Merge branch 'mlx5_realtime_ts' into rdma.git for-next

Aharon Landau says:

====================
In case device supports only real-time timestamp, the kernel will fail to
create QP despite rdma-core requested such timestamp type.

It is because device returns free-running timestamp, and the conversion
from free-running to real-time is performed in the user space.

This series fixes it, by returning real-time timestamp.
====================

* mlx5_realtime_ts:
  RDMA/mlx5: Support real-time timestamp directly from the device
  RDMA/mlx5: Refactor get_ts_format functions to simplify code
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

drivers/infiniband/hw/mlx5/cq.c

@@ -725,7 +725,8 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 		return -EFAULT;
 
 	if ((ucmd.flags & ~(MLX5_IB_CREATE_CQ_FLAGS_CQE_128B_PAD |
-			    MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX)))
+			    MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX |
+			    MLX5_IB_CREATE_CQ_FLAGS_REAL_TIME_TS)))
 		return -EINVAL;
 
 	if ((ucmd.cqe_size != 64 && ucmd.cqe_size != 128) ||
@@ -826,6 +827,9 @@ static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 		cq->private_flags |= MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD;
 	}
 
+	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_REAL_TIME_TS)
+		cq->private_flags |= MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS;
+
 	MLX5_SET(create_cq_in, *cqb, uid, context->devx_uid);
 	return 0;
 

drivers/infiniband/hw/mlx5/main.c

@@ -1816,6 +1816,12 @@ static int set_ucontext_resp(struct ib_ucontext *uctx,
 	if (MLX5_CAP_GEN(dev->mdev, ece_support))
 		resp->comp_mask |= MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_ECE;
 
+	if (rt_supported(MLX5_CAP_GEN(dev->mdev, sq_ts_format)) &&
+	    rt_supported(MLX5_CAP_GEN(dev->mdev, rq_ts_format)) &&
+	    rt_supported(MLX5_CAP_ROCE(dev->mdev, qp_ts_format)))
+		resp->comp_mask |=
+			MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_REAL_TIME_TS;
+
 	resp->num_dyn_bfregs = bfregi->num_dyn_bfregs;
 
 	if (MLX5_CAP_GEN(dev->mdev, drain_sigerr))

drivers/infiniband/hw/mlx5/mlx5_ib.h

@@ -549,6 +549,7 @@ static inline const struct mlx5_umr_wr *umr_wr(const struct ib_send_wr *wr)
 
 enum mlx5_ib_cq_pr_flags {
 	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
+	MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS = 1 << 1,
 };
 
 struct mlx5_ib_cq {
@@ -1611,4 +1612,10 @@ static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
 		(MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
 		 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
 }
+
+static inline bool rt_supported(int ts_cap)
+{
+	return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
+	       ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
+}
 #endif /* MLX5_IB_H */

drivers/infiniband/hw/mlx5/qp.c

@@ -1173,69 +1173,79 @@ static void destroy_flow_rule_vport_sq(struct mlx5_ib_sq *sq)
 	sq->flow_rule = NULL;
 }
 
-static int get_rq_ts_format(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *send_cq)
+static bool fr_supported(int ts_cap)
 {
-	bool fr_supported =
-		MLX5_CAP_GEN(dev->mdev, rq_ts_format) ==
-			MLX5_RQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING ||
-		MLX5_CAP_GEN(dev->mdev, rq_ts_format) ==
-			MLX5_RQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
+	return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING ||
+	       ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
+}
 
