[PATCH rdma-next 0/6] RDMA mlx5 driver updates rdma-next-2017-10-19

[PATCH rdma-next 0/6] RDMA mlx5 driver updates rdma-next-2017-10-19

[Leon Romanovsky]

This patch set adds two local to mlx5 features.

The first feature comes from Guy, who updated the mlx5 driver to support
CQE compression for 128B-sized CQEs in addition to already supported
64B-sized CQEs: commit 1cbe6fc86ccf ("IB/mlx5: Add support for CQE compressing").

The second feature comes from Maor, this feature enables receive stateless
offloading support for incoming tunneling traffic for raw ethernet QP.

The patches are available in the git repository at:
  git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git tags/rdma-next-2017-10-19-1

	Thanks
---------------------------------------

Guy Levi (3):
  IB/mlx5: Add 128B CQE compression and padding HW bits
  IB/mlx5: Support 128B CQE compression feature
  IB/mlx5: Support padded 128B CQE feature

Maor Gottlieb (3):
  IB/mlx5: Update tunnel offloads bits
  IB/mlx5: Add tunneling offloads support
  IB/mlx5: Add support for RSS on the inner packet

 drivers/infiniband/hw/mlx5/cq.c      | 33 +++++++++++++++++++-----
 drivers/infiniband/hw/mlx5/main.c    | 26 +++++++++++++++++--
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  7 +++++
 drivers/infiniband/hw/mlx5/qp.c      | 50 ++++++++++++++++++++++++++++++++----
 include/linux/mlx5/cq.h              |  6 +++--
 include/linux/mlx5/mlx5_ifc.h        |  8 ++++--
 include/uapi/rdma/mlx5-abi.h         | 29 ++++++++++++++++++---
 7 files changed, 138 insertions(+), 21 deletions(-)

--
2.14.2

--
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

[PATCH rdma-next 1/6] IB/mlx5: Add 128B CQE compression and padding HW bits

[Leon Romanovsky]

From: Guy Levi <guyle-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Adding new bits in mlx5_ifc_cmd_hca_cap to get the hardware capabilities
for:
 - compression_128: Support 128B CQE compression
 - cqe_128_always: Support 128B CQE padding

Signed-off-by: Guy Levi <guyle-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 include/linux/mlx5/mlx5_ifc.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index db655db45b77..d49928c314ed 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -1047,7 +1047,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 	u8	   num_of_uars_per_page[0x20];
 	u8         reserved_at_540[0x40];
 
-	u8         reserved_at_580[0x3f];
+	u8         reserved_at_580[0x3d];
+	u8         cqe_128_always[0x1];
+	u8         cqe_compression_128[0x1];
 	u8         cqe_compression[0x1];
 
 	u8         cqe_compression_timeout[0x10];
-- 
2.14.2

--
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

[PATCH rdma-next 2/6] IB/mlx5: Support 128B CQE compression feature

[Leon Romanovsky]

From: Guy Levi <guyle-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

In commit 1cbe6fc86ccf ("IB/mlx5: Add support for CQE compressing") the
concept of CQE compression was introduced and added a support for 64B
CQE size. This change update the code to support 128B CQE size as well.

Signed-off-by: Guy Levi <guyle-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/hw/mlx5/cq.c   | 6 ++++--
 drivers/infiniband/hw/mlx5/main.c | 8 ++++++--
 include/uapi/rdma/mlx5-abi.h      | 7 ++++++-
 3 files changed, 16 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index b8a116d0e063..51871f049c57 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -804,8 +804,10 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 	*index = to_mucontext(context)->bfregi.sys_pages[0];
 
 	if (ucmd.cqe_comp_en == 1) {
-		if (unlikely((*cqe_size != 64) ||
-			     !MLX5_CAP_GEN(dev->mdev, cqe_compression))) {
+		if (!((*cqe_size == 128 &&
+		       MLX5_CAP_GEN(dev->mdev, cqe_compression_128)) ||
+		      (*cqe_size == 64  &&
+		       MLX5_CAP_GEN(dev->mdev, cqe_compression)))) {
 			err = -EOPNOTSUPP;
 			mlx5_ib_warn(dev, "CQE compression is not supported for size %d!\n",
 				     *cqe_size);
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index f66f693c058b..249c01c24eac 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -827,8 +827,12 @@ static int mlx5_ib_query_device(struct ib_device *ibdev,
 			sizeof(resp.mlx5_ib_support_multi_pkt_send_wqes);
 	}
 
