fs_core.c

   1/*
   2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the
   8 * OpenIB.org BSD license below:
   9 *
  10 *     Redistribution and use in source and binary forms, with or
  11 *     without modification, are permitted provided that the following
  12 *     conditions are met:
  13 *
  14 *      - Redistributions of source code must retain the above
  15 *        copyright notice, this list of conditions and the following
  16 *        disclaimer.
  17 *
  18 *      - Redistributions in binary form must reproduce the above
  19 *        copyright notice, this list of conditions and the following
  20 *        disclaimer in the documentation and/or other materials
  21 *        provided with the distribution.
  22 *
  23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30 * SOFTWARE.
  31 */
  32
  33#include <linux/mutex.h>
  34#include <linux/mlx5/driver.h>
  35#include <linux/mlx5/vport.h>
  36#include <linux/mlx5/eswitch.h>
  37#include <net/devlink.h>
  38
  39#include "mlx5_core.h"
  40#include "fs_core.h"
  41#include "fs_cmd.h"
  42#include "fs_ft_pool.h"
  43#include "diag/fs_tracepoint.h"
  44#include "devlink.h"
  45
  46#define INIT_TREE_NODE_ARRAY_SIZE(...)	(sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  47					 sizeof(struct init_tree_node))
  48
  49#define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  50		 ...) {.type = FS_TYPE_PRIO,\
  51	.min_ft_level = min_level_val,\
  52	.num_levels = num_levels_val,\
  53	.num_leaf_prios = num_prios_val,\
  54	.caps = caps_val,\
  55	.children = (struct init_tree_node[]) {__VA_ARGS__},\
  56	.ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  57}
  58
  59#define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  60	ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  61		 __VA_ARGS__)\
  62
  63#define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE,	\
  64	.def_miss_action = def_miss_act,\
  65	.children = (struct init_tree_node[]) {__VA_ARGS__},\
  66	.ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  67}
  68
  69#define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  70				   sizeof(long))
  71
  72#define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  73
  74#define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  75			       .caps = (long[]) {__VA_ARGS__} }
  76
  77#define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  78					   FS_CAP(flow_table_properties_nic_receive.modify_root), \
  79					   FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  80					   FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  81
  82#define FS_CHAINING_CAPS_EGRESS                                                \
  83	FS_REQUIRED_CAPS(                                                      \
  84		FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  85		FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  86		FS_CAP(flow_table_properties_nic_transmit                      \
  87			       .identified_miss_table_mode),                   \
  88		FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  89
  90#define FS_CHAINING_CAPS_RDMA_TX                                                \
  91	FS_REQUIRED_CAPS(                                                       \
  92		FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \
  93		FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root),    \
  94		FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  95			       .identified_miss_table_mode),                    \
  96		FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  97			       .flow_table_modify))
  98
  99#define LEFTOVERS_NUM_LEVELS 1
 100#define LEFTOVERS_NUM_PRIOS 1
 101
 102#define RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1
 103#define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1
 104
 105#define BY_PASS_PRIO_NUM_LEVELS 1
 106#define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
 107			   LEFTOVERS_NUM_PRIOS)
 108
 109#define KERNEL_RX_MACSEC_NUM_PRIOS  1
 110#define KERNEL_RX_MACSEC_NUM_LEVELS 3
 111#define KERNEL_RX_MACSEC_MIN_LEVEL (BY_PASS_MIN_LEVEL + KERNEL_RX_MACSEC_NUM_PRIOS)
 112
 113#define ETHTOOL_PRIO_NUM_LEVELS 1
 114#define ETHTOOL_NUM_PRIOS 11
 115#define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 116/* Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}}, IPsec policy,
 117 * IPsec policy miss, {IPsec RoCE MPV,Alias table},IPsec RoCE policy
 118 */
 119#define KERNEL_NIC_PRIO_NUM_LEVELS 11
 120#define KERNEL_NIC_NUM_PRIOS 1
 121/* One more level for tc, and one more for promisc */
 122#define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 2)
 123
 124#define KERNEL_NIC_PROMISC_NUM_PRIOS 1
 125#define KERNEL_NIC_PROMISC_NUM_LEVELS 1
 126
 127#define KERNEL_NIC_TC_NUM_PRIOS  1
 128#define KERNEL_NIC_TC_NUM_LEVELS 3
 129
 130#define ANCHOR_NUM_LEVELS 1
 131#define ANCHOR_NUM_PRIOS 1
 132#define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 133
 134#define OFFLOADS_MAX_FT 2
 135#define OFFLOADS_NUM_PRIOS 2
 136#define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
 137
 138#define LAG_PRIO_NUM_LEVELS 1
 139#define LAG_NUM_PRIOS 1
 140#define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + KERNEL_RX_MACSEC_MIN_LEVEL + 1)
 141
 142#define KERNEL_TX_IPSEC_NUM_PRIOS  1
 143#define KERNEL_TX_IPSEC_NUM_LEVELS 4
 144#define KERNEL_TX_IPSEC_MIN_LEVEL        (KERNEL_TX_IPSEC_NUM_LEVELS)
 145
 146#define KERNEL_TX_MACSEC_NUM_PRIOS  1
 147#define KERNEL_TX_MACSEC_NUM_LEVELS 2
 148#define KERNEL_TX_MACSEC_MIN_LEVEL       (KERNEL_TX_IPSEC_MIN_LEVEL + KERNEL_TX_MACSEC_NUM_PRIOS)
 149
 150struct node_caps {
 151	size_t	arr_sz;
 152	long	*caps;
 153};
 154
 155static struct init_tree_node {
 156	enum fs_node_type	type;
 157	struct init_tree_node *children;
 158	int ar_size;
 159	struct node_caps caps;
 160	int min_ft_level;
 161	int num_leaf_prios;
 162	int prio;
 163	int num_levels;
 164	enum mlx5_flow_table_miss_action def_miss_action;
 165} root_fs = {
 166	.type = FS_TYPE_NAMESPACE,
 167	.ar_size = 8,
 168	  .children = (struct init_tree_node[]){
 169		  ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 170			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 171				  ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 172						    BY_PASS_PRIO_NUM_LEVELS))),
 173		  ADD_PRIO(0, KERNEL_RX_MACSEC_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 174			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 175				  ADD_MULTIPLE_PRIO(KERNEL_RX_MACSEC_NUM_PRIOS,
 176						    KERNEL_RX_MACSEC_NUM_LEVELS))),
 177		  ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 178			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 179				  ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 180						    LAG_PRIO_NUM_LEVELS))),
 181		  ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 182			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 183				  ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 184						    OFFLOADS_MAX_FT))),
 185		  ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 186			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 187				  ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 188						    ETHTOOL_PRIO_NUM_LEVELS))),
 189		  ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 190			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 191				  ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 192						    KERNEL_NIC_TC_NUM_LEVELS),
 193				  ADD_MULTIPLE_PRIO(KERNEL_NIC_PROMISC_NUM_PRIOS,
 194						    KERNEL_NIC_PROMISC_NUM_LEVELS),
 195				  ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 196						    KERNEL_NIC_PRIO_NUM_LEVELS))),
 197		  ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 198			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 199				  ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 200						    LEFTOVERS_NUM_LEVELS))),
 201		  ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 202			   ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 203				  ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 204						    ANCHOR_NUM_LEVELS))),
 205	}
 206};
 207
 208static struct init_tree_node egress_root_fs = {
 209	.type = FS_TYPE_NAMESPACE,
 210	.ar_size = 3,
 211	.children = (struct init_tree_node[]) {
 212		ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 213			 FS_CHAINING_CAPS_EGRESS,
 214			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 215				ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 216						  BY_PASS_PRIO_NUM_LEVELS))),
 217		ADD_PRIO(0, KERNEL_TX_IPSEC_MIN_LEVEL, 0,
 218			 FS_CHAINING_CAPS_EGRESS,
 219			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 220				ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
 221						  KERNEL_TX_IPSEC_NUM_LEVELS))),
 222		ADD_PRIO(0, KERNEL_TX_MACSEC_MIN_LEVEL, 0,
 223			 FS_CHAINING_CAPS_EGRESS,
 224			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 225				ADD_MULTIPLE_PRIO(KERNEL_TX_MACSEC_NUM_PRIOS,
 226						  KERNEL_TX_MACSEC_NUM_LEVELS))),
 227	}
 228};
 229
 230enum {
 231	RDMA_RX_IPSEC_PRIO,
 232	RDMA_RX_MACSEC_PRIO,
 233	RDMA_RX_COUNTERS_PRIO,
 234	RDMA_RX_BYPASS_PRIO,
 235	RDMA_RX_KERNEL_PRIO,
 236};
 237
 238#define RDMA_RX_IPSEC_NUM_PRIOS 1
 239#define RDMA_RX_IPSEC_NUM_LEVELS 4
 240#define RDMA_RX_IPSEC_MIN_LEVEL  (RDMA_RX_IPSEC_NUM_LEVELS)
 241
 242#define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS
 243#define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1)
 244#define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2)
 245
 246#define RDMA_RX_MACSEC_NUM_PRIOS 1
 247#define RDMA_RX_MACSEC_PRIO_NUM_LEVELS 2
 248#define RDMA_RX_MACSEC_MIN_LEVEL  (RDMA_RX_COUNTERS_MIN_LEVEL + RDMA_RX_MACSEC_NUM_PRIOS)
 249
 250static struct init_tree_node rdma_rx_root_fs = {
 251	.type = FS_TYPE_NAMESPACE,
 252	.ar_size = 5,
 253	.children = (struct init_tree_node[]) {
 254		[RDMA_RX_IPSEC_PRIO] =
 255		ADD_PRIO(0, RDMA_RX_IPSEC_MIN_LEVEL, 0,
 256			 FS_CHAINING_CAPS,
 257			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 258				ADD_MULTIPLE_PRIO(RDMA_RX_IPSEC_NUM_PRIOS,
 259						  RDMA_RX_IPSEC_NUM_LEVELS))),
 260		[RDMA_RX_MACSEC_PRIO] =
 261		ADD_PRIO(0, RDMA_RX_MACSEC_MIN_LEVEL, 0,
 262			 FS_CHAINING_CAPS,
 263			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 264				ADD_MULTIPLE_PRIO(RDMA_RX_MACSEC_NUM_PRIOS,
 265						  RDMA_RX_MACSEC_PRIO_NUM_LEVELS))),
 266		[RDMA_RX_COUNTERS_PRIO] =
 267		ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0,
 268			 FS_CHAINING_CAPS,
 269			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 270				ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS,
 271						  RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))),
 272		[RDMA_RX_BYPASS_PRIO] =
 273		ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0,
 274			 FS_CHAINING_CAPS,
 275			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 276				ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 277						  BY_PASS_PRIO_NUM_LEVELS))),
 278		[RDMA_RX_KERNEL_PRIO] =
 279		ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0,
 280			 FS_CHAINING_CAPS,
 281			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 282				ADD_MULTIPLE_PRIO(1, 1))),
 283	}
 284};
 285
 286enum {
 287	RDMA_TX_COUNTERS_PRIO,
 288	RDMA_TX_IPSEC_PRIO,
 289	RDMA_TX_MACSEC_PRIO,
 290	RDMA_TX_BYPASS_PRIO,
 291};
 292
 293#define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS
 294#define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1)
 295
 296#define RDMA_TX_IPSEC_NUM_PRIOS 2
 297#define RDMA_TX_IPSEC_PRIO_NUM_LEVELS 1
 298#define RDMA_TX_IPSEC_MIN_LEVEL  (RDMA_TX_COUNTERS_MIN_LEVEL + RDMA_TX_IPSEC_NUM_PRIOS)
 299
 300#define RDMA_TX_MACSEC_NUM_PRIOS 1
 301#define RDMA_TX_MACESC_PRIO_NUM_LEVELS 1
 302#define RDMA_TX_MACSEC_MIN_LEVEL  (RDMA_TX_COUNTERS_MIN_LEVEL + RDMA_TX_MACSEC_NUM_PRIOS)
 303
 304static struct init_tree_node rdma_tx_root_fs = {
 305	.type = FS_TYPE_NAMESPACE,
 306	.ar_size = 4,
 307	.children = (struct init_tree_node[]) {
 308		[RDMA_TX_COUNTERS_PRIO] =
 309		ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0,
 310			 FS_CHAINING_CAPS,
 311			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 312				ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS,
 313						  RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))),
 314		[RDMA_TX_IPSEC_PRIO] =
 315		ADD_PRIO(0, RDMA_TX_IPSEC_MIN_LEVEL, 0,
 316			 FS_CHAINING_CAPS,
 317			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 318				ADD_MULTIPLE_PRIO(RDMA_TX_IPSEC_NUM_PRIOS,
 319						  RDMA_TX_IPSEC_PRIO_NUM_LEVELS))),
 320		[RDMA_TX_MACSEC_PRIO] =
 321		ADD_PRIO(0, RDMA_TX_MACSEC_MIN_LEVEL, 0,
 322			 FS_CHAINING_CAPS,
 323			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 324				ADD_MULTIPLE_PRIO(RDMA_TX_MACSEC_NUM_PRIOS,
 325						  RDMA_TX_MACESC_PRIO_NUM_LEVELS))),
 326		[RDMA_TX_BYPASS_PRIO] =
 327		ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0,
 328			 FS_CHAINING_CAPS_RDMA_TX,
 329			 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 330				ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL,
 331						  BY_PASS_PRIO_NUM_LEVELS))),
 332	}
 333};
 334
 335enum fs_i_lock_class {
 336	FS_LOCK_GRANDPARENT,
 337	FS_LOCK_PARENT,
 338	FS_LOCK_CHILD
 339};
 340
 341static const struct rhashtable_params rhash_fte = {
 342	.key_len = sizeof_field(struct fs_fte, val),
 343	.key_offset = offsetof(struct fs_fte, val),
 344	.head_offset = offsetof(struct fs_fte, hash),
 345	.automatic_shrinking = true,
 346	.min_size = 1,
 347};
 348
 349static const struct rhashtable_params rhash_fg = {
 350	.key_len = sizeof_field(struct mlx5_flow_group, mask),
 351	.key_offset = offsetof(struct mlx5_flow_group, mask),
 352	.head_offset = offsetof(struct mlx5_flow_group, hash),
 353	.automatic_shrinking = true,
 354	.min_size = 1,
 355
 356};
 357
 358static void del_hw_flow_table(struct fs_node *node);
 359static void del_hw_flow_group(struct fs_node *node);
 360static void del_hw_fte(struct fs_node *node);
 361static void del_sw_flow_table(struct fs_node *node);
 362static void del_sw_flow_group(struct fs_node *node);
 363static void del_sw_fte(struct fs_node *node);
 364static void del_sw_prio(struct fs_node *node);
 365static void del_sw_ns(struct fs_node *node);
 366/* Delete rule (destination) is special case that
 367 * requires to lock the FTE for all the deletion process.
