib_flow_attr_type {
enum ib_flow_spec_type {
/* L2 headers*/
IB_FLOW_SPEC_ETH = 0x20,
+ IB_FLOW_SPEC_IB = 0x22,
/* L3 header*/
IB_FLOW_SPEC_IPV4 = 0x30,
/* L4 headers*/
IB_FLOW_SPEC_TCP = 0x40,
IB_FLOW_SPEC_UDP = 0x41
};
-
+#define IB_FLOW_SPEC_LAYER_MASK 0xF0
#define IB_FLOW_SPEC_SUPPORT_LAYERS 4
/* Flow steering rule priority is set according to it's domain.
@@ -1132,6 +1133,18 @@ struct ib_flow_spec_eth {
struct ib_flow_eth_filter mask;
};
+struct ib_flow_ib_filter {
+ __be16 dlid;
+ __u8 sl;
+};
+
+struct ib_flow_spec_ib {
+ enum ib_flow_spec_type type;
+ u16 size;
+ struct ib_flow_ib_filter val;
+ struct ib_flow_ib_filter mask;
+};
+
struct ib_flow_ipv4_filter {
__be32 src_ip;
__be32 dst_ip;
@@ -1162,6 +1175,7 @@ union ib_flow_spec {
u16 size;
};
struct ib_flow_spec_eth eth;
+ struct ib_flow_spec_ib ib;
struct ib_flow_spec_ipv4 ipv4;
struct ib_flow_spec_tcp_udp tcp_udp;
};
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH for-next 3/6] net/mlx4_core: Add support for steer-able IB UD QPs
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
2013-11-07 13:25 ` [PATCH for-next 1/6] IB/core: Add flow-steering support for IPoIB UD traffic Or Gerlitz
2013-11-07 13:25 ` [PATCH for-next 2/6] IB/core: Add support for IB L2 device managed steering Or Gerlitz
@ 2013-11-07 13:25 ` Or Gerlitz
2013-11-07 13:25 ` [PATCH for-next 4/6] IB/mlx4_ib: Enable device managed steering support for IB ports too Or Gerlitz
` (4 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Or Gerlitz @ 2013-11-07 13:25 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, matanb-VPRAkNaXOzVWk0Htik3J/w,
Or Gerlitz
From: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
This patch adds support for allocating IB UD QPs that we could steer
traffic from them. In order to do that, we introduce a new firmware
command FLOW_STEERING_IB_UC_QP_RANGE and a capability bit.
This command isn't supported for VFs.
Signed-off-by: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/net/ethernet/mellanox/mlx4/cmd.c | 9 +++++++++
drivers/net/ethernet/mellanox/mlx4/fw.c | 10 ++++++++++
drivers/net/ethernet/mellanox/mlx4/mcg.c | 17 +++++++++++++++++
drivers/net/ethernet/mellanox/mlx4/mlx4.h | 5 +++++
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 10 ++++++++++
include/linux/mlx4/cmd.h | 1 +
include/linux/mlx4/device.h | 6 +++++-
7 files changed, 57 insertions(+), 1 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index ea20182..d7f9003 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -1371,6 +1371,15 @@ static struct mlx4_cmd_info cmd_info[] = {
.verify = NULL,
.wrapper = mlx4_QP_FLOW_STEERING_DETACH_wrapper
},
+ {
+ .opcode = MLX4_FLOW_STEERING_IB_UC_QP_RANGE,
+ .has_inbox = false,
+ .has_outbox = false,
+ .out_is_imm = false,
+ .encode_slave_id = false,
+ .verify = NULL,
+ .wrapper = mlx4_FLOW_STEERING_IB_UC_QP_RANGE_wrapper
+ },
};
static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 0d63daa..69558e7 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -472,6 +472,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
#define QUERY_DEV_CAP_MAX_XRC_OFFSET 0x67
#define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET 0x68
#define QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET 0x70
+#define QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET 0x74
#define QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET 0x76
#define QUERY_DEV_CAP_FLOW_STEERING_MAX_QP_OFFSET 0x77
#define QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET 0x80
@@ -562,6 +563,9 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
if (field & 0x80)
dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FS_EN;
dev_cap->fs_log_max_ucast_qp_range_size = field & 0x1f;
+ MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
+ if (field & 0x80)
+ dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DMFS_IPOIB;
MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_MAX_QP_OFFSET);
