crypto/qat: add minimum enqueue threshold

This patch adds minimum enqueue threshold to Intel
QuickAssist Technology PMD.
It is an optimisation, configured by a command line option,
which can be used to reduce MMIO write occurrences.

Signed-off-by: Fiona Trahe <fiona.trahe@intel.com>
Signed-off-by: Arek Kusztal <arkadiuszx.kusztal@intel.com>

doc/guides/cryptodevs/qat.rst

@@ -243,6 +243,29 @@ allocated while for GEN1 devices, 12 buffers are allocated, plus 1472 bytes over
 	larger than the input size).
 
 
+Running QAT PMD with minimum threshold for burst size
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+If only a small number or packets can be enqueued. Each enqueue causes an expensive MMIO write.
+These MMIO write occurrences can be optimised by setting any of the following parameters:
+
+- qat_sym_enq_threshold
+- qat_asym_enq_threshold
+- qat_comp_enq_threshold
+
+When any of these parameters is set rte_cryptodev_enqueue_burst function will
+return 0 (thereby avoiding an MMIO) if the device is congested and number of packets
+possible to enqueue is smaller.
+To use this feature the user must set the parameter on process start as a device additional parameter::
+
+  -w 03:01.1,qat_sym_enq_threshold=32,qat_comp_enq_threshold=16
+
+All parameters can be used with the same device regardless of order. Parameters are separated
+by comma. When the same parameter is used more than once first occurrence of the parameter
+is used.
+Maximum threshold that can be set is 32.
+
+
 Device and driver naming
 ~~~~~~~~~~~~~~~~~~~~~~~~
 

drivers/common/qat/qat_common.c

@@ -94,6 +94,9 @@ void qat_stats_get(struct qat_pci_device *dev,
 		stats->dequeued_count += qp[i]->stats.dequeued_count;
 		stats->enqueue_err_count += qp[i]->stats.enqueue_err_count;
 		stats->dequeue_err_count += qp[i]->stats.dequeue_err_count;
+		stats->threshold_hit_count += qp[i]->stats.threshold_hit_count;
+		QAT_LOG(DEBUG, "Threshold was used for qp %d %"PRIu64" times",
+				i, stats->threshold_hit_count);
 	}
 }
 

drivers/common/qat/qat_common.h

@@ -61,6 +61,9 @@ struct qat_common_stats {
 	/**< Total error count on operations enqueued */
 	uint64_t dequeue_err_count;
 	/**< Total error count on operations dequeued */
+	uint64_t threshold_hit_count;
+	/**< Total number of times min qp threshold condition was fulfilled */
+
 };
 
 struct qat_pci_device;

drivers/common/qat/qat_device.c

@@ -3,6 +3,8 @@
  */
 
 #include <rte_string_fns.h>
+#include <rte_devargs.h>
+#include <ctype.h>
 
 #include "qat_device.h"
 #include "adf_transport_access_macros.h"
@@ -105,12 +107,71 @@ qat_get_qat_dev_from_pci_dev(struct rte_pci_device *pci_dev)
 	return qat_pci_get_named_dev(name);
 }
 
+static void qat_dev_parse_cmd(const char *str, struct qat_dev_cmd_param
+		*qat_dev_cmd_param)
+{
+	int i = 0;
+	const char *param;
+
+	while (1) {
+		char value_str[4] = { };
+
+		param = qat_dev_cmd_param[i].name;
+		if (param == NULL)
+			return;
+		long value = 0;
+		const char *arg = strstr(str, param);
+		const char *arg2 = NULL;
+
+		if (arg) {
