[dpdk-dev] [PATCH v5 3/8] crypto/bcmfs: add queue pair management API

[Vikas Gupta <vikas.gupta@broadcom.com>]

Add queue pair management APIs which will be used by Crypto device to
manage h/w queues. A bcmfs device structure owns multiple queue-pairs
based on the mapped address allocated to it.

Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Raveendra Padasalagi <raveendra.padasalagi@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
---
 drivers/crypto/bcmfs/bcmfs_device.c  |   4 +
 drivers/crypto/bcmfs/bcmfs_device.h  |   5 +
 drivers/crypto/bcmfs/bcmfs_hw_defs.h |  32 +++
 drivers/crypto/bcmfs/bcmfs_qp.c      | 345 +++++++++++++++++++++++++++
 drivers/crypto/bcmfs/bcmfs_qp.h      | 122 ++++++++++
 drivers/crypto/bcmfs/meson.build     |   3 +-
 6 files changed, 510 insertions(+), 1 deletion(-)
 create mode 100644 drivers/crypto/bcmfs/bcmfs_hw_defs.h
 create mode 100644 drivers/crypto/bcmfs/bcmfs_qp.c
 create mode 100644 drivers/crypto/bcmfs/bcmfs_qp.h

diff --git a/drivers/crypto/bcmfs/bcmfs_device.c b/drivers/crypto/bcmfs/bcmfs_device.c
index 0ccddea202..a01a5c79d5 100644
--- a/drivers/crypto/bcmfs/bcmfs_device.c
+++ b/drivers/crypto/bcmfs/bcmfs_device.c
@@ -12,6 +12,7 @@
 
 #include "bcmfs_device.h"
 #include "bcmfs_logs.h"
+#include "bcmfs_qp.h"
 #include "bcmfs_vfio.h"
 
 struct bcmfs_device_attr {
@@ -77,6 +78,9 @@ fsdev_allocate_one_dev(struct rte_vdev_device *vdev,
 	if (bcmfs_attach_vfio(fsdev))
 		goto cleanup;
 
+	/* Maximum number of QPs supported */
+	fsdev->max_hw_qps = fsdev->mmap_size / BCMFS_HW_QUEUE_IO_ADDR_LEN;
+
 	TAILQ_INSERT_TAIL(&fsdev_list, fsdev, next);
 
 	return fsdev;
diff --git a/drivers/crypto/bcmfs/bcmfs_device.h b/drivers/crypto/bcmfs/bcmfs_device.h
index f99d57d4bd..dede5b82dc 100644
--- a/drivers/crypto/bcmfs/bcmfs_device.h
+++ b/drivers/crypto/bcmfs/bcmfs_device.h
@@ -11,6 +11,7 @@
 #include <rte_bus_vdev.h>
 
 #include "bcmfs_logs.h"
+#include "bcmfs_qp.h"
 
 /* max number of dev nodes */
 #define BCMFS_MAX_NODES		4
@@ -44,6 +45,10 @@ struct bcmfs_device {
 	uint8_t *mmap_addr;
 	/* mapped size */
 	uint32_t mmap_size;
+	/* max number of h/w queue pairs detected */
+	uint16_t max_hw_qps;
+	/* current qpairs in use */
+	struct bcmfs_qp *qps_in_use[BCMFS_MAX_HW_QUEUES];
 };
 
