Support mlx5 flow steering with RAW data

index 49a75d31185e..eba113cf1117 100644
@@ -309,89 +309,17 @@ static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 	return NULL;
 }
 
-static bool check_last_reserved(const u32 *match_criteria)
-{
-	char *match_criteria_reserved =
-		MLX5_ADDR_OF(fte_match_param, match_criteria, MLX5_FTE_MATCH_PARAM_RESERVED);
-
-	return	!match_criteria_reserved[0] &&
-		!memcmp(match_criteria_reserved, match_criteria_reserved + 1,
-			MLX5_FLD_SZ_BYTES(fte_match_param,
-					  MLX5_FTE_MATCH_PARAM_RESERVED) - 1);
-}
-
-static bool check_valid_mask(u8 match_criteria_enable, const u32 *match_criteria)
-{
-	if (match_criteria_enable & ~(
-		(1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS)   |
-		(1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) |
-		(1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) |
-		(1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS_2)))
-		return false;
-
-	if (!(match_criteria_enable &
-	      1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS)) {
-		char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
-						  match_criteria, outer_headers);
-
-		if (fg_type_mask[0] ||
-		    memcmp(fg_type_mask, fg_type_mask + 1,
-			   MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4) - 1))
-			return false;
-	}
-
-	if (!(match_criteria_enable &
-	      1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS)) {
-		char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
-						  match_criteria, misc_parameters);
-
-		if (fg_type_mask[0] ||
-		    memcmp(fg_type_mask, fg_type_mask + 1,
-			   MLX5_ST_SZ_BYTES(fte_match_set_misc) - 1))
-			return false;
-	}
-
-	if (!(match_criteria_enable &
-	      1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS)) {
-		char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
-						  match_criteria, inner_headers);
-
-		if (fg_type_mask[0] ||
-		    memcmp(fg_type_mask, fg_type_mask + 1,
-			   MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4) - 1))
-			return false;
-	}
-
-	if (!(match_criteria_enable &
-	      1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS_2)) {
-		char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
-						  match_criteria, misc_parameters_2);
-
-		if (fg_type_mask[0] ||
-		    memcmp(fg_type_mask, fg_type_mask + 1,
-			   MLX5_ST_SZ_BYTES(fte_match_set_misc2) - 1))
-			return false;
-	}
-
-	return check_last_reserved(match_criteria);
-}
-
 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 {
 	int i;
 
-	if (!check_valid_mask(spec->match_criteria_enable, spec->match_criteria)) {
-		pr_warn("mlx5_core: Match criteria given mismatches match_criteria_enable\n");
-		return false;
-	}
-
 	for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 		if (spec->match_value[i] & ~spec->match_criteria[i]) {
 			pr_warn("mlx5_core: match_value differs from match_criteria\n");
 			return false;
 		}
 
-	return check_last_reserved(spec->match_value);
+	return true;
 }
 
 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)

@@ -1158,9 +1086,6 @@ struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 	struct mlx5_flow_group *fg;
 	int err;
 
-	if (!check_valid_mask(match_criteria_enable, match_criteria))
-		return ERR_PTR(-EINVAL);
-
 	if (ft->autogroup.active)
 		return ERR_PTR(-EPERM);
 

index b3820a34e773..0f11fff32a9b 100644
@@ -240,6 +240,9 @@ const char *parse_fs_dst(struct trace_seq *p,
 	case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE:
 		trace_seq_printf(p, "ft=%p\n", dst->ft);
 		break;
+	case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM:
+		trace_seq_printf(p, "ft_num=%u\n", dst->ft_num);
+		break;
 	case MLX5_FLOW_DESTINATION_TYPE_TIR:
 		trace_seq_printf(p, "tir=%u\n", dst->tir_num);
 		break;

index 5a00deff5457..3a04551696c0 100644
@@ -362,18 +362,20 @@ static int mlx5_cmd_set_fte(struct mlx5_core_dev *dev,
 		int list_size = 0;
 
 		list_for_each_entry(dst, &fte->node.children, node.list) {
-			unsigned int id;
+			unsigned int id, type = dst->dest_attr.type;
 
-			if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
+			if (type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
 				continue;
 