-	if (send_cq->create_flags & IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION) {
-		if (!fr_supported) {
-			mlx5_ib_dbg(dev, "Free running TS format is not supported\n");
+static int get_ts_format(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+			 bool fr_sup, bool rt_sup)
+{
+	if (cq->private_flags & MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS) {
+		if (!rt_sup) {
+			mlx5_ib_dbg(dev,
+				    "Real time TS format is not supported\n");
+			return -EOPNOTSUPP;
+		}
+		return MLX5_TIMESTAMP_FORMAT_REAL_TIME;
+	}
+	if (cq->create_flags & IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION) {
+		if (!fr_sup) {
+			mlx5_ib_dbg(dev,
+				    "Free running TS format is not supported\n");
 			return -EOPNOTSUPP;
 		}
-		return MLX5_RQC_TIMESTAMP_FORMAT_FREE_RUNNING;
+		return MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
 	}
-	return fr_supported ? MLX5_RQC_TIMESTAMP_FORMAT_FREE_RUNNING :
-			      MLX5_RQC_TIMESTAMP_FORMAT_DEFAULT;
+	return fr_sup ? MLX5_TIMESTAMP_FORMAT_FREE_RUNNING :
+			MLX5_TIMESTAMP_FORMAT_DEFAULT;
+}
+
+static int get_rq_ts_format(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *recv_cq)
+{
+	u8 ts_cap = MLX5_CAP_GEN(dev->mdev, rq_ts_format);
+
+	return get_ts_format(dev, recv_cq, fr_supported(ts_cap),
+			     rt_supported(ts_cap));
 }
 
 static int get_sq_ts_format(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *send_cq)
 {
-	bool fr_supported =
-		MLX5_CAP_GEN(dev->mdev, sq_ts_format) ==
-			MLX5_SQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING ||
-		MLX5_CAP_GEN(dev->mdev, sq_ts_format) ==
-			MLX5_SQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
+	u8 ts_cap = MLX5_CAP_GEN(dev->mdev, sq_ts_format);
 
-	if (send_cq->create_flags & IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION) {
-		if (!fr_supported) {
-			mlx5_ib_dbg(dev, "Free running TS format is not supported\n");
-			return -EOPNOTSUPP;
-		}
-		return MLX5_SQC_TIMESTAMP_FORMAT_FREE_RUNNING;
-	}
-	return fr_supported ? MLX5_SQC_TIMESTAMP_FORMAT_FREE_RUNNING :
-			      MLX5_SQC_TIMESTAMP_FORMAT_DEFAULT;
+	return get_ts_format(dev, send_cq, fr_supported(ts_cap),
+			     rt_supported(ts_cap));
 }
 
 static int get_qp_ts_format(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *send_cq,
 			    struct mlx5_ib_cq *recv_cq)
 {
-	bool fr_supported =
-		MLX5_CAP_ROCE(dev->mdev, qp_ts_format) ==
-			MLX5_QP_TIMESTAMP_FORMAT_CAP_FREE_RUNNING ||
-		MLX5_CAP_ROCE(dev->mdev, qp_ts_format) ==
-			MLX5_QP_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
-	int ts_format = fr_supported ? MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING :
-				       MLX5_QPC_TIMESTAMP_FORMAT_DEFAULT;
-
-	if (recv_cq &&
-	    recv_cq->create_flags & IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION)
-		ts_format = MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING;
-
-	if (send_cq &&
-	    send_cq->create_flags & IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION)
-		ts_format = MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING;
-
-	if (ts_format == MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING &&
-	    !fr_supported) {
-		mlx5_ib_dbg(dev, "Free running TS format is not supported\n");
+	u8 ts_cap = MLX5_CAP_ROCE(dev->mdev, qp_ts_format);
+	bool fr_sup = fr_supported(ts_cap);
+	bool rt_sup = rt_supported(ts_cap);
+	u8 default_ts = fr_sup ? MLX5_TIMESTAMP_FORMAT_FREE_RUNNING :
+				 MLX5_TIMESTAMP_FORMAT_DEFAULT;
+	int send_ts_format =
+		send_cq ? get_ts_format(dev, send_cq, fr_sup, rt_sup) :
+			  default_ts;
+	int recv_ts_format =
+		recv_cq ? get_ts_format(dev, recv_cq, fr_sup, rt_sup) :
+			  default_ts;
+
+	if (send_ts_format < 0 || recv_ts_format < 0)
+		return -EOPNOTSUPP;
+
+	if (send_ts_format != MLX5_TIMESTAMP_FORMAT_DEFAULT &&
+	    recv_ts_format != MLX5_TIMESTAMP_FORMAT_DEFAULT &&
+	    send_ts_format != recv_ts_format) {
+		mlx5_ib_dbg(
+			dev,
+			"The send ts_format does not match the receive ts_format\n");
 		return -EOPNOTSUPP;
 	}
-	return ts_format;
+
+	return send_ts_format == default_ts ? recv_ts_format : send_ts_format;
 }
 