-	if (field_avail(typeof(resp), reserved, uhw->outlen))
-		resp.response_length += sizeof(resp.reserved);
+	if (field_avail(typeof(resp), flags, uhw->outlen)) {
+		resp.response_length += sizeof(resp.flags);
+		if (MLX5_CAP_GEN(mdev, cqe_compression_128))
+			resp.flags |=
+				MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_COMP;
+	}
 
 	if (field_avail(typeof(resp), sw_parsing_caps,
 			uhw->outlen)) {
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index b1d5b87ba3fd..a8fc1f0956d0 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -203,6 +203,11 @@ struct mlx5_ib_striding_rq_caps {
 	__u32 supported_qpts;
 };
 
+enum mlx5_ib_query_dev_resp_flags {
+	/* Support 128B CQE compression */
+	MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_COMP = 1 << 0,
+};
+
 struct mlx5_ib_query_device_resp {
 	__u32	comp_mask;
 	__u32	response_length;
@@ -211,7 +216,7 @@ struct mlx5_ib_query_device_resp {
 	struct	mlx5_ib_cqe_comp_caps cqe_comp_caps;
 	struct	mlx5_packet_pacing_caps packet_pacing_caps;
 	__u32	mlx5_ib_support_multi_pkt_send_wqes;
-	__u32	reserved;
+	__u32	flags; /* Use enum mlx5_ib_query_dev_resp_flags */
 	struct mlx5_ib_sw_parsing_caps sw_parsing_caps;
 	struct mlx5_ib_striding_rq_caps striding_rq_caps;
 };
-- 
2.14.2

--
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

[PATCH rdma-next 3/6] IB/mlx5: Support padded 128B CQE feature

[Leon Romanovsky]

From: Guy Levi <guyle-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

In some benchmarks and some CPU architectures, writing the CQE on a full
cache line size improves performance by saving memory access operations
(read-modify-write) relative to partial cache line change. This patch
lets the user to configure the device to pad the CQE up to 128B in case
its content is less than 128B. Currently the driver supports only padding
for a CQE size of 128B.

Signed-off-by: Guy Levi <guyle-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/hw/mlx5/cq.c      | 27 +++++++++++++++++++++++----
 drivers/infiniband/hw/mlx5/main.c    |  4 ++++
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  5 +++++
 include/linux/mlx5/cq.h              |  6 ++++--
 include/uapi/rdma/mlx5-abi.h         |  7 ++++++-
 5 files changed, 42 insertions(+), 7 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index 51871f049c57..01b218a3c277 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -754,13 +754,13 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 	int err;
 
 	ucmdlen = udata->inlen < sizeof(ucmd) ?
-		  (sizeof(ucmd) - sizeof(ucmd.reserved)) : sizeof(ucmd);
+		  (sizeof(ucmd) - sizeof(ucmd.flags)) : sizeof(ucmd);
 
 	if (ib_copy_from_udata(&ucmd, udata, ucmdlen))
 		return -EFAULT;
 
 	if (ucmdlen == sizeof(ucmd) &&
-	    ucmd.reserved != 0)
+	    (ucmd.flags & ~(MLX5_IB_CREATE_CQ_FLAGS_CQE_128B_PAD)))
 		return -EINVAL;
 
 	if (ucmd.cqe_size != 64 && ucmd.cqe_size != 128)
@@ -830,6 +830,19 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 			 ilog2(ucmd.cqe_comp_res_format));
 	}
 
+	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_CQE_128B_PAD) {
+		if (*cqe_size != 128 ||
+		    !MLX5_CAP_GEN(dev->mdev, cqe_128_always)) {
+			err = -EOPNOTSUPP;
+			mlx5_ib_warn(dev,
+				     "CQE padding is not supported for CQE size of %dB!\n",
+				     *cqe_size);
+			goto err_cqb;
+		}
+
+		cq->private_flags |= MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD;
+	}
+
 	return 0;
 
 err_cqb:
@@ -989,7 +1002,10 @@ struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev,
 	cq->cqe_size = cqe_size;
 
 	cqc = MLX5_ADDR_OF(create_cq_in, cqb, cq_context);
-	MLX5_SET(cqc, cqc, cqe_sz, cqe_sz_to_mlx_sz(cqe_size));
+	MLX5_SET(cqc, cqc, cqe_sz,
+		 cqe_sz_to_mlx_sz(cqe_size,
+				  cq->private_flags &
+				  MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD));
 	MLX5_SET(cqc, cqc, log_cq_size, ilog2(entries));
 	MLX5_SET(cqc, cqc, uar_page, index);
 	MLX5_SET(cqc, cqc, c_eqn, eqn);
@@ -1339,7 +1355,10 @@ int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
 