+			arg2 = arg + strlen(param);
+			if (*arg2 != '=') {
+				QAT_LOG(DEBUG, "parsing error '=' sign"
+						" should immediately follow %s",
+						param);
+				arg2 = NULL;
+			} else
+				arg2++;
+		} else {
+			QAT_LOG(DEBUG, "%s not provided", param);
+		}
+		if (arg2) {
+			int iter = 0;
+			while (iter < 2) {
+				if (!isdigit(*(arg2 + iter)))
+					break;
+				iter++;
+			}
+			if (!iter) {
+				QAT_LOG(DEBUG, "parsing error %s"
+					       " no number provided",
+					       param);
+			} else {
+				memcpy(value_str, arg2, iter);
+				value = strtol(value_str, NULL, 10);
+				if (value > MAX_QP_THRESHOLD_SIZE) {
+					QAT_LOG(DEBUG, "Exceeded max size of"
+						" threshold, setting to %d",
+						MAX_QP_THRESHOLD_SIZE);
+					value = MAX_QP_THRESHOLD_SIZE;
+				}
+				QAT_LOG(DEBUG, "parsing %s = %ld",
+						param, value);
+			}
+		}
+		qat_dev_cmd_param[i].val = value;
+		i++;
+	}
+}
+
 struct qat_pci_device *
-qat_pci_device_allocate(struct rte_pci_device *pci_dev)
+qat_pci_device_allocate(struct rte_pci_device *pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param)
 {
 	struct qat_pci_device *qat_dev;
 	uint8_t qat_dev_id;
 	char name[QAT_DEV_NAME_MAX_LEN];
+	struct rte_devargs *devargs = pci_dev->device.devargs;
 
 	rte_pci_device_name(&pci_dev->addr, name, sizeof(name));
 	snprintf(name+strlen(name), QAT_DEV_NAME_MAX_LEN-strlen(name), "_qat");
@@ -148,6 +209,9 @@ qat_pci_device_allocate(struct rte_pci_device *pci_dev)
 		return NULL;
 	}
 
+	if (devargs && devargs->drv_str)
+		qat_dev_parse_cmd(devargs->drv_str, qat_dev_cmd_param);
+
 	rte_spinlock_init(&qat_dev->arb_csr_lock);
 
 	qat_dev->attached = QAT_ATTACHED;
@@ -199,37 +263,45 @@ qat_pci_dev_destroy(struct qat_pci_device *qat_pci_dev,
 static int qat_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 		struct rte_pci_device *pci_dev)
 {
-	int ret = 0;
+	int sym_ret = 0, asym_ret = 0, comp_ret = 0;
 	int num_pmds_created = 0;
 	struct qat_pci_device *qat_pci_dev;
+	struct qat_dev_cmd_param qat_dev_cmd_param[] = {
+			{ SYM_ENQ_THRESHOLD_NAME, 0 },
+			{ ASYM_ENQ_THRESHOLD_NAME, 0 },
+			{ COMP_ENQ_THRESHOLD_NAME, 0 },
+			{ NULL, 0 },
+	};
 
 	QAT_LOG(DEBUG, "Found QAT device at %02x:%02x.%x",
 			pci_dev->addr.bus,
 			pci_dev->addr.devid,
 			pci_dev->addr.function);
 
-	qat_pci_dev = qat_pci_device_allocate(pci_dev);
+	qat_pci_dev = qat_pci_device_allocate(pci_dev, qat_dev_cmd_param);
 	if (qat_pci_dev == NULL)
 		return -ENODEV;
 
-	ret = qat_sym_dev_create(qat_pci_dev);
-	if (ret == 0)
+	sym_ret = qat_sym_dev_create(qat_pci_dev, qat_dev_cmd_param);
+	if (sym_ret == 0) {
 		num_pmds_created++;
+
+	}
 	else
 		QAT_LOG(WARNING,
 				"Failed to create QAT SYM PMD on device %s",
 				qat_pci_dev->name);
 
-	ret = qat_comp_dev_create(qat_pci_dev);
-	if (ret == 0)
+	comp_ret = qat_comp_dev_create(qat_pci_dev, qat_dev_cmd_param);
+	if (comp_ret == 0)
 		num_pmds_created++;
 	else
 		QAT_LOG(WARNING,
 				"Failed to create QAT COMP PMD on device %s",
 				qat_pci_dev->name);
 
-	ret = qat_asym_dev_create(qat_pci_dev);