 static int create_raw_packet_qp_sq(struct mlx5_ib_dev *dev,

drivers/net/ethernet/mellanox/mlx5/core/en_main.c

@@ -644,8 +644,8 @@ int mlx5e_create_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param)
 		return -ENOMEM;
 
 	ts_format = mlx5_is_real_time_rq(mdev) ?
-		    MLX5_RQC_TIMESTAMP_FORMAT_REAL_TIME :
-		    MLX5_RQC_TIMESTAMP_FORMAT_FREE_RUNNING;
+			    MLX5_TIMESTAMP_FORMAT_REAL_TIME :
+			    MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
 	rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
 	wq  = MLX5_ADDR_OF(rqc, rqc, wq);
 
@@ -1188,8 +1188,8 @@ static int mlx5e_create_sq(struct mlx5_core_dev *mdev,
 		return -ENOMEM;
 
 	ts_format = mlx5_is_real_time_sq(mdev) ?
-		    MLX5_SQC_TIMESTAMP_FORMAT_REAL_TIME :
-		    MLX5_SQC_TIMESTAMP_FORMAT_FREE_RUNNING;
+			    MLX5_TIMESTAMP_FORMAT_REAL_TIME :
+			    MLX5_TIMESTAMP_FORMAT_FREE_RUNNING;
 	sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
 	wq = MLX5_ADDR_OF(sqc, sqc, wq);
 

drivers/net/ethernet/mellanox/mlx5/core/lib/clock.h

@@ -37,16 +37,18 @@ static inline bool mlx5_is_real_time_rq(struct mlx5_core_dev *mdev)
 {
 	u8 rq_ts_format_cap = MLX5_CAP_GEN(mdev, rq_ts_format);
 
-	return (rq_ts_format_cap == MLX5_RQ_TIMESTAMP_FORMAT_CAP_REAL_TIME  ||
-		rq_ts_format_cap == MLX5_RQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME);
+	return (rq_ts_format_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
+		rq_ts_format_cap ==
+			MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME);
 }
 
 static inline bool mlx5_is_real_time_sq(struct mlx5_core_dev *mdev)
 {
 	u8 sq_ts_format_cap = MLX5_CAP_GEN(mdev, sq_ts_format);
 
-	return (sq_ts_format_cap == MLX5_SQ_TIMESTAMP_FORMAT_CAP_REAL_TIME  ||
-		sq_ts_format_cap == MLX5_SQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME);
+	return (sq_ts_format_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
+		sq_ts_format_cap ==
+			MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME);
 }
 
 typedef ktime_t (*cqe_ts_to_ns)(struct mlx5_clock *, u64);

include/linux/mlx5/mlx5_ifc.h

@@ -953,9 +953,9 @@ struct mlx5_ifc_per_protocol_networking_offload_caps_bits {
 };
 
 enum {
-	MLX5_QP_TIMESTAMP_FORMAT_CAP_FREE_RUNNING               = 0x0,
-	MLX5_QP_TIMESTAMP_FORMAT_CAP_REAL_TIME                  = 0x1,
-	MLX5_QP_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME = 0x2,
+	MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING               = 0x0,
+	MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME                  = 0x1,
+	MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME = 0x2,
 };
 
 struct mlx5_ifc_roce_cap_bits {
@@ -1296,18 +1296,6 @@ enum {
 	MLX5_STEERING_FORMAT_CONNECTX_6DX = 1,
 };
 