-			MLX5_SET(dest_format_struct, in_dests, destination_type,
-				 dst->dest_attr.type);
-			if (dst->dest_attr.type ==
-			    MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) {
+			switch (type) {
+			case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM:
+				id = dst->dest_attr.ft_num;
+				type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
+				break;
+			case MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE:
 				id = dst->dest_attr.ft->id;
-			} else if (dst->dest_attr.type ==
-				   MLX5_FLOW_DESTINATION_TYPE_VPORT) {
+				break;
+			case MLX5_FLOW_DESTINATION_TYPE_VPORT:
 				id = dst->dest_attr.vport.num;
 				MLX5_SET(dest_format_struct, in_dests,
 					 destination_eswitch_owner_vhca_id_valid,

@@ -381,9 +383,12 @@ static int mlx5_cmd_set_fte(struct mlx5_core_dev *dev,
 				MLX5_SET(dest_format_struct, in_dests,
 					 destination_eswitch_owner_vhca_id,
 					 dst->dest_attr.vport.vhca_id);
-			} else {
+				break;
+			default:
 				id = dst->dest_attr.tir_num;
 			}
+
+			MLX5_SET(dest_format_struct, in_dests, destination_type, type);
 			MLX5_SET(dest_format_struct, in_dests, destination_id, id);
 			in_dests += MLX5_ST_SZ_BYTES(dest_format_struct);
 			list_size++;

index eba113cf1117..69aa298a0b1c 100644
@@ -1356,7 +1356,9 @@ static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
 		     d1->ft == d2->ft) ||
 		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
-		     d1->tir_num == d2->tir_num))
+		     d1->tir_num == d2->tir_num) ||
+		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
+		     d1->ft_num == d2->ft_num))
 			return true;
 	}
 

index 757b4a30281e..a45febcf6b51 100644
@@ -89,6 +89,7 @@ struct mlx5_flow_destination {
 	enum mlx5_flow_destination_type	type;
 	union {
 		u32			tir_num;
+		u32			ft_num;
 		struct mlx5_flow_table	*ft;
 		struct mlx5_fc		*counter;
 		struct {

index 44a6ce01c3bb..fb89cc519703 100644
@@ -1181,6 +1181,7 @@ enum mlx5_flow_destination_type {
 
 	MLX5_FLOW_DESTINATION_TYPE_PORT         = 0x99,
 	MLX5_FLOW_DESTINATION_TYPE_COUNTER      = 0x100,
+	MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM = 0x101,
 };
 
 struct mlx5_ifc_dest_format_struct_bits {

index c753a34cd984..adb9209c4710 100644
@@ -376,8 +376,7 @@ static const struct uverbs_attr_spec uverbs_flow_action_esp_keymat[] = {
 static const struct uverbs_attr_spec uverbs_flow_action_esp_replay[] = {
 	[IB_UVERBS_FLOW_ACTION_ESP_REPLAY_NONE] = {
 		.type = UVERBS_ATTR_TYPE_PTR_IN,
-		/* No need to specify any data */
-		UVERBS_ATTR_SIZE(0, 0),
+		UVERBS_ATTR_NO_DATA(),
 	},
 	[IB_UVERBS_FLOW_ACTION_ESP_REPLAY_BMP] = {
 		.type = UVERBS_ATTR_TYPE_PTR_IN,

137 lines not on this page · raw message

index 67e86c8304a2..3d9544ec996f 100644
@@ -46,6 +46,7 @@
 #include <rdma/ib_user_verbs.h>
 #include <rdma/mlx5-abi.h>
 #include <rdma/uverbs_ioctl.h>
+#include <rdma/mlx5_user_ioctl_cmds.h>
 
 #define mlx5_ib_dbg(dev, format, arg...)				\
 pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\

@@ -179,6 +180,16 @@ struct mlx5_ib_flow_handler {
 	struct ib_counters		*ibcounters;
 };
 