 368 */
 369static void del_sw_hw_rule(struct fs_node *node);
 370static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 371				struct mlx5_flow_destination *d2);
 372static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 373static struct mlx5_flow_rule *
 374find_flow_rule(struct fs_fte *fte,
 375	       struct mlx5_flow_destination *dest);
 376
 377static void tree_init_node(struct fs_node *node,
 378			   void (*del_hw_func)(struct fs_node *),
 379			   void (*del_sw_func)(struct fs_node *))
 380{
 381	refcount_set(&node->refcount, 1);
 382	INIT_LIST_HEAD(&node->list);
 383	INIT_LIST_HEAD(&node->children);
 384	init_rwsem(&node->lock);
 385	node->del_hw_func = del_hw_func;
 386	node->del_sw_func = del_sw_func;
 387	node->active = false;
 388}
 389
 390static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 391{
 392	if (parent)
 393		refcount_inc(&parent->refcount);
 394	node->parent = parent;
 395
 396	/* Parent is the root */
 397	if (!parent)
 398		node->root = node;
 399	else
 400		node->root = parent->root;
 401}
 402
 403static int tree_get_node(struct fs_node *node)
 404{
 405	return refcount_inc_not_zero(&node->refcount);
 406}
 407
 408static void nested_down_read_ref_node(struct fs_node *node,
 409				      enum fs_i_lock_class class)
 410{
 411	if (node) {
 412		down_read_nested(&node->lock, class);
 413		refcount_inc(&node->refcount);
 414	}
 415}
 416
 417static void nested_down_write_ref_node(struct fs_node *node,
 418				       enum fs_i_lock_class class)
 419{
 420	if (node) {
 421		down_write_nested(&node->lock, class);
 422		refcount_inc(&node->refcount);
 423	}
 424}
 425
 426static void down_write_ref_node(struct fs_node *node, bool locked)
 427{
 428	if (node) {
 429		if (!locked)
 430			down_write(&node->lock);
 431		refcount_inc(&node->refcount);
 432	}
 433}
 434
 435static void up_read_ref_node(struct fs_node *node)
 436{
 437	refcount_dec(&node->refcount);
 438	up_read(&node->lock);
 439}
 440
 441static void up_write_ref_node(struct fs_node *node, bool locked)
 442{
 443	refcount_dec(&node->refcount);
 444	if (!locked)
 445		up_write(&node->lock);
 446}
 447
 448static void tree_put_node(struct fs_node *node, bool locked)
 449{
 450	struct fs_node *parent_node = node->parent;
 451
 452	if (refcount_dec_and_test(&node->refcount)) {
 453		if (node->del_hw_func)
 454			node->del_hw_func(node);
 455		if (parent_node) {
 456			down_write_ref_node(parent_node, locked);
 457			list_del_init(&node->list);
 458		}
 459		node->del_sw_func(node);
 460		if (parent_node)
 461			up_write_ref_node(parent_node, locked);
 462		node = NULL;
 463	}
 464	if (!node && parent_node)
 465		tree_put_node(parent_node, locked);
 466}
 467
 468static int tree_remove_node(struct fs_node *node, bool locked)
 469{
 470	if (refcount_read(&node->refcount) > 1) {
 471		refcount_dec(&node->refcount);
 472		return -EEXIST;
 473	}
 474	tree_put_node(node, locked);
 475	return 0;
 476}
 477
 478static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 479				 unsigned int prio)
 480{
 481	struct fs_prio *iter_prio;
 482
 483	fs_for_each_prio(iter_prio, ns) {
 484		if (iter_prio->prio == prio)
 485			return iter_prio;
 486	}
 487
 488	return NULL;
 489}
 490
 491static bool is_fwd_next_action(u32 action)
 492{
 493	return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
 494			 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
 495}
 496
 497static bool is_fwd_dest_type(enum mlx5_flow_destination_type type)
 498{
 499	return type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM ||
 500		type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE ||
 501		type == MLX5_FLOW_DESTINATION_TYPE_UPLINK ||
 502		type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
 503		type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER ||
 504		type == MLX5_FLOW_DESTINATION_TYPE_TIR ||
 505		type == MLX5_FLOW_DESTINATION_TYPE_RANGE ||
 506		type == MLX5_FLOW_DESTINATION_TYPE_TABLE_TYPE;
 507}
 508
 509static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 510{
 511	int i;
 512
 513	for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 514		if (spec->match_value[i] & ~spec->match_criteria[i]) {
 515			pr_warn("mlx5_core: match_value differs from match_criteria\n");
 516			return false;
 517		}
 518
 519	return true;
 520}
 521
 522struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 523{
 524	struct fs_node *root;
 525	struct mlx5_flow_namespace *ns;
 526
 527	root = node->root;
 528
 529	if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 530		pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 531		return NULL;
 532	}
 533
 534	ns = container_of(root, struct mlx5_flow_namespace, node);
 535	return container_of(ns, struct mlx5_flow_root_namespace, ns);
 536}
 537
 538static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 539{
 540	struct mlx5_flow_root_namespace *root = find_root(node);
 541
 542	if (root)
 543		return root->dev->priv.steering;
 544	return NULL;
 545}
 546
 547static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 548{
 549	struct mlx5_flow_root_namespace *root = find_root(node);
 550
 551	if (root)
 552		return root->dev;
 553	return NULL;
 554}
 555
 556static void del_sw_ns(struct fs_node *node)
 557{
 558	kfree(node);
 559}
 560
 561static void del_sw_prio(struct fs_node *node)
 562{
 563	kfree(node);
 564}
 565
 566static void del_hw_flow_table(struct fs_node *node)
 567{
 568	struct mlx5_flow_root_namespace *root;
 569	struct mlx5_flow_table *ft;
 570	struct mlx5_core_dev *dev;
 571	int err;
 572
 573	fs_get_obj(ft, node);
 574	dev = get_dev(&ft->node);
 575	root = find_root(&ft->node);
 576	trace_mlx5_fs_del_ft(ft);
 577
 578	if (node->active) {
 579		err = root->cmds->destroy_flow_table(root, ft);
 580		if (err)
 581			mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 582	}
 583}
 584
 585static void del_sw_flow_table(struct fs_node *node)
 586{
 587	struct mlx5_flow_table *ft;
 588	struct fs_prio *prio;
 589
 590	fs_get_obj(ft, node);
 591
 592	rhltable_destroy(&ft->fgs_hash);
 593	if (ft->node.parent) {
 594		fs_get_obj(prio, ft->node.parent);
 595		prio->num_ft--;
 596	}
 597	kfree(ft);
 598}
 599
 600static void modify_fte(struct fs_fte *fte)
 601{
 602	struct mlx5_flow_root_namespace *root;
 603	struct mlx5_flow_table *ft;
 604	struct mlx5_flow_group *fg;
 605	struct mlx5_core_dev *dev;
 606	int err;
 607
 608	fs_get_obj(fg, fte->node.parent);
 609	fs_get_obj(ft, fg->node.parent);
 610	dev = get_dev(&fte->node);
 611
 612	root = find_root(&ft->node);
 613	err = root->cmds->update_fte(root, ft, fg, fte->act_dests.modify_mask, fte);
 614	if (err)
 615		mlx5_core_warn(dev,
 616			       "%s can't del rule fg id=%d fte_index=%d\n",
 617			       __func__, fg->id, fte->index);
 618	fte->act_dests.modify_mask = 0;
 619}
 620
 621static void del_sw_hw_dup_rule(struct fs_node *node)
 622{
 623	struct mlx5_flow_rule *rule;
 624	struct fs_fte *fte;
 625
 626	fs_get_obj(rule, node);
 627	fs_get_obj(fte, rule->node.parent);
 628	trace_mlx5_fs_del_rule(rule);
 629
 630	if (is_fwd_next_action(rule->sw_action)) {
 631		mutex_lock(&rule->dest_attr.ft->lock);
 632		list_del(&rule->next_ft);
 633		mutex_unlock(&rule->dest_attr.ft->lock);
 634	}
 635
 636	/* If a pending rule is being deleted it means
 637	 * this is a NO APPEND rule, so there are no partial deletions,
 638	 * all the rules of the mlx5_flow_handle are going to be deleted
 639	 * and the rules aren't shared with any other mlx5_flow_handle instance
 640	 * so no need to do any bookkeeping like in del_sw_hw_rule().
 641	 */
 642
 643	kfree(rule);
 644}
 645
 646static void del_sw_hw_rule(struct fs_node *node)
 647{
 648	struct mlx5_flow_rule *rule;
 649	struct fs_fte *fte;
 650
 651	fs_get_obj(rule, node);
 652	fs_get_obj(fte, rule->node.parent);
 653	trace_mlx5_fs_del_rule(rule);
 654	if (is_fwd_next_action(rule->sw_action)) {
 655		mutex_lock(&rule->dest_attr.ft->lock);
 656		list_del(&rule->next_ft);
 657		mutex_unlock(&rule->dest_attr.ft->lock);
 658	}
 659
 660	if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER) {
 661		--fte->act_dests.dests_size;
 662		fte->act_dests.modify_mask |=
 663			BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 664			BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 665		fte->act_dests.action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 666		mlx5_fc_local_put(rule->dest_attr.counter);
 667		goto out;
 668	}
 669
 670	if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT) {
 671		--fte->act_dests.dests_size;
 672		fte->act_dests.modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 673		fte->act_dests.action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
 674		goto out;
 675	}
 676
 677	if (is_fwd_dest_type(rule->dest_attr.type)) {
 678		--fte->act_dests.dests_size;
 679		--fte->act_dests.fwd_dests;
 680
 681		if (!fte->act_dests.fwd_dests)
 682			fte->act_dests.action.action &=
 683				~MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
 684		fte->act_dests.modify_mask |=
 685			BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 686		goto out;
 687	}
 688out:
 689	kfree(rule);
 690}
 691
 692static void switch_to_pending_act_dests(struct fs_fte *fte)
 693{
 694	struct fs_node *iter;
 695
 696	memcpy(&fte->act_dests, &fte->dup->act_dests, sizeof(fte->act_dests));
 697
 698	list_bulk_move_tail(&fte->node.children,
 699			    fte->dup->children.next,
 700			    fte->dup->children.prev);
 701
 702	list_for_each_entry(iter, &fte->node.children, list)
 703		iter->del_sw_func = del_sw_hw_rule;
 704
 705	/* Make sure the fte isn't deleted
 706	 * as mlx5_del_flow_rules() decreases the refcount
 707	 * of the fte to trigger deletion.