 #endif /* _BCMFS_DEVICE_H_ */
diff --git a/drivers/crypto/bcmfs/bcmfs_hw_defs.h b/drivers/crypto/bcmfs/bcmfs_hw_defs.h
new file mode 100644
index 0000000000..7d5bb5d8fe
--- /dev/null
+++ b/drivers/crypto/bcmfs/bcmfs_hw_defs.h
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2020 Broadcom
+ * All rights reserved.
+ */
+
+#ifndef _BCMFS_HW_DEFS_H_
+#define _BCMFS_HW_DEFS_H_
+
+#include <rte_atomic.h>
+#include <rte_byteorder.h>
+#include <rte_common.h>
+#include <rte_io.h>
+
+#ifndef BIT
+#define BIT(nr)         (1UL << (nr))
+#endif
+
+#define FS_RING_REGS_SIZE		0x10000
+#define FS_RING_DESC_SIZE		8
+#define FS_RING_BD_ALIGN_ORDER		12
+#define FS_RING_BD_DESC_PER_REQ		32
+#define FS_RING_CMPL_ALIGN_ORDER	13
+#define FS_RING_CMPL_SIZE		(1024 * FS_RING_DESC_SIZE)
+#define FS_RING_MAX_REQ_COUNT		1024
+#define FS_RING_PAGE_SHFT		12
+#define FS_RING_PAGE_SIZE		BIT(FS_RING_PAGE_SHFT)
+
+/* Minimum and maximum number of requests supported */
+#define FS_RM_MAX_REQS			4096
+#define FS_RM_MIN_REQS			32
+
+#endif /* BCMFS_HW_DEFS_H_ */
diff --git a/drivers/crypto/bcmfs/bcmfs_qp.c b/drivers/crypto/bcmfs/bcmfs_qp.c
new file mode 100644
index 0000000000..864e7bb746
--- /dev/null
+++ b/drivers/crypto/bcmfs/bcmfs_qp.c
@@ -0,0 +1,345 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(C) 2020 Broadcom.
+ * All rights reserved.
+ */
+
+#include <inttypes.h>
+
+#include <rte_atomic.h>
+#include <rte_bitmap.h>
+#include <rte_common.h>
+#include <rte_dev.h>
+#include <rte_malloc.h>
+#include <rte_memzone.h>
+#include <rte_prefetch.h>
+#include <rte_string_fns.h>
+
+#include "bcmfs_logs.h"
+#include "bcmfs_qp.h"
+#include "bcmfs_hw_defs.h"
+
+/* TX or submission queue name */
+static const char *txq_name = "tx";
+/* Completion or receive queue name */
+static const char *cmplq_name = "cmpl";
+
+/* Helper function */
+static int
+bcmfs_qp_check_queue_alignment(uint64_t phys_addr,
+			       uint32_t align)
+{
+	if (((align - 1) & phys_addr) != 0)
+		return -EINVAL;
+	return 0;
+}
+
+static void
+bcmfs_queue_delete(struct bcmfs_queue *queue,
+		   uint16_t queue_pair_id)
+{
+	const struct rte_memzone *mz;
+	int status = 0;
+
+	if (queue == NULL) {
+		BCMFS_LOG(DEBUG, "Invalid queue");
+		return;
+	}
+	BCMFS_LOG(DEBUG, "Free ring %d type %d, memzone: %s",
+		  queue_pair_id, queue->q_type, queue->memz_name);
+
+	mz = rte_memzone_lookup(queue->memz_name);
+	if (mz != NULL)	{
+		/* Write an unused pattern to the queue memory. */
+		memset(queue->base_addr, 0x9B, queue->queue_size);
+		status = rte_memzone_free(mz);
+		if (status != 0)
+			BCMFS_LOG(ERR, "Error %d on freeing queue %s",
+					status, queue->memz_name);
+	} else {
+		BCMFS_LOG(DEBUG, "queue %s doesn't exist",