dev_cap->fs_max_num_qp_per_entry = field;
MLX4_GET(stat_rate, outbox, QUERY_DEV_CAP_RATE_SUPPORT_OFFSET);
@@ -819,6 +823,12 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
MLX4_PUT(outbox->buf, field,
QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
}
+
+ /* turn off ipoib managed steering for guests */
+ MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
+ field &= ~0x80;
+ MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET);
+
return 0;
}
diff --git a/drivers/net/ethernet/mellanox/mlx4/mcg.c b/drivers/net/ethernet/mellanox/mlx4/mcg.c
index 55f6245..98f55d9 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mcg.c
+++ b/drivers/net/ethernet/mellanox/mlx4/mcg.c
@@ -897,6 +897,23 @@ int mlx4_flow_detach(struct mlx4_dev *dev, u64 reg_id)
}
EXPORT_SYMBOL_GPL(mlx4_flow_detach);
+int mlx4_FLOW_STEERING_IB_UC_QP_RANGE(struct mlx4_dev *dev, u32 min_range_qpn,
+ u32 max_range_qpn)
+{
+ int err;
+ u64 in_param;
+
+ in_param = ((u64) min_range_qpn) << 32;
+ in_param |= ((u64) max_range_qpn) & 0xFFFFFFFF;
+
+ err = mlx4_cmd(dev, in_param, 0, 0,
+ MLX4_FLOW_STEERING_IB_UC_QP_RANGE,
+ MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
+
+ return err;
+}
+EXPORT_SYMBOL_GPL(mlx4_FLOW_STEERING_IB_UC_QP_RANGE);
+
int mlx4_qp_attach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
int block_mcast_loopback, enum mlx4_protocol prot,
enum mlx4_steer_type steer)
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index 348bb8c..069ca2d 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -1219,6 +1219,11 @@ int mlx4_QP_FLOW_STEERING_DETACH_wrapper(struct mlx4_dev *dev, int slave,
struct mlx4_cmd_mailbox *inbox,
struct mlx4_cmd_mailbox *outbox,
struct mlx4_cmd_info *cmd);
+int mlx4_FLOW_STEERING_IB_UC_QP_RANGE_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd);
int mlx4_get_mgm_entry_size(struct mlx4_dev *dev);
int mlx4_get_qp_per_mgm(struct mlx4_dev *dev);
diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
index dd68763..3a26de6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
+++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
@@ -3387,6 +3387,16 @@ int mlx4_QUERY_IF_STAT_wrapper(struct mlx4_dev *dev, int slave,
return err;
}
+int mlx4_FLOW_STEERING_IB_UC_QP_RANGE_wrapper(struct mlx4_dev *dev, int slave,
+ struct mlx4_vhcr *vhcr,
+ struct mlx4_cmd_mailbox *inbox,
+ struct mlx4_cmd_mailbox *outbox,
+ struct mlx4_cmd_info *cmd)
+{
+ return -EPERM;
+}
+
+
static void detach_qp(struct mlx4_dev *dev, int slave, struct res_qp *rqp)
{
struct res_gid *rgid;
diff --git a/include/linux/mlx4/cmd.h b/include/linux/mlx4/cmd.h
index cd1fdf7..09111b1 100644
--- a/include/linux/mlx4/cmd.h
+++ b/include/linux/mlx4/cmd.h
@@ -161,6 +161,7 @@ enum {
/* register/delete flow steering network rules */
MLX4_QP_FLOW_STEERING_ATTACH = 0x65,
MLX4_QP_FLOW_STEERING_DETACH = 0x66,
+ MLX4_FLOW_STEERING_IB_UC_QP_RANGE = 0x64,
};
enum {
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 24ce6bd..4e4ae47 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -159,7 +159,8 @@ enum {
MLX4_DEV_CAP_FLAG2_TS = 1LL << 5,
MLX4_DEV_CAP_FLAG2_VLAN_CONTROL = 1LL << 6,
MLX4_DEV_CAP_FLAG2_FSM = 1LL << 7,
- MLX4_DEV_CAP_FLAG2_UPDATE_QP = 1LL << 8
+ MLX4_DEV_CAP_FLAG2_UPDATE_QP = 1LL << 8,
+ MLX4_DEV_CAP_FLAG2_DMFS_IPOIB = 1LL << 9
};
enum {
@@ -1125,6 +1126,9 @@ int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave, u8 port, int
void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid);
__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave);
+int mlx4_FLOW_STEERING_IB_UC_QP_RANGE(struct mlx4_dev *dev, u32 min_range_qpn,
+ u32 max_range_qpn);
+
cycle_t mlx4_read_clock(struct mlx4_dev *dev);
#endif /* MLX4_DEVICE_H */
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH for-next 4/6] IB/mlx4_ib: Enable device managed steering support for IB ports too
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
` (2 preceding siblings ...)