 708	 */
 709	tree_get_node(&fte->node);
 710}
 711
 712static void del_hw_fte(struct fs_node *node)
 713{
 714	struct mlx5_flow_root_namespace *root;
 715	struct mlx5_flow_table *ft;
 716	struct mlx5_flow_group *fg;
 717	struct mlx5_core_dev *dev;
 718	bool pending_used = false;
 719	struct fs_fte *fte;
 720	int err;
 721
 722	fs_get_obj(fte, node);
 723	fs_get_obj(fg, fte->node.parent);
 724	fs_get_obj(ft, fg->node.parent);
 725
 726	trace_mlx5_fs_del_fte(fte);
 727	WARN_ON(fte->act_dests.dests_size);
 728	dev = get_dev(&ft->node);
 729	root = find_root(&ft->node);
 730
 731	if (fte->dup && !list_empty(&fte->dup->children)) {
 732		switch_to_pending_act_dests(fte);
 733		pending_used = true;
 734	} else {
 735		/* Avoid double call to del_hw_fte */
 736		node->del_hw_func = NULL;
 737	}
 738
 739	if (node->active) {
 740		if (pending_used) {
 741			err = root->cmds->update_fte(root, ft, fg,
 742						     fte->act_dests.modify_mask, fte);
 743			if (err)
 744				mlx5_core_warn(dev,
 745					       "flow steering can't update to pending rule in index %d of flow group id %d\n",
 746					       fte->index, fg->id);
 747			fte->act_dests.modify_mask = 0;
 748		} else {
 749			err = root->cmds->delete_fte(root, ft, fte);
 750			if (err)
 751				mlx5_core_warn(dev,
 752					       "flow steering can't delete fte in index %d of flow group id %d\n",
 753					       fte->index, fg->id);
 754			node->active = false;
 755		}
 756	}
 757}
 758
 759static void del_sw_fte(struct fs_node *node)
 760{
 761	struct mlx5_flow_steering *steering = get_steering(node);
 762	struct mlx5_flow_group *fg;
 763	struct fs_fte *fte;
 764	int err;
 765
 766	fs_get_obj(fte, node);
 767	fs_get_obj(fg, fte->node.parent);
 768
 769	err = rhashtable_remove_fast(&fg->ftes_hash,
 770				     &fte->hash,
 771				     rhash_fte);
 772	WARN_ON(err);
 773	ida_free(&fg->fte_allocator, fte->index - fg->start_index);
 774	kvfree(fte->dup);
 775	kmem_cache_free(steering->ftes_cache, fte);
 776}
 777
 778static void del_hw_flow_group(struct fs_node *node)
 779{
 780	struct mlx5_flow_root_namespace *root;
 781	struct mlx5_flow_group *fg;
 782	struct mlx5_flow_table *ft;
 783	struct mlx5_core_dev *dev;
 784
 785	fs_get_obj(fg, node);
 786	fs_get_obj(ft, fg->node.parent);
 787	dev = get_dev(&ft->node);
 788	trace_mlx5_fs_del_fg(fg);
 789
 790	root = find_root(&ft->node);
 791	if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 792		mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 793			       fg->id, ft->id);
 794}
 795
 796static void del_sw_flow_group(struct fs_node *node)
 797{
 798	struct mlx5_flow_steering *steering = get_steering(node);
 799	struct mlx5_flow_group *fg;
 800	struct mlx5_flow_table *ft;
 801	int err;
 802
 803	fs_get_obj(fg, node);
 804	fs_get_obj(ft, fg->node.parent);
 805
 806	rhashtable_destroy(&fg->ftes_hash);
 807	ida_destroy(&fg->fte_allocator);
 808	if (ft->autogroup.active &&
 809	    fg->max_ftes == ft->autogroup.group_size &&
 810	    fg->start_index < ft->autogroup.max_fte)
 811		ft->autogroup.num_groups--;
 812	err = rhltable_remove(&ft->fgs_hash,
 813			      &fg->hash,
 814			      rhash_fg);
 815	WARN_ON(err);
 816	kmem_cache_free(steering->fgs_cache, fg);
 817}
 818
 819static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 820{
 821	int index;
 822	int ret;
 823
 824	index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL);
 825	if (index < 0)
 826		return index;
 827
 828	fte->index = index + fg->start_index;
 829retry_insert:
 830	ret = rhashtable_insert_fast(&fg->ftes_hash,
 831				     &fte->hash,
 832				     rhash_fte);
 833	if (ret) {
 834		if (ret == -EBUSY) {
 835			cond_resched();
 836			goto retry_insert;
 837		}
 838		goto err_ida_remove;
 839	}
 840
 841	tree_add_node(&fte->node, &fg->node);
 842	list_add_tail(&fte->node.list, &fg->node.children);
 843	return 0;
 844
 845err_ida_remove:
 846	ida_free(&fg->fte_allocator, index);
 847	return ret;
 848}
 849
 850static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 851				const struct mlx5_flow_spec *spec,
 852				struct mlx5_flow_act *flow_act)
 853{
 854	struct mlx5_flow_steering *steering = get_steering(&ft->node);
 855	struct fs_fte *fte;
 856
 857	fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 858	if (!fte)
 859		return ERR_PTR(-ENOMEM);
 860
 861	memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 862	fte->node.type =  FS_TYPE_FLOW_ENTRY;
 863	fte->act_dests.action = *flow_act;
 864	fte->act_dests.flow_context = spec->flow_context;
 865
 866	tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 867
 868	return fte;
 869}
 870
 871static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 872			       struct mlx5_flow_group *fg)
 873{
 874	rhashtable_destroy(&fg->ftes_hash);
 875	kmem_cache_free(steering->fgs_cache, fg);
 876}
 877
 878static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 879						u8 match_criteria_enable,
 880						const void *match_criteria,
 881						int start_index,
 882						int end_index)
 883{
 884	struct mlx5_flow_group *fg;
 885	int ret;
 886
 887	fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 888	if (!fg)
 889		return ERR_PTR(-ENOMEM);
 890
 891	ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 892	if (ret) {
 893		kmem_cache_free(steering->fgs_cache, fg);
 894		return ERR_PTR(ret);
 895	}
 896
 897	ida_init(&fg->fte_allocator);
 898	fg->mask.match_criteria_enable = match_criteria_enable;
 899	memcpy(&fg->mask.match_criteria, match_criteria,
 900	       sizeof(fg->mask.match_criteria));
 901	fg->node.type =  FS_TYPE_FLOW_GROUP;
 902	fg->start_index = start_index;
 903	fg->max_ftes = end_index - start_index + 1;
 904
 905	return fg;
 906}
 907
 908static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 909						       u8 match_criteria_enable,
 910						       const void *match_criteria,
 911						       int start_index,
 912						       int end_index,
 913						       struct list_head *prev)
 914{
 915	struct mlx5_flow_steering *steering = get_steering(&ft->node);
 916	struct mlx5_flow_group *fg;
 917	int ret;
 918
 919	fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 920			      start_index, end_index);
 921	if (IS_ERR(fg))
 922		return fg;
 923
 924	/* initialize refcnt, add to parent list */
 925	ret = rhltable_insert(&ft->fgs_hash,
 926			      &fg->hash,
 927			      rhash_fg);
 928	if (ret) {
 929		dealloc_flow_group(steering, fg);
 930		return ERR_PTR(ret);
 931	}
 932
 933	tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 934	tree_add_node(&fg->node, &ft->node);
 935	/* Add node to group list */
 936	list_add(&fg->node.list, prev);
 937	atomic_inc(&ft->node.version);
 938
 939	return fg;
 940}
 941
 942static struct mlx5_flow_table *
 943alloc_flow_table(struct mlx5_flow_table_attr *ft_attr, u16 vport,
 944		 enum fs_flow_table_type table_type,
 945		 enum fs_flow_table_op_mod op_mod)
 946{
 947	struct mlx5_flow_table *ft;
 948	int ret;
 949
 950	ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 951	if (!ft)
 952		return ERR_PTR(-ENOMEM);
 953
 954	ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 955	if (ret) {
 956		kfree(ft);
 957		return ERR_PTR(ret);
 958	}
 959
 960	ft->level = ft_attr->level;
 961	ft->node.type = FS_TYPE_FLOW_TABLE;
 962	ft->op_mod = op_mod;
 963	ft->type = table_type;
 964	ft->vport = vport;
 965	ft->esw_owner_vhca_id = ft_attr->esw_owner_vhca_id;
 966	ft->flags = ft_attr->flags;
 967	INIT_LIST_HEAD(&ft->fwd_rules);
 968	mutex_init(&ft->lock);
 969
 970	return ft;
 971}
 972
 973/* If reverse is false, then we search for the first flow table in the
 974 * root sub-tree from start(closest from right), else we search for the
 975 * last flow table in the root sub-tree till start(closest from left).
 976 */
 977static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 978							 struct list_head *start,
 979							 bool reverse)
 980{
 981#define list_advance_entry(pos, reverse)		\
 982	((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 983
 984#define list_for_each_advance_continue(pos, head, reverse)	\
 985	for (pos = list_advance_entry(pos, reverse);		\
 986	     &pos->list != (head);				\
 987	     pos = list_advance_entry(pos, reverse))
 988
 989	struct fs_node *iter = list_entry(start, struct fs_node, list);
 990	struct mlx5_flow_table *ft = NULL;
 991
 992	if (!root)
 993		return NULL;
 994
 995	list_for_each_advance_continue(iter, &root->children, reverse) {
 996		if (iter->type == FS_TYPE_FLOW_TABLE) {
 997			fs_get_obj(ft, iter);
 998			return ft;
 999		}
1000		ft = find_closest_ft_recursive(iter, &iter->children, reverse);
1001		if (ft)
1002			return ft;
1003	}
1004
1005	return ft;
1006}
1007
1008static struct fs_node *find_prio_chains_parent(struct fs_node *parent,
1009					       struct fs_node **child)
1010{
1011	struct fs_node *node = NULL;
1012
1013	while (parent && parent->type != FS_TYPE_PRIO_CHAINS) {
1014		node = parent;
1015		parent = parent->parent;
1016	}
1017
1018	if (child)
1019		*child = node;
1020
1021	return parent;
1022}
1023
1024/* If reverse is false then return the first flow table next to the passed node
1025 * in the tree, else return the last flow table before the node in the tree.
1026 * If skip is true, skip the flow tables in the same prio_chains prio.