+struct mlx5_ib_flow_matcher {
+	struct mlx5_ib_match_params matcher_mask;
+	int			mask_len;
+	enum mlx5_ib_flow_type	flow_type;
+	u16			priority;
+	struct mlx5_core_dev	*mdev;
+	atomic_t		usecnt;
+	u8			match_criteria_enable;
+};
+
 struct mlx5_ib_flow_db {
 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];

index 017ccf75890c..7f230d1ec2b8 100644
@@ -204,6 +204,8 @@ struct uverbs_object_tree_def {
 #define UVERBS_ATTR_SIZE(_min_len, _len)			\
 	.u.ptr.min_len = _min_len, .u.ptr.len = _len
 
+#define UVERBS_ATTR_NO_DATA() UVERBS_ATTR_SIZE(0, 0)
+
 /*
  * Specifies a uapi structure that cannot be extended. The user must always
  * supply the whole structure and nothing more. The structure must be declared

index 1a05bb4b0b34..233d5d140179 100644
@@ -33,6 +33,7 @@
 #ifndef MLX5_USER_IOCTL_CMDS_H
 #define MLX5_USER_IOCTL_CMDS_H
 
+#include <linux/types.h>
 #include <rdma/ib_user_ioctl_cmds.h>
 
 enum mlx5_ib_create_flow_action_attrs {

@@ -112,10 +113,40 @@ enum mlx5_ib_devx_umem_methods {
 	MLX5_IB_METHOD_DEVX_UMEM_DEREG,
 };
 
-enum mlx5_ib_devx_objects {
+enum mlx5_ib_objects {
 	MLX5_IB_OBJECT_DEVX = (1U << UVERBS_ID_NS_SHIFT),
 	MLX5_IB_OBJECT_DEVX_OBJ,
 	MLX5_IB_OBJECT_DEVX_UMEM,
+	MLX5_IB_OBJECT_FLOW_MATCHER,
+};
+
+enum mlx5_ib_flow_matcher_create_attrs {
+	MLX5_IB_ATTR_FLOW_MATCHER_CREATE_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_ATTR_FLOW_MATCHER_MATCH_MASK,
+	MLX5_IB_ATTR_FLOW_MATCHER_FLOW_TYPE,
+	MLX5_IB_ATTR_FLOW_MATCHER_MATCH_CRITERIA,
+};
+
+enum mlx5_ib_flow_matcher_destroy_attrs {
+	MLX5_IB_ATTR_FLOW_MATCHER_DESTROY_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+};
+
+enum mlx5_ib_flow_matcher_methods {
+	MLX5_IB_METHOD_FLOW_MATCHER_CREATE = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_METHOD_FLOW_MATCHER_DESTROY,
+};
+
+#define MLX5_IB_DW_MATCH_PARAM 0x80
+
+struct mlx5_ib_match_params {
+	__u32	match_params[MLX5_IB_DW_MATCH_PARAM];
+};
+
+enum mlx5_ib_flow_type {
+	MLX5_IB_FLOW_TYPE_NORMAL,
+	MLX5_IB_FLOW_TYPE_SNIFFER,
+	MLX5_IB_FLOW_TYPE_ALL_DEFAULT,
+	MLX5_IB_FLOW_TYPE_MC_DEFAULT,
 };
 
 #endif

index e4c298ac5946..312f9dbec09f 100644
@@ -2771,6 +2771,9 @@ void ib_uverbs_flow_resources_free(struct ib_uflow_resources *uflow_res)
 {
 	unsigned int i;
 
+	if (!uflow_res)
+		return;
+
 	for (i = 0; i < uflow_res->collection_num; i++)
 		atomic_dec(&uflow_res->collection[i]->usecnt);
 

@@ -3590,6 +3593,7 @@ int ib_uverbs_ex_create_flow(struct ib_uverbs_file *file,
 	}
 	atomic_inc(&qp->usecnt);
 	flow_id->qp = qp;
+	flow_id->device = qp->device;
 	flow_id->uobject = uobj;
 	uobj->object = flow_id;
 	uflow = container_of(uobj, typeof(*uflow), uobject);

index 718c8430d364..c1e0492cc78a 100644
@@ -54,9 +54,10 @@ static int uverbs_free_flow(struct ib_uobject *uobject,
 	struct ib_qp *qp = flow->qp;
 	int ret;
 