2013-11-07 13:25 ` [PATCH for-next 3/6] net/mlx4_core: Add support for steer-able IB UD QPs Or Gerlitz
@ 2013-11-07 13:25 ` Or Gerlitz
2013-11-07 13:25 ` [PATCH for-next 5/6] IB/mlx4_ib: Add mechanism to support flow-steering over IB links Or Gerlitz
` (3 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Or Gerlitz @ 2013-11-07 13:25 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, matanb-VPRAkNaXOzVWk0Htik3J/w,
Or Gerlitz
From: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Up until now, flow steering wasn't supported when using IB ports.
This patch enables the support for flow steering if all hardware ports
support that, e.g under the new MLX4_DEV_CAP_FLAG2_DMFS_IPOIB mlx4 device
capability.
Signed-off-by: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/infiniband/hw/mlx4/main.c | 31 +++++++++++++++++++------------
drivers/infiniband/hw/mlx4/mlx4_ib.h | 1 +
2 files changed, 20 insertions(+), 12 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index f061264..093fc85 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -92,21 +92,27 @@ static union ib_gid zgid;
static int check_flow_steering_support(struct mlx4_dev *dev)
{
+ int eth_num_ports = 0;
int ib_num_ports = 0;
- int i;
-
- mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
- ib_num_ports++;
- if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
- if (ib_num_ports || mlx4_is_mfunc(dev)) {
- pr_warn("Device managed flow steering is unavailable "
- "for IB ports or in multifunction env.\n");
- return 0;
+ int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
+
+ if (dmfs) {
+ int i;
+ mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
+ eth_num_ports++;
+ mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
+ ib_num_ports++;
+ dmfs &= (!ib_num_ports ||
+ (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
+ (!eth_num_ports ||
+ (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
+ if (ib_num_ports && mlx4_is_mfunc(dev)) {
+ pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
+ dmfs = 0;
}
- return 1;
}
- return 0;
+ return dmfs;
}
static int mlx4_ib_query_device(struct ib_device *ibdev,
@@ -165,7 +171,7 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
else
props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
- if (check_flow_steering_support(dev->dev))
+ if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
}
@@ -1688,6 +1694,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
}
if (check_flow_steering_support(dev)) {
+ ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 036b663..e9fb396 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -494,6 +494,7 @@ struct mlx4_ib_dev {
struct kobject *dev_ports_parent[MLX4_MFUNC_MAX];
struct mlx4_ib_iov_port iov_ports[MLX4_MAX_PORTS];
struct pkey_mgt pkeys;
+ int steering_support;
};
struct ib_event_work {
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH for-next 5/6] IB/mlx4_ib: Add mechanism to support flow-steering over IB links
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
` (3 preceding siblings ...)
2013-11-07 13:25 ` [PATCH for-next 4/6] IB/mlx4_ib: Enable device managed steering support for IB ports too Or Gerlitz
@ 2013-11-07 13:25 ` Or Gerlitz
2013-11-07 13:25 ` [PATCH for-next 6/6] IB/mlx4_ib: Add support for steer-able IB UD QPs Or Gerlitz
` (2 subsequent siblings)
7 siblings, 0 replies; 10+ messages in thread
From: Or Gerlitz @ 2013-11-07 13:25 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, matanb-VPRAkNaXOzVWk0Htik3J/w,
Or Gerlitz
From: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
The mlx4 device requires adding IB flow spec to rules that apply over
infiniband link layer. This patch adds a mechanism to add such a rule.