1027 */
1028static struct mlx5_flow_table *find_closest_ft(struct fs_node *node, bool reverse,
1029					       bool skip)
1030{
1031	struct fs_node *prio_chains_parent = NULL;
1032	struct mlx5_flow_table *ft = NULL;
1033	struct fs_node *curr_node;
1034	struct fs_node *parent;
1035
1036	if (skip)
1037		prio_chains_parent = find_prio_chains_parent(node, NULL);
1038	parent = node->parent;
1039	curr_node = node;
1040	while (!ft && parent) {
1041		if (parent != prio_chains_parent)
1042			ft = find_closest_ft_recursive(parent, &curr_node->list,
1043						       reverse);
1044		curr_node = parent;
1045		parent = curr_node->parent;
1046	}
1047	return ft;
1048}
1049
1050/* Assuming all the tree is locked by mutex chain lock */
1051static struct mlx5_flow_table *find_next_chained_ft(struct fs_node *node)
1052{
1053	return find_closest_ft(node, false, true);
1054}
1055
1056/* Assuming all the tree is locked by mutex chain lock */
1057static struct mlx5_flow_table *find_prev_chained_ft(struct fs_node *node)
1058{
1059	return find_closest_ft(node, true, true);
1060}
1061
1062static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
1063						struct mlx5_flow_act *flow_act)
1064{
1065	struct fs_prio *prio;
1066	bool next_ns;
1067
1068	next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
1069	fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
1070
1071	return find_next_chained_ft(&prio->node);
1072}
1073
1074static int connect_fts_in_prio(struct mlx5_core_dev *dev,
1075			       struct fs_prio *prio,
1076			       struct mlx5_flow_table *ft)
1077{
1078	struct mlx5_flow_root_namespace *root = find_root(&prio->node);
1079	struct mlx5_flow_table *iter;
1080	int err;
1081
1082	fs_for_each_ft(iter, prio) {
1083		err = root->cmds->modify_flow_table(root, iter, ft);
1084		if (err) {
1085			mlx5_core_err(dev,
1086				      "Failed to modify flow table id %d, type %d, err %d\n",
1087				      iter->id, iter->type, err);
1088			/* The driver is out of sync with the FW */
1089			return err;
1090		}
1091	}
1092	return 0;
1093}
1094
1095static struct mlx5_flow_table *find_closet_ft_prio_chains(struct fs_node *node,
1096							  struct fs_node *parent,
1097							  struct fs_node **child,
1098							  bool reverse)
1099{
1100	struct mlx5_flow_table *ft;
1101
1102	ft = find_closest_ft(node, reverse, false);
1103
1104	if (ft && parent == find_prio_chains_parent(&ft->node, child))
1105		return ft;
1106
1107	return NULL;
1108}
1109
1110/* Connect flow tables from previous priority of prio to ft */
1111static int connect_prev_fts(struct mlx5_core_dev *dev,
1112			    struct mlx5_flow_table *ft,
1113			    struct fs_prio *prio)
1114{
1115	struct fs_node *prio_parent, *parent = NULL, *child, *node;
1116	struct mlx5_flow_table *prev_ft;
1117	int err = 0;
1118
1119	prio_parent = find_prio_chains_parent(&prio->node, &child);
1120
1121	/* return directly if not under the first sub ns of prio_chains prio */
1122	if (prio_parent && !list_is_first(&child->list, &prio_parent->children))
1123		return 0;
1124
1125	prev_ft = find_prev_chained_ft(&prio->node);
1126	while (prev_ft) {
1127		struct fs_prio *prev_prio;
1128
1129		fs_get_obj(prev_prio, prev_ft->node.parent);
1130		err = connect_fts_in_prio(dev, prev_prio, ft);
1131		if (err)
1132			break;
1133
1134		if (!parent) {
1135			parent = find_prio_chains_parent(&prev_prio->node, &child);
1136			if (!parent)
1137				break;
1138		}
1139
1140		node = child;
1141		prev_ft = find_closet_ft_prio_chains(node, parent, &child, true);
1142	}
1143	return err;
1144}
1145
1146static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
1147				 *prio)
1148{
1149	struct mlx5_flow_root_namespace *root = find_root(&prio->node);
1150	struct mlx5_ft_underlay_qp *uqp;
1151	int min_level = INT_MAX;
1152	int err = 0;
1153	u32 qpn;
1154
1155	if (root->root_ft)
1156		min_level = root->root_ft->level;
1157
1158	if (ft->level >= min_level)
1159		return 0;
1160
1161	if (list_empty(&root->underlay_qpns)) {
1162		/* Don't set any QPN (zero) in case QPN list is empty */
1163		qpn = 0;
1164		err = root->cmds->update_root_ft(root, ft, qpn, false);
1165	} else {
1166		list_for_each_entry(uqp, &root->underlay_qpns, list) {
1167			qpn = uqp->qpn;
1168			err = root->cmds->update_root_ft(root, ft,
1169							 qpn, false);
1170			if (err)
1171				break;
1172		}
1173	}
1174
1175	if (err)
1176		mlx5_core_warn(root->dev,
1177			       "Update root flow table of id(%u) qpn(%d) failed\n",
1178			       ft->id, qpn);
1179	else
1180		root->root_ft = ft;
1181
1182	return err;
1183}
1184
1185static bool rule_is_pending(struct fs_fte *fte, struct mlx5_flow_rule *rule)
1186{
1187	struct mlx5_flow_rule *tmp_rule;
1188	struct fs_node *iter;
1189
1190	if (!fte->dup || list_empty(&fte->dup->children))
1191		return false;
1192
1193	list_for_each_entry(iter, &fte->dup->children, list) {
1194		tmp_rule = container_of(iter, struct mlx5_flow_rule, node);
1195
1196		if (tmp_rule == rule)
1197			return true;
1198	}
1199
1200	return false;
1201}
1202
1203static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
1204					 struct mlx5_flow_destination *dest)
1205{
1206	struct mlx5_flow_root_namespace *root;
1207	struct fs_fte_action *act_dests;
1208	struct mlx5_flow_table *ft;
1209	struct mlx5_flow_group *fg;
1210	bool pending = false;
1211	struct fs_fte *fte;
1212	int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1213	int err = 0;
1214
1215	fs_get_obj(fte, rule->node.parent);
1216
1217	pending = rule_is_pending(fte, rule);
1218	if (pending)
1219		act_dests = &fte->dup->act_dests;
1220	else
1221		act_dests = &fte->act_dests;
1222
1223	if (!(act_dests->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1224		return -EINVAL;
1225	down_write_ref_node(&fte->node, false);
1226	fs_get_obj(fg, fte->node.parent);
1227	fs_get_obj(ft, fg->node.parent);
1228
1229	memcpy(&rule->dest_attr, dest, sizeof(*dest));
1230	root = find_root(&ft->node);
1231	if (!pending)
1232		err = root->cmds->update_fte(root, ft, fg,
1233					     modify_mask, fte);
1234	up_write_ref_node(&fte->node, false);
1235
1236	return err;
1237}
1238
1239int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
1240				 struct mlx5_flow_destination *new_dest,
1241				 struct mlx5_flow_destination *old_dest)
1242{
1243	int i;
1244
1245	if (!old_dest) {
1246		if (handle->num_rules != 1)
1247			return -EINVAL;
1248		return _mlx5_modify_rule_destination(handle->rule[0],
1249						     new_dest);
1250	}
1251
1252	for (i = 0; i < handle->num_rules; i++) {
1253		if (mlx5_flow_dests_cmp(old_dest, &handle->rule[i]->dest_attr))
1254			return _mlx5_modify_rule_destination(handle->rule[i],
1255							     new_dest);
1256	}
1257
1258	return -EINVAL;
1259}
1260
1261/* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
1262static int connect_fwd_rules(struct mlx5_core_dev *dev,
1263			     struct mlx5_flow_table *new_next_ft,
1264			     struct mlx5_flow_table *old_next_ft)
1265{
1266	struct mlx5_flow_destination dest = {};
1267	struct mlx5_flow_rule *iter;
1268	int err = 0;
1269
1270	/* new_next_ft and old_next_ft could be NULL only
1271	 * when we create/destroy the anchor flow table.
1272	 */
1273	if (!new_next_ft || !old_next_ft)
1274		return 0;
1275
1276	dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1277	dest.ft = new_next_ft;
1278
1279	mutex_lock(&old_next_ft->lock);
1280	list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1281	mutex_unlock(&old_next_ft->lock);
1282	list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1283		if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1284		    iter->ft->ns == new_next_ft->ns)
1285			continue;
1286
1287		err = _mlx5_modify_rule_destination(iter, &dest);
1288		if (err)
1289			pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1290			       new_next_ft->id);
1291	}
1292	return 0;
1293}
1294
1295static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1296			      struct fs_prio *prio)
1297{
1298	struct mlx5_flow_table *next_ft, *first_ft;
1299	int err = 0;
1300
1301	/* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1302
1303	first_ft = list_first_entry_or_null(&prio->node.children,
1304					    struct mlx5_flow_table, node.list);
1305	if (!first_ft || first_ft->level > ft->level) {
1306		err = connect_prev_fts(dev, ft, prio);
1307		if (err)
1308			return err;
1309
1310		next_ft = first_ft ? first_ft : find_next_chained_ft(&prio->node);
1311		err = connect_fwd_rules(dev, ft, next_ft);
1312		if (err)
1313			return err;
1314	}
1315
1316	if (MLX5_CAP_FLOWTABLE(dev,
1317			       flow_table_properties_nic_receive.modify_root))
1318		err = update_root_ft_create(ft, prio);
1319	return err;
1320}
1321
1322static void list_add_flow_table(struct mlx5_flow_table *ft,
1323				struct fs_prio *prio)
1324{
1325	struct list_head *prev = &prio->node.children;
1326	struct mlx5_flow_table *iter;
1327
1328	fs_for_each_ft(iter, prio) {
1329		if (iter->level > ft->level)
1330			break;
1331		prev = &iter->node.list;
1332	}
1333	list_add(&ft->node.list, prev);
1334}
1335
1336static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1337							struct mlx5_flow_table_attr *ft_attr,
1338							enum fs_flow_table_op_mod op_mod,
1339							u16 vport)
1340{
1341	struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1342	bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1343	struct mlx5_flow_table *next_ft;
1344	struct fs_prio *fs_prio = NULL;
1345	struct mlx5_flow_table *ft;
1346	int err;
1347
1348	if (!root) {
1349		pr_err("mlx5: flow steering failed to find root of namespace\n");
1350		return ERR_PTR(-ENODEV);
1351	}
1352
1353	mutex_lock(&root->chain_lock);
1354	fs_prio = find_prio(ns, ft_attr->prio);
1355	if (!fs_prio) {
1356		err = -EINVAL;
1357		goto unlock_root;
1358	}
1359	if (!unmanaged) {
1360		/* The level is related to the
1361		 * priority level range.
1362		 */
1363		if (ft_attr->level >= fs_prio->num_levels) {
1364			err = -ENOSPC;
1365			goto unlock_root;
1366		}
1367
1368		ft_attr->level += fs_prio->start_level;
1369	}
1370
1371	/* The level is related to the
1372	 * priority level range.
1373	 */
1374	ft = alloc_flow_table(ft_attr, vport, root->table_type, op_mod);
1375	if (IS_ERR(ft)) {
1376		err = PTR_ERR(ft);
1377		goto unlock_root;
1378	}
1379
1380	tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1381	next_ft = unmanaged ? ft_attr->next_ft :
1382			      find_next_chained_ft(&fs_prio->node);
1383	ft->def_miss_action = ns->def_miss_action;
1384	ft->ns = ns;
1385	err = root->cmds->create_flow_table(root, ft, ft_attr, next_ft);
1386	if (err)
1387		goto free_ft;
1388
1389	if (!unmanaged) {
1390		err = connect_flow_table(root->dev, ft, fs_prio);
1391		if (err)
1392			goto destroy_ft;
1393	}
1394
1395	ft->node.active = true;
1396	down_write_ref_node(&fs_prio->node, false);
1397	if (!unmanaged) {
1398		tree_add_node(&ft->node, &fs_prio->node);
1399		list_add_flow_table(ft, fs_prio);
1400	} else {
1401		ft->node.root = fs_prio->node.root;
1402	}
1403	fs_prio->num_ft++;
1404	up_write_ref_node(&fs_prio->node, false);
1405	mutex_unlock(&root->chain_lock);
1406	trace_mlx5_fs_add_ft(ft);
1407	return ft;
1408destroy_ft:
1409	root->cmds->destroy_flow_table(root, ft);
1410free_ft:
1411	rhltable_destroy(&ft->fgs_hash);
1412	kfree(ft);
1413unlock_root:
1414	mutex_unlock(&root->chain_lock);
1415	return ERR_PTR(err);
1416}
1417
1418struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1419					       struct mlx5_flow_table_attr *ft_attr)
1420{
1421	return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1422}
1423EXPORT_SYMBOL(mlx5_create_flow_table);
1424
1425u32 mlx5_flow_table_id(struct mlx5_flow_table *ft)
1426{
1427	return ft->id;
1428}
1429EXPORT_SYMBOL(mlx5_flow_table_id);
1430
1431struct mlx5_flow_table *
1432mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1433			     struct mlx5_flow_table_attr *ft_attr, u16 vport)
1434{
1435	return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1436}
1437
1438struct mlx5_flow_table*
1439mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1440				 int prio, u32 level)
1441{
1442	struct mlx5_flow_table_attr ft_attr = {};
1443
1444	ft_attr.level = level;
1445	ft_attr.prio  = prio;
1446	ft_attr.max_fte = 1;
1447
1448	return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1449}
1450EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1451
1452#define MAX_FLOW_GROUP_SIZE BIT(24)
1453struct mlx5_flow_table*
1454mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1455				    struct mlx5_flow_table_attr *ft_attr)
1456{
1457	int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1458	int max_num_groups = ft_attr->autogroup.max_num_groups;
1459	struct mlx5_flow_table *ft;
1460	int autogroups_max_fte;
1461
1462	ft = mlx5_create_vport_flow_table(ns, ft_attr, ft_attr->vport);
1463	if (IS_ERR(ft))
1464		return ft;
1465
1466	autogroups_max_fte = ft->max_fte - num_reserved_entries;
1467	if (max_num_groups > autogroups_max_fte)
1468		goto err_validate;
1469	if (num_reserved_entries > ft->max_fte)
1470		goto err_validate;
1471
1472	/* Align the number of groups according to the largest group size */
1473	if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1474		max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1475
1476	ft->autogroup.active = true;
1477	ft->autogroup.required_groups = max_num_groups;
1478	ft->autogroup.max_fte = autogroups_max_fte;
1479	/* We save place for flow groups in addition to max types */
1480	ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1481
1482	return ft;
1483
1484err_validate:
1485	mlx5_destroy_flow_table(ft);
1486	return ERR_PTR(-ENOSPC);
1487}
1488EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1489
1490struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1491					       u32 *fg_in)
1492{
1493	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1494	void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1495					    fg_in, match_criteria);
1496	u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1497					    fg_in,
1498					    match_criteria_enable);
1499	int start_index = MLX5_GET(create_flow_group_in, fg_in,
1500				   start_flow_index);
1501	int end_index = MLX5_GET(create_flow_group_in, fg_in,
1502				 end_flow_index);
1503	struct mlx5_flow_group *fg;
1504	int err;
1505
1506	if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1507		return ERR_PTR(-EPERM);
1508
1509	down_write_ref_node(&ft->node, false);
1510	fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1511				     start_index, end_index,
1512				     ft->node.children.prev);
1513	up_write_ref_node(&ft->node, false);
1514	if (IS_ERR(fg))
1515		return fg;
1516
1517	err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1518	if (err) {
1519		tree_put_node(&fg->node, false);
1520		return ERR_PTR(err);
1521	}
1522	trace_mlx5_fs_add_fg(fg);
1523	fg->node.active = true;
1524
1525	return fg;
1526}
1527EXPORT_SYMBOL(mlx5_create_flow_group);
1528
1529static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1530{
1531	struct mlx5_flow_rule *rule;
1532
1533	rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1534	if (!rule)
1535		return NULL;
1536
1537	INIT_LIST_HEAD(&rule->next_ft);
1538	rule->node.type = FS_TYPE_FLOW_DEST;
1539	if (dest)
1540		memcpy(&rule->dest_attr, dest, sizeof(*dest));
1541	else
1542		rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE;
1543
1544	return rule;
1545}
1546
1547static struct mlx5_flow_handle *alloc_handle(int num_rules)
1548{
1549	struct mlx5_flow_handle *handle;
1550
1551	handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1552	if (!handle)
1553		return NULL;
1554
1555	handle->num_rules = num_rules;
1556
1557	return handle;
1558}
1559
1560static void destroy_flow_handle_dup(struct mlx5_flow_handle *handle,
1561				    int i)
1562{
1563	for (; --i >= 0;) {
1564		list_del(&handle->rule[i]->node.list);
1565		kfree(handle->rule[i]);
1566	}
1567	kfree(handle);
1568}
1569
1570static void destroy_flow_handle(struct fs_fte *fte,
1571				struct mlx5_flow_handle *handle,
1572				struct mlx5_flow_destination *dest,
1573				int i)
1574{
1575	for (; --i >= 0;) {
1576		if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1577			fte->act_dests.dests_size--;
1578			list_del(&handle->rule[i]->node.list);
1579			kfree(handle->rule[i]);
1580		}
1581	}
1582	kfree(handle);
1583}
1584
1585static struct mlx5_flow_handle *
1586create_flow_handle_dup(struct list_head *children,
1587		       struct mlx5_flow_destination *dest,
1588		       int dest_num,
1589		       struct fs_fte_action *act_dests)
1590{
1591	static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1592	static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1593	struct mlx5_flow_rule *rule = NULL;
1594	struct mlx5_flow_handle *handle;
1595	int i = 0;
1596	int type;
1597
1598	handle = alloc_handle((dest_num) ? dest_num : 1);
1599	if (!handle)
1600		return NULL;
1601
1602	do {
1603		rule = alloc_rule(dest + i);
1604		if (!rule)
1605			goto free_rules;
1606
1607		/* Add dest to dests list- we need flow tables to be in the
1608		 * end of the list for forward to next prio rules.