-	ret = qp->device->destroy_flow(flow);
+	ret = flow->device->destroy_flow(flow);
 	if (!ret) {
-		atomic_dec(&qp->usecnt);
+		if (qp)
+			atomic_dec(&qp->usecnt);
 		ib_uverbs_flow_resources_free(uflow->resources);
 	}
 

index 2696f1d730a1..d02bbbafbd3c 100644
@@ -2079,6 +2079,7 @@ struct ib_flow_attr {
 
 struct ib_flow {
 	struct ib_qp		*qp;
+	struct ib_device	*device;
 	struct ib_uobject	*uobject;
 };
 

index 60ac1fbe940e..32c7d8c71d85 100644
@@ -89,6 +89,28 @@ void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev,
 	mlx5_cmd_exec(dev->mdev, in, sizeof(in), out, sizeof(out));
 }
 
+bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id, int *dest_type)
+{
+	struct devx_obj *devx_obj = obj;
+	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox, opcode);
+
+	switch (opcode) {
+	case MLX5_CMD_OP_DESTROY_TIR:
+		*dest_type = MLX5_FLOW_DESTINATION_TYPE_TIR;
+		*dest_id = MLX5_GET(general_obj_in_cmd_hdr, devx_obj->dinbox,
+				    obj_id);
+		return true;
+
+	case MLX5_CMD_OP_DESTROY_FLOW_TABLE:
+		*dest_type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
+		*dest_id = MLX5_GET(destroy_flow_table_in, devx_obj->dinbox,
+				    table_id);
+		return true;
+	default:
+		return false;
+	}
+}
+
 static int devx_is_valid_obj_id(struct devx_obj *obj, const void *in)
 {
 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);

120 lines not on this page · raw message

index b7f94bc3811a..53e30449477e 100644
@@ -3643,6 +3643,15 @@ static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
 	return ERR_PTR(err);
 }
 
+struct mlx5_ib_flow_handler *
+mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
+			struct mlx5_ib_flow_matcher *fs_matcher,
+			void *cmd_in, int inlen, int dest_id,
+			int dest_type)
+{
+	return ERR_PTR(-EOPNOTSUPP);
+}
+
 static u32 mlx5_ib_flow_action_flags_to_accel_xfrm_flags(u32 mlx5_flags)
 {
 	u32 flags = 0;

index 3d9544ec996f..7c502052cb87 100644
@@ -1236,6 +1236,12 @@ int mlx5_ib_devx_create(struct mlx5_ib_dev *dev,
 void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev,
 			  struct mlx5_ib_ucontext *context);
 const struct uverbs_object_tree_def *mlx5_ib_get_devx_tree(void);
+struct mlx5_ib_flow_handler *mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
+						     struct mlx5_ib_flow_matcher *fs_matcher,
+						     void *cmd_in,
+						     int inlen, int dest_id,
+						     int dest_type);
+bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id, int *dest_type);
 #else
 static inline int
 mlx5_ib_devx_create(struct mlx5_ib_dev *dev,

@@ -1244,6 +1250,15 @@ static inline void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev,
 					struct mlx5_ib_ucontext *context) {}
 static inline const struct uverbs_object_tree_def *
 mlx5_ib_get_devx_tree(void) { return NULL; }
+static inline struct mlx5_ib_flow_handler *
+mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
+			struct mlx5_ib_flow_matcher *fs_matcher,
+			void *cmd_in,
+			int inlen, int dest_id,
+			int dest_type) { return -EOPNOTSUPP; };
+static inline bool
+mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id,
+			  int *dest_type) { return false; };
 #endif
 static inline void init_query_mad(struct ib_smp *mad)
 {

index d02bbbafbd3c..d360e6f381c8 100644
@@ -4137,6 +4137,20 @@ ib_get_vector_affinity(struct ib_device *device, int comp_vector)
 
 }
 
+static inline void ib_set_flow(struct ib_uobject *uobj, struct ib_flow *ibflow,
+			       struct ib_qp *qp, struct ib_device *device)
+{
+	uobj->object = ibflow;
+	ibflow->uobject = uobj;
+
+	if (qp) {
+		atomic_inc(&qp->usecnt);
+		ibflow->qp = qp;
+	}
+
+	ibflow->device = device;
+}
+
 /**
  * rdma_roce_rescan_device - Rescan all of the network devices in the system
  * and add their gids, as needed, to the relevant RoCE devices.