If higher levels e.g. IP/UDP/TCP flow specs are provided, the device
require us to add an empty wildecarded IB rule. Furthermore, the device
requires the qpn to be put in the rule.
Add here specific parsing support for IB empty rules and the
ability to self-generate missing specs based on existing ones.
Signed-off-by: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/infiniband/hw/mlx4/main.c | 135 ++++++++++++++++++++++++++++++++++++-
1 files changed, 134 insertions(+), 1 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 093fc85..ecfd48b 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -55,6 +55,7 @@
#define DRV_RELDATE "April 4, 2008"
#define MLX4_IB_FLOW_MAX_PRIO 0xFFF
+#define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
MODULE_AUTHOR("Roland Dreier");
MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
@@ -828,6 +829,7 @@ struct mlx4_ib_steering {
};
static int parse_flow_attr(struct mlx4_dev *dev,
+ u32 qp_num,
union ib_flow_spec *ib_spec,
struct _rule_hw *mlx4_spec)
{
@@ -843,6 +845,14 @@ static int parse_flow_attr(struct mlx4_dev *dev,
mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
break;
+ case IB_FLOW_SPEC_IB:
+ type = MLX4_NET_TRANS_RULE_ID_IB;
+ mlx4_spec->ib.l3_qpn =
+ cpu_to_be32(qp_num);
+ mlx4_spec->ib.qpn_mask =
+ cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
+ break;
+
case IB_FLOW_SPEC_IPV4:
type = MLX4_NET_TRANS_RULE_ID_IPV4;
@@ -874,6 +884,115 @@ static int parse_flow_attr(struct mlx4_dev *dev,
return mlx4_hw_rule_sz(dev, type);
}
+struct default_rules {
+ __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
+ __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
+ __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
+ __u8 link_layer;
+};
+static const struct default_rules default_table[] = {
+ {
+ .mandatory_fields = {IB_FLOW_SPEC_IPV4},
+ .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
+ .rules_create_list = {IB_FLOW_SPEC_IB},
+ .link_layer = IB_LINK_LAYER_INFINIBAND
+ }
+};
+
+static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
+ struct ib_flow_attr *flow_attr)
+{
+ int i, j, k;
+ void *ib_flow;
+ const struct default_rules *pdefault_rules = default_table;
+ u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
+
+ for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
+ pdefault_rules++) {
+ __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
+ memset(&field_types, 0, sizeof(field_types));
+
+ if (link_layer != pdefault_rules->link_layer)
+ continue;
+
+ ib_flow = flow_attr + 1;
+ /* we assume the specs are sorted */
+ for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
+ j < flow_attr->num_of_specs; k++) {
+ union ib_flow_spec *current_flow =
+ (union ib_flow_spec *)ib_flow;
+
+ /* same layer but different type */
+ if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
+ (pdefault_rules->mandatory_fields[k] &
+ IB_FLOW_SPEC_LAYER_MASK)) &&
+ (current_flow->type !=
+ pdefault_rules->mandatory_fields[k]))
+ goto out;
+
+ /* same layer, try match next one */
+ if (current_flow->type ==
+ pdefault_rules->mandatory_fields[k]) {
+ j++;
+ ib_flow +=
+ ((union ib_flow_spec *)ib_flow)->size;
+ }
+ }
+
+ ib_flow = flow_attr + 1;
+ for (j = 0; j < flow_attr->num_of_specs;
+ j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
+ for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
+ /* same layer and same type */
+ if (((union ib_flow_spec *)ib_flow)->type ==
+ pdefault_rules->mandatory_not_fields[k])
+ goto out;
+
+ return i;
+ }
+out:
+ return -1;
+}
+
+static int __mlx4_ib_create_default_rules(
+ struct mlx4_ib_dev *mdev,
+ struct ib_qp *qp,
+ const struct default_rules *pdefault_rules,
+ struct _rule_hw *mlx4_spec) {
+ int size = 0;
+ int i;
+
+ for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
+ sizeof(pdefault_rules->rules_create_list[0]); i++) {
+ int ret;
+ union ib_flow_spec ib_spec;