1609		 */
1610		tree_init_node(&rule->node, NULL, del_sw_hw_dup_rule);
1611		if (dest &&
1612		    dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1613			list_add(&rule->node.list, children);
1614		else
1615			list_add_tail(&rule->node.list, children);
1616
1617		if (dest) {
1618			act_dests->dests_size++;
1619
1620			if (is_fwd_dest_type(dest[i].type))
1621				act_dests->fwd_dests++;
1622
1623			type = dest[i].type ==
1624				MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1625			act_dests->modify_mask |= type ? count : dst;
1626		}
1627		handle->rule[i] = rule;
1628	} while (++i < dest_num);
1629
1630	return handle;
1631
1632free_rules:
1633	destroy_flow_handle_dup(handle, i);
1634	act_dests->dests_size = 0;
1635	act_dests->fwd_dests = 0;
1636
1637	return NULL;
1638}
1639
1640static struct mlx5_flow_handle *
1641create_flow_handle(struct fs_fte *fte,
1642		   struct mlx5_flow_destination *dest,
1643		   int dest_num,
1644		   int *modify_mask,
1645		   bool *new_rule)
1646{
1647	struct mlx5_flow_handle *handle;
1648	struct mlx5_flow_rule *rule = NULL;
1649	static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1650	static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1651	int type;
1652	int i = 0;
1653
1654	handle = alloc_handle((dest_num) ? dest_num : 1);
1655	if (!handle)
1656		return ERR_PTR(-ENOMEM);
1657
1658	do {
1659		if (dest) {
1660			rule = find_flow_rule(fte, dest + i);
1661			if (rule) {
1662				refcount_inc(&rule->node.refcount);
1663				goto rule_found;
1664			}
1665		}
1666
1667		*new_rule = true;
1668		rule = alloc_rule(dest + i);
1669		if (!rule)
1670			goto free_rules;
1671
1672		/* Add dest to dests list- we need flow tables to be in the
1673		 * end of the list for forward to next prio rules.
1674		 */
1675		tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1676		if (dest &&
1677		    dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1678			list_add(&rule->node.list, &fte->node.children);
1679		else
1680			list_add_tail(&rule->node.list, &fte->node.children);
1681		if (dest) {
1682			fte->act_dests.dests_size++;
1683
1684			if (is_fwd_dest_type(dest[i].type))
1685				fte->act_dests.fwd_dests++;
1686
1687			type = dest[i].type ==
1688				MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1689			*modify_mask |= type ? count : dst;
1690		}
1691rule_found:
1692		handle->rule[i] = rule;
1693	} while (++i < dest_num);
1694
1695	return handle;
1696
1697free_rules:
1698	destroy_flow_handle(fte, handle, dest, i);
1699	return ERR_PTR(-ENOMEM);
1700}
1701
1702/* fte should not be deleted while calling this function */
1703static struct mlx5_flow_handle *
1704add_rule_fte(struct fs_fte *fte,
1705	     struct mlx5_flow_group *fg,
1706	     struct mlx5_flow_destination *dest,
1707	     int dest_num,
1708	     bool update_action)
1709{
1710	struct mlx5_flow_root_namespace *root;
1711	struct mlx5_flow_handle *handle;
1712	struct mlx5_flow_table *ft;
1713	int modify_mask = 0;
1714	int err;
1715	bool new_rule = false;
1716
1717	handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1718				    &new_rule);
1719	if (IS_ERR(handle) || !new_rule)
1720		goto out;
1721
1722	if (update_action)
1723		modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1724
1725	fs_get_obj(ft, fg->node.parent);
1726	root = find_root(&fg->node);
1727	if (!(fte->status & FS_FTE_STATUS_EXISTING))
1728		err = root->cmds->create_fte(root, ft, fg, fte);
1729	else
1730		err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1731	if (err)
1732		goto free_handle;
1733
1734	fte->node.active = true;
1735	fte->status |= FS_FTE_STATUS_EXISTING;
1736	atomic_inc(&fg->node.version);
1737
1738out:
1739	return handle;
1740
1741free_handle:
1742	destroy_flow_handle(fte, handle, dest, handle->num_rules);
1743	return ERR_PTR(err);
1744}
1745
1746static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1747						     const struct mlx5_flow_spec *spec)
1748{
1749	struct list_head *prev = &ft->node.children;
1750	u32 max_fte = ft->autogroup.max_fte;
1751	unsigned int candidate_index = 0;
1752	unsigned int group_size = 0;
1753	struct mlx5_flow_group *fg;
1754
1755	if (!ft->autogroup.active)
1756		return ERR_PTR(-ENOENT);
1757
1758	if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1759		group_size = ft->autogroup.group_size;
1760
1761	/*  max_fte == ft->autogroup.max_types */
1762	if (group_size == 0)
1763		group_size = 1;
1764
1765	/* sorted by start_index */
1766	fs_for_each_fg(fg, ft) {
1767		if (candidate_index + group_size > fg->start_index)
1768			candidate_index = fg->start_index + fg->max_ftes;
1769		else
1770			break;
1771		prev = &fg->node.list;
1772	}
1773
1774	if (candidate_index + group_size > max_fte)
1775		return ERR_PTR(-ENOSPC);
1776
1777	fg = alloc_insert_flow_group(ft,
1778				     spec->match_criteria_enable,
1779				     spec->match_criteria,
1780				     candidate_index,
1781				     candidate_index + group_size - 1,
1782				     prev);
1783	if (IS_ERR(fg))
1784		goto out;
1785
1786	if (group_size == ft->autogroup.group_size)
1787		ft->autogroup.num_groups++;
1788
1789out:
1790	return fg;
1791}
1792
1793static int create_auto_flow_group(struct mlx5_flow_table *ft,
1794				  struct mlx5_flow_group *fg)
1795{
1796	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1797	int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1798	void *match_criteria_addr;
1799	u8 src_esw_owner_mask_on;
1800	void *misc;
1801	int err;
1802	u32 *in;
1803
1804	in = kvzalloc(inlen, GFP_KERNEL);
1805	if (!in)
1806		return -ENOMEM;
1807
1808	MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1809		 fg->mask.match_criteria_enable);
1810	MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1811	MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1812		 fg->max_ftes - 1);
1813
1814	misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1815			    misc_parameters);
1816	src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1817					 source_eswitch_owner_vhca_id);
1818	MLX5_SET(create_flow_group_in, in,
1819		 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1820
1821	match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1822					   in, match_criteria);
1823	memcpy(match_criteria_addr, fg->mask.match_criteria,
1824	       sizeof(fg->mask.match_criteria));
1825
1826	err = root->cmds->create_flow_group(root, ft, in, fg);
1827	if (!err) {
1828		fg->node.active = true;
1829		trace_mlx5_fs_add_fg(fg);
1830	}
1831
1832	kvfree(in);
1833	return err;
1834}
1835
1836int mlx5_fs_get_packet_reformat_id(struct mlx5_pkt_reformat *pkt_reformat,
1837				   u32 *id)
1838{
1839	switch (pkt_reformat->owner) {
1840	case MLX5_FLOW_RESOURCE_OWNER_FW:
1841		*id = pkt_reformat->id;
1842		return 0;
1843	case MLX5_FLOW_RESOURCE_OWNER_SW:
1844		return mlx5_fs_dr_action_get_pkt_reformat_id(pkt_reformat, id);
1845	case MLX5_FLOW_RESOURCE_OWNER_HWS:
1846		return mlx5_fs_hws_action_get_pkt_reformat_id(pkt_reformat, id);
1847	default:
1848		return -EINVAL;
1849	}
1850}
1851
1852static bool mlx5_pkt_reformat_cmp(struct mlx5_pkt_reformat *p1,
1853				  struct mlx5_pkt_reformat *p2)
1854{
1855	int err1, err2;
1856	u32 id1, id2;
1857
1858	if (p1->owner != p2->owner)
1859		return false;
1860
1861	err1 = mlx5_fs_get_packet_reformat_id(p1, &id1);
1862	err2 = mlx5_fs_get_packet_reformat_id(p2, &id2);
1863
1864	return !err1 && !err2 && id1 == id2;
1865}
1866
1867static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1868				struct mlx5_flow_destination *d2)
1869{
1870	if (d1->type == d2->type) {
1871		if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
1872		      d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) &&
1873		     d1->vport.num == d2->vport.num &&
1874		     d1->vport.flags == d2->vport.flags &&
1875		     ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1876		      (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1877		     ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1878		      mlx5_pkt_reformat_cmp(d1->vport.pkt_reformat,
1879					    d2->vport.pkt_reformat) : true)) ||
1880		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1881		     d1->ft == d2->ft) ||
1882		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1883		     d1->tir_num == d2->tir_num) ||
1884		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1885		     d1->ft_num == d2->ft_num) ||
1886		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1887		     d1->sampler_id == d2->sampler_id) ||
1888		    (d1->type == MLX5_FLOW_DESTINATION_TYPE_RANGE &&
1889		     d1->range.field == d2->range.field &&
1890		     d1->range.hit_ft == d2->range.hit_ft &&
1891		     d1->range.miss_ft == d2->range.miss_ft &&
1892		     d1->range.min == d2->range.min &&
1893		     d1->range.max == d2->range.max))
1894			return true;
1895	}
1896
1897	return false;
1898}
1899
1900static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1901					     struct mlx5_flow_destination *dest)
1902{
1903	struct mlx5_flow_rule *rule;
1904
1905	list_for_each_entry(rule, &fte->node.children, node.list) {
1906		if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1907			return rule;
1908	}
1909	return NULL;
1910}
1911
1912static bool check_conflicting_actions_vlan(const struct mlx5_fs_vlan *vlan0,
1913					   const struct mlx5_fs_vlan *vlan1)
1914{
1915	return vlan0->ethtype != vlan1->ethtype ||
1916	       vlan0->vid != vlan1->vid ||
1917	       vlan0->prio != vlan1->prio;
1918}
1919
1920static bool check_conflicting_actions(const struct mlx5_flow_act *act1,
1921				      const struct mlx5_flow_act *act2)
1922{
1923	u32 action1 = act1->action;
1924	u32 action2 = act2->action;
1925	u32 xored_actions;
1926
1927	xored_actions = action1 ^ action2;
1928
1929	/* if one rule only wants to count, it's ok */
1930	if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1931	    action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1932		return false;
1933
1934	if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1935			     MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1936			     MLX5_FLOW_CONTEXT_ACTION_DECAP |
1937			     MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1938			     MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1939			     MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1940			     MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1941			     MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1942		return true;