-	if (ret == 0)
+	asym_ret = qat_asym_dev_create(qat_pci_dev, qat_dev_cmd_param);
+	if (asym_ret == 0)
 		num_pmds_created++;
 	else
 		QAT_LOG(WARNING,
@@ -264,13 +336,15 @@ static struct rte_pci_driver rte_qat_pmd = {
 };
 
 __rte_weak int
-qat_sym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused)
+qat_sym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused,
+		struct qat_dev_cmd_param *qat_dev_cmd_param __rte_unused)
 {
 	return 0;
 }
 
 __rte_weak int
-qat_asym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused)
+qat_asym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused,
+		struct qat_dev_cmd_param *qat_dev_cmd_param __rte_unused)
 {
 	return 0;
 }
@@ -288,7 +362,8 @@ qat_asym_dev_destroy(struct qat_pci_device *qat_pci_dev __rte_unused)
 }
 
 __rte_weak int
-qat_comp_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused)
+qat_comp_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused,
+		struct qat_dev_cmd_param *qat_dev_cmd_param __rte_unused)
 {
 	return 0;
 }

drivers/common/qat/qat_device.h

@@ -16,6 +16,16 @@
 
 #define QAT_DEV_NAME_MAX_LEN	64
 
+#define SYM_ENQ_THRESHOLD_NAME "qat_sym_enq_threshold"
+#define ASYM_ENQ_THRESHOLD_NAME "qat_asym_enq_threshold"
+#define COMP_ENQ_THRESHOLD_NAME "qat_comp_enq_threshold"
+#define MAX_QP_THRESHOLD_SIZE	32
+
+struct qat_dev_cmd_param {
+	const char *name;
+	uint16_t val;
+};
+
 enum qat_comp_num_im_buffers {
 	QAT_NUM_INTERM_BUFS_GEN1 = 12,
 	QAT_NUM_INTERM_BUFS_GEN2 = 20,
@@ -94,7 +104,8 @@ struct qat_gen_hw_data {
 extern struct qat_gen_hw_data qat_gen_config[];
 
 struct qat_pci_device *
-qat_pci_device_allocate(struct rte_pci_device *pci_dev);
+qat_pci_device_allocate(struct rte_pci_device *pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
 qat_pci_device_release(struct rte_pci_device *pci_dev);
@@ -104,10 +115,12 @@ qat_get_qat_dev_from_pci_dev(struct rte_pci_device *pci_dev);
 
 /* declaration needed for weak functions */
 int
-qat_sym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused);
+qat_sym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
-qat_asym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused);
+qat_asym_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
 qat_sym_dev_destroy(struct qat_pci_device *qat_pci_dev __rte_unused);
@@ -116,7 +129,8 @@ int
 qat_asym_dev_destroy(struct qat_pci_device *qat_pci_dev __rte_unused);
 
 int
-qat_comp_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused);
+qat_comp_dev_create(struct qat_pci_device *qat_pci_dev __rte_unused,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
 qat_comp_dev_destroy(struct qat_pci_device *qat_pci_dev __rte_unused);

drivers/common/qat/qat_qp.c

@@ -608,8 +608,19 @@ qat_enqueue_op_burst(void *qp, void **ops, uint16_t nb_ops)
 			if (nb_ops_possible == 0)
 				return 0;
 		}
+		/* QAT has plenty of work queued already, so don't waste cycles
+		 * enqueueing, wait til the application has gathered a bigger
+		 * burst or some completed ops have been dequeued
+		 */
+		if (tmp_qp->min_enq_burst_threshold && inflights >
