[PATCH mlx5-next] RDMA/mlx5: Allow CQ creation without attached EQs

Subsystems: infiniband subsystem, mellanox mlx5 ib driver, mellanox mlx5 core vpi driver, the rest

STALE2002d LANDED

Landed in mainline as 7232c132d13a on 2021-02-16.

2 messages, 2 authors, 2021-02-12 · open the first message on its own page

[PATCH mlx5-next] RDMA/mlx5: Allow CQ creation without attached EQs

From: Leon Romanovsky <leon@kernel.org>
Date: 2021-02-11 08:57:10

From: Tal Gilboa <talgi@nvidia.com>

The traditional DevX CQ creation flow goes through mlx5_core_create_cq()
which checks that the given EQN corresponds to an existing EQ. For some
mlx5 devices this behaviour is too strict, they expect EQN assignment
during modify CQ stage.

Allow them to create CQ through general command interface.

Signed-off-by: Tal Gilboa <talgi@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/devx.c | 13 ++++++++++++-
 include/linux/mlx5/mlx5_ifc.h     |  5 +++--
 2 files changed, 15 insertions(+), 3 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index 526057a33edb..8152d0ddac2d 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -1439,6 +1439,16 @@ static void devx_cq_comp(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe)
 	rcu_read_unlock();
 }

+static bool is_apu_thread_cq(struct mlx5_ib_dev *dev, const void *in)
+{
+	if (!MLX5_CAP_GEN(dev->mdev, apu) ||
+	    !MLX5_GET(cqc, MLX5_ADDR_OF(create_cq_in, in, cq_context),
+		      apu_thread_cq))
+		return false;
+
+	return true;
+}
+
 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
 	struct uverbs_attr_bundle *attrs)
 {
@@ -1492,7 +1502,8 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
 		obj->flags |= DEVX_OBJ_FLAGS_DCT;
 		err = mlx5_core_create_dct(dev, &obj->core_dct, cmd_in,
 					   cmd_in_len, cmd_out, cmd_out_len);
-	} else if (opcode == MLX5_CMD_OP_CREATE_CQ) {
+	} else if (opcode == MLX5_CMD_OP_CREATE_CQ &&
+		   !is_apu_thread_cq(dev, cmd_in)) {
 		obj->flags |= DEVX_OBJ_FLAGS_CQ;
 		obj->core_cq.comp = devx_cq_comp;
 		err = mlx5_core_create_cq(dev->mdev, &obj->core_cq,
diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index ffe2c7231ae4..816893f34e79 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -1659,7 +1659,8 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 	u8         sf_set_partition[0x1];
 	u8         reserved_at_682[0x1];
 	u8         log_max_sf[0x5];
-	u8         reserved_at_688[0x8];
+	u8         apu[0x1];
+	u8         reserved_at_689[0x7];
 	u8         log_min_sf_size[0x8];
 	u8         max_num_sf_partitions[0x8];
@@ -3874,7 +3875,7 @@ struct mlx5_ifc_cqc_bits {
 	u8         status[0x4];
 	u8         reserved_at_4[0x2];
 	u8         dbr_umem_valid[0x1];
-	u8         reserved_at_7[0x1];
+	u8         apu_thread_cq[0x1];
 	u8         cqe_sz[0x3];
 	u8         cc[0x1];
 	u8         reserved_at_c[0x1];
--
2.29.2

Re: [PATCH mlx5-next] RDMA/mlx5: Allow CQ creation without attached EQs

From: Jason Gunthorpe <jgg@nvidia.com>
Date: 2021-02-12 18:09:17

On Thu, Feb 11, 2021 at 10:55:49AM +0200, Leon Romanovsky wrote:
From: Tal Gilboa <talgi@nvidia.com>

The traditional DevX CQ creation flow goes through mlx5_core_create_cq()
which checks that the given EQN corresponds to an existing EQ. For some
mlx5 devices this behaviour is too strict, they expect EQN assignment
during modify CQ stage.

Allow them to create CQ through general command interface.

Signed-off-by: Tal Gilboa <talgi@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/devx.c | 13 ++++++++++++-
 include/linux/mlx5/mlx5_ifc.h     |  5 +++--
 2 files changed, 15 insertions(+), 3 deletions(-)
Applied to for-next

Thanks,
Jason
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help