+				queue->memz_name);
+	}
+}
+
+static const struct rte_memzone *
+queue_dma_zone_reserve(const char *queue_name, uint32_t queue_size,
+		       int socket_id, unsigned int align)
+{
+	const struct rte_memzone *mz;
+
+	mz = rte_memzone_lookup(queue_name);
+	if (mz != NULL) {
+		if (((size_t)queue_size <= mz->len) &&
+		    (socket_id == SOCKET_ID_ANY ||
+		     socket_id == mz->socket_id)) {
+			BCMFS_LOG(DEBUG, "re-use memzone already "
+					"allocated for %s", queue_name);
+			return mz;
+		}
+
+		BCMFS_LOG(ERR, "Incompatible memzone already "
+				"allocated %s, size %u, socket %d. "
+				"Requested size %u, socket %u",
+				queue_name, (uint32_t)mz->len,
+				mz->socket_id, queue_size, socket_id);
+		return NULL;
+	}
+
+	BCMFS_LOG(DEBUG, "Allocate memzone for %s, size %u on socket %u",
+		  queue_name, queue_size, socket_id);
+	return rte_memzone_reserve_aligned(queue_name, queue_size,
+		socket_id, RTE_MEMZONE_IOVA_CONTIG, align);
+}
+
+static int
+bcmfs_queue_create(struct bcmfs_queue *queue,
+		   struct bcmfs_qp_config *qp_conf,
+		   uint16_t queue_pair_id,
+		   enum bcmfs_queue_type qtype)
+{
+	const struct rte_memzone *qp_mz;
+	char q_name[16];
+	unsigned int align;
+	uint32_t queue_size_bytes;
+	int ret;
+
+	if (qtype == BCMFS_RM_TXQ) {
+		strlcpy(q_name, txq_name, sizeof(q_name));
+		align = 1U << FS_RING_BD_ALIGN_ORDER;
+		queue_size_bytes = qp_conf->nb_descriptors *
+				   qp_conf->max_descs_req * FS_RING_DESC_SIZE;
+		queue_size_bytes = RTE_ALIGN_MUL_CEIL(queue_size_bytes,
+						      FS_RING_PAGE_SIZE);
+		/* make queue size to multiple for 4K pages */
+	} else if (qtype == BCMFS_RM_CPLQ) {
+		strlcpy(q_name, cmplq_name, sizeof(q_name));
+		align = 1U << FS_RING_CMPL_ALIGN_ORDER;
+
+		/*
+		 * Memory size for cmpl + MSI
+		 * For MSI allocate here itself and so we allocate twice
+		 */
+		queue_size_bytes = 2 * FS_RING_CMPL_SIZE;
+	} else {
+		BCMFS_LOG(ERR, "Invalid queue selection");
+		return -EINVAL;
+	}
+
+	queue->q_type = qtype;
+
+	/*
+	 * Allocate a memzone for the queue - create a unique name.
+	 */
+	snprintf(queue->memz_name, sizeof(queue->memz_name),
+		 "%s_%d_%s_%d_%s", "bcmfs", qtype, "qp_mem",
+		 queue_pair_id, q_name);
+	qp_mz = queue_dma_zone_reserve(queue->memz_name, queue_size_bytes,
+				       0, align);
+	if (qp_mz == NULL) {
+		BCMFS_LOG(ERR, "Failed to allocate ring memzone");
+		return -ENOMEM;
+	}
+
+	if (bcmfs_qp_check_queue_alignment(qp_mz->iova, align)) {
+		BCMFS_LOG(ERR, "Invalid alignment on queue create "
+					" 0x%" PRIx64 "\n",
+					queue->base_phys_addr);
+		ret = -EFAULT;
+		goto queue_create_err;
+	}
+
+	queue->base_addr = (char *)qp_mz->addr;
+	queue->base_phys_addr = qp_mz->iova;
