[dpdk-dev] [PATCH v4 10/13] regex/mlx5: fastpath setup

[Ori Kam <orika@mellanox.com>]

From: Yuval Avnery <yuvalav@mellanox.com>

Allocated and register input/output buffers and metadata.

Signed-off-by: Yuval Avnery <yuvalav@mellanox.com>
Acked-by: Ori Kam <orika@mellanox.com>
---
 drivers/common/mlx5/mlx5_prm.h           |  36 ++++++
 drivers/regex/mlx5/Makefile              |   1 +
 drivers/regex/mlx5/meson.build           |   1 +
 drivers/regex/mlx5/mlx5_regex.h          |   8 ++
 drivers/regex/mlx5/mlx5_regex_control.c  |   2 +
 drivers/regex/mlx5/mlx5_regex_fastpath.c | 206 +++++++++++++++++++++++++++++++
 6 files changed, 254 insertions(+)
 create mode 100644 drivers/regex/mlx5/mlx5_regex_fastpath.c

diff --git a/drivers/common/mlx5/mlx5_prm.h b/drivers/common/mlx5/mlx5_prm.h
index f6f4ec0..bfbc58b 100644
--- a/drivers/common/mlx5/mlx5_prm.h
+++ b/drivers/common/mlx5/mlx5_prm.h
@@ -373,6 +373,42 @@ struct mlx5_cqe {
 	uint8_t op_own;
 };
 
+/* MMO metadata segment */
+
+#define	MLX5_OPCODE_MMO	0x2f
+#define	MLX5_OPC_MOD_MMO_REGEX 0x4
+
+struct mlx5_wqe_metadata_seg {
+	uint32_t mmo_control_31_0; /* mmo_control_63_32 is in ctrl_seg.imm */
+	uint32_t lkey;
+	uint64_t addr;
+};
+
+struct mlx5_ifc_regexp_mmo_control_bits {
+	uint8_t reserved_at_31[0x2];
+	uint8_t le[0x1];
+	uint8_t reserved_at_28[0x1];
+	uint8_t subset_id_0[0xc];
+	uint8_t reserved_at_16[0x4];
+	uint8_t subset_id_1[0xc];
+	uint8_t ctrl[0x4];
+	uint8_t subset_id_2[0xc];
+	uint8_t reserved_at_16_1[0x4];
+	uint8_t subset_id_3[0xc];
+};
+
+struct mlx5_ifc_regexp_metadata_bits {
+	uint8_t rof_version[0x10];
+	uint8_t latency_count[0x10];
+	uint8_t instruction_count[0x10];
+	uint8_t primary_thread_count[0x10];
+	uint8_t match_count[0x8];
+	uint8_t detected_match_count[0x8];
+	uint8_t status[0x10];
+	uint8_t job_id[0x20];
+	uint8_t reserved[0x80];
+};
+
 /* Adding direct verbs to data-path. */
 
 /* CQ sequence number mask. */
diff --git a/drivers/regex/mlx5/Makefile b/drivers/regex/mlx5/Makefile
index 806db99..02947d5 100644
--- a/drivers/regex/mlx5/Makefile
+++ b/drivers/regex/mlx5/Makefile
@@ -11,6 +11,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_REGEX_PMD) += mlx5_regex.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_REGEX_PMD) += mlx5_rxp.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_REGEX_PMD) += mlx5_regex_devx.c
 SRCS-$(CONFIG_RTE_LIBRTE_MLX5_REGEX_PMD) += mlx5_regex_control.c
+SRCS-$(CONFIG_RTE_LIBRTE_MLX5_REGEX_PMD) += mlx5_regex_fastpath.c
 
 # Basic CFLAGS.
 CFLAGS += -O3
diff --git a/drivers/regex/mlx5/meson.build b/drivers/regex/mlx5/meson.build
index 6f01df7..7f800f2 100644
--- a/drivers/regex/mlx5/meson.build
+++ b/drivers/regex/mlx5/meson.build
@@ -14,6 +14,7 @@ sources = files(
 	'mlx5_rxp.c',
 	'mlx5_regex_devx.c',
 	'mlx5_regex_control.c',
+	'mlx5_regex_fastpath.c',
 )
 cflags_options = [
 	'-std=c11',
diff --git a/drivers/regex/mlx5/mlx5_regex.h b/drivers/regex/mlx5/mlx5_regex.h
index bf285a1..21bb02b 100644
--- a/drivers/regex/mlx5/mlx5_regex.h
+++ b/drivers/regex/mlx5/mlx5_regex.h
@@ -40,6 +40,11 @@ struct mlx5_regex_qp {
 	struct mlx5_regex_sq *sqs; /* Pointer to sq array. */
 	uint16_t nb_obj; /* Number of sq objects. */
 	struct mlx5_regex_cq cq; /* CQ struct. */
+	uint32_t free_sqs;
+	struct mlx5_regex_job *jobs;
+	struct ibv_mr *metadata;
+	struct ibv_mr *inputs;
+	struct ibv_mr *outputs;
 };
 