+ switch (pdefault_rules->rules_create_list[i]) {
+ case 0:
+ /* no rule */
+ continue;
+ case IB_FLOW_SPEC_IB:
+ ib_spec.type = IB_FLOW_SPEC_IB;
+ ib_spec.size = sizeof(struct ib_flow_spec_ib);
+
+ break;
+ default:
+ /* invalid rule */
+ return -EINVAL;
+ }
+ /* We must put empty rule, qpn is being ignored */
+ ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
+ mlx4_spec);
+ if (ret < 0) {
+ pr_info("invalid parsing\n");
+ return -EINVAL;
+ }
+
+ mlx4_spec = (void *)mlx4_spec + ret;
+ size += ret;
+ }
+ return size;
+}
+
static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
int domain,
enum mlx4_net_trans_promisc_mode flow_type,
@@ -887,6 +1006,7 @@ static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_att
struct mlx4_net_trans_rule_hw_ctrl *ctrl;
size_t rule_size = sizeof(struct mlx4_net_trans_rule_hw_ctrl) +
(sizeof(struct _rule_hw) * flow_attr->num_of_specs);
+ int default_flow;
static const u16 __mlx4_domain[] = {
[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
@@ -922,8 +1042,21 @@ static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_att
ib_flow = flow_attr + 1;
size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
+ /* Add default flows */
+ default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
+ if (default_flow >= 0) {
+ ret = __mlx4_ib_create_default_rules(
+ mdev, qp, default_table + default_flow,
+ mailbox->buf + size);
+ if (ret < 0) {
+ mlx4_free_cmd_mailbox(mdev->dev, mailbox);
+ return -EINVAL;
+ }
+ size += ret;
+ }
for (i = 0; i < flow_attr->num_of_specs; i++) {
- ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
+ ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
+ mailbox->buf + size);
if (ret < 0) {
mlx4_free_cmd_mailbox(mdev->dev, mailbox);
return -EINVAL;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH for-next 6/6] IB/mlx4_ib: Add support for steer-able IB UD QPs
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
` (4 preceding siblings ...)
2013-11-07 13:25 ` [PATCH for-next 5/6] IB/mlx4_ib: Add mechanism to support flow-steering over IB links Or Gerlitz
@ 2013-11-07 13:25 ` Or Gerlitz
2013-11-16 19:35 ` [PATCH for-next 0/6] Add flow steering support for IPoIB Or Gerlitz
2014-01-14 22:09 ` Roland Dreier
7 siblings, 0 replies; 10+ messages in thread
From: Or Gerlitz @ 2013-11-07 13:25 UTC (permalink / raw)
To: roland-DgEjT+Ai2ygdnm+yROfE0A
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, matanb-VPRAkNaXOzVWk0Htik3J/w,
Or Gerlitz
From: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
This patch adds support for steer-able (NETIF) QPs creation.
When we create the device, we allocate a range of steer-able QPs.
Afterwards, when a QP is created with the NETIF flag, it's allocated
from this range. Allocation is managed by bitmap allocator.
Internal steering rule for those QPs is automatically-generated upon
their creation.
Signed-off-by: Matan Barak <matanb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
drivers/infiniband/hw/mlx4/main.c | 106 +++++++++++++++++++++++++++++++++-
drivers/infiniband/hw/mlx4/mlx4_ib.h | 11 ++++
drivers/infiniband/hw/mlx4/qp.c | 53 +++++++++++++++--
3 files changed, 163 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index ecfd48b..5573eb4 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1858,8 +1858,35 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
spin_lock_init(&ibdev->sm_lock);
mutex_init(&ibdev->cap_mask_mutex);
+ if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
+ ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
+ err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
+ MLX4_IB_UC_STEER_QPN_ALIGN,
+ &ibdev->steer_qpn_base);
+ if (err)
+ goto err_counter;
+
+ ibdev->ib_uc_qpns_bitmap =
+ kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
+ sizeof(long),
+ GFP_KERNEL);
+ if (!ibdev->ib_uc_qpns_bitmap) {
+ dev_err(&dev->pdev->dev, "bit map alloc failed\n");