+	queue->queue_size = queue_size_bytes;
+
+	return 0;
+
+queue_create_err:
+	rte_memzone_free(qp_mz);
+
+	return ret;
+}
+
+int
+bcmfs_qp_release(struct bcmfs_qp **qp_addr)
+{
+	struct bcmfs_qp *qp = *qp_addr;
+
+	if (qp == NULL) {
+		BCMFS_LOG(DEBUG, "qp already freed");
+		return 0;
+	}
+
+	/* Don't free memory if there are still responses to be processed */
+	if ((qp->stats.enqueued_count - qp->stats.dequeued_count) == 0) {
+		/* Stop the h/w ring */
+		qp->ops->stopq(qp);
+		/* Delete the queue pairs */
+		bcmfs_queue_delete(&qp->tx_q, qp->qpair_id);
+		bcmfs_queue_delete(&qp->cmpl_q, qp->qpair_id);
+	} else {
+		return -EAGAIN;
+	}
+
+	rte_bitmap_reset(qp->ctx_bmp);
+	rte_free(qp->ctx_bmp_mem);
+	rte_free(qp->ctx_pool);
+
+	rte_free(qp);
+	*qp_addr = NULL;
+
+	return 0;
+}
+
+int
+bcmfs_qp_setup(struct bcmfs_qp **qp_addr,
+	       uint16_t queue_pair_id,
+	       struct bcmfs_qp_config *qp_conf)
+{
+	struct bcmfs_qp *qp;
+	uint32_t bmp_size;
+	uint32_t nb_descriptors = qp_conf->nb_descriptors;
+	uint16_t i;
+	int rc;
+
+	if (nb_descriptors < FS_RM_MIN_REQS) {
+		BCMFS_LOG(ERR, "Can't create qp for %u descriptors",
+			  nb_descriptors);
+		return -EINVAL;
+	}
+
+	if (nb_descriptors > FS_RM_MAX_REQS)
+		nb_descriptors = FS_RM_MAX_REQS;
+
+	if (qp_conf->iobase == NULL) {
+		BCMFS_LOG(ERR, "IO onfig space null");
+		return -EINVAL;
+	}
+
+	qp = rte_zmalloc_socket("BCM FS PMD qp metadata",
+				sizeof(*qp), RTE_CACHE_LINE_SIZE,
+				qp_conf->socket_id);
+	if (qp == NULL) {
+		BCMFS_LOG(ERR, "Failed to alloc mem for qp struct");
+		return -ENOMEM;
+	}
+
+	qp->qpair_id = queue_pair_id;
+	qp->ioreg = qp_conf->iobase;
+	qp->nb_descriptors = nb_descriptors;
+
+	qp->stats.enqueued_count = 0;
+	qp->stats.dequeued_count = 0;
+
+	rc = bcmfs_queue_create(&qp->tx_q, qp_conf, qp->qpair_id,
+				BCMFS_RM_TXQ);
+	if (rc) {
+		BCMFS_LOG(ERR, "Tx queue create failed queue_pair_id %u",
+			  queue_pair_id);
+		goto create_err;
+	}
+
+	rc = bcmfs_queue_create(&qp->cmpl_q, qp_conf, qp->qpair_id,
+				BCMFS_RM_CPLQ);
+	if (rc) {
+		BCMFS_LOG(ERR, "Cmpl queue create failed queue_pair_id= %u",
+			  queue_pair_id);
+		goto q_create_err;
+	}
+
+	/* ctx saving bitmap */
+	bmp_size = rte_bitmap_get_memory_footprint(nb_descriptors);
+
+	/* Allocate memory for bitmap */
+	qp->ctx_bmp_mem = rte_zmalloc("ctx_bmp_mem", bmp_size,
+				      RTE_CACHE_LINE_SIZE);
+	if (qp->ctx_bmp_mem == NULL) {
+		rc = -ENOMEM;
+		goto qp_create_err;
+	}
+
+	/* Initialize pool resource bitmap array */
+	qp->ctx_bmp = rte_bitmap_init(nb_descriptors, qp->ctx_bmp_mem,
+				      bmp_size);
+	if (qp->ctx_bmp == NULL) {
+		rc = -EINVAL;
+		goto bmap_mem_free;
+	}
+
+	/* Mark all pools available */
+	for (i = 0; i < nb_descriptors; i++)