1943
1944	if (action1 & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT &&
1945	    act1->pkt_reformat != act2->pkt_reformat)
1946		return true;
1947
1948	if (action1 & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR &&
1949	    act1->modify_hdr != act2->modify_hdr)
1950		return true;
1951
1952	if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH &&
1953	    check_conflicting_actions_vlan(&act1->vlan[0], &act2->vlan[0]))
1954		return true;
1955
1956	if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2 &&
1957	    check_conflicting_actions_vlan(&act1->vlan[1], &act2->vlan[1]))
1958		return true;
1959
1960	return false;
1961}
1962
1963static int check_conflicting_ftes(struct fs_fte *fte,
1964				  const struct mlx5_flow_context *flow_context,
1965				  const struct mlx5_flow_act *flow_act)
1966{
1967	if (check_conflicting_actions(flow_act, &fte->act_dests.action)) {
1968		mlx5_core_warn(get_dev(&fte->node),
1969			       "Found two FTEs with conflicting actions\n");
1970		return -EEXIST;
1971	}
1972
1973	if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1974	    fte->act_dests.flow_context.flow_tag != flow_context->flow_tag) {
1975		mlx5_core_warn(get_dev(&fte->node),
1976			       "FTE flow tag %u already exists with different flow tag %u\n",
1977			       fte->act_dests.flow_context.flow_tag,
1978			       flow_context->flow_tag);
1979		return -EEXIST;
1980	}
1981
1982	return 0;
1983}
1984
1985static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1986					    const struct mlx5_flow_spec *spec,
1987					    struct mlx5_flow_act *flow_act,
1988					    struct mlx5_flow_destination *dest,
1989					    int dest_num,
1990					    struct fs_fte *fte)
1991{
1992	struct mlx5_flow_handle *handle;
1993	int old_action;
1994	int i;
1995	int ret;
1996
1997	ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1998	if (ret)
1999		return ERR_PTR(ret);
2000
2001	old_action = fte->act_dests.action.action;
2002	fte->act_dests.action.action |= flow_act->action;
2003	handle = add_rule_fte(fte, fg, dest, dest_num,
2004			      old_action != flow_act->action);
2005	if (IS_ERR(handle)) {
2006		fte->act_dests.action.action = old_action;
2007		return handle;
2008	}
2009	trace_mlx5_fs_set_fte(fte, false);
2010
2011	/* Link newly added rules into the tree. */
2012	for (i = 0; i < handle->num_rules; i++) {
2013		if (!handle->rule[i]->node.parent) {
2014			tree_add_node(&handle->rule[i]->node, &fte->node);
2015			trace_mlx5_fs_add_rule(handle->rule[i]);
2016		}
2017	}
2018	return handle;
2019}
2020
2021static bool counter_is_valid(u32 action)
2022{
2023	return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
2024			  MLX5_FLOW_CONTEXT_ACTION_ALLOW |
2025			  MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
2026}
2027
2028static bool dest_is_valid(struct mlx5_flow_destination *dest,
2029			  struct mlx5_flow_act *flow_act,
2030			  struct mlx5_flow_table *ft)
2031{
2032	bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
2033	u32 action = flow_act->action;
2034
2035	if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
2036		return counter_is_valid(action);
2037
2038	if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
2039		return true;
2040
2041	if (ignore_level) {
2042		if (ft->type != FS_FT_FDB &&
2043		    ft->type != FS_FT_NIC_RX &&
2044		    ft->type != FS_FT_NIC_TX)
2045			return false;
2046
2047		if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
2048		    ft->type != dest->ft->type)
2049			return false;
2050	}
2051
2052	if (!dest || ((dest->type ==
2053	    MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
2054	    (dest->ft->level <= ft->level && !ignore_level)))
2055		return false;
2056	return true;
2057}
2058
2059struct match_list {
2060	struct list_head	list;
2061	struct mlx5_flow_group *g;
2062};
2063
2064static void free_match_list(struct match_list *head, bool ft_locked)
2065{
2066	struct match_list *iter, *match_tmp;
2067
2068	list_for_each_entry_safe(iter, match_tmp, &head->list,
2069				 list) {
2070		tree_put_node(&iter->g->node, ft_locked);
2071		list_del(&iter->list);
2072		kfree(iter);
2073	}
2074}
2075
2076static int build_match_list(struct match_list *match_head,
2077			    struct mlx5_flow_table *ft,
2078			    const struct mlx5_flow_spec *spec,
2079			    struct mlx5_flow_group *fg,
2080			    bool ft_locked)
2081{
2082	struct rhlist_head *tmp, *list;
2083	struct mlx5_flow_group *g;
2084
2085	rcu_read_lock();
2086	INIT_LIST_HEAD(&match_head->list);
2087	/* Collect all fgs which has a matching match_criteria */
2088	list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
2089	/* RCU is atomic, we can't execute FW commands here */
2090	rhl_for_each_entry_rcu(g, tmp, list, hash) {
2091		struct match_list *curr_match;
2092
2093		if (fg && fg != g)
2094			continue;
2095
2096		if (unlikely(!tree_get_node(&g->node)))
2097			continue;
2098
2099		curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
2100		if (!curr_match) {
2101			rcu_read_unlock();
2102			free_match_list(match_head, ft_locked);
2103			return -ENOMEM;
2104		}
2105		curr_match->g = g;
2106		list_add_tail(&curr_match->list, &match_head->list);
2107	}
2108	rcu_read_unlock();
2109	return 0;
2110}
2111
2112static u64 matched_fgs_get_version(struct list_head *match_head)
2113{
2114	struct match_list *iter;
2115	u64 version = 0;
2116
2117	list_for_each_entry(iter, match_head, list)
2118		version += (u64)atomic_read(&iter->g->node.version);
2119	return version;
2120}
2121
2122static struct fs_fte *
2123lookup_fte_locked(struct mlx5_flow_group *g,
2124		  const u32 *match_value,
2125		  bool take_write)
2126{
2127	struct fs_fte *fte_tmp;
2128
2129	if (take_write)
2130		nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2131	else
2132		nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
2133	fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
2134					 rhash_fte);
2135	if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
2136		fte_tmp = NULL;
2137		goto out;
2138	}
2139
2140	nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
2141
2142	if (!fte_tmp->node.active) {
2143		up_write_ref_node(&fte_tmp->node, false);
2144
2145		if (take_write)
2146			up_write_ref_node(&g->node, false);
2147		else
2148			up_read_ref_node(&g->node);
2149
2150		tree_put_node(&fte_tmp->node, false);
2151
2152		return NULL;
2153	}
2154
2155out:
2156	if (take_write)
2157		up_write_ref_node(&g->node, false);
2158	else
2159		up_read_ref_node(&g->node);
2160	return fte_tmp;
2161}
2162
2163/* Native capability lacks support for adding an additional match with the same value
2164 * to the same flow group. To accommodate the NO APPEND flag in these scenarios,
2165 * we include the new rule in the existing flow table entry (fte) without immediate
2166 * hardware commitment. When a request is made to delete the corresponding hardware rule,
2167 * we then commit the pending rule to hardware.
2168 */
2169static struct mlx5_flow_handle *
2170add_rule_dup_match_fte(struct fs_fte *fte,
2171		       const struct mlx5_flow_spec *spec,
2172		       struct mlx5_flow_act *flow_act,
2173		       struct mlx5_flow_destination *dest,
2174		       int dest_num)
2175{
2176	struct mlx5_flow_handle *handle;
2177	struct fs_fte_dup *dup;
2178	int i = 0;
2179
2180	if (!fte->dup) {
2181		dup = kvzalloc(sizeof(*dup), GFP_KERNEL);
2182		if (!dup)
2183			return ERR_PTR(-ENOMEM);
2184		/* dup will be freed when the fte is freed
2185		 * this way we don't allocate / free dup on every rule deletion
2186		 * or creation
2187		 */
2188		INIT_LIST_HEAD(&dup->children);
2189		fte->dup = dup;
2190	}
2191
2192	if (!list_empty(&fte->dup->children)) {
2193		mlx5_core_warn(get_dev(&fte->node),
2194			       "Can have only a single duplicate rule\n");
2195
2196		return ERR_PTR(-EEXIST);
2197	}
2198
2199	fte->dup->act_dests.action = *flow_act;
2200	fte->dup->act_dests.flow_context = spec->flow_context;
2201	fte->dup->act_dests.dests_size = 0;
2202	fte->dup->act_dests.fwd_dests = 0;
2203	fte->dup->act_dests.modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
2204
2205	handle = create_flow_handle_dup(&fte->dup->children,
2206					dest, dest_num,
2207					&fte->dup->act_dests);
2208	if (!handle)
2209		return ERR_PTR(-ENOMEM);
2210
2211	for (i = 0; i < handle->num_rules; i++) {
2212		tree_add_node(&handle->rule[i]->node, &fte->node);
2213		trace_mlx5_fs_add_rule(handle->rule[i]);
2214	}
2215
2216	return handle;
2217}
2218
2219static struct mlx5_flow_handle *
2220try_add_to_existing_fg(struct mlx5_flow_table *ft,
2221		       struct list_head *match_head,
2222		       const struct mlx5_flow_spec *spec,
2223		       struct mlx5_flow_act *flow_act,
2224		       struct mlx5_flow_destination *dest,
2225		       int dest_num,
2226		       int ft_version)
2227{
2228	struct mlx5_flow_steering *steering = get_steering(&ft->node);
2229	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2230	struct mlx5_flow_group *g;
2231	struct mlx5_flow_handle *rule;
2232	struct match_list *iter;
2233	bool take_write = false;
2234	bool try_again = false;
2235	struct fs_fte *fte;
2236	u64  version = 0;
2237	int err;
2238
2239	fte = alloc_fte(ft, spec, flow_act);
2240	if (IS_ERR(fte))
2241		return  ERR_PTR(-ENOMEM);
2242
2243search_again_locked:
2244	if (flow_act->flags & FLOW_ACT_NO_APPEND &&
2245	    (root->cmds->get_capabilities(root, root->table_type) &
2246	     MLX5_FLOW_STEERING_CAP_DUPLICATE_MATCH))
2247		goto skip_search;
2248	version = matched_fgs_get_version(match_head);
2249	/* Try to find an fte with identical match value and attempt update its
2250	 * action.
2251	 */
2252	list_for_each_entry(iter, match_head, list) {
2253		struct fs_fte *fte_tmp;
2254
2255		g = iter->g;
2256		fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
2257		if (!fte_tmp)
2258			continue;
2259		if (flow_act->flags & FLOW_ACT_NO_APPEND)
2260			rule = add_rule_dup_match_fte(fte_tmp, spec, flow_act, dest, dest_num);
2261		else
2262			rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
2263		/* No error check needed here, because insert_fte() is not called */
2264		up_write_ref_node(&fte_tmp->node, false);
2265		tree_put_node(&fte_tmp->node, false);
2266		kmem_cache_free(steering->ftes_cache, fte);
2267		return rule;
2268	}
2269
2270skip_search:
2271	/* No group with matching fte found, or we skipped the search.
2272	 * Try to add a new fte to any matching fg.