 	MLX5_SET(cqc, cqc, log_page_size,
 		 page_shift - MLX5_ADAPTER_PAGE_SHIFT);
-	MLX5_SET(cqc, cqc, cqe_sz, cqe_sz_to_mlx_sz(cqe_size));
+	MLX5_SET(cqc, cqc, cqe_sz,
+		 cqe_sz_to_mlx_sz(cqe_size,
+				  cq->private_flags &
+				  MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD));
 	MLX5_SET(cqc, cqc, log_cq_size, ilog2(entries));
 
 	MLX5_SET(modify_cq_in, in, op_mod, MLX5_CQ_OPMOD_RESIZE);
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 249c01c24eac..107c038da328 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -829,9 +829,13 @@ static int mlx5_ib_query_device(struct ib_device *ibdev,
 
 	if (field_avail(typeof(resp), flags, uhw->outlen)) {
 		resp.response_length += sizeof(resp.flags);
+
 		if (MLX5_CAP_GEN(mdev, cqe_compression_128))
 			resp.flags |=
 				MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_COMP;
+
+		if (MLX5_CAP_GEN(mdev, cqe_128_always))
+			resp.flags |= MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_PAD;
 	}
 
 	if (field_avail(typeof(resp), sw_parsing_caps,
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index b30945c3a1fb..2063d1dfc101 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -446,6 +446,10 @@ struct mlx5_shared_mr_info {
 	struct ib_umem		*umem;
 };
 
+enum mlx5_ib_cq_pr_flags {
+	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
+};
+
 struct mlx5_ib_cq {
 	struct ib_cq		ibcq;
 	struct mlx5_core_cq	mcq;
@@ -468,6 +472,7 @@ struct mlx5_ib_cq {
 	struct list_head	wc_list;
 	enum ib_cq_notify_flags notify_flags;
 	struct work_struct	notify_work;
+	u16			private_flags; /* Use mlx5_ib_cq_pr_flags */
 };
 
 struct mlx5_ib_wc {
diff --git a/include/linux/mlx5/cq.h b/include/linux/mlx5/cq.h
index 95898847c7d4..cc718e245b1e 100644
--- a/include/linux/mlx5/cq.h
+++ b/include/linux/mlx5/cq.h
@@ -125,11 +125,13 @@ struct mlx5_cq_modify_params {
 enum {
 	CQE_SIZE_64 = 0,
 	CQE_SIZE_128 = 1,
+	CQE_SIZE_128_PAD = 2,
 };
 
-static inline int cqe_sz_to_mlx_sz(u8 size)
+static inline int cqe_sz_to_mlx_sz(u8 size, int padding_128_en)
 {
-	return size == 64 ? CQE_SIZE_64 : CQE_SIZE_128;
+	return padding_128_en ? CQE_SIZE_128_PAD :
+				size == 64 ? CQE_SIZE_64 : CQE_SIZE_128;
 }
 
 static inline void mlx5_cq_set_ci(struct mlx5_core_cq *cq)
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index a8fc1f0956d0..201a60f032dd 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -206,6 +206,7 @@ struct mlx5_ib_striding_rq_caps {
 enum mlx5_ib_query_dev_resp_flags {
 	/* Support 128B CQE compression */
 	MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_COMP = 1 << 0,
+	MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_PAD  = 1 << 1,
 };
 
 struct mlx5_ib_query_device_resp {
@@ -221,13 +222,17 @@ struct mlx5_ib_query_device_resp {
 	struct mlx5_ib_striding_rq_caps striding_rq_caps;
 };
 
+enum mlx5_ib_create_cq_flags {
+	MLX5_IB_CREATE_CQ_FLAGS_CQE_128B_PAD	= 1 << 0,
+};
+
 struct mlx5_ib_create_cq {
 	__u64	buf_addr;
 	__u64	db_addr;
 	__u32	cqe_size;
 	__u8    cqe_comp_en;
 	__u8    cqe_comp_res_format;
-	__u16	reserved; /* explicit padding (optional on i386) */
+	__u16	flags;
 };
 
 struct mlx5_ib_create_cq_resp {
-- 
2.14.2

--
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

[PATCH rdma-next 4/6] IB/mlx5: Update tunnel offloads bits

[Leon Romanovsky]

From: Maor Gottlieb <maorg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

This patch updates the mlx5_ifc with the following:
- Fix tunnel_stateless_gre typo.
- max_geneve_opt_len - Maximum geneve options length.
- tunnel_stateless_geneve_rx - If set, receive Stateless
  Offloads for Geneve tunneled (inner) packets are supported.

