mlx5: Introduce devx APIs to create and destroy EQs

Introduce mlx5dv_devx_create_eq() and mlx5dv_devx_destroy_eq() to create
and destroy EQ objects, respectively.

Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>

debian/ibverbs-providers.symbols

@@ -151,6 +151,8 @@ libmlx5.so.1 ibverbs-providers #MINVER#
  mlx5dv_dr_aso_other_domain_link@MLX5_1.22 38
  mlx5dv_dr_aso_other_domain_unlink@MLX5_1.22 38
  mlx5dv_devx_alloc_msi_vector@MLX5_1.23 40
+ mlx5dv_devx_create_eq@MLX5_1.23 40
+ mlx5dv_devx_destroy_eq@MLX5_1.23 40
  mlx5dv_devx_free_msi_vector@MLX5_1.23 40
 libefa.so.1 ibverbs-providers #MINVER#
 * Build-Depends-Package: libibverbs-dev

providers/mlx5/libmlx5.map

@@ -215,5 +215,7 @@ MLX5_1.22 {
 MLX5_1.23 {
 	global:
 		mlx5dv_devx_alloc_msi_vector;
+		mlx5dv_devx_create_eq;
+		mlx5dv_devx_destroy_eq;
 		mlx5dv_devx_free_msi_vector;
 } MLX5_1.22;

providers/mlx5/man/CMakeLists.txt

@@ -14,6 +14,7 @@ rdma_man_pages(
   mlx5dv_devx_alloc_msi_vector.3.md
   mlx5dv_devx_alloc_uar.3.md
   mlx5dv_devx_create_cmd_comp.3.md
+  mlx5dv_devx_create_eq.3.md
   mlx5dv_devx_create_event_channel.3.md
   mlx5dv_devx_get_event.3.md
   mlx5dv_devx_obj_create.3.md
@@ -61,6 +62,7 @@ rdma_alias_man_pages(
  mlx5dv_devx_alloc_msi_vector.3 mlx5dv_devx_free_msi_vector.3
  mlx5dv_devx_alloc_uar.3 mlx5dv_devx_free_uar.3
  mlx5dv_devx_create_cmd_comp.3 mlx5dv_devx_destroy_cmd_comp.3
+ mlx5dv_devx_create_eq.3 mlx5dv_devx_destroy_eq.3
  mlx5dv_devx_create_event_channel.3 mlx5dv_devx_destroy_event_channel.3
  mlx5dv_devx_create_cmd_comp.3 mlx5dv_devx_get_async_cmd_comp.3
  mlx5dv_devx_obj_create.3 mlx5dv_devx_general_cmd.3

providers/mlx5/man/mlx5dv_devx_create_eq.3.md

@@ -0,0 +1,83 @@
+---
+date: 2022-01-12
+footer: mlx5
+header: "mlx5 Programmer's Manual"
+tagline: Verbs
+layout: page
+license: 'Licensed under the OpenIB.org BSD license (FreeBSD Variant) - See COPYING.md'
+section: 3
+title: mlx5dv_devx_create_eq
+---
+
+# NAME
+
+mlx5dv_devx_create_eq - Create an EQ object
+
+mlx5dv_devx_destroy_eq - Destroy an EQ object
+
+# SYNOPSIS
+
+```c
+#include <infiniband/mlx5dv.h>
+
+struct mlx5dv_devx_eq *
+mlx5dv_devx_create_eq(struct ibv_context *ibctx, const void *in, size_t inlen,
+		      void *out, size_t outlen);
+
+int mlx5dv_devx_destroy_eq(struct mlx5dv_devx_eq *eq);
+
+```
+
+# DESCRIPTION
+
+Create / Destroy an EQ object. Upon creation, the caller prepares the in/out mail boxes based on the device
+specification format; For the input mailbox, caller needs to prepare all fields except "eqc.log_page_size"
+and the pas list, which will be set by the driver. The "eqc.intr" field should be used from the output of
+mlx5dv_devx_alloc_msi_vector().
+
+# ARGUMENTS
+*ibctx*
+:	RDMA device context to create the action on.
+
+*in*
+:	A buffer which contains the command's input data provided in a device specification format.
+
+*inlen*
+:	The size of *in* buffer in bytes.
+
+*out*
+:	A buffer which contains the command's output data according to the device specification format.
+
+*outlen*
+:	The size of *out* buffer in bytes.
+
+*eq*
+:	The  EQ object to work on.
+
+```c
+struct mlx5dv_devx_eq {
+    void *vaddr;
+};
+```
+
+*vaddr*
+:	EQ VA that was allocated in the driver for.
+
+# NOTES
+
+mlx5dv_devx_query_eqn() will not support vectors which are used by mlx5dv_devx_create_eq().
+
+# RETURN VALUE
+
+Upon success *mlx5dv_devx_create_eq* will return a new *struct mlx5dv_devx_eq*;
+On error NULL will be returned and errno will be set.
+
+Upon success *mlx5dv_devx_destroy_eq* will return 0, on error errno will be returned.
+
+# SEE ALSO
+
+*mlx5dv_devx_alloc_msi_vector(3)*, *mlx5dv_devx_query_eqn(3)*
+
+# AUTHOR
+
+Mark Zhang <markzhang@nvidia.com>

providers/mlx5/mlx5.h

@@ -918,6 +918,14 @@ struct mlx5_devx_msi_vector {
 	struct ibv_context *ibctx;
 };
 
