[PATCH rdma-next] IB/mlx5: Add support for extended atomic operations

[PATCH rdma-next] IB/mlx5: Add support for extended atomic operations

[Leon Romanovsky]

From: Yonatan Cohen <yonatanc@mellanox.com>

Extended atomic operations cmp&swp and fetch&add is a Mellanox
feature extending the standard atomic operation to use, varied
operand sizes, as apposed to normal atomic operation that use
an 8 byte operand only.
Extended atomics allows masking the results and arguments.

This patch configures QP to support extended atomic operation
with the maximum size possible, as exposed by HCA capabilities.

Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 96 +++++++++++++++++++++++++++++++++++------
 include/linux/mlx5/driver.h     | 23 +++++-----
 2 files changed, 95 insertions(+), 24 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 1a6f34ea2f5d..027697c5b75b 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1733,6 +1733,53 @@ static void configure_requester_scat_cqe(struct mlx5_ib_dev *dev,
 		MLX5_SET(qpc, qpc, cs_req, MLX5_REQ_SCAT_DATA32_CQE);
 }

+static int atomic_size_to_mode(int size_mask)
+{
+	/* driver does not support atomic_size > 256B
+	 * and does not know how to translate bigger sizes
+	 */
+	int supported_size_mask = size_mask & 0x1ff;
+	int log_max_size;
+
+	if (!supported_size_mask)
+		return -EOPNOTSUPP;
+
+	log_max_size = __fls(supported_size_mask);
+
+	if (log_max_size > 3)
+		return log_max_size;
+
+	return MLX5_ATOMIC_MODE_8B;
+}
+
+static int get_atomic_mode(struct mlx5_ib_dev *dev,
+			   enum ib_qp_type qp_type)
+{
+	u8 atomic_operations = MLX5_CAP_ATOMIC(dev->mdev, atomic_operations);
+	u8 atomic = MLX5_CAP_GEN(dev->mdev, atomic);
+	int atomic_mode = -EOPNOTSUPP;
+	int atomic_size_mask;
+
+	if (!atomic)
+		return -EOPNOTSUPP;
+
+	if (qp_type == MLX5_IB_QPT_DCT)
+		atomic_size_mask = MLX5_CAP_ATOMIC(dev->mdev, atomic_size_dc);
+	else
+		atomic_size_mask = MLX5_CAP_ATOMIC(dev->mdev, atomic_size_qp);
+
+	if ((atomic_operations & MLX5_ATOMIC_OPS_EXTENDED_CMP_SWAP) ||
+	    (atomic_operations & MLX5_ATOMIC_OPS_EXTENDED_FETCH_ADD))
+		atomic_mode = atomic_size_to_mode(atomic_size_mask);
+
+	if (atomic_mode <= 0 &&
+	    (atomic_operations & MLX5_ATOMIC_OPS_CMP_SWAP &&
+	     atomic_operations & MLX5_ATOMIC_OPS_FETCH_ADD))
+		atomic_mode = MLX5_ATOMIC_MODE_IB_COMP;
+
+	return atomic_mode;
+}
+
 static inline bool check_flags_mask(uint64_t input, uint64_t supported)
 {
 	return (input & ~supported) == 0;
@@ -2562,13 +2609,15 @@ int mlx5_ib_destroy_qp(struct ib_qp *qp)
 	return 0;
 }

-static __be32 to_mlx5_access_flags(struct mlx5_ib_qp *qp, const struct ib_qp_attr *attr,
-				   int attr_mask)
+static int to_mlx5_access_flags(struct mlx5_ib_qp *qp,
+				const struct ib_qp_attr *attr,
+				int attr_mask, __be32 *hw_access_flags)
 {
-	u32 hw_access_flags = 0;
 	u8 dest_rd_atomic;
 	u32 access_flags;

+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.device);
+
 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
 		dest_rd_atomic = attr->max_dest_rd_atomic;
 	else
@@ -2583,13 +2632,25 @@ static __be32 to_mlx5_access_flags(struct mlx5_ib_qp *qp, const struct ib_qp_att
 		access_flags &= IB_ACCESS_REMOTE_WRITE;

 	if (access_flags & IB_ACCESS_REMOTE_READ)
-		hw_access_flags |= MLX5_QP_BIT_RRE;
-	if (access_flags & IB_ACCESS_REMOTE_ATOMIC)
-		hw_access_flags |= (MLX5_QP_BIT_RAE | MLX5_ATOMIC_MODE_CX);
+		*hw_access_flags |= MLX5_QP_BIT_RRE;
+	if ((access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
+	    qp->ibqp.qp_type == IB_QPT_RC) {
+		int atomic_mode;
+
+		atomic_mode = get_atomic_mode(dev, qp->ibqp.qp_type);
+		if (atomic_mode < 0)
+			return -EOPNOTSUPP;
+
+		*hw_access_flags |= MLX5_QP_BIT_RAE;
+		*hw_access_flags |= atomic_mode << MLX5_ATOMIC_MODE_OFFSET;
+	}
+
 	if (access_flags & IB_ACCESS_REMOTE_WRITE)
-		hw_access_flags |= MLX5_QP_BIT_RWE;
+		*hw_access_flags |= MLX5_QP_BIT_RWE;
+
+	*hw_access_flags = cpu_to_be32(*hw_access_flags);