Signed-off-by: Maor Gottlieb <maorg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 include/linux/mlx5/mlx5_ifc.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index d49928c314ed..d4c29c183d28 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -614,7 +614,9 @@ struct mlx5_ifc_per_protocol_networking_offload_caps_bits {
 	u8         swp[0x1];
 	u8         swp_csum[0x1];
 	u8         swp_lso[0x1];
-	u8         reserved_at_23[0x1d];
+	u8         reserved_at_23[0x1b];
+	u8         max_geneve_opt_len[0x1];
+	u8         tunnel_stateless_geneve_rx[0x1];
 
 	u8         reserved_at_40[0x10];
 	u8         lro_min_mss_size[0x10];
-- 
2.14.2

--
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

[PATCH rdma-next 5/6] IB/mlx5: Add tunneling offloads support

[Leon Romanovsky]

From: Maor Gottlieb <maorg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

The device can support receive Stateless Offloads for the inner
packet's fields only when the packet is processed by TIR which is
enabled to support tunneling. Otherwise, the device treats the
packet as an ordinary non-tunneling packet and receive offloads
can be done only for the outer packet's field.
In order to enable receive Stateless Offloading support for incoming
tunneling traffic the TIR should be created with tunneled_offload_en.
Tunneling offloads is supported only be raw ethernet QP.

This patch includes:
* New QP creation flag for tunneling offloads.
* Reports device capabilities.

Signed-off-by: Maor Gottlieb <maorg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/hw/mlx5/main.c    | 14 +++++++++++++
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  2 ++
 drivers/infiniband/hw/mlx5/qp.c      | 39 +++++++++++++++++++++++++++++++-----
 include/uapi/rdma/mlx5-abi.h         | 11 +++++++++-
 4 files changed, 60 insertions(+), 6 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 107c038da328..c44dd0896f80 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -875,6 +875,20 @@ static int mlx5_ib_query_device(struct ib_device *ibdev,
 		}
 	}
 
+	if (field_avail(typeof(resp), tunnel_offloads_caps,
+			uhw->outlen)) {
+		resp.response_length += sizeof(resp.tunnel_offloads_caps);
+		if (MLX5_CAP_ETH(mdev, tunnel_stateless_vxlan))
+			resp.tunnel_offloads_caps |=
+				MLX5_IB_TUNNELED_OFFLOADS_VXLAN;
+		if (MLX5_CAP_ETH(mdev, tunnel_stateless_geneve_rx))
+			resp.tunnel_offloads_caps |=
+				MLX5_IB_TUNNELED_OFFLOADS_GENEVE;
+		if (MLX5_CAP_ETH(mdev, tunnel_stateless_gre))
+			resp.tunnel_offloads_caps |=
+				MLX5_IB_TUNNELED_OFFLOADS_GRE;
+	}
+
 	if (uhw->outlen) {
 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 2063d1dfc101..14b155413d7c 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -399,6 +399,7 @@ struct mlx5_ib_qp {
 	struct list_head	cq_send_list;
 	u32			rate_limit;
 	u32                     underlay_qpn;
+	bool			tunnel_offload_en;
 };
 
 struct mlx5_ib_cq_buf {
@@ -422,6 +423,7 @@ enum mlx5_ib_qp_flags {
 	MLX5_IB_QP_CVLAN_STRIPPING		= 1 << 9,
 	MLX5_IB_QP_UNDERLAY			= 1 << 10,
 	MLX5_IB_QP_SCATTER_END_PADDING		= 1 << 11,
+	MLX5_IB_QP_TUNNEL_OFFLOAD		= 1 << 12,
 };
 
 struct mlx5_umr_wr {
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 511d4fdcaaae..0f1106040a45 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1204,8 +1204,16 @@ static void destroy_raw_packet_qp_rq(struct mlx5_ib_dev *dev,
 	mlx5_core_destroy_rq_tracked(dev->mdev, &rq->base.mqp);
 }
 