+struct mlx5_devx_eq {
+	struct mlx5dv_devx_eq dv_eq;
+	struct ibv_context *ibctx;
+	uint64_t iova;
+	size_t size;
+	int eqn;
+};
+
 struct ibv_flow *
 _mlx5dv_create_flow(struct mlx5dv_flow_matcher *flow_matcher,
 		    struct mlx5dv_flow_match_parameters *match_value,
@@ -1568,6 +1576,10 @@ struct mlx5_dv_context_ops {
 	int (*map_ah_to_qp)(struct ibv_ah *ah, uint32_t qp_num);
 	struct mlx5dv_devx_msi_vector *(*devx_alloc_msi_vector)(struct ibv_context *ibctx);
 	int (*devx_free_msi_vector)(struct mlx5dv_devx_msi_vector *msi);
+	struct mlx5dv_devx_eq *(*devx_create_eq)(struct ibv_context *ibctx,
+						 const void *in, size_t inlen,
+						 void *out, size_t outlen);
+	int (*devx_destroy_eq)(struct mlx5dv_devx_eq *eq);
 };
 
 struct mlx5_dv_context_ops *mlx5_get_dv_ops(struct ibv_context *context);

providers/mlx5/mlx5_vfio.c

@@ -3185,6 +3185,116 @@ static int vfio_devx_free_msi_vector(struct mlx5dv_devx_msi_vector *msi)
 	return ret;
 }
 
+static struct mlx5dv_devx_eq *
+vfio_devx_create_eq(struct ibv_context *ibctx, const void *in, size_t inlen,
+		    void *out, size_t outlen)
+{
+	struct mlx5_vfio_context *ctx = to_mvfio_ctx(ibctx);
+	struct mlx5_devx_eq *eq;
+	void *eqc, *in_pas;
+	size_t inlen_pas;
+	uint64_t size;
+	__be64 *pas;
+	int err;
+
+	eqc = DEVX_ADDR_OF(create_eq_in, in, eq_context_entry);
+	if ((inlen < DEVX_ST_SZ_BYTES(create_eq_in)) ||
+	    (DEVX_GET(create_eq_in, in, opcode) != MLX5_CMD_OP_CREATE_EQ) ||
+	    (DEVX_GET(eqc, eqc, intr) == MLX5_VFIO_CMD_VEC_IDX)) {
+		errno = EINVAL;
+		return NULL;
+	}
+
+	size = max(roundup_pow_of_two(
+			   (1ULL << DEVX_GET(eqc, eqc, log_eq_size)) * MLX5_EQE_SIZE),
+		   ctx->iova_min_page_size);
+	if (size > SIZE_MAX) {
+		errno = ERANGE;
+		return NULL;
+	}
+
+	eq = calloc(1, sizeof(*eq));
+	if (!eq) {
+		errno = ENOMEM;
+		return NULL;
+	}
+
+	eq->size = size;
+	err = posix_memalign(&eq->dv_eq.vaddr,
+			     MLX5_ADAPTER_PAGE_SIZE, eq->size);
+	if (err) {
+		errno = err;
+		goto err_va;
+	}
+
+	err = iset_alloc_range(ctx->iova_alloc, eq->size, &eq->iova);
+	if (err)
+		goto err_range;
+
+	err = mlx5_vfio_register_mem(ctx, eq->dv_eq.vaddr, eq->iova, eq->size);
+	if (err)
+		goto err_reg;
+
+	inlen_pas = inlen + DEVX_FLD_SZ_BYTES(create_eq_in, pas[0]) * 1;
+	in_pas = calloc(1, inlen_pas);
+	if (!in_pas) {
+		errno = ENOMEM;
+		goto err_inpas;
+	}
+
+	memcpy(in_pas, in, inlen);
+	eqc = DEVX_ADDR_OF(create_eq_in, in_pas, eq_context_entry);
+	DEVX_SET(eqc, eqc, log_page_size,
+		 ilog32(eq->size - 1) - MLX5_ADAPTER_PAGE_SHIFT);
+
+	pas = (__be64 *)DEVX_ADDR_OF(create_eq_in, in_pas, pas);
+	pas[0] = htobe64(eq->iova);
+
+	err = mlx5_vfio_cmd_exec(ctx, in_pas, inlen_pas, out, outlen, 0);
+	if (err)
+		goto err_cmd;
+
+	free(in_pas);
+	eq->ibctx = ibctx;
+	eq->eqn = DEVX_GET(create_eq_out, out, eq_number);
+	return &eq->dv_eq;
+
+err_cmd:
+	free(in_pas);
+err_inpas:
+	mlx5_vfio_unregister_mem(ctx, eq->iova, eq->size);
+err_reg:
+	iset_insert_range(ctx->iova_alloc, eq->iova, eq->size);
+err_range:
+	free(eq->dv_eq.vaddr);
+err_va:
+	free(eq);
+	return NULL;
+}
+
+static int vfio_devx_destroy_eq(struct mlx5dv_devx_eq *dveq)
+{
+	struct mlx5_devx_eq *eq =
+		container_of(dveq, struct mlx5_devx_eq, dv_eq);
+	struct mlx5_vfio_context *ctx = to_mvfio_ctx(eq->ibctx);
+	uint32_t out[DEVX_ST_SZ_DW(destroy_eq_out)] = {};
+	uint32_t in[DEVX_ST_SZ_DW(destroy_eq_in)] = {};
+	int err;
+
+	DEVX_SET(destroy_eq_in, in, opcode, MLX5_CMD_OP_DESTROY_EQ);
+	DEVX_SET(destroy_eq_in, in, eq_number, eq->eqn);
+
+	err = mlx5_vfio_cmd_exec(ctx, in, sizeof(in), out, sizeof(out), 0);
+	if (err)
+		return err;
+
+	mlx5_vfio_unregister_mem(ctx, eq->iova, eq->size);
+	iset_insert_range(ctx->iova_alloc, eq->iova, eq->size);
+	free(eq);
+
+	return 0;
+}
+
 static struct mlx5_dv_context_ops mlx5_vfio_dv_ctx_ops = {
 	.devx_general_cmd = vfio_devx_general_cmd,
 	.devx_obj_create = vfio_devx_obj_create,
@@ -3200,6 +3310,8 @@ static struct mlx5_dv_context_ops mlx5_vfio_dv_ctx_ops = {
 	.init_obj = vfio_init_obj,
 	.devx_alloc_msi_vector = vfio_devx_alloc_msi_vector,
 	.devx_free_msi_vector = vfio_devx_free_msi_vector,
+	.devx_create_eq = vfio_devx_create_eq,
+	.devx_destroy_eq = vfio_devx_destroy_eq,
 };
 