index 2eb1767042af..b3b21733cc55 100644
@@ -97,22 +97,14 @@
 		.methods = &UVERBS_OBJECT_METHODS(_object_id)                  \
 	}
 
-/* Used by drivers to declare a complete parsing tree for a single method that
- * differs only in having additional driver specific attributes.
+/* Used by drivers to declare a complete parsing tree for new methods
  */
-#define ADD_UVERBS_ATTRIBUTES_SIMPLE(_name, _object_id, _method_id, ...)       \
-	static const struct uverbs_attr_def *const UVERBS_METHOD_ATTRS(        \
-		_method_id)[] = { __VA_ARGS__ };                               \
-	static const struct uverbs_method_def UVERBS_METHOD(_method_id) = {    \
-		.id = _method_id,                                              \
-		.num_attrs = ARRAY_SIZE(UVERBS_METHOD_ATTRS(_method_id)),      \
-		.attrs = &UVERBS_METHOD_ATTRS(_method_id),                     \
-	};                                                                     \
+#define ADD_UVERBS_METHODS(_name, _object_id, ...)                             \
 	static const struct uverbs_method_def *const UVERBS_OBJECT_METHODS(    \
-		_object_id)[] = { &UVERBS_METHOD(_method_id) };                \
+		_object_id)[] = { __VA_ARGS__ };                               \
 	static const struct uverbs_object_def _name##_struct = {               \
 		.id = _object_id,                                              \
-		.num_methods = 1,                                              \
+		.num_methods = ARRAY_SIZE(UVERBS_OBJECT_METHODS(_object_id)),  \
 		.methods = &UVERBS_OBJECT_METHODS(_object_id)                  \
 	};                                                                     \
 	static const struct uverbs_object_def *const _name##_ptrs[] = {        \

@@ -123,4 +115,17 @@
 		.objects = &_name##_ptrs,                                      \
 	}
 
+/* Used by drivers to declare a complete parsing tree for a single method that
+ * differs only in having additional driver specific attributes.
+ */
+#define ADD_UVERBS_ATTRIBUTES_SIMPLE(_name, _object_id, _method_id, ...)       \
+	static const struct uverbs_attr_def *const UVERBS_METHOD_ATTRS(        \
+		_method_id)[] = { __VA_ARGS__ };                               \
+	static const struct uverbs_method_def UVERBS_METHOD(_method_id) = {    \
+		.id = _method_id,                                              \
+		.num_attrs = ARRAY_SIZE(UVERBS_METHOD_ATTRS(_method_id)),      \
+		.attrs = &UVERBS_METHOD_ATTRS(_method_id),                     \
+	};                                                                     \
+	ADD_UVERBS_METHODS(_name, _object_id, &UVERBS_METHOD(_method_id))
+
 #endif

index 233d5d140179..9c51801b9e64 100644
@@ -149,4 +149,21 @@ enum mlx5_ib_flow_type {
 	MLX5_IB_FLOW_TYPE_MC_DEFAULT,
 };
 
+enum mlx5_ib_create_flow_attrs {
+	MLX5_IB_ATTR_CREATE_FLOW_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_ATTR_CREATE_FLOW_MATCH_VALUE,
+	MLX5_IB_ATTR_CREATE_FLOW_DEST_QP,
+	MLX5_IB_ATTR_CREATE_FLOW_DEST_DEVX,
+	MLX5_IB_ATTR_CREATE_FLOW_MATCHER,
+};
+
+enum mlx5_ib_destoy_flow_attrs {
+	MLX5_IB_ATTR_DESTROY_FLOW_HANDLE = (1U << UVERBS_ID_NS_SHIFT),
+};
+
+enum mlx5_ib_flow_methods {
+	MLX5_IB_METHOD_CREATE_FLOW = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_METHOD_DESTROY_FLOW,
+};
+
 #endif

index 53e30449477e..dba3d529f49f 100644
@@ -2980,11 +2980,11 @@ static void counters_clear_description(struct ib_counters *counters)
 
 static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
 {
-	struct mlx5_ib_dev *dev = to_mdev(flow_id->qp->device);
 	struct mlx5_ib_flow_handler *handler = container_of(flow_id,
 							  struct mlx5_ib_flow_handler,
 							  ibflow);
 	struct mlx5_ib_flow_handler *iter, *tmp;
+	struct mlx5_ib_dev *dev = handler->dev;
 
 	mutex_lock(&dev->flow_db->lock);
 

