net/sfc: support flow API filters

Only pattern items VOID, ETH and actions VOID, QUEUE is now
supported.

Signed-off-by: Roman Zhukov <roman.zhukov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>

doc/guides/nics/features/sfc_efx.ini

@@ -19,6 +19,7 @@ RSS hash             = Y
 RSS key update       = Y
 RSS reta update      = Y
 Flow control         = Y
+Flow API             = Y
 VLAN offload         = P
 L3 checksum offload  = Y
 L4 checksum offload  = Y

doc/guides/nics/sfc_efx.rst

@@ -81,6 +81,8 @@ SFC EFX PMD has support for:
 
 - Transmit VLAN insertion (if running firmware variant supports it)
 
+- Flow API
+
 
 Non-supported Features
 ----------------------
@@ -114,6 +116,28 @@ required in the receive buffer.
 It should be taken into account when mbuf pool for receive is created.
 
 
+Flow API support
+----------------
+
+Supported attributes:
+
+- Ingress
+
+Supported pattern items:
+
+- VOID
+
+- ETH (exact match of source/destination addresses, EtherType)
+
+Supported actions:
+
+- VOID
+
+- QUEUE
+
+Validating flow rules depends on the firmware variant.
+
+
 Supported NICs
 --------------
 

doc/guides/rel_notes/release_17_05.rst

@@ -84,6 +84,11 @@ New Features
   Added support for Hardware TSO for tunneled and non-tunneled packets.
   Tunneling protocols supported are GRE and VXLAN.
 
+* **Updated the sfc_efx driver.**
+
+  * Generic flow API support for Ethernet, VLAN, IPv4, IPv6, UDP and TCP
+    pattern items with QUEUE action for ingress traffic.
+
 * **Added vmxnet3 version 3 support.**
 
   Added support for vmxnet3 version 3 which includes several

drivers/net/sfc/Makefile

@@ -91,6 +91,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_SFC_EFX_PMD) += sfc_rx.c
 SRCS-$(CONFIG_RTE_LIBRTE_SFC_EFX_PMD) += sfc_tx.c
 SRCS-$(CONFIG_RTE_LIBRTE_SFC_EFX_PMD) += sfc_tso.c
 SRCS-$(CONFIG_RTE_LIBRTE_SFC_EFX_PMD) += sfc_filter.c
+SRCS-$(CONFIG_RTE_LIBRTE_SFC_EFX_PMD) += sfc_flow.c
 
 VPATH += $(SRCDIR)/base
 

drivers/net/sfc/sfc.c

@@ -320,10 +320,17 @@ sfc_start(struct sfc_adapter *sa)
 	if (rc != 0)
 		goto fail_tx_start;
 
+	rc = sfc_flow_start(sa);
+	if (rc != 0)
+		goto fail_flows_insert;
+
 	sa->state = SFC_ADAPTER_STARTED;
 	sfc_log_init(sa, "done");
 	return 0;
 
+fail_flows_insert:
+	sfc_tx_stop(sa);
+
 fail_tx_start:
 	sfc_rx_stop(sa);
 
@@ -368,6 +375,7 @@ sfc_stop(struct sfc_adapter *sa)
 
 	sa->state = SFC_ADAPTER_STOPPING;
 
+	sfc_flow_stop(sa);
 	sfc_tx_stop(sa);
 	sfc_rx_stop(sa);
 	sfc_port_stop(sa);
@@ -640,6 +648,8 @@ sfc_attach(struct sfc_adapter *sa)
 	sfc_log_init(sa, "fini nic");
 	efx_nic_fini(enp);
 
+	sfc_flow_init(sa);
+
 	sa->state = SFC_ADAPTER_INITIALIZED;
 
 	sfc_log_init(sa, "done");
@@ -698,5 +708,6 @@ sfc_detach(struct sfc_adapter *sa)
 
 	sfc_mem_bar_fini(sa);
 