 static void mlx5_vfio_uninit_context(struct mlx5_vfio_context *ctx)

providers/mlx5/mlx5dv.h

@@ -2145,6 +2145,15 @@ mlx5dv_devx_alloc_msi_vector(struct ibv_context *ibctx);
 
 int mlx5dv_devx_free_msi_vector(struct mlx5dv_devx_msi_vector *msi);
 
+struct mlx5dv_devx_eq {
+	void *vaddr;
+};
+
+struct mlx5dv_devx_eq *
+mlx5dv_devx_create_eq(struct ibv_context *ibctx, const void *in, size_t inlen,
+		      void *out, size_t outlen);
+
+int mlx5dv_devx_destroy_eq(struct mlx5dv_devx_eq *eq);
 
 #ifdef __cplusplus
 }

providers/mlx5/verbs.c

@@ -7360,6 +7360,32 @@ int mlx5dv_devx_free_msi_vector(struct mlx5dv_devx_msi_vector *dvmsi)
 	return dvops->devx_free_msi_vector(dvmsi);
 }
 
+struct mlx5dv_devx_eq *
+mlx5dv_devx_create_eq(struct ibv_context *ibctx, const void *in, size_t inlen,
+		      void *out, size_t outlen)
+{
+	struct mlx5_dv_context_ops *dvops = mlx5_get_dv_ops(ibctx);
+
+	if (!dvops || !dvops->devx_create_eq) {
+		errno = EOPNOTSUPP;
+		return NULL;
+	}
+
+	return dvops->devx_create_eq(ibctx, in, inlen, out, outlen);
+}
+
+int mlx5dv_devx_destroy_eq(struct mlx5dv_devx_eq *dveq)
+{
+	struct mlx5_devx_eq *eq =
+		container_of(dveq, struct mlx5_devx_eq, dv_eq);
+	struct mlx5_dv_context_ops *dvops = mlx5_get_dv_ops(eq->ibctx);
+
+	if (!dvops || !dvops->devx_destroy_eq)
+		return EOPNOTSUPP;
+
+	return dvops->devx_destroy_eq(dveq);
+}
+
 void mlx5_set_dv_ctx_ops(struct mlx5_dv_context_ops *ops)
 {
 	ops->devx_general_cmd = _mlx5dv_devx_general_cmd;