 struct mlx5_regex_db {
@@ -91,4 +96,7 @@ int mlx5_devx_regex_database_program(void *ctx, uint8_t engine,
 /* mlx5_regex_control.c */
 int mlx5_regex_qp_setup(struct rte_regexdev *dev, uint16_t qp_ind,
 			const struct rte_regexdev_qp_conf *cfg);
+
+/* mlx5_regex_fastpath.c */
+int mlx5_regexdev_setup_fastpath(struct mlx5_regex_priv *priv, uint32_t qp_id);
 #endif /* MLX5_REGEX_H */
diff --git a/drivers/regex/mlx5/mlx5_regex_control.c b/drivers/regex/mlx5/mlx5_regex_control.c
index d378f48..9b3f39e 100644
--- a/drivers/regex/mlx5/mlx5_regex_control.c
+++ b/drivers/regex/mlx5/mlx5_regex_control.c
@@ -352,6 +352,8 @@
 			goto error;
 		}
 	}
+
+	mlx5_regexdev_setup_fastpath(priv, qp_ind);
 	return 0;
 
 error:
diff --git a/drivers/regex/mlx5/mlx5_regex_fastpath.c b/drivers/regex/mlx5/mlx5_regex_fastpath.c
new file mode 100644
index 0000000..23809c9
--- /dev/null
+++ b/drivers/regex/mlx5/mlx5_regex_fastpath.c
@@ -0,0 +1,206 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2020 Mellanox Technologies, Ltd
+ */
+#include <unistd.h>
+#include <sys/mman.h>
+
+#include <rte_malloc.h>
+#include <rte_log.h>
+#include <rte_errno.h>
+#include <rte_bus_pci.h>
+#include <rte_pci.h>
+#include <rte_regexdev_driver.h>
+#include <rte_mbuf.h>
+
+
+#include <infiniband/mlx5dv.h>
+#include <mlx5_glue.h>
+#include <mlx5_common.h>
+#include <mlx5_prm.h>
+#include <strings.h>
+
+#include "mlx5_regex_utils.h"
+#include "mlx5_rxp.h"
+#include "mlx5_regex.h"
+
+/* Verbs header. */
+/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
+#ifdef PEDANTIC
+#pragma GCC diagnostic ignored "-Wpedantic"
+#endif
+#include <infiniband/mlx5dv.h>
+#ifdef PEDANTIC
+#pragma GCC diagnostic error "-Wpedantic"
+#endif
+
+#define MAX_WQE_INDEX 0xffff
+#define MLX5_REGEX_METADATA_SIZE 64
+#define MLX5_REGEX_MAX_INPUT (1<<14)
+#define MLX5_REGEX_MAX_OUTPUT (1<<11)
+
+#define MLX5_REGEX_WQE_METADATA_OFFSET 16
+#define MLX5_REGEX_WQE_GATHER_OFFSET 32
+#define MLX5_REGEX_WQE_SCATTER_OFFSET 48
+
+static inline uint32_t
+sq_size_get(struct mlx5_regex_sq *sq)
+{
+	return (1U << sq->log_nb_desc);
+}
+static inline uint32_t
+cq_size_get(struct mlx5_regex_cq *cq)
+{
+	return (1U << cq->log_nb_desc);
+}
+
+struct mlx5_regex_job {
+	uint64_t user_id;
+	uint8_t *input;
+	volatile uint8_t *output;
+	volatile uint8_t *metadata;
+} __rte_cached_aligned;
+
+static inline void
+set_data_seg(struct mlx5_wqe_data_seg *seg,
+	     uint32_t length, uint32_t lkey,
+	     uintptr_t address)
+{
+	seg->byte_count = rte_cpu_to_be_32(length);
+	seg->lkey = rte_cpu_to_be_32(lkey);
+	seg->addr = rte_cpu_to_be_64(address);
+}
+
+static inline void
+set_metadata_seg(struct mlx5_wqe_metadata_seg *seg,
+		 uint32_t mmo_control_31_0, uint32_t lkey,
+		 uintptr_t address)
+{
+	seg->mmo_control_31_0 = htobe32(mmo_control_31_0);
+	seg->lkey = rte_cpu_to_be_32(lkey);
+	seg->addr = rte_cpu_to_be_64(address);
+}
+
+static void
+setup_sqs(struct mlx5_regex_qp *queue)
+{
+	size_t sqid, entry;
+	uint32_t job_id;
+	for (sqid = 0; sqid < queue->nb_obj; sqid++) {