+	sfc_flow_fini(sa);
 	sa->state = SFC_ADAPTER_UNINITIALIZED;
 }

drivers/net/sfc/sfc_ethdev.c

@@ -40,7 +40,7 @@
 #include "sfc_ev.h"
 #include "sfc_rx.h"
 #include "sfc_tx.h"
-
+#include "sfc_flow.h"
 
 static void
 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
@@ -1210,8 +1210,8 @@ bad_reta_entry:
 
 static int
 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
-		    __rte_unused enum rte_filter_op filter_op,
-		    __rte_unused void *arg)
+		    enum rte_filter_op filter_op,
+		    void *arg)
 {
 	struct sfc_adapter *sa = dev->data->dev_private;
 	int rc = ENOTSUP;
@@ -1246,6 +1246,14 @@ sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
 	case RTE_ETH_FILTER_HASH:
 		sfc_err(sa, "Hash filters not supported");
 		break;
+	case RTE_ETH_FILTER_GENERIC:
+		if (filter_op != RTE_ETH_FILTER_GET) {
+			rc = EINVAL;
+		} else {
+			*(const void **)arg = &sfc_flow_ops;
+			rc = 0;
+		}
+		break;
 	default:
 		sfc_err(sa, "Unknown filter type %u", filter_type);
 		break;

drivers/net/sfc/sfc_filter.h

@@ -32,6 +32,8 @@
 
 #include "efx.h"
 
+#include "sfc_flow.h"
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -41,6 +43,8 @@ struct sfc_filter {
 	size_t				supported_match_num;
 	/** Driver cache of supported filter match masks */
 	uint32_t			*supported_match;
+	/** List of flow rules */
+	struct sfc_flow_list		flow_list;
 };
 
 struct sfc_adapter;