+ goto err_steer_qp_release;
+ }
+
+ bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
+
+ err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
+ dev, ibdev->steer_qpn_base,
+ ibdev->steer_qpn_base +
+ ibdev->steer_qpn_count - 1);
+ if (err)
+ goto err_steer_free_bitmap;
+ }
+
if (ib_register_device(&ibdev->ib_dev, NULL))
- goto err_counter;
+ goto err_steer_free_bitmap;
if (mlx4_ib_mad_init(ibdev))
goto err_reg;
@@ -1910,6 +1937,13 @@ err_mad:
err_reg:
ib_unregister_device(&ibdev->ib_dev);
+err_steer_free_bitmap:
+ kfree(ibdev->ib_uc_qpns_bitmap);
+
+err_steer_qp_release:
+ if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
+ mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
+ ibdev->steer_qpn_count);
err_counter:
for (; i; --i)
if (ibdev->counters[i - 1] != -1)
@@ -1930,6 +1964,69 @@ err_dealloc:
return NULL;
}
+int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
+{
+ int offset;
+
+ WARN_ON(!dev->ib_uc_qpns_bitmap);
+
+ offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
+ dev->steer_qpn_count,
+ get_count_order(count));
+ if (offset < 0)
+ return offset;
+
+ *qpn = dev->steer_qpn_base + offset;
+ return 0;
+}
+
+void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
+{
+ if (!qpn ||
+ dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
+ return;
+
+ BUG_ON(qpn < dev->steer_qpn_base);
+
+ bitmap_release_region(dev->ib_uc_qpns_bitmap,
+ qpn - dev->steer_qpn_base,
+ get_count_order(count));
+}
+
+int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
+ int is_attach)
+{
+ int err;
+ size_t flow_size;
+ struct ib_flow_attr *flow = NULL;
+ struct ib_flow_spec_ib *ib_spec;
+
+ if (is_attach) {
+ flow_size = sizeof(struct ib_flow_attr) +
+ sizeof(struct ib_flow_spec_ib);
+ flow = kzalloc(flow_size, GFP_KERNEL);
+ if (!flow)
+ return -ENOMEM;
+ flow->port = mqp->port;
+ flow->num_of_specs = 1;
+ flow->size = flow_size;
+ ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
+ ib_spec->type = IB_FLOW_SPEC_IB;
+ ib_spec->size = sizeof(struct ib_flow_spec_ib);
+ /* Add an empty rule for IB L2 */
+ memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
+
+ err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
+ IB_FLOW_DOMAIN_NIC,
+ MLX4_FS_REGULAR,
+ &mqp->reg_id);
+ } else {
+ err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
+ }
+ kfree(flow);
+ return err;
+}
+
static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
{
struct mlx4_ib_dev *ibdev = ibdev_ptr;
@@ -1943,6 +2040,13 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
pr_warn("failure unregistering notifier\n");
ibdev->iboe.nb.notifier_call = NULL;
}
+
+ if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
+ mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
+ ibdev->steer_qpn_count);
+ kfree(ibdev->ib_uc_qpns_bitmap);
+ }
+
iounmap(ibdev->uar_map);
for (p = 0; p < ibdev->num_ports; ++p)
if (ibdev->counters[p] != -1)
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index e9fb396..837f9aa 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -68,6 +68,8 @@ enum {
/*module param to indicate if SM assigns the alias_GUID*/
extern int mlx4_ib_sm_guid_assign;
+#define MLX4_IB_UC_STEER_QPN_ALIGN 1
+#define MLX4_IB_UC_MAX_NUM_QPS 256
struct mlx4_ib_ucontext {
struct ib_ucontext ibucontext;
struct mlx4_uar uar;
@@ -153,6 +155,7 @@ struct mlx4_ib_wq {
enum mlx4_ib_qp_flags {
MLX4_IB_QP_LSO = IB_QP_CREATE_IPOIB_UD_LSO,
MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
+ MLX4_IB_QP_NETIF = IB_QP_CREATE_NETIF_QP,
MLX4_IB_SRIOV_TUNNEL_QP = 1 << 30,
MLX4_IB_SRIOV_SQP = 1 << 31,
};
@@ -270,6 +273,7 @@ struct mlx4_ib_qp {
struct list_head gid_list;
struct list_head steering_rules;
struct mlx4_ib_buf *sqp_proxy_rcv;
+ u64 reg_id;
};
@@ -494,6 +498,9 @@ struct mlx4_ib_dev {
struct kobject *dev_ports_parent[MLX4_MFUNC_MAX];
struct mlx4_ib_iov_port iov_ports[MLX4_MAX_PORTS];
struct pkey_mgt pkeys;