+				QAT_QP_MIN_INFL_THRESHOLD && nb_ops_possible <
+				tmp_qp->min_enq_burst_threshold) {
+			tmp_qp->stats.threshold_hit_count++;
+			return 0;
+		}
 	}
 
+
 	while (nb_ops_sent != nb_ops_possible) {
 		ret = tmp_qp->build_request(*ops, base_addr + tail,
 				tmp_qp->op_cookies[tail / queue->msg_size],

drivers/common/qat/qat_qp.h

@@ -12,6 +12,8 @@ struct qat_pci_device;
 #define QAT_CSR_HEAD_WRITE_THRESH 32U
 /* number of requests to accumulate before writing head CSR */
 
+#define QAT_QP_MIN_INFL_THRESHOLD	256
+
 typedef int (*build_request_t)(void *op,
 		uint8_t *req, void *op_cookie,
 		enum qat_device_gen qat_dev_gen);
@@ -77,6 +79,7 @@ struct qat_qp {
 	uint32_t enqueued;
 	uint32_t dequeued __rte_aligned(4);
 	uint16_t max_inflights;
+	uint16_t min_enq_burst_threshold;
 } __rte_cache_aligned;
 
 extern const struct qat_qp_hw_data qat_gen1_qps[][ADF_MAX_QPS_ON_ANY_SERVICE];

drivers/compress/qat/qat_comp_pmd.c

@@ -139,6 +139,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id,
 								= *qp_addr;
 
 	qp = (struct qat_qp *)*qp_addr;
+	qp->min_enq_burst_threshold = qat_private->min_enq_burst_threshold;
 
 	for (i = 0; i < qp->nb_descriptors; i++) {
 
@@ -660,8 +661,10 @@ static const struct rte_driver compdev_qat_driver = {
 	.alias = qat_comp_drv_name
 };
 int
-qat_comp_dev_create(struct qat_pci_device *qat_pci_dev)
+qat_comp_dev_create(struct qat_pci_device *qat_pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param)
 {
+	int i = 0;
 	if (qat_pci_dev->qat_dev_gen == QAT_GEN3) {
 		QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx");
 		return 0;
@@ -719,6 +722,15 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev)
 		break;
 	}
 
+	while (1) {
+		if (qat_dev_cmd_param[i].name == NULL)
+			break;
+		if (!strcmp(qat_dev_cmd_param[i].name, COMP_ENQ_THRESHOLD_NAME))
+			comp_dev->min_enq_burst_threshold =
+					qat_dev_cmd_param[i].val;
+		i++;
+	}
+
 	QAT_LOG(DEBUG,
 		    "Created QAT COMP device %s as compressdev instance %d",
 			name, compressdev->data->dev_id);

drivers/compress/qat/qat_comp_pmd.h

@@ -32,10 +32,12 @@ struct qat_comp_dev_private {
 	/**< The device's pool for qat_comp_xforms */
 	struct rte_mempool *streampool;
 	/**< The device's pool for qat_comp_streams */
+	uint16_t min_enq_burst_threshold;
 };
 
 int
-qat_comp_dev_create(struct qat_pci_device *qat_pci_dev);
+qat_comp_dev_create(struct qat_pci_device *qat_pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
 qat_comp_dev_destroy(struct qat_pci_device *qat_pci_dev);

drivers/crypto/qat/qat_asym_pmd.c

@@ -160,6 +160,7 @@ static int qat_asym_qp_setup(struct rte_cryptodev *dev, uint16_t qp_id,
 							= *qp_addr;
 
 	qp = (struct qat_qp *)*qp_addr;
+	qp->min_enq_burst_threshold = qat_private->min_enq_burst_threshold;
 
 	for (i = 0; i < qp->nb_descriptors; i++) {
 		int j;
@@ -235,8 +236,10 @@ static const struct rte_driver cryptodev_qat_asym_driver = {
 };
 
 int
-qat_asym_dev_create(struct qat_pci_device *qat_pci_dev)
+qat_asym_dev_create(struct qat_pci_device *qat_pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param)