+		rte_bitmap_set(qp->ctx_bmp, i);
+
+	/* Allocate memory for context */
+	qp->ctx_pool = rte_zmalloc("qp_ctx_pool",
+				   sizeof(unsigned long) *
+				   nb_descriptors, 0);
+	if (qp->ctx_pool == NULL) {
+		BCMFS_LOG(ERR, "ctx allocation pool fails");
+		rc = -ENOMEM;
+		goto bmap_free;
+	}
+
+	/* Start h/w ring */
+	qp->ops->startq(qp);
+
+	*qp_addr = qp;
+
+	return 0;
+
+bmap_free:
+	rte_bitmap_reset(qp->ctx_bmp);
+bmap_mem_free:
+	rte_free(qp->ctx_bmp_mem);
+qp_create_err:
+	bcmfs_queue_delete(&qp->cmpl_q, queue_pair_id);
+q_create_err:
+	bcmfs_queue_delete(&qp->tx_q, queue_pair_id);
+create_err:
+	rte_free(qp);
+
+	return rc;
+}
+
+uint16_t
+bcmfs_enqueue_op_burst(void *qp, void **ops, uint16_t nb_ops)
+{
+	struct bcmfs_qp *tmp_qp = (struct bcmfs_qp *)qp;
+	register uint32_t nb_ops_sent = 0;
+	uint16_t nb_ops_possible = nb_ops;
+	int ret;
+
+	if (unlikely(nb_ops == 0))
+		return 0;
+
+	while (nb_ops_sent != nb_ops_possible) {
+		ret = tmp_qp->ops->enq_one_req(qp, *ops);
+		if (ret != 0) {
+			tmp_qp->stats.enqueue_err_count++;
+			/* This message cannot be enqueued */
+			if (nb_ops_sent == 0)
+				return 0;
+			goto ring_db;
+		}
+
+		ops++;
+		nb_ops_sent++;
+	}
+
+ring_db:
+	tmp_qp->stats.enqueued_count += nb_ops_sent;
+	tmp_qp->ops->ring_db(tmp_qp);
+
+	return nb_ops_sent;
+}
+
+uint16_t
+bcmfs_dequeue_op_burst(void *qp, void **ops, uint16_t nb_ops)
+{
+	struct bcmfs_qp *tmp_qp = (struct bcmfs_qp *)qp;
+	uint32_t deq = tmp_qp->ops->dequeue(tmp_qp, ops, nb_ops);
+
+	tmp_qp->stats.dequeued_count += deq;
+
+	return deq;
+}
diff --git a/drivers/crypto/bcmfs/bcmfs_qp.h b/drivers/crypto/bcmfs/bcmfs_qp.h
new file mode 100644
index 0000000000..52c487956e
--- /dev/null
+++ b/drivers/crypto/bcmfs/bcmfs_qp.h
@@ -0,0 +1,122 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2020 Broadcom
+ * All rights reserved.
+ */
+
+#ifndef _BCMFS_QP_H_
+#define _BCMFS_QP_H_
+
+#include <rte_memzone.h>
+
+/* Maximum number of h/w queues supported by device */
+#define BCMFS_MAX_HW_QUEUES		32
+
+/* H/W queue IO address space len */
+#define BCMFS_HW_QUEUE_IO_ADDR_LEN	(64 * 1024)
+
+/* Maximum size of device ops name */
+#define BCMFS_HW_OPS_NAMESIZE		32
+
+enum bcmfs_queue_type {
+	/* TX or submission queue */
+	BCMFS_RM_TXQ,
+	 /* Completion or receive queue */
+	BCMFS_RM_CPLQ
+};
+
+struct bcmfs_qp_stats {
+	/* Count of all operations enqueued */
+	uint64_t enqueued_count;
+	/* Count of all operations dequeued */
+	uint64_t dequeued_count;
+	/* Total error count on operations enqueued */
+	uint64_t enqueue_err_count;
+	/* Total error count on operations dequeued */
+	uint64_t dequeue_err_count;
+};
+
+struct bcmfs_qp_config {
+	/* Socket to allocate memory on */