-	return cpu_to_be32(hw_access_flags);
+	return 0;
 }

 enum {
@@ -3287,8 +3348,15 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 				cpu_to_be32(fls(attr->max_dest_rd_atomic - 1) << 21);
 	}

-	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC))
-		context->params2 |= to_mlx5_access_flags(qp, attr, attr_mask);
+	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC)) {
+		__be32 access_flags = 0;
+
+		err = to_mlx5_access_flags(qp, attr, attr_mask, &access_flags);
+		if (err)
+			goto out;
+
+		context->params2 |= access_flags;
+	}

 	if (attr_mask & IB_QP_MIN_RNR_TIMER)
 		context->rnr_nextrecvpsn |= cpu_to_be32(attr->min_rnr_timer << 24);
@@ -3504,10 +3572,14 @@ static int mlx5_ib_modify_dct(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
 			MLX5_SET(dctc, dctc, rwe, 1);
 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC) {
-			if (!mlx5_ib_dc_atomic_is_supported(dev))
+			int atomic_mode;
+
+			atomic_mode = get_atomic_mode(dev, MLX5_IB_QPT_DCT);
+			if (atomic_mode < 0)
 				return -EOPNOTSUPP;
+
+			MLX5_SET(dctc, dctc, atomic_mode, atomic_mode);
 			MLX5_SET(dctc, dctc, rae, 1);
-			MLX5_SET(dctc, dctc, atomic_mode, MLX5_ATOMIC_MODE_DCT_CX);
 		}
 		MLX5_SET(dctc, dctc, pkey_index, attr->pkey_index);
 		MLX5_SET(dctc, dctc, port, attr->port_num);
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index e2fa59f98ff8..024668c993f2 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -97,14 +97,15 @@ enum {
 };

 enum {
-	MLX5_ATOMIC_MODE_IB_COMP	= 1 << 16,
-	MLX5_ATOMIC_MODE_CX		= 2 << 16,
-	MLX5_ATOMIC_MODE_8B		= 3 << 16,
-	MLX5_ATOMIC_MODE_16B		= 4 << 16,
-	MLX5_ATOMIC_MODE_32B		= 5 << 16,
-	MLX5_ATOMIC_MODE_64B		= 6 << 16,
-	MLX5_ATOMIC_MODE_128B		= 7 << 16,
-	MLX5_ATOMIC_MODE_256B		= 8 << 16,
+	MLX5_ATOMIC_MODE_OFFSET = 16,
+	MLX5_ATOMIC_MODE_IB_COMP = 1,
+	MLX5_ATOMIC_MODE_CX = 2,
+	MLX5_ATOMIC_MODE_8B = 3,
+	MLX5_ATOMIC_MODE_16B = 4,
+	MLX5_ATOMIC_MODE_32B = 5,
+	MLX5_ATOMIC_MODE_64B = 6,
+	MLX5_ATOMIC_MODE_128B = 7,
+	MLX5_ATOMIC_MODE_256B = 8,
 };

 enum {
@@ -162,13 +163,11 @@ enum mlx5_dcbx_oper_mode {
 	MLX5E_DCBX_PARAM_VER_OPER_AUTO  = 0x3,
 };

-enum mlx5_dct_atomic_mode {
-	MLX5_ATOMIC_MODE_DCT_CX         = 2,
-};
-
 enum {
 	MLX5_ATOMIC_OPS_CMP_SWAP	= 1 << 0,
 	MLX5_ATOMIC_OPS_FETCH_ADD	= 1 << 1,
+	MLX5_ATOMIC_OPS_EXTENDED_CMP_SWAP = 1 << 2,
+	MLX5_ATOMIC_OPS_EXTENDED_FETCH_ADD = 1 << 3,
 };

 enum mlx5_page_fault_resume_flags {

Re: [PATCH rdma-next] IB/mlx5: Add support for extended atomic operations

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 876 bytes --]

On Wed, 2018-10-10 at 09:25 +0300, Leon Romanovsky wrote:
> From: Yonatan Cohen <yonatanc@mellanox.com>
> 
> Extended atomic operations cmp&swp and fetch&add is a Mellanox
> feature extending the standard atomic operation to use, varied
> operand sizes, as apposed to normal atomic operation that use
> an 8 byte operand only.
> Extended atomics allows masking the results and arguments.
> 
> This patch configures QP to support extended atomic operation
> with the maximum size possible, as exposed by HCA capabilities.
> 
> Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
> Reviewed-by: Guy Levi <guyle@mellanox.com>
> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>

Thanks, applied to for-next.

-- 
Doug Ledford <dledford@redhat.com>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 833 bytes --]