drivers/net/sfc/sfc_flow.c

@@ -0,0 +1,704 @@
+/*-
+ * Copyright (c) 2017 Solarflare Communications Inc.
+ * All rights reserved.
+ *
+ * This software was jointly developed between OKTET Labs (under contract
+ * for Solarflare) and Solarflare Communications, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+ * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
+ * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
+ * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <rte_tailq.h>
+#include <rte_common.h>
+#include <rte_ethdev.h>
+#include <rte_eth_ctrl.h>
+#include <rte_ether.h>
+#include <rte_flow.h>
+#include <rte_flow_driver.h>
+
+#include "efx.h"
+
+#include "sfc.h"
+#include "sfc_rx.h"
+#include "sfc_filter.h"
+#include "sfc_flow.h"
+#include "sfc_log.h"
+
+/*
+ * At now flow API is implemented in such a manner that each
+ * flow rule is converted to a hardware filter.
+ * All elements of flow rule (attributes, pattern items, actions)
+ * correspond to one or more fields in the efx_filter_spec_s structure
+ * that is responsible for the hardware filter.
+ */
+
+enum sfc_flow_item_layers {
+	SFC_FLOW_ITEM_ANY_LAYER,
+	SFC_FLOW_ITEM_START_LAYER,
+	SFC_FLOW_ITEM_L2,
+};
+
+typedef int (sfc_flow_item_parse)(const struct rte_flow_item *item,
+				  efx_filter_spec_t *spec,
+				  struct rte_flow_error *error);
+
+struct sfc_flow_item {
+	enum rte_flow_item_type type;		/* Type of item */
+	enum sfc_flow_item_layers layer;	/* Layer of item */
+	enum sfc_flow_item_layers prev_layer;	/* Previous layer of item */
+	sfc_flow_item_parse *parse;		/* Parsing function */
+};
+
+static sfc_flow_item_parse sfc_flow_parse_void;
+static sfc_flow_item_parse sfc_flow_parse_eth;
+
+static boolean_t
+sfc_flow_is_zero(const uint8_t *buf, unsigned int size)
+{
+	uint8_t sum = 0;
+	unsigned int i;
+
+	for (i = 0; i < size; i++)
+		sum |= buf[i];
+
+	return (sum == 0) ? B_TRUE : B_FALSE;
+}
+
+/*
+ * Validate item and prepare structures spec and mask for parsing
+ */
+static int
+sfc_flow_parse_init(const struct rte_flow_item *item,
+		    const void **spec_ptr,
+		    const void **mask_ptr,
+		    const void *supp_mask,
+		    const void *def_mask,
+		    unsigned int size,
+		    struct rte_flow_error *error)
+{
+	const uint8_t *spec;
+	const uint8_t *mask;
+	const uint8_t *last;
+	uint8_t match;
+	uint8_t supp;
+	unsigned int i;
+
+	if (item == NULL) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+				   "NULL item");
+		return -rte_errno;
+	}
+
+	if ((item->last != NULL || item->mask != NULL) && item->spec == NULL) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ITEM, item,
+				   "Mask or last is set without spec");
+		return -rte_errno;
+	}
+
+	/*
+	 * If "mask" is not set, default mask is used,
+	 * but if default mask is NULL, "mask" should be set
+	 */
+	if (item->mask == NULL) {
+		if (def_mask == NULL) {
+			rte_flow_error_set(error, EINVAL,
+				RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+				"Mask should be specified");
+			return -rte_errno;
+		}
+
+		mask = (const uint8_t *)def_mask;
+	} else {
+		mask = (const uint8_t *)item->mask;
+	}
+
+	spec = (const uint8_t *)item->spec;
+	last = (const uint8_t *)item->last;
+
+	if (spec == NULL)
+		goto exit;
+
+	/*
+	 * If field values in "last" are either 0 or equal to the corresponding
+	 * values in "spec" then they are ignored
+	 */
+	if (last != NULL &&
+	    !sfc_flow_is_zero(last, size) &&
+	    memcmp(last, spec, size) != 0) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_ITEM, item,
+				   "Ranging is not supported");
+		return -rte_errno;
+	}
+
+	if (supp_mask == NULL) {
+		rte_flow_error_set(error, EINVAL,
+			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+			"Supported mask for item should be specified");
+		return -rte_errno;
+	}
+
+	/* Check that mask and spec not asks for more match than supp_mask */
+	for (i = 0; i < size; i++) {
+		match = spec[i] | mask[i];
+		supp = ((const uint8_t *)supp_mask)[i];
+
+		if ((match | supp) != supp) {
+			rte_flow_error_set(error, ENOTSUP,
+					   RTE_FLOW_ERROR_TYPE_ITEM, item,
+					   "Item's field is not supported");
+			return -rte_errno;
+		}
+	}
+
+exit:
+	*spec_ptr = spec;
+	*mask_ptr = mask;
+	return 0;
+}
+
+/*
+ * Protocol parsers.
+ * Masking is not supported, so masks in items should be either
+ * full or empty (zeroed) and set only for supported fields which
+ * are specified in the supp_mask.
+ */
+
+static int
+sfc_flow_parse_void(__rte_unused const struct rte_flow_item *item,