-enum {
-	MLX5_SQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING               = 0x0,
-	MLX5_SQ_TIMESTAMP_FORMAT_CAP_REAL_TIME                  = 0x1,
-	MLX5_SQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME = 0x2,
-};
-
-enum {
-	MLX5_RQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING               = 0x0,
-	MLX5_RQ_TIMESTAMP_FORMAT_CAP_REAL_TIME                  = 0x1,
-	MLX5_RQ_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME = 0x2,
-};
-
 struct mlx5_ifc_cmd_hca_cap_bits {
 	u8         reserved_at_0[0x1f];
 	u8         vhca_resource_manager[0x1];
@@ -2944,9 +2932,9 @@ enum {
 };
 
 enum {
-	MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING = 0x0,
-	MLX5_QPC_TIMESTAMP_FORMAT_DEFAULT      = 0x1,
-	MLX5_QPC_TIMESTAMP_FORMAT_REAL_TIME    = 0x2,
+	MLX5_TIMESTAMP_FORMAT_FREE_RUNNING = 0x0,
+	MLX5_TIMESTAMP_FORMAT_DEFAULT      = 0x1,
+	MLX5_TIMESTAMP_FORMAT_REAL_TIME    = 0x2,
 };
 
 struct mlx5_ifc_qpc_bits {
@@ -3396,12 +3384,6 @@ enum {
 	MLX5_SQC_STATE_ERR  = 0x3,
 };
 
-enum {
-	MLX5_SQC_TIMESTAMP_FORMAT_FREE_RUNNING = 0x0,
-	MLX5_SQC_TIMESTAMP_FORMAT_DEFAULT      = 0x1,
-	MLX5_SQC_TIMESTAMP_FORMAT_REAL_TIME    = 0x2,
-};
-
 struct mlx5_ifc_sqc_bits {
 	u8         rlky[0x1];
 	u8         cd_master[0x1];
@@ -3507,12 +3489,6 @@ enum {
 	MLX5_RQC_STATE_ERR  = 0x3,
 };
 
-enum {
-	MLX5_RQC_TIMESTAMP_FORMAT_FREE_RUNNING = 0x0,
-	MLX5_RQC_TIMESTAMP_FORMAT_DEFAULT      = 0x1,
-	MLX5_RQC_TIMESTAMP_FORMAT_REAL_TIME    = 0x2,
-};
-
 struct mlx5_ifc_rqc_bits {
 	u8         rlky[0x1];
 	u8	   delay_drop_en[0x1];

include/linux/mlx5/qp.h

@@ -550,8 +550,8 @@ static inline const char *mlx5_qp_state_str(int state)
 static inline int mlx5_get_qp_default_ts(struct mlx5_core_dev *dev)
 {
 	return !MLX5_CAP_ROCE(dev, qp_ts_format) ?
-		       MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING :
-		       MLX5_QPC_TIMESTAMP_FORMAT_DEFAULT;
+		       MLX5_TIMESTAMP_FORMAT_FREE_RUNNING :
+		       MLX5_TIMESTAMP_FORMAT_DEFAULT;
 }
 
 #endif /* MLX5_QP_H */

include/uapi/rdma/mlx5-abi.h

@@ -102,6 +102,7 @@ enum mlx5_ib_alloc_ucontext_resp_mask {
 	MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_DUMP_FILL_MKEY    = 1UL << 1,
 	MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_ECE               = 1UL << 2,
 	MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_SQD2RTS           = 1UL << 3,
+	MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_REAL_TIME_TS	   = 1UL << 4,
 };
 
 enum mlx5_user_cmds_supp_uhw {
@@ -271,6 +272,7 @@ struct mlx5_ib_query_device_resp {
 enum mlx5_ib_create_cq_flags {
 	MLX5_IB_CREATE_CQ_FLAGS_CQE_128B_PAD	= 1 << 0,
 	MLX5_IB_CREATE_CQ_FLAGS_UAR_PAGE_INDEX  = 1 << 1,
+	MLX5_IB_CREATE_CQ_FLAGS_REAL_TIME_TS	= 1 << 2,
 };
 
 struct mlx5_ib_create_cq {