+	int socket_id;
+	/* Mapped iobase for qp */
+	void *iobase;
+	/* nb_descriptors or requests a h/w queue can accommodate */
+	uint16_t nb_descriptors;
+	/* Maximum number of h/w descriptors needed by a request */
+	uint16_t max_descs_req;
+};
+
+struct bcmfs_queue {
+	/* Base virt address */
+	void *base_addr;
+	/* Base iova */
+	rte_iova_t base_phys_addr;
+	/* Queue type */
+	enum bcmfs_queue_type q_type;
+	/* Queue size based on nb_descriptors and max_descs_reqs */
+	uint32_t queue_size;
+	union {
+		/* s/w pointer for tx h/w queue*/
+		uint32_t tx_write_ptr;
+		/* s/w pointer for completion h/w queue*/
+		uint32_t cmpl_read_ptr;
+	};
+	/* Memzone name */
+	char memz_name[RTE_MEMZONE_NAMESIZE];
+};
+
+struct bcmfs_qp {
+	/* Queue-pair ID */
+	uint16_t qpair_id;
+	/* Mapped IO address */
+	void *ioreg;
+	/* A TX queue */
+	struct bcmfs_queue tx_q;
+	/* A Completion queue */
+	struct bcmfs_queue cmpl_q;
+	/* Number of requests queue can accommodate */
+	uint32_t nb_descriptors;
+	/* Number of pending requests and enqueued to h/w queue */
+	uint16_t nb_pending_requests;
+	/* A pool which act as a hash for <request-ID and virt address> pair */
+	unsigned long *ctx_pool;
+	/* virt address for mem allocated for bitmap */
+	void *ctx_bmp_mem;
+	/* Bitmap */
+	struct rte_bitmap *ctx_bmp;
+	/* Associated stats */
+	struct bcmfs_qp_stats stats;
+	/* h/w ops associated with qp */
+	struct bcmfs_hw_queue_pair_ops *ops;
+
+} __rte_cache_aligned;
+
+/* Structure defining h/w queue pair operations */
+struct bcmfs_hw_queue_pair_ops {
+	/* ops name */
+	char name[BCMFS_HW_OPS_NAMESIZE];
+	/* Enqueue an object */
+	int (*enq_one_req)(struct bcmfs_qp *qp, void *obj);
+	/* Ring doorbell */
+	void (*ring_db)(struct bcmfs_qp *qp);
+	/* Dequeue objects */
+	uint16_t (*dequeue)(struct bcmfs_qp *qp, void **obj,
+			    uint16_t nb_ops);
+	/* Start the h/w queue */
+	int (*startq)(struct bcmfs_qp *qp);
+	/* Stop the h/w queue */
+	void (*stopq)(struct bcmfs_qp *qp);
+};
+
+uint16_t
+bcmfs_enqueue_op_burst(void *qp, void **ops, uint16_t nb_ops);
+uint16_t
+bcmfs_dequeue_op_burst(void *qp, void **ops, uint16_t nb_ops);
+int
+bcmfs_qp_release(struct bcmfs_qp **qp_addr);
+int
+bcmfs_qp_setup(struct bcmfs_qp **qp_addr,
+	       uint16_t queue_pair_id,
+	       struct bcmfs_qp_config *bcmfs_conf);
+
+#endif /* _BCMFS_QP_H_ */
diff --git a/drivers/crypto/bcmfs/meson.build b/drivers/crypto/bcmfs/meson.build
index fd39eba20e..7e2bcbf14b 100644
--- a/drivers/crypto/bcmfs/meson.build
+++ b/drivers/crypto/bcmfs/meson.build
@@ -7,5 +7,6 @@ deps += ['eal', 'bus_vdev']
 sources = files(
 		'bcmfs_logs.c',
 		'bcmfs_device.c',
-		'bcmfs_vfio.c'
+		'bcmfs_vfio.c',
+		'bcmfs_qp.c'
 		)
-- 
2.17.1