2273	 */
2274
2275	/* Check the ft version, for case that new flow group
2276	 * was added while the fgs weren't locked
2277	 */
2278	if (atomic_read(&ft->node.version) != ft_version) {
2279		rule = ERR_PTR(-EAGAIN);
2280		goto out;
2281	}
2282
2283	/* Check the fgs version. If version have changed it could be that an
2284	 * FTE with the same match value was added while the fgs weren't
2285	 * locked.
2286	 */
2287	if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
2288	    version != matched_fgs_get_version(match_head)) {
2289		take_write = true;
2290		goto search_again_locked;
2291	}
2292
2293	list_for_each_entry(iter, match_head, list) {
2294		g = iter->g;
2295
2296		nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2297
2298		if (!g->node.active) {
2299			try_again = true;
2300			up_write_ref_node(&g->node, false);
2301			continue;
2302		}
2303
2304		err = insert_fte(g, fte);
2305		if (err) {
2306			up_write_ref_node(&g->node, false);
2307			if (err == -ENOSPC)
2308				continue;
2309			kmem_cache_free(steering->ftes_cache, fte);
2310			return ERR_PTR(err);
2311		}
2312
2313		nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
2314		up_write_ref_node(&g->node, false);
2315		rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
2316		up_write_ref_node(&fte->node, false);
2317		if (IS_ERR(rule))
2318			tree_put_node(&fte->node, false);
2319		return rule;
2320	}
2321	err = try_again ? -EAGAIN : -ENOENT;
2322	rule = ERR_PTR(err);
2323out:
2324	kmem_cache_free(steering->ftes_cache, fte);
2325	return rule;
2326}
2327
2328static struct mlx5_flow_handle *
2329_mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2330		     const struct mlx5_flow_spec *spec,
2331		     struct mlx5_flow_act *flow_act,
2332		     struct mlx5_flow_destination *dest,
2333		     int dest_num)
2334
2335{
2336	struct mlx5_flow_steering *steering = get_steering(&ft->node);
2337	struct mlx5_flow_handle *rule;
2338	struct match_list match_head;
2339	struct mlx5_flow_group *g;
2340	bool take_write = false;
2341	struct fs_fte *fte;
2342	int version;
2343	int err;
2344	int i;
2345
2346	if (!check_valid_spec(spec))
2347		return ERR_PTR(-EINVAL);
2348
2349	if (flow_act->fg && ft->autogroup.active)
2350		return ERR_PTR(-EINVAL);
2351
2352	if (dest && dest_num <= 0)
2353		return ERR_PTR(-EINVAL);
2354
2355	for (i = 0; i < dest_num; i++) {
2356		if (!dest_is_valid(&dest[i], flow_act, ft))
2357			return ERR_PTR(-EINVAL);
2358	}
2359	nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
2360search_again_locked:
2361	version = atomic_read(&ft->node.version);
2362
2363	/* Collect all fgs which has a matching match_criteria */
2364	err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write);
2365	if (err) {
2366		if (take_write)
2367			up_write_ref_node(&ft->node, false);
2368		else
2369			up_read_ref_node(&ft->node);
2370		return ERR_PTR(err);
2371	}
2372
2373	if (!take_write)
2374		up_read_ref_node(&ft->node);
2375
2376	rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
2377				      dest_num, version);
2378	free_match_list(&match_head, take_write);
2379	if (!IS_ERR(rule) ||
2380	    (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
2381		if (take_write)
2382			up_write_ref_node(&ft->node, false);
2383		return rule;
2384	}
2385
2386	if (!take_write) {
2387		nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
2388		take_write = true;
2389	}
2390
2391	if (PTR_ERR(rule) == -EAGAIN ||
2392	    version != atomic_read(&ft->node.version))
2393		goto search_again_locked;
2394
2395	g = alloc_auto_flow_group(ft, spec);
2396	if (IS_ERR(g)) {
2397		rule = ERR_CAST(g);
2398		up_write_ref_node(&ft->node, false);
2399		return rule;
2400	}
2401
2402	fte = alloc_fte(ft, spec, flow_act);
2403	if (IS_ERR(fte)) {
2404		up_write_ref_node(&ft->node, false);
2405		err = PTR_ERR(fte);
2406		goto err_alloc_fte;
2407	}
2408
2409	nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2410	up_write_ref_node(&ft->node, false);
2411
2412	err = create_auto_flow_group(ft, g);
2413	if (err)
2414		goto err_release_fg;
2415
2416	err = insert_fte(g, fte);
2417	if (err)
2418		goto err_release_fg;
2419
2420	nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
2421	up_write_ref_node(&g->node, false);
2422	rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
2423	up_write_ref_node(&fte->node, false);
2424	if (IS_ERR(rule))
2425		tree_put_node(&fte->node, false);
2426	tree_put_node(&g->node, false);
2427	return rule;
2428
2429err_release_fg:
2430	up_write_ref_node(&g->node, false);
2431	kmem_cache_free(steering->ftes_cache, fte);
2432err_alloc_fte:
2433	tree_put_node(&g->node, false);
2434	return ERR_PTR(err);
2435}
2436
2437static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
2438{
2439	return ((ft->type == FS_FT_NIC_RX) &&
2440		(MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
2441}
2442
2443struct mlx5_flow_handle *
2444mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2445		    const struct mlx5_flow_spec *spec,
2446		    struct mlx5_flow_act *flow_act,
2447		    struct mlx5_flow_destination *dest,
2448		    int num_dest)
2449{
2450	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2451	static const struct mlx5_flow_spec zero_spec = {};
2452	struct mlx5_flow_destination *gen_dest = NULL;
2453	struct mlx5_flow_table *next_ft = NULL;
2454	struct mlx5_flow_handle *handle = NULL;
2455	u32 sw_action = flow_act->action;
2456	int i;
2457
2458	if (!spec)
2459		spec = &zero_spec;
2460
2461	if (!is_fwd_next_action(sw_action))
2462		return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2463
2464	if (!fwd_next_prio_supported(ft))
2465		return ERR_PTR(-EOPNOTSUPP);
2466
2467	mutex_lock(&root->chain_lock);
2468	next_ft = find_next_fwd_ft(ft, flow_act);
2469	if (!next_ft) {
2470		handle = ERR_PTR(-EOPNOTSUPP);
2471		goto unlock;
2472	}
2473
2474	gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
2475			   GFP_KERNEL);
2476	if (!gen_dest) {
2477		handle = ERR_PTR(-ENOMEM);
2478		goto unlock;
2479	}
2480	for (i = 0; i < num_dest; i++)
2481		gen_dest[i] = dest[i];
2482	gen_dest[i].type =
2483		MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2484	gen_dest[i].ft = next_ft;
2485	dest = gen_dest;
2486	num_dest++;
2487	flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
2488			      MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
2489	flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2490	handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2491	if (IS_ERR(handle))
2492		goto unlock;
2493
2494	if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
2495		mutex_lock(&next_ft->lock);
2496		list_add(&handle->rule[num_dest - 1]->next_ft,
2497			 &next_ft->fwd_rules);
2498		mutex_unlock(&next_ft->lock);
2499		handle->rule[num_dest - 1]->sw_action = sw_action;
2500		handle->rule[num_dest - 1]->ft = ft;
2501	}
2502unlock:
2503	mutex_unlock(&root->chain_lock);
2504	kfree(gen_dest);
2505	return handle;
2506}
2507EXPORT_SYMBOL(mlx5_add_flow_rules);
2508
2509void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2510{
2511	struct fs_fte *fte;
2512	int i;
2513
2514	/* In order to consolidate the HW changes we lock the FTE for other
2515	 * changes, and increase its refcount, in order not to perform the
2516	 * "del" functions of the FTE. Will handle them here.
2517	 * The removal of the rules is done under locked FTE.
2518	 * After removing all the handle's rules, if there are remaining
2519	 * rules, it means we just need to modify the FTE in FW, and
2520	 * unlock/decrease the refcount we increased before.
2521	 * Otherwise, it means the FTE should be deleted. First delete the
2522	 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2523	 * the FTE, which will handle the last decrease of the refcount, as
2524	 * well as required handling of its parent.
2525	 */
2526	fs_get_obj(fte, handle->rule[0]->node.parent);
2527	down_write_ref_node(&fte->node, false);
2528	for (i = handle->num_rules - 1; i >= 0; i--)
2529		tree_remove_node(&handle->rule[i]->node, true);
2530	if (list_empty(&fte->node.children)) {
2531		fte->node.del_hw_func(&fte->node);
2532		up_write_ref_node(&fte->node, false);
2533		tree_put_node(&fte->node, false);
2534	} else if (fte->act_dests.dests_size) {
2535		if (fte->act_dests.modify_mask)
2536			modify_fte(fte);
2537		up_write_ref_node(&fte->node, false);
2538	} else {
2539		up_write_ref_node(&fte->node, false);
2540	}
2541	kfree(handle);
2542}
2543EXPORT_SYMBOL(mlx5_del_flow_rules);
2544
2545/* Assuming prio->node.children(flow tables) is sorted by level */
2546static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2547{
2548	struct fs_node *prio_parent, *child;
2549	struct fs_prio *prio;
2550
2551	fs_get_obj(prio, ft->node.parent);
2552
2553	if (!list_is_last(&ft->node.list, &prio->node.children))
2554		return list_next_entry(ft, node.list);
2555
2556	prio_parent = find_prio_chains_parent(&prio->node, &child);
2557
2558	if (prio_parent && list_is_first(&child->list, &prio_parent->children))
2559		return find_closest_ft(&prio->node, false, false);
2560
2561	return find_next_chained_ft(&prio->node);
2562}
2563
2564static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2565{
2566	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2567	struct mlx5_ft_underlay_qp *uqp;
2568	struct mlx5_flow_table *new_root_ft = NULL;
2569	int err = 0;
2570	u32 qpn;
2571
2572	if (root->root_ft != ft)
2573		return 0;
2574
2575	new_root_ft = find_next_ft(ft);
2576	if (!new_root_ft) {
2577		root->root_ft = NULL;
2578		return 0;
2579	}
2580
2581	if (list_empty(&root->underlay_qpns)) {
2582		/* Don't set any QPN (zero) in case QPN list is empty */
2583		qpn = 0;
2584		err = root->cmds->update_root_ft(root, new_root_ft,
2585						 qpn, false);
2586	} else {
2587		list_for_each_entry(uqp, &root->underlay_qpns, list) {
2588			qpn = uqp->qpn;
2589			err = root->cmds->update_root_ft(root,
2590							 new_root_ft, qpn,
2591							 false);
2592			if (err)
2593				break;
2594		}
2595	}
2596
2597	if (err)
2598		mlx5_core_warn(root->dev,
2599			       "Update root flow table of id(%u) qpn(%d) failed\n",
2600			       ft->id, qpn);
2601	else
2602		root->root_ft = new_root_ft;
2603
2604	return 0;
2605}
2606
2607/* Connect flow table from previous priority to
2608 * the next flow table.