+static bool tunnel_offload_supported(struct mlx5_core_dev *dev)
+{
+	return  (MLX5_CAP_ETH(dev, tunnel_stateless_vxlan) ||
+		 MLX5_CAP_ETH(dev, tunnel_stateless_gre) ||
+		 MLX5_CAP_ETH(dev, tunnel_stateless_geneve_rx));
+}
+
 static int create_raw_packet_qp_tir(struct mlx5_ib_dev *dev,
-				    struct mlx5_ib_rq *rq, u32 tdn)
+				    struct mlx5_ib_rq *rq, u32 tdn,
+				    bool tunnel_offload_en)
 {
 	u32 *in;
 	void *tirc;
@@ -1221,6 +1229,8 @@ static int create_raw_packet_qp_tir(struct mlx5_ib_dev *dev,
 	MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
 	MLX5_SET(tirc, tirc, inline_rqn, rq->base.mqp.qpn);
 	MLX5_SET(tirc, tirc, transport_domain, tdn);
+	if (tunnel_offload_en)
+		MLX5_SET(tirc, tirc, tunneled_offload_en, 1);
 
 	err = mlx5_core_create_tir(dev->mdev, in, inlen, &rq->tirn);
 
@@ -1273,7 +1283,8 @@ static int create_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 			goto err_destroy_sq;
 
 
-		err = create_raw_packet_qp_tir(dev, rq, tdn);
+		err = create_raw_packet_qp_tir(dev, rq, tdn,
+					       qp->tunnel_offload_en);
 		if (err)
 			goto err_destroy_rq;
 	}
@@ -1360,7 +1371,7 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	if (udata->outlen < min_resp_len)
 		return -EINVAL;
 
-	required_cmd_sz = offsetof(typeof(ucmd), reserved1) + sizeof(ucmd.reserved1);
+	required_cmd_sz = offsetof(typeof(ucmd), flags) + sizeof(ucmd.flags);
 	if (udata->inlen < required_cmd_sz) {
 		mlx5_ib_dbg(dev, "invalid inlen\n");
 		return -EINVAL;
@@ -1383,8 +1394,14 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 		return -EOPNOTSUPP;
 	}
 
-	if (memchr_inv(ucmd.reserved, 0, sizeof(ucmd.reserved)) || ucmd.reserved1) {
-		mlx5_ib_dbg(dev, "invalid reserved\n");
+	if (ucmd.flags & ~MLX5_QP_FLAG_TUNNEL_OFFLOADS) {
+		mlx5_ib_dbg(dev, "invalid flags\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS &&
+	    !tunnel_offload_supported(dev->mdev)) {
+		mlx5_ib_dbg(dev, "tunnel offloads isn't supported\n");
 		return -EOPNOTSUPP;
 	}
 
@@ -1407,6 +1424,10 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	MLX5_SET(tirc, tirc, transport_domain, tdn);
 
 	hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
+
+	if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)
+		MLX5_SET(tirc, tirc, tunneled_offload_en, 1);
+
 	switch (ucmd.rx_hash_function) {
 	case MLX5_RX_HASH_FUNC_TOEPLITZ:
 	{
@@ -1606,6 +1627,14 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 
 		qp->wq_sig = !!(ucmd.flags & MLX5_QP_FLAG_SIGNATURE);
 		qp->scat_cqe = !!(ucmd.flags & MLX5_QP_FLAG_SCATTER_CQE);
+		if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS) {
+			if (init_attr->qp_type != IB_QPT_RAW_PACKET ||
+			    !tunnel_offload_supported(mdev)) {
+				mlx5_ib_dbg(dev, "Tunnel offload isn't supported\n");
+				return -EOPNOTSUPP;
+			}
+			qp->tunnel_offload_en = true;
+		}
 
 		if (init_attr->create_flags & IB_QP_CREATE_SOURCE_QPN) {
 			if (init_attr->qp_type != IB_QPT_UD ||
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index 201a60f032dd..791655ec4aff 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -39,6 +39,7 @@
 enum {
 	MLX5_QP_FLAG_SIGNATURE		= 1 << 0,
 	MLX5_QP_FLAG_SCATTER_CQE	= 1 << 1,
+	MLX5_QP_FLAG_TUNNEL_OFFLOADS	= 1 << 2,
 };
 
 enum {
@@ -209,6 +210,12 @@ enum mlx5_ib_query_dev_resp_flags {
 	MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_PAD  = 1 << 1,
 };
 