+		    __rte_unused efx_filter_spec_t *efx_spec,
+		    __rte_unused struct rte_flow_error *error)
+{
+	return 0;
+}
+
+/**
+ * Convert Ethernet item to EFX filter specification.
+ *
+ * @param item[in]
+ *   Item specification. Only source and destination addresses and
+ *   Ethernet type fields are supported. If the mask is NULL, default
+ *   mask will be used. Ranging is not supported.
+ * @param efx_spec[in, out]
+ *   EFX filter specification to update.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ */
+static int
+sfc_flow_parse_eth(const struct rte_flow_item *item,
+		   efx_filter_spec_t *efx_spec,
+		   struct rte_flow_error *error)
+{
+	int rc;
+	const struct rte_flow_item_eth *spec = NULL;
+	const struct rte_flow_item_eth *mask = NULL;
+	const struct rte_flow_item_eth supp_mask = {
+		.dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
+		.src.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
+		.type = 0xffff,
+	};
+
+	rc = sfc_flow_parse_init(item,
+				 (const void **)&spec,
+				 (const void **)&mask,
+				 &supp_mask,
+				 &rte_flow_item_eth_mask,
+				 sizeof(struct rte_flow_item_eth),
+				 error);
+	if (rc != 0)
+		return rc;
+
+	/* If "spec" is not set, could be any Ethernet */
+	if (spec == NULL)
+		return 0;
+
+	if (is_same_ether_addr(&mask->dst, &supp_mask.dst)) {
+		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_MAC;
+		rte_memcpy(efx_spec->efs_loc_mac, spec->dst.addr_bytes,
+			   EFX_MAC_ADDR_LEN);
+	} else if (!is_zero_ether_addr(&mask->dst)) {
+		goto fail_bad_mask;
+	}
+
+	if (is_same_ether_addr(&mask->src, &supp_mask.src)) {
+		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_MAC;
+		rte_memcpy(efx_spec->efs_rem_mac, spec->src.addr_bytes,
+			   EFX_MAC_ADDR_LEN);
+	} else if (!is_zero_ether_addr(&mask->src)) {
+		goto fail_bad_mask;
+	}
+
+	/*
+	 * Ether type is in big-endian byte order in item and
+	 * in little-endian in efx_spec, so byte swap is used
+	 */
+	if (mask->type == supp_mask.type) {
+		efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
+		efx_spec->efs_ether_type = rte_bswap16(spec->type);
+	} else if (mask->type != 0) {
+		goto fail_bad_mask;
+	}
+
+	return 0;
+
+fail_bad_mask:
+	rte_flow_error_set(error, EINVAL,
+			   RTE_FLOW_ERROR_TYPE_ITEM, item,
+			   "Bad mask in the ETH pattern item");
+	return -rte_errno;
+}
+
+static const struct sfc_flow_item sfc_flow_items[] = {
+	{
+		.type = RTE_FLOW_ITEM_TYPE_VOID,
+		.prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
+		.layer = SFC_FLOW_ITEM_ANY_LAYER,
+		.parse = sfc_flow_parse_void,
+	},
+	{
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.prev_layer = SFC_FLOW_ITEM_START_LAYER,
+		.layer = SFC_FLOW_ITEM_L2,
+		.parse = sfc_flow_parse_eth,
+	},
+};
+
+/*
+ * Protocol-independent flow API support
+ */
+static int
+sfc_flow_parse_attr(const struct rte_flow_attr *attr,
+		    struct rte_flow *flow,
+		    struct rte_flow_error *error)
+{
+	if (attr == NULL) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ATTR, NULL,
+				   "NULL attribute");
+		return -rte_errno;
+	}
+	if (attr->group != 0) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
+				   "Groups are not supported");
+		return -rte_errno;
+	}
+	if (attr->priority != 0) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, attr,
+				   "Priorities are not supported");
+		return -rte_errno;
+	}
+	if (attr->egress != 0) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attr,
+				   "Egress is not supported");
+		return -rte_errno;
+	}
+	if (attr->ingress == 0) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, attr,
+				   "Only ingress is supported");
+		return -rte_errno;
+	}
+
+	flow->spec.efs_flags |= EFX_FILTER_FLAG_RX;
+	flow->spec.efs_rss_context = EFX_FILTER_SPEC_RSS_CONTEXT_DEFAULT;
+
+	return 0;
+}
+
+/* Get item from array sfc_flow_items */
+static const struct sfc_flow_item *
+sfc_flow_get_item(enum rte_flow_item_type type)
+{
+	unsigned int i;
+
+	for (i = 0; i < RTE_DIM(sfc_flow_items); i++)
+		if (sfc_flow_items[i].type == type)
+			return &sfc_flow_items[i];
+
+	return NULL;
+}
+
+static int
+sfc_flow_parse_pattern(const struct rte_flow_item pattern[],
+		       struct rte_flow *flow,
+		       struct rte_flow_error *error)
+{
+	int rc;
+	unsigned int prev_layer = SFC_FLOW_ITEM_ANY_LAYER;
+	const struct sfc_flow_item *item;
+
+	if (pattern == NULL) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
+				   "NULL pattern");
+		return -rte_errno;
+	}
+
+	for (; pattern != NULL &&