+ unsigned long *ib_uc_qpns_bitmap;
+ int steer_qpn_count;
+ int steer_qpn_base;
int steering_support;
};
@@ -753,5 +760,9 @@ void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device);
__be64 mlx4_ib_gen_node_guid(void);
+int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn);
+void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count);
+int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
+ int is_attach);
#endif /* MLX4_IB_H */
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 4f10af2..387fbf2 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -716,6 +716,14 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)
qp->flags |= MLX4_IB_QP_LSO;
+ if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
+ if (dev->steering_support ==
+ MLX4_STEERING_MODE_DEVICE_MANAGED)
+ qp->flags |= MLX4_IB_QP_NETIF;
+ else
+ goto err;
+ }
+
err = set_kernel_sq_size(dev, &init_attr->cap, qp_type, qp);
if (err)
goto err;
@@ -765,7 +773,11 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
if (init_attr->qp_type == IB_QPT_RAW_PACKET)
err = mlx4_qp_reserve_range(dev->dev, 1, 1 << 8, &qpn);
else
- err = mlx4_qp_reserve_range(dev->dev, 1, 1, &qpn);
+ if (qp->flags & MLX4_IB_QP_NETIF)
+ err = mlx4_ib_steer_qp_alloc(dev, 1, &qpn);
+ else
+ err = mlx4_qp_reserve_range(dev->dev, 1, 1,
+ &qpn);
if (err)
goto err_proxy;
}
@@ -790,8 +802,12 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
return 0;
err_qpn:
- if (!sqpn)
- mlx4_qp_release_range(dev->dev, qpn, 1);
+ if (!sqpn) {
+ if (qp->flags & MLX4_IB_QP_NETIF)
+ mlx4_ib_steer_qp_free(dev, qpn, 1);
+ else
+ mlx4_qp_release_range(dev->dev, qpn, 1);
+ }
err_proxy:
if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI)
free_proxy_bufs(pd->device, qp);
@@ -932,8 +948,12 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
mlx4_qp_free(dev->dev, &qp->mqp);
- if (!is_sqp(dev, qp) && !is_tunnel_qp(dev, qp))
- mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1);
+ if (!is_sqp(dev, qp) && !is_tunnel_qp(dev, qp)) {
+ if (qp->flags & MLX4_IB_QP_NETIF)
+ mlx4_ib_steer_qp_free(dev, qp->mqp.qpn, 1);
+ else
+ mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1);
+ }
mlx4_mtt_cleanup(dev->dev, &qp->mtt);
@@ -987,9 +1007,16 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
*/
if (init_attr->create_flags & ~(MLX4_IB_QP_LSO |
MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK |
- MLX4_IB_SRIOV_TUNNEL_QP | MLX4_IB_SRIOV_SQP))
+ MLX4_IB_SRIOV_TUNNEL_QP |
+ MLX4_IB_SRIOV_SQP |
+ MLX4_IB_QP_NETIF))
return ERR_PTR(-EINVAL);
+ if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
+ if (init_attr->qp_type != IB_QPT_UD)
+ return ERR_PTR(-EINVAL);
+ }
+
if (init_attr->create_flags &&
(udata ||
((init_attr->create_flags & ~MLX4_IB_SRIOV_SQP) &&
@@ -1235,6 +1262,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
struct mlx4_qp_context *context;
enum mlx4_qp_optpar optpar = 0;
int sqd_event;
+ int steer_qp = 0;
int err = -EINVAL;
context = kzalloc(sizeof *context, GFP_KERNEL);
@@ -1319,6 +1347,11 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
optpar |= MLX4_QP_OPTPAR_COUNTER_INDEX;
} else
context->pri_path.counter_index = 0xff;
+
+ if (qp->flags & MLX4_IB_QP_NETIF) {
+ mlx4_ib_steer_qp_reg(dev, qp, 1);
+ steer_qp = 1;
+ }
}
if (attr_mask & IB_QP_PKEY_INDEX) {
@@ -1547,9 +1580,14 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
qp->sq_next_wqe = 0;
if (qp->rq.wqe_cnt)
*qp->db.db = 0;
+
+ if (qp->flags & MLX4_IB_QP_NETIF)
+ mlx4_ib_steer_qp_reg(dev, qp, 0);
}
out:
+ if (err && steer_qp)
+ mlx4_ib_steer_qp_reg(dev, qp, 0);
kfree(context);
return err;
}
@@ -2762,6 +2800,9 @@ done:
if (qp->flags & MLX4_IB_QP_LSO)
qp_init_attr->create_flags |= IB_QP_CREATE_IPOIB_UD_LSO;
+ if (qp->flags & MLX4_IB_QP_NETIF)
+ qp_init_attr->create_flags |= IB_QP_CREATE_NETIF_QP;
+
qp_init_attr->sq_sig_type =
qp->sq_signal_bits == cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE) ?
IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [PATCH for-next 0/6] Add flow steering support for IPoIB
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
` (5 preceding siblings ...)
2013-11-07 13:25 ` [PATCH for-next 6/6] IB/mlx4_ib: Add support for steer-able IB UD QPs Or Gerlitz
@ 2013-11-16 19:35 ` Or Gerlitz
[not found] ` <CAJZOPZK=mRXakostsuFt3XzP3ZF5FW+4YurTSRz1JbPBgzurCg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2014-01-14 22:09 ` Roland Dreier
7 siblings, 1 reply; 10+ messages in thread
From: Or Gerlitz @ 2013-11-16 19:35 UTC (permalink / raw)
To: Roland Dreier; +Cc: linux-rdma, Matan Barak
On Thu, Nov 7,Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> wrote:
> This series adds on the flow-steering patches that were merged in 3.12
> Now flow steering rules can be set such that they steer UD
> traffic targeted to IPoIB netdevice into applications IB UD QP.
> For that end, IPoIB provides hint to the low-level device through
> a QP creation flag that this QP may be subject to flow-steering.
> FWIW, this patch set has nothing to do with the uverbs extensions series.
Any comment?! is this going in for 3.13?
Note that without this series which enables flow-steering also on IB
ports, folks that have VPI environments, e.g dual ported HCA with one
port playing Ethernet and the 2nd port IB, can't enjoy flow-steering
since we have only one dev cap bit and the flow-steering patches that
went into 3.12 only implemented the feature for raw qps, so the driver
(mlx4) could enable the device capability when both ports are IB.
Or.
>
> Matan and Or.
>
> Matan Barak (6):
> IB/core: Add flow-steering support for IPoIB UD traffic
> IB/core: Add support for IB L2 device managed steering
> net/mlx4_core: Add support for steer-able IB UD QPs
> IB/mlx4_ib: Enable device managed steering support for IB ports too
> IB/mlx4_ib: Add mechanism to support flow-steering over IB links
> IB/mlx4_ib: Add support for steer-able IB UD QPs
>
> drivers/infiniband/hw/mlx4/main.c | 272 +++++++++++++++++++-
> drivers/infiniband/hw/mlx4/mlx4_ib.h | 12 +
> drivers/infiniband/hw/mlx4/qp.c | 53 ++++-
> drivers/infiniband/ulp/ipoib/ipoib_verbs.c | 3 +
> drivers/net/ethernet/mellanox/mlx4/cmd.c | 9 +
> drivers/net/ethernet/mellanox/mlx4/fw.c | 10 +
> drivers/net/ethernet/mellanox/mlx4/mcg.c | 17 ++
> drivers/net/ethernet/mellanox/mlx4/mlx4.h | 5 +
> .../net/ethernet/mellanox/mlx4/resource_tracker.c | 10 +
> include/linux/mlx4/cmd.h | 1 +
> include/linux/mlx4/device.h | 6 +-
> include/rdma/ib_verbs.h | 17 ++-
> 12 files changed, 393 insertions(+), 22 deletions(-)
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH for-next 0/6] Add flow steering support for IPoIB
[not found] ` <1383830717-17312-1-git-send-email-ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
` (6 preceding siblings ...)
2013-11-16 19:35 ` [PATCH for-next 0/6] Add flow steering support for IPoIB Or Gerlitz
@ 2014-01-14 22:09 ` Roland Dreier
7 siblings, 0 replies; 10+ messages in thread
From: Roland Dreier @ 2014-01-14 22:09 UTC (permalink / raw)
To: Or Gerlitz; +Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, matanb
Applied this IB UD flow steering stuff, sorry for the delay.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 10+ messages in thread