IB/mlx5: Enable Eth VFs to query their min-inline value for user-space

For some mlx5 HW models (CX4, CX4Lx), the VF driver needs to put part
of the packet headers on the TX descriptor so the e-switch can do proper
matching and steering. This is called "min-inline", it's advertized to
the VF by the FW and also enforced on them by the HW, such that if they
don't obey, their packets are dropped.

SRIOV VF libmlx5 instances should take into account the min-inline
value of their vports. For that end, we provide this value through
the vendor response part of init_ucontext command.

The min inline value is reported in a way which will let newer libmlx5
instances realize that they are running over an older kernel and act
accordingly (e.g apply some educated guess).

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 8727116a4cabfa539b3022e0a92d80609c95e775..9d8535385bb8bded0e6dc5161396034cdf524b9b 100644 (file)
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -53,6 +53,7 @@
 #include <linux/in.h>
 #include <linux/etherdevice.h>
 #include <linux/mlx5/fs.h>
+#include <linux/mlx5/vport.h>
 #include "mlx5_ib.h"
 
 #define DRIVER_NAME "mlx5_ib"
@@ -1202,6 +1203,14 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev,
                resp.response_length += sizeof(resp.cmds_supp_uhw);
        }
 
+       if (field_avail(typeof(resp), eth_min_inline, udata->outlen)) {
+               if (mlx5_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET) {
+                       mlx5_query_min_inline(dev->mdev, &resp.eth_min_inline);
+                       resp.eth_min_inline++;
+               }
+               resp.response_length += sizeof(resp.eth_min_inline);
+       }
+
        /*
         * We don't want to expose information from the PCI bar that is located
         * after 4096 bytes, so if the arch only supports larger pages, let's

diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index 85dc966ea70be14a07863c1d6ccc88e3431c517d..da7cd62bace746879e154829b5ecad5f6cd87c00 100644 (file)
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -90,6 +90,17 @@ enum mlx5_user_cmds_supp_uhw {
        MLX5_USER_CMDS_SUPP_UHW_CREATE_AH    = 1 << 1,
 };
 
+/* The eth_min_inline response value is set to off-by-one vs the FW
+ * returned value to allow user-space to deal with older kernels.
+ */
+enum mlx5_user_inline_mode {
+       MLX5_USER_INLINE_MODE_NA,
+       MLX5_USER_INLINE_MODE_NONE,
+       MLX5_USER_INLINE_MODE_L2,
+       MLX5_USER_INLINE_MODE_IP,
+       MLX5_USER_INLINE_MODE_TCP_UDP,
+};
+
 struct mlx5_ib_alloc_ucontext_resp {
        __u32   qp_tab_size;
        __u32   bf_reg_size;
@@ -106,7 +117,8 @@ struct mlx5_ib_alloc_ucontext_resp {
        __u32   response_length;
        __u8    cqe_version;
        __u8    cmds_supp_uhw;
-       __u16   reserved2;
+       __u8    eth_min_inline;
+       __u8    reserved2;
        __u64   hca_core_clock_offset;
        __u32   log_uar_size;
        __u32   num_uars_per_page;