+		struct mlx5_regex_sq *sq = &queue->sqs[sqid];
+		uint8_t *wqe = (uint8_t *)sq->wqe;
+		for (entry = 0 ; entry < sq_size_get(sq); entry++) {
+			job_id = sqid * sq_size_get(sq) + entry;
+			struct mlx5_regex_job *job = &queue->jobs[job_id];
+
+			set_metadata_seg((struct mlx5_wqe_metadata_seg *)
+					 (wqe + MLX5_REGEX_WQE_METADATA_OFFSET),
+					 0, queue->metadata->lkey,
+					 (uintptr_t)job->metadata);
+			set_data_seg((struct mlx5_wqe_data_seg *)
+				     (wqe + MLX5_REGEX_WQE_GATHER_OFFSET),
+				     0, queue->inputs->lkey,
+				     (uintptr_t)job->input);
+			set_data_seg((struct mlx5_wqe_data_seg *)
+				     (wqe + MLX5_REGEX_WQE_SCATTER_OFFSET),
+				     MLX5_REGEX_MAX_OUTPUT,
+				     queue->outputs->lkey,
+				     (uintptr_t)job->output);
+			wqe += 64;
+		}
+		queue->free_sqs |= 1 << sqid;
+	}
+}
+
+static int
+setup_buffers(struct mlx5_regex_qp *qp, struct ibv_pd *pd)
+{
+	uint32_t i;
+	int err;
+
+	void *ptr = rte_calloc(__func__, qp->nb_desc,
+			       MLX5_REGEX_METADATA_SIZE,
+			       MLX5_REGEX_METADATA_SIZE);
+	if (!ptr)
+		return -ENOMEM;
+
+	qp->metadata = mlx5_glue->reg_mr(pd, ptr,
+					 MLX5_REGEX_METADATA_SIZE*qp->nb_desc,
+					 IBV_ACCESS_LOCAL_WRITE);
+	if (!qp->metadata) {
+		rte_free(ptr);
+		return -EINVAL;
+	}
+	ptr = rte_calloc(__func__, qp->nb_desc,
+			 MLX5_REGEX_MAX_INPUT,
+			 MLX5_REGEX_MAX_INPUT);
+
+	if (!ptr) {
+		err = -ENOMEM;
+		goto err_input;
+	}
+	qp->inputs = mlx5_glue->reg_mr(pd, ptr,
+				       MLX5_REGEX_MAX_INPUT*qp->nb_desc,
+				       IBV_ACCESS_LOCAL_WRITE);
+	if (!qp->inputs) {
+		rte_free(ptr);
+		err = -EINVAL;
+		goto err_input;
+	}
+
+	ptr = rte_calloc(__func__, qp->nb_desc,
+			 MLX5_REGEX_MAX_OUTPUT,
+			 MLX5_REGEX_MAX_OUTPUT);
+	if (!ptr) {
+		err = -ENOMEM;
+		goto err_output;
+	}
+	qp->outputs = mlx5_glue->reg_mr(pd, ptr,
+					MLX5_REGEX_MAX_OUTPUT*qp->nb_desc,
+					IBV_ACCESS_LOCAL_WRITE);
+	if (!qp->outputs) {
+		rte_free(ptr);
+		err = -EINVAL;
+		goto err_output;
+	}
+
+	/* distribute buffers to jobs */
+	for (i = 0; i < qp->nb_desc; i++) {
+		qp->jobs[i].input =
+			(uint8_t *)qp->inputs->addr +
+			(i % qp->nb_desc) * MLX5_REGEX_MAX_INPUT;
+		qp->jobs[i].output =
+			(uint8_t *)qp->outputs->addr +
+			(i % qp->nb_desc) * MLX5_REGEX_MAX_OUTPUT;
+		qp->jobs[i].metadata =
+			(uint8_t *)qp->metadata->addr +
+			(i % qp->nb_desc) * MLX5_REGEX_METADATA_SIZE;
+	}
+	return 0;
+
+err_output:
+	ptr = qp->inputs->addr;
+	rte_free(ptr);
+	mlx5_glue->dereg_mr(qp->inputs);
+err_input:
+	ptr = qp->metadata->addr;
+	rte_free(ptr);
+	mlx5_glue->dereg_mr(qp->metadata);
+	return err;
+}
+
+int
+mlx5_regexdev_setup_fastpath(struct mlx5_regex_priv *priv, uint32_t qp_id)
+{
+	struct mlx5_regex_qp *qp = &priv->qps[qp_id];
+	int err;
+
+	qp->jobs = rte_calloc(__func__, qp->nb_desc, sizeof(*qp->jobs),
+			      sizeof(*qp->jobs));
+	if (!qp->jobs)
+		return -ENOMEM;
+	err = setup_buffers(qp, priv->pd);
+	if (err)
+		return err;
+	setup_sqs(qp);
+	return 0;
+}
-- 
1.8.3.1