+	       pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
+		item = sfc_flow_get_item(pattern->type);
+		if (item == NULL) {
+			rte_flow_error_set(error, ENOTSUP,
+					   RTE_FLOW_ERROR_TYPE_ITEM, pattern,
+					   "Unsupported pattern item");
+			return -rte_errno;
+		}
+
+		/*
+		 * Omitting one or several protocol layers at the beginning
+		 * of pattern is supported
+		 */
+		if (item->prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
+		    prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
+		    item->prev_layer != prev_layer) {
+			rte_flow_error_set(error, ENOTSUP,
+					   RTE_FLOW_ERROR_TYPE_ITEM, pattern,
+					   "Unexpected sequence of pattern items");
+			return -rte_errno;
+		}
+
+		rc = item->parse(pattern, &flow->spec, error);
+		if (rc != 0)
+			return rc;
+
+		if (item->layer != SFC_FLOW_ITEM_ANY_LAYER)
+			prev_layer = item->layer;
+	}
+
+	if (pattern == NULL) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+				   "NULL item");
+		return -rte_errno;
+	}
+
+	return 0;
+}
+
+static int
+sfc_flow_parse_queue(struct sfc_adapter *sa,
+		     const struct rte_flow_action_queue *queue,
+		     struct rte_flow *flow)
+{
+	struct sfc_rxq *rxq;
+
+	if (queue->index >= sa->rxq_count)
+		return -EINVAL;
+
+	rxq = sa->rxq_info[queue->index].rxq;
+	flow->spec.efs_dmaq_id = (uint16_t)rxq->hw_index;
+
+	return 0;
+}
+
+static int
+sfc_flow_parse_actions(struct sfc_adapter *sa,
+		       const struct rte_flow_action actions[],
+		       struct rte_flow *flow,
+		       struct rte_flow_error *error)
+{
+	int rc;
+	boolean_t is_specified = B_FALSE;
+
+	if (actions == NULL) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
+				   "NULL actions");
+		return -rte_errno;
+	}
+
+	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+		switch (actions->type) {
+		case RTE_FLOW_ACTION_TYPE_VOID:
+			break;
+
+		case RTE_FLOW_ACTION_TYPE_QUEUE:
+			rc = sfc_flow_parse_queue(sa, actions->conf, flow);
+			if (rc != 0) {
+				rte_flow_error_set(error, EINVAL,
+					RTE_FLOW_ERROR_TYPE_ACTION, actions,
+					"Bad QUEUE action");
+				return -rte_errno;
+			}
+
+			is_specified = B_TRUE;
+			break;
+
+		default:
+			rte_flow_error_set(error, ENOTSUP,
+					   RTE_FLOW_ERROR_TYPE_ACTION, actions,
+					   "Action is not supported");
+			return -rte_errno;
+		}
+	}
+
+	if (!is_specified) {
+		rte_flow_error_set(error, EINVAL,
+				   RTE_FLOW_ERROR_TYPE_ACTION_NUM, actions,
+				   "Action is unspecified");
+		return -rte_errno;
+	}
+
+	return 0;
+}
+
+static int
+sfc_flow_parse(struct rte_eth_dev *dev,
+	       const struct rte_flow_attr *attr,
+	       const struct rte_flow_item pattern[],
+	       const struct rte_flow_action actions[],
+	       struct rte_flow *flow,
+	       struct rte_flow_error *error)
+{
+	struct sfc_adapter *sa = dev->data->dev_private;
+	int rc;
+
+	memset(&flow->spec, 0, sizeof(flow->spec));
+
+	rc = sfc_flow_parse_attr(attr, flow, error);
+	if (rc != 0)
+		goto fail_bad_value;
+
+	rc = sfc_flow_parse_pattern(pattern, flow, error);
+	if (rc != 0)
+		goto fail_bad_value;
+
+	rc = sfc_flow_parse_actions(sa, actions, flow, error);
+	if (rc != 0)
+		goto fail_bad_value;
+
+	if (!sfc_filter_is_match_supported(sa, flow->spec.efs_match_flags)) {
+		rte_flow_error_set(error, ENOTSUP,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+				   "Flow rule pattern is not supported");
+		return -rte_errno;
+	}
+
+fail_bad_value:
+	return rc;
+}
+
+static int
+sfc_flow_validate(struct rte_eth_dev *dev,
+		  const struct rte_flow_attr *attr,
+		  const struct rte_flow_item pattern[],
+		  const struct rte_flow_action actions[],
+		  struct rte_flow_error *error)
+{
+	struct rte_flow flow;
+
+	return sfc_flow_parse(dev, attr, pattern, actions, &flow, error);
+}
+
+static struct rte_flow *
+sfc_flow_create(struct rte_eth_dev *dev,
+		const struct rte_flow_attr *attr,
+		const struct rte_flow_item pattern[],
+		const struct rte_flow_action actions[],
+		struct rte_flow_error *error)
+{
+	struct sfc_adapter *sa = dev->data->dev_private;
+	struct rte_flow *flow = NULL;
+	int rc;
+
+	flow = rte_zmalloc("sfc_rte_flow", sizeof(*flow), 0);
+	if (flow == NULL) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+				   "Failed to allocate memory");
+		goto fail_no_mem;
+	}
+
+	rc = sfc_flow_parse(dev, attr, pattern, actions, flow, error);
+	if (rc != 0)
+		goto fail_bad_value;
+
+	TAILQ_INSERT_TAIL(&sa->filter.flow_list, flow, entries);
+
+	sfc_adapter_lock(sa);
+