+enum mlx5_ib_tunnel_offloads {
+	MLX5_IB_TUNNELED_OFFLOADS_VXLAN  = 1 << 0,
+	MLX5_IB_TUNNELED_OFFLOADS_GRE    = 1 << 1,
+	MLX5_IB_TUNNELED_OFFLOADS_GENEVE = 1 << 2
+};
+
 struct mlx5_ib_query_device_resp {
 	__u32	comp_mask;
 	__u32	response_length;
@@ -220,6 +227,8 @@ struct mlx5_ib_query_device_resp {
 	__u32	flags; /* Use enum mlx5_ib_query_dev_resp_flags */
 	struct mlx5_ib_sw_parsing_caps sw_parsing_caps;
 	struct mlx5_ib_striding_rq_caps striding_rq_caps;
+	__u32	tunnel_offloads_caps; /* enum mlx5_ib_tunnel_offloads */
+	__u32	reserved;
 };
 
 enum mlx5_ib_create_cq_flags {
@@ -304,7 +313,7 @@ struct mlx5_ib_create_qp_rss {
 	__u8 reserved[6];
 	__u8 rx_hash_key[128]; /* valid only for Toeplitz */
 	__u32   comp_mask;
-	__u32   reserved1;
+	__u32	flags;
 };
 
 struct mlx5_ib_create_qp_resp {
-- 
2.14.2

--
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

[PATCH rdma-next 6/6] IB/mlx5: Add support for RSS on the inner packet

[Leon Romanovsky]

From: Maor Gottlieb <maorg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

Some user space application would like to do RSS on the inner
packet fields instead on the outer.
When MLX5_RX_HASH_INNER is set with one or more of the other
hash fields, then the RSS will be done using the inner packet.

Signed-off-by: Maor Gottlieb <maorg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Reviewed-by: Mark Bloch <markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
---
 drivers/infiniband/hw/mlx5/qp.c | 11 +++++++++++
 include/uapi/rdma/mlx5-abi.h    |  4 +++-
 2 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 0f1106040a45..c6735a2979a8 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1405,6 +1405,12 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 		return -EOPNOTSUPP;
 	}
 
+	if (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_INNER &&
+	    !(ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)) {
+		mlx5_ib_dbg(dev, "Tunnel offloads must be set for inner RSS\n");
+		return -EOPNOTSUPP;
+	}
+
 	err = ib_copy_to_udata(udata, &resp, min_resp_len);
 	if (err) {
 		mlx5_ib_dbg(dev, "copy failed\n");
@@ -1428,6 +1434,11 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)
 		MLX5_SET(tirc, tirc, tunneled_offload_en, 1);
 
+	if (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_INNER)
+		hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_inner);
+	else
+		hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
+
 	switch (ucmd.rx_hash_function) {
 	case MLX5_RX_HASH_FUNC_TOEPLITZ:
 	{
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index 791655ec4aff..442b46b74a3a 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -303,7 +303,9 @@ enum mlx5_rx_hash_fields {
 	MLX5_RX_HASH_SRC_PORT_TCP	= 1 << 4,
 	MLX5_RX_HASH_DST_PORT_TCP	= 1 << 5,
 	MLX5_RX_HASH_SRC_PORT_UDP	= 1 << 6,
-	MLX5_RX_HASH_DST_PORT_UDP	= 1 << 7
+	MLX5_RX_HASH_DST_PORT_UDP	= 1 << 7,
+	/* Save bits for future fields */
+	MLX5_RX_HASH_INNER		= 1 << 31
 };
 
 struct mlx5_ib_create_qp_rss {
-- 
2.14.2

--
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

Re: [PATCH rdma-next 0/6] RDMA mlx5 driver updates rdma-next-2017-10-19

[Doug Ledford]

On Thu, 2017-10-19 at 08:25 +0300, Leon Romanovsky wrote:
> This patch set adds two local to mlx5 features.
> 
> The first feature comes from Guy, who updated the mlx5 driver to
> support
> CQE compression for 128B-sized CQEs in addition to already supported
> 64B-sized CQEs: commit 1cbe6fc86ccf ("IB/mlx5: Add support for CQE
> compressing").
> 
> The second feature comes from Maor, this feature enables receive
> stateless
> offloading support for incoming tunneling traffic for raw ethernet
> QP.
> 
> The patches are available in the git repository at:
>   git.kernel.org/pub/scm/linux/kernel/git/leon/linux-rdma.git
> tags/rdma-next-2017-10-19-1
> 
>         Thanks

Thanks, series applied (with a fixup because I haven't taken the PCI
write pad patch series yet).

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
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