@@ -3002,6 +3002,8 @@ static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
 		counters_clear_description(handler->ibcounters);
 
 	mutex_unlock(&dev->flow_db->lock);
+	if (handler->flow_matcher)
+		atomic_dec(&handler->flow_matcher->usecnt);
 	kfree(handler);
 
 	return 0;

@@ -3022,6 +3024,26 @@ enum flow_table_type {
 
 #define MLX5_FS_MAX_TYPES	 6
 #define MLX5_FS_MAX_ENTRIES	 BIT(16)
+
+static struct mlx5_ib_flow_prio *_get_prio(struct mlx5_flow_namespace *ns,
+					   struct mlx5_ib_flow_prio *prio,
+					   int priority,
+					   int num_entries, int num_groups)
+{
+	struct mlx5_flow_table *ft;
+
+	ft = mlx5_create_auto_grouped_flow_table(ns, priority,
+						 num_entries,
+						 num_groups,
+						 0, 0);
+	if (IS_ERR(ft))
+		return ERR_CAST(ft);
+
+	prio->flow_table = ft;
+	prio->refcount = 0;
+	return prio;
+}
+
 static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
 						struct ib_flow_attr *flow_attr,
 						enum flow_table_type ft_type)

@@ -3034,7 +3056,6 @@ static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
 	int num_entries;
 	int num_groups;
 	int priority;
-	int err = 0;
 
 	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
 						       log_max_ft_size));

@@ -3084,21 +3105,10 @@ static struct mlx5_ib_flow_prio *get_flow_table(struct mlx5_ib_dev *dev,
 		return ERR_PTR(-ENOMEM);
 
 	ft = prio->flow_table;
-	if (!ft) {
-		ft = mlx5_create_auto_grouped_flow_table(ns, priority,
-							 num_entries,
-							 num_groups,
-							 0, 0);
-
-		if (!IS_ERR(ft)) {
-			prio->refcount = 0;
-			prio->flow_table = ft;
-		} else {
-			err = PTR_ERR(ft);
-		}
-	}
+	if (!ft)
+		return _get_prio(ns, prio, priority, num_entries, num_groups);
 
-	return err ? ERR_PTR(err) : prio;
+	return prio;
 }
 
 static void set_underlay_qp(struct mlx5_ib_dev *dev,

@@ -3357,6 +3367,7 @@ static struct mlx5_ib_flow_handler *_create_flow_rule(struct mlx5_ib_dev *dev,
 
 	ft_prio->refcount++;
 	handler->prio = ft_prio;
+	handler->dev = dev;
 
 	ft_prio->flow_table = ft;
 free:

@@ -3643,13 +3654,184 @@ static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
 	return ERR_PTR(err);
 }
 