 {
+	int i = 0;
 	struct rte_cryptodev_pmd_init_params init_params = {
 			.name = "",
 			.socket_id = qat_pci_dev->pci_dev->device.numa_node,
@@ -281,6 +284,15 @@ qat_asym_dev_create(struct qat_pci_device *qat_pci_dev)
 	internals->asym_dev_id = cryptodev->data->dev_id;
 	internals->qat_dev_capabilities = qat_gen1_asym_capabilities;
 
+	while (1) {
+		if (qat_dev_cmd_param[i].name == NULL)
+			break;
+		if (!strcmp(qat_dev_cmd_param[i].name, ASYM_ENQ_THRESHOLD_NAME))
+			internals->min_enq_burst_threshold =
+					qat_dev_cmd_param[i].val;
+		i++;
+	}
+
 	QAT_LOG(DEBUG, "Created QAT ASYM device %s as cryptodev instance %d",
 			cryptodev->data->name, internals->asym_dev_id);
 	return 0;

drivers/crypto/qat/qat_asym_pmd.h

@@ -26,6 +26,7 @@ struct qat_asym_dev_private {
 	/**< Device instance for this rte_cryptodev */
 	const struct rte_cryptodev_capabilities *qat_dev_capabilities;
 	/* QAT device asymmetric crypto capabilities */
+	uint16_t min_enq_burst_threshold;
 };
 
 uint16_t
@@ -42,7 +43,8 @@ int qat_asym_session_configure(struct rte_cryptodev *dev,
 		struct rte_mempool *mempool);
 
 int
-qat_asym_dev_create(struct qat_pci_device *qat_pci_dev);
+qat_asym_dev_create(struct qat_pci_device *qat_pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
 qat_asym_dev_destroy(struct qat_pci_device *qat_pci_dev);

drivers/crypto/qat/qat_sym_pmd.c

@@ -169,6 +169,7 @@ static int qat_sym_qp_setup(struct rte_cryptodev *dev, uint16_t qp_id,
 							= *qp_addr;
 
 	qp = (struct qat_qp *)*qp_addr;
+	qp->min_enq_burst_threshold = qat_private->min_enq_burst_threshold;
 
 	for (i = 0; i < qp->nb_descriptors; i++) {
 
@@ -237,8 +238,10 @@ static const struct rte_driver cryptodev_qat_sym_driver = {
 };
 
 int
-qat_sym_dev_create(struct qat_pci_device *qat_pci_dev)
+qat_sym_dev_create(struct qat_pci_device *qat_pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param __rte_unused)
 {
+	int i = 0;
 	struct rte_cryptodev_pmd_init_params init_params = {
 			.name = "",
 			.socket_id = qat_pci_dev->pci_dev->device.numa_node,
@@ -302,6 +305,15 @@ qat_sym_dev_create(struct qat_pci_device *qat_pci_dev)
 		break;
 	}
 
+	while (1) {
+		if (qat_dev_cmd_param[i].name == NULL)
+			break;
+		if (!strcmp(qat_dev_cmd_param[i].name, SYM_ENQ_THRESHOLD_NAME))
+			internals->min_enq_burst_threshold =
+					qat_dev_cmd_param[i].val;
+		i++;
+	}
+
 	QAT_LOG(DEBUG, "Created QAT SYM device %s as cryptodev instance %d",
 			cryptodev->data->name, internals->sym_dev_id);
 	return 0;

drivers/crypto/qat/qat_sym_pmd.h

@@ -28,10 +28,12 @@ struct qat_sym_dev_private {
 	/**< Device instance for this rte_cryptodev */
 	const struct rte_cryptodev_capabilities *qat_dev_capabilities;
 	/* QAT device symmetric crypto capabilities */
+	uint16_t min_enq_burst_threshold;
 };
 
 int
-qat_sym_dev_create(struct qat_pci_device *qat_pci_dev);
+qat_sym_dev_create(struct qat_pci_device *qat_pci_dev,
+		struct qat_dev_cmd_param *qat_dev_cmd_param);
 
 int
 qat_sym_dev_destroy(struct qat_pci_device *qat_pci_dev);