+	if (sa->state == SFC_ADAPTER_STARTED) {
+		rc = efx_filter_insert(sa->nic, &flow->spec);
+		if (rc != 0) {
+			rte_flow_error_set(error, rc,
+				RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+				"Failed to insert filter");
+			goto fail_filter_insert;
+		}
+	}
+
+	sfc_adapter_unlock(sa);
+
+	return flow;
+
+fail_filter_insert:
+	TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
+
+fail_bad_value:
+	rte_free(flow);
+	sfc_adapter_unlock(sa);
+
+fail_no_mem:
+	return NULL;
+}
+
+static int
+sfc_flow_remove(struct sfc_adapter *sa,
+		struct rte_flow *flow,
+		struct rte_flow_error *error)
+{
+	int rc = 0;
+
+	SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+	if (sa->state == SFC_ADAPTER_STARTED) {
+		rc = efx_filter_remove(sa->nic, &flow->spec);
+		if (rc != 0)
+			rte_flow_error_set(error, rc,
+				RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
+				"Failed to destroy flow rule");
+	}
+
+	TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
+	rte_free(flow);
+
+	return rc;
+}
+
+static int
+sfc_flow_destroy(struct rte_eth_dev *dev,
+		 struct rte_flow *flow,
+		 struct rte_flow_error *error)
+{
+	struct sfc_adapter *sa = dev->data->dev_private;
+	struct rte_flow *flow_ptr;
+	int rc = EINVAL;
+
+	sfc_adapter_lock(sa);
+
+	TAILQ_FOREACH(flow_ptr, &sa->filter.flow_list, entries) {
+		if (flow_ptr == flow)
+			rc = 0;
+	}
+	if (rc != 0) {
+		rte_flow_error_set(error, rc,
+				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
+				   "Failed to find flow rule to destroy");
+		goto fail_bad_value;
+	}
+
+	rc = sfc_flow_remove(sa, flow, error);
+
+fail_bad_value:
+	sfc_adapter_unlock(sa);
+
+	return -rc;
+}
+
+static int
+sfc_flow_flush(struct rte_eth_dev *dev,
+	       struct rte_flow_error *error)
+{
+	struct sfc_adapter *sa = dev->data->dev_private;
+	struct rte_flow *flow;
+	int rc = 0;
+	int ret = 0;
+
+	sfc_adapter_lock(sa);
+
+	while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
+		rc = sfc_flow_remove(sa, flow, error);
+		if (rc != 0)
+			ret = rc;
+	}
+
+	sfc_adapter_unlock(sa);
+
+	return -ret;
+}
+
+const struct rte_flow_ops sfc_flow_ops = {
+	.validate = sfc_flow_validate,
+	.create = sfc_flow_create,
+	.destroy = sfc_flow_destroy,
+	.flush = sfc_flow_flush,
+	.query = NULL,
+};
+
+void
+sfc_flow_init(struct sfc_adapter *sa)
+{
+	SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+	TAILQ_INIT(&sa->filter.flow_list);
+}
+
+void
+sfc_flow_fini(struct sfc_adapter *sa)
+{
+	struct rte_flow *flow;
+
+	SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+	while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
+		TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
+		rte_free(flow);
+	}
+}
+
+void
+sfc_flow_stop(struct sfc_adapter *sa)
+{
+	struct rte_flow *flow;
+
+	SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+	TAILQ_FOREACH(flow, &sa->filter.flow_list, entries)
+		efx_filter_remove(sa->nic, &flow->spec);
+}
+
+int
+sfc_flow_start(struct sfc_adapter *sa)
+{
+	struct rte_flow *flow;
+	int rc = 0;
+
+	sfc_log_init(sa, "entry");
+
+	SFC_ASSERT(sfc_adapter_is_locked(sa));
+
+	TAILQ_FOREACH(flow, &sa->filter.flow_list, entries) {
+		rc = efx_filter_insert(sa->nic, &flow->spec);
+		if (rc != 0)
+			goto fail_bad_flow;
+	}
+
+	sfc_log_init(sa, "done");
+
+fail_bad_flow:
+	return rc;
+}

drivers/net/sfc/sfc_flow.h

@@ -0,0 +1,62 @@
+/*-
+ * Copyright (c) 2017 Solarflare Communications Inc.
+ * All rights reserved.
+ *
+ * This software was jointly developed between OKTET Labs (under contract
+ * for Solarflare) and Solarflare Communications, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+ * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
+ * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
+ * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _SFC_FLOW_H
+#define _SFC_FLOW_H
+
+#include <rte_tailq.h>
+#include <rte_flow_driver.h>
+
+#include "efx.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* PMD-specific definition of the opaque type from rte_flow.h */
+struct rte_flow {
+	efx_filter_spec_t spec;		/* filter specification */
+	TAILQ_ENTRY(rte_flow) entries;	/* flow list entries */
+};
+
+TAILQ_HEAD(sfc_flow_list, rte_flow);
+
+extern const struct rte_flow_ops sfc_flow_ops;
+
+struct sfc_adapter;
+
+void sfc_flow_init(struct sfc_adapter *sa);
+void sfc_flow_fini(struct sfc_adapter *sa);
+int sfc_flow_start(struct sfc_adapter *sa);
+void sfc_flow_stop(struct sfc_adapter *sa);
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* _SFC_FLOW_H */