+static struct mlx5_ib_flow_prio *_get_flow_table(struct mlx5_ib_dev *dev,
+						 int priority, bool mcast)
+{
+	int max_table_size;
+	struct mlx5_flow_namespace *ns = NULL;
+	struct mlx5_ib_flow_prio *prio;
+
+	max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
+			     log_max_ft_size));
+	if (max_table_size < MLX5_FS_MAX_ENTRIES)
+		return ERR_PTR(-ENOMEM);
+
+	if (mcast)
+		priority = MLX5_IB_FLOW_MCAST_PRIO;
+	else
+		priority = ib_prio_to_core_prio(priority, false);
+
+	ns = mlx5_get_flow_namespace(dev->mdev, MLX5_FLOW_NAMESPACE_BYPASS);
+	if (!ns)
+		return ERR_PTR(-ENOTSUPP);
+
+	prio = &dev->flow_db->prios[priority];
+
+	if (prio->flow_table)
+		return prio;
+
+	return _get_prio(ns, prio, priority, MLX5_FS_MAX_ENTRIES,
+			 MLX5_FS_MAX_TYPES);
+}
+
+static struct mlx5_ib_flow_handler *
+_create_raw_flow_rule(struct mlx5_ib_dev *dev,
+		      struct mlx5_ib_flow_prio *ft_prio,
+		      struct mlx5_flow_destination *dst,
+		      struct mlx5_ib_flow_matcher  *fs_matcher,
+		      void *cmd_in, int inlen)
+{
+	struct mlx5_ib_flow_handler *handler;
+	struct mlx5_flow_act flow_act = {.flow_tag = MLX5_FS_DEFAULT_FLOW_TAG};
+	struct mlx5_flow_spec *spec;
+	struct mlx5_flow_table *ft = ft_prio->flow_table;
+	int err = 0;
+
+	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
+	handler = kzalloc(sizeof(*handler), GFP_KERNEL);
+	if (!handler || !spec) {
+		err = -ENOMEM;
+		goto free;
+	}
+
+	INIT_LIST_HEAD(&handler->list);
+
+	memcpy(spec->match_value, cmd_in, inlen);
+	memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
+	       fs_matcher->mask_len);
+	spec->match_criteria_enable = fs_matcher->match_criteria_enable;
+
+	flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
+	handler->rule = mlx5_add_flow_rules(ft, spec,
+					    &flow_act, dst, 1);
+
+	if (IS_ERR(handler->rule)) {
+		err = PTR_ERR(handler->rule);
+		goto free;
+	}
+
+	ft_prio->refcount++;
+	handler->prio = ft_prio;
+	handler->dev = dev;
+	ft_prio->flow_table = ft;
+
+free:
+	if (err)
+		kfree(handler);
+	kvfree(spec);
+	return err ? ERR_PTR(err) : handler;
+}
+
+static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
+				void *match_v)
+{
+	void *match_c;
+	void *match_v_set_lyr_2_4, *match_c_set_lyr_2_4;
+	void *dmac, *dmac_mask;
+	void *ipv4, *ipv4_mask;
+
+	if (!(fs_matcher->match_criteria_enable &
+	      (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
+		return false;
+
+	match_c = fs_matcher->matcher_mask.match_params;
+	match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
+					   outer_headers);
+	match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
+					   outer_headers);
+
+	dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
+			    dmac_47_16);
+	dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
+				 dmac_47_16);
+
+	if (is_multicast_ether_addr(dmac) &&
+	    is_multicast_ether_addr(dmac_mask))
+		return true;
+
+	ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
+			    dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
+
+	ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
+				 dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
+
+	if (ipv4_is_multicast(*(__be32 *)(ipv4)) &&
+	    ipv4_is_multicast(*(__be32 *)(ipv4_mask)))
+		return true;
+
+	return false;
+}
+
 struct mlx5_ib_flow_handler *
 mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
 			struct mlx5_ib_flow_matcher *fs_matcher,
 			void *cmd_in, int inlen, int dest_id,
 			int dest_type)
 {
-	return ERR_PTR(-EOPNOTSUPP);
+	struct mlx5_flow_destination *dst;
+	struct mlx5_ib_flow_prio *ft_prio;
+	int priority = fs_matcher->priority;
+	struct mlx5_ib_flow_handler *handler;
+	bool mcast;
+	int err;
+
+	if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
+		return ERR_PTR(-EOPNOTSUPP);
+
+	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
+		return ERR_PTR(-ENOMEM);
+
+	if (dest_type != MLX5_FLOW_DESTINATION_TYPE_TIR)
+		return ERR_PTR(-ENOTSUPP);
+
+	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
+	if (!dst)
+		return ERR_PTR(-ENOMEM);
+
+	mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
+	mutex_lock(&dev->flow_db->lock);
+
+	ft_prio = _get_flow_table(dev, priority, mcast);
+	if (IS_ERR(ft_prio)) {
+		err = PTR_ERR(ft_prio);
+		goto unlock;
+	}
+
+	dst->type = dest_type;
+	dst->tir_num = dest_id;
+	handler = _create_raw_flow_rule(dev, ft_prio, dst, fs_matcher, cmd_in,
+					inlen);
+
+	if (IS_ERR(handler)) {
+		err = PTR_ERR(handler);
+		goto destroy_ft;
+	}
+
+	mutex_unlock(&dev->flow_db->lock);
+	atomic_inc(&fs_matcher->usecnt);
+	handler->flow_matcher = fs_matcher;
+
+	kfree(dst);
+
+	return handler;
+
+destroy_ft:
+	put_flow_table(dev, ft_prio, false);
+unlock:
+	mutex_unlock(&dev->flow_db->lock);
+	kfree(dst);
+
+	return ERR_PTR(err);
 }
 
 static u32 mlx5_ib_flow_action_flags_to_accel_xfrm_flags(u32 mlx5_flags)

index 7c502052cb87..7690dc6a897e 100644
@@ -178,6 +178,8 @@ struct mlx5_ib_flow_handler {
 	struct mlx5_ib_flow_prio	*prio;
 	struct mlx5_flow_handle		*rule;
 	struct ib_counters		*ibcounters;
+	struct mlx5_ib_dev		*dev;
+	struct mlx5_ib_flow_matcher	*flow_matcher;
 };
 
 struct mlx5_ib_flow_matcher {

index dba3d529f49f..d750562cb2c5 100644
@@ -3791,9 +3791,6 @@ mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
 	if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
 		return ERR_PTR(-ENOMEM);
 
-	if (dest_type != MLX5_FLOW_DESTINATION_TYPE_TIR)
-		return ERR_PTR(-ENOTSUPP);
-
 	dst = kzalloc(sizeof(*dst), GFP_KERNEL);
 	if (!dst)
 		return ERR_PTR(-ENOMEM);

@@ -3807,8 +3804,14 @@ mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
 		goto unlock;
 	}
 
-	dst->type = dest_type;
-	dst->tir_num = dest_id;
+	if (dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR) {
+		dst->type = dest_type;
+		dst->tir_num = dest_id;
+	} else {
+		dst->type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
+		dst->ft_num = dest_id;
+	}
+
 	handler = _create_raw_flow_rule(dev, ft_prio, dst, fs_matcher, cmd_in,
 					inlen);
 

index 76e76b87e5e9..ff5a02713b87 100644
@@ -243,3 +243,12 @@ DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER,
 DECLARE_UVERBS_OBJECT_TREE(flow_objects,
 			&UVERBS_OBJECT(MLX5_IB_OBJECT_FLOW_MATCHER));
 
+int mlx5_ib_get_flow_trees(const struct uverbs_object_tree_def **root)
+{
+	int i = 0;
+
+	root[i++] = &flow_objects;
+	root[i++] = &mlx5_ib_fs;
+
+	return i;
+}

index d750562cb2c5..71f3e9677622 100644
@@ -5537,7 +5537,7 @@ ADD_UVERBS_ATTRIBUTES_SIMPLE(
 			   UVERBS_ATTR_TYPE(u64),
 			   UA_MANDATORY));
 
-#define NUM_TREES	3
+#define NUM_TREES	5
 static int populate_specs_root(struct mlx5_ib_dev *dev)
 {
 	const struct uverbs_object_tree_def *default_root[NUM_TREES + 1] = {

@@ -5557,6 +5557,8 @@ static int populate_specs_root(struct mlx5_ib_dev *dev)
 	    !WARN_ON(num_trees >= ARRAY_SIZE(default_root)))
 		default_root[num_trees++] = mlx5_ib_get_devx_tree();
 
+	num_trees += mlx5_ib_get_flow_trees(default_root + num_trees);
+
 	dev->ib_dev.driver_specs_root =
 		uverbs_alloc_spec_tree(num_trees, default_root);
 

index 7690dc6a897e..623f5d8bd48e 100644
@@ -1244,6 +1244,7 @@ struct mlx5_ib_flow_handler *mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
 						     int inlen, int dest_id,
 						     int dest_type);
 bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id, int *dest_type);
+int mlx5_ib_get_flow_trees(const struct uverbs_object_tree_def **root);
 #else
 static inline int
 mlx5_ib_devx_create(struct mlx5_ib_dev *dev,

@@ -1261,6 +1262,8 @@ mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
 static inline bool
 mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id,
 			  int *dest_type) { return false; };
+static inline int
+mlx5_ib_get_flow_trees(const struct uverbs_object_tree_def **root) { return 0; };
 #endif
 static inline void init_query_mad(struct ib_smp *mad)
 {