2609 */
2610static int disconnect_flow_table(struct mlx5_flow_table *ft)
2611{
2612	struct mlx5_core_dev *dev = get_dev(&ft->node);
2613	struct mlx5_flow_table *next_ft;
2614	struct fs_prio *prio;
2615	int err = 0;
2616
2617	err = update_root_ft_destroy(ft);
2618	if (err)
2619		return err;
2620
2621	fs_get_obj(prio, ft->node.parent);
2622	if  (!(list_first_entry(&prio->node.children,
2623				struct mlx5_flow_table,
2624				node.list) == ft))
2625		return 0;
2626
2627	next_ft = find_next_ft(ft);
2628	err = connect_fwd_rules(dev, next_ft, ft);
2629	if (err)
2630		return err;
2631
2632	err = connect_prev_fts(dev, next_ft, prio);
2633	if (err)
2634		mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2635			       ft->id);
2636	return err;
2637}
2638
2639int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2640{
2641	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2642	int err = 0;
2643
2644	mutex_lock(&root->chain_lock);
2645	if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2646		err = disconnect_flow_table(ft);
2647	if (err) {
2648		mutex_unlock(&root->chain_lock);
2649		return err;
2650	}
2651	if (tree_remove_node(&ft->node, false))
2652		mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2653			       ft->id);
2654	mutex_unlock(&root->chain_lock);
2655
2656	return err;
2657}
2658EXPORT_SYMBOL(mlx5_destroy_flow_table);
2659
2660void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2661{
2662	if (tree_remove_node(&fg->node, false))
2663		mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2664			       fg->id);
2665}
2666EXPORT_SYMBOL(mlx5_destroy_flow_group);
2667
2668struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2669						int n)
2670{
2671	struct mlx5_flow_steering *steering = dev->priv.steering;
2672
2673	if (!steering || !steering->fdb_sub_ns)
2674		return NULL;
2675
2676	return steering->fdb_sub_ns[n];
2677}
2678EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2679
2680static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type)
2681{
2682	switch (type) {
2683	case MLX5_FLOW_NAMESPACE_BYPASS:
2684	case MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC:
2685	case MLX5_FLOW_NAMESPACE_LAG:
2686	case MLX5_FLOW_NAMESPACE_OFFLOADS:
2687	case MLX5_FLOW_NAMESPACE_ETHTOOL:
2688	case MLX5_FLOW_NAMESPACE_KERNEL:
2689	case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2690	case MLX5_FLOW_NAMESPACE_ANCHOR:
2691		return true;
2692	default:
2693		return false;
2694	}
2695}
2696
2697struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2698						    enum mlx5_flow_namespace_type type)
2699{
2700	struct mlx5_flow_steering *steering = dev->priv.steering;
2701	struct mlx5_flow_root_namespace *root_ns;
2702	int prio = 0;
2703	struct fs_prio *fs_prio;
2704	struct mlx5_flow_namespace *ns;
2705
2706	if (!steering)
2707		return NULL;
2708
2709	switch (type) {
2710	case MLX5_FLOW_NAMESPACE_FDB:
2711		if (steering->fdb_root_ns)
2712			return &steering->fdb_root_ns->ns;
2713		return NULL;
2714	case MLX5_FLOW_NAMESPACE_PORT_SEL:
2715		if (steering->port_sel_root_ns)
2716			return &steering->port_sel_root_ns->ns;
2717		return NULL;
2718	case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2719		if (steering->sniffer_rx_root_ns)
2720			return &steering->sniffer_rx_root_ns->ns;
2721		return NULL;
2722	case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2723		if (steering->sniffer_tx_root_ns)
2724			return &steering->sniffer_tx_root_ns->ns;
2725		return NULL;
2726	case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
2727		root_ns = steering->fdb_root_ns;
2728		prio =  FDB_BYPASS_PATH;
2729		break;
2730	case MLX5_FLOW_NAMESPACE_EGRESS:
2731	case MLX5_FLOW_NAMESPACE_EGRESS_IPSEC:
2732	case MLX5_FLOW_NAMESPACE_EGRESS_MACSEC:
2733		root_ns = steering->egress_root_ns;
2734		prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2735		break;
2736	case MLX5_FLOW_NAMESPACE_RDMA_RX:
2737		root_ns = steering->rdma_rx_root_ns;
2738		prio = RDMA_RX_BYPASS_PRIO;
2739		break;
2740	case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL:
2741		root_ns = steering->rdma_rx_root_ns;
2742		prio = RDMA_RX_KERNEL_PRIO;
2743		break;
2744	case MLX5_FLOW_NAMESPACE_RDMA_TX:
2745		root_ns = steering->rdma_tx_root_ns;
2746		prio = RDMA_TX_BYPASS_PRIO;
2747		break;
2748	case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS:
2749		root_ns = steering->rdma_rx_root_ns;
2750		prio = RDMA_RX_COUNTERS_PRIO;
2751		break;
2752	case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS:
2753		root_ns = steering->rdma_tx_root_ns;
2754		prio = RDMA_TX_COUNTERS_PRIO;
2755		break;
2756	case MLX5_FLOW_NAMESPACE_RDMA_RX_IPSEC:
2757		root_ns = steering->rdma_rx_root_ns;
2758		prio = RDMA_RX_IPSEC_PRIO;
2759		break;
2760	case MLX5_FLOW_NAMESPACE_RDMA_TX_IPSEC:
2761		root_ns = steering->rdma_tx_root_ns;
2762		prio = RDMA_TX_IPSEC_PRIO;
2763		break;
2764	case MLX5_FLOW_NAMESPACE_RDMA_RX_MACSEC:
2765		root_ns = steering->rdma_rx_root_ns;
2766		prio = RDMA_RX_MACSEC_PRIO;
2767		break;
2768	case MLX5_FLOW_NAMESPACE_RDMA_TX_MACSEC:
2769		root_ns = steering->rdma_tx_root_ns;
2770		prio = RDMA_TX_MACSEC_PRIO;
2771		break;
2772	default: /* Must be NIC RX */
2773		WARN_ON(!is_nic_rx_ns(type));
2774		root_ns = steering->root_ns;
2775		prio = type;
2776		break;
2777	}
2778
2779	if (!root_ns)
2780		return NULL;
2781
2782	fs_prio = find_prio(&root_ns->ns, prio);
2783	if (!fs_prio)
2784		return NULL;
2785
2786	ns = list_first_entry(&fs_prio->node.children,
2787			      typeof(*ns),
2788			      node.list);
2789
2790	return ns;
2791}
2792EXPORT_SYMBOL(mlx5_get_flow_namespace);
2793
2794struct mlx5_vport_acl_root_ns {
2795	u16 vport_idx;
2796	struct mlx5_flow_root_namespace *root_ns;
2797};
2798
2799struct mlx5_flow_namespace *
2800mlx5_get_flow_vport_namespace(struct mlx5_core_dev *dev,
2801			      enum mlx5_flow_namespace_type type, int vport_idx)
2802{
2803	struct mlx5_flow_steering *steering = dev->priv.steering;
2804	struct mlx5_vport_acl_root_ns *vport_ns;
2805
2806	if (!steering)
2807		return NULL;
2808
2809	switch (type) {
2810	case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2811		vport_ns = xa_load(&steering->esw_egress_root_ns, vport_idx);
2812		if (vport_ns)
2813			return &vport_ns->root_ns->ns;
2814		else
2815			return NULL;
2816	case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2817		vport_ns = xa_load(&steering->esw_ingress_root_ns, vport_idx);
2818		if (vport_ns)
2819			return &vport_ns->root_ns->ns;
2820		else
2821			return NULL;
2822	case MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX:
2823		if (vport_idx >= steering->rdma_transport_rx_vports)
2824			return NULL;
2825		if (steering->rdma_transport_rx_root_ns &&
2826		    steering->rdma_transport_rx_root_ns[vport_idx])
2827			return &steering->rdma_transport_rx_root_ns[vport_idx]->ns;
2828		else
2829			return NULL;
2830	case MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX:
2831		if (vport_idx >= steering->rdma_transport_tx_vports)
2832			return NULL;
2833
2834		if (steering->rdma_transport_tx_root_ns &&
2835		    steering->rdma_transport_tx_root_ns[vport_idx])
2836			return &steering->rdma_transport_tx_root_ns[vport_idx]->ns;
2837		else
2838			return NULL;
2839	default:
2840		return NULL;
2841	}
2842}
2843EXPORT_SYMBOL(mlx5_get_flow_vport_namespace);
2844
2845static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2846				       unsigned int prio,
2847				       int num_levels,
2848				       enum fs_node_type type)
2849{
2850	struct fs_prio *fs_prio;
2851
2852	fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2853	if (!fs_prio)
2854		return ERR_PTR(-ENOMEM);
2855
2856	fs_prio->node.type = type;
2857	tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2858	tree_add_node(&fs_prio->node, &ns->node);
2859	fs_prio->num_levels = num_levels;
2860	fs_prio->prio = prio;
2861	list_add_tail(&fs_prio->node.list, &ns->node.children);
2862
2863	return fs_prio;
2864}
2865
2866static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2867					      unsigned int prio,
2868					      int num_levels)
2869{
2870	return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2871}
2872
2873static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2874				      unsigned int prio, int num_levels)
2875{
2876	return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2877}
2878
2879static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2880						     *ns)
2881{
2882	ns->node.type = FS_TYPE_NAMESPACE;
2883
2884	return ns;
2885}
2886
2887static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2888						       int def_miss_act)
2889{
2890	struct mlx5_flow_namespace	*ns;
2891
2892	ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2893	if (!ns)
2894		return ERR_PTR(-ENOMEM);
2895
2896	fs_init_namespace(ns);
2897	ns->def_miss_action = def_miss_act;
2898	tree_init_node(&ns->node, NULL, del_sw_ns);
2899	tree_add_node(&ns->node, &prio->node);
2900	list_add_tail(&ns->node.list, &prio->node.children);
2901
2902	return ns;
2903}
2904
2905static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2906			     struct init_tree_node *prio_metadata)
2907{
2908	struct fs_prio *fs_prio;
2909	int i;
2910
2911	for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2912		fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2913		if (IS_ERR(fs_prio))
2914			return PTR_ERR(fs_prio);
2915	}
2916	return 0;
2917}
2918
2919#define FLOW_TABLE_BIT_SZ 1
2920#define GET_FLOW_TABLE_CAP(dev, offset) \
2921	((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) +	\
2922			offset / 32)) >>					\
2923	  (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2924static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2925{
2926	int i;
2927
2928	for (i = 0; i < caps->arr_sz; i++) {
2929		if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2930			return false;
2931	}
2932	return true;
2933}
2934
2935static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2936				    struct init_tree_node *init_node,
2937				    struct fs_node *fs_parent_node,
2938				    struct init_tree_node *init_parent_node,
2939				    int prio)
2940{
2941	int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2942					      flow_table_properties_nic_receive.
2943					      max_ft_level);
2944	struct mlx5_flow_namespace *fs_ns;
2945	struct fs_prio *fs_prio;
2946	struct fs_node *base;
2947	int i;
2948	int err;
2949
2950	if (init_node->type == FS_TYPE_PRIO) {
2951		if ((init_node->min_ft_level > max_ft_level) ||
2952		    !has_required_caps(steering->dev, &init_node->caps))
2953			return 0;
2954
2955		fs_get_obj(fs_ns, fs_parent_node);
2956		if (init_node->num_leaf_prios)
2957			return create_leaf_prios(fs_ns, prio, init_node);
2958		fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2959		if (IS_ERR(fs_prio))
2960			return PTR_ERR(fs_prio);
2961		base = &fs_prio->node;
2962	} else if (init_node->type == FS_TYPE_NAMESPACE) {
2963		fs_get_obj(fs_prio, fs_parent_node);
2964		fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2965		if (IS_ERR(fs_ns))
2966			return PTR_ERR(fs_ns);
2967		base = &fs_ns->node;
2968	} else {
2969		return -EINVAL;
2970	}
2971	prio = 0;
2972	for (i = 0; i < init_node->ar_size; i++) {
2973		err = init_root_tree_recursive(steering, &init_node->children[i],
2974					       base, init_node, prio);
2975		if (err)
2976			return err;
2977		if (init_node->children[i].type == FS_TYPE_PRIO &&
2978		    init_node->children[i].num_leaf_prios) {
2979			prio += init_node->children[i].num_leaf_prios;
2980		}
2981	}
2982
2983	return 0;
2984}
2985
2986static int init_root_tree(struct mlx5_flow_steering *steering,
2987			  struct init_tree_node *init_node,
2988			  struct fs_node *fs_parent_node)
2989{
2990	int err;
2991	int i;
2992
2993	for (i = 0; i < init_node->ar_size; i++) {
2994		err = init_root_tree_recursive(steering, &init_node->children[i],
2995					       fs_parent_node,
2996					       init_node, i);
2997		if (err)
2998			return err;
2999	}
3000	return 0;
3001}
3002
3003static void del_sw_root_ns(struct fs_node *node)
3004{
3005	struct mlx5_flow_root_namespace *root_ns;
3006	struct mlx5_flow_namespace *ns;
3007
3008	fs_get_obj(ns, node);
3009	root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
3010	mutex_destroy(&root_ns->chain_lock);
3011	kfree(node);
3012}
3013
3014static struct mlx5_flow_root_namespace
3015*create_root_ns(struct mlx5_flow_steering *steering,
3016		enum fs_flow_table_type table_type)
3017{
3018	const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
3019	struct mlx5_flow_root_namespace *root_ns;
3020	struct mlx5_flow_namespace *ns;
3021
3022	/* Create the root namespace */
3023	root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
3024	if (!root_ns)
3025		return NULL;
3026
3027	root_ns->dev = steering->dev;
3028	root_ns->table_type = table_type;
3029	root_ns->cmds = cmds;
3030
3031	INIT_LIST_HEAD(&root_ns->underlay_qpns);
3032
3033	ns = &root_ns->ns;
3034	fs_init_namespace(ns);
3035	mutex_init(&root_ns->chain_lock);
3036	tree_init_node(&ns->node, NULL, del_sw_root_ns);
3037	tree_add_node(&ns->node, NULL);
3038
3039	return root_ns;
3040}
3041
3042static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
3043
3044static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
3045{
3046	struct fs_prio *prio;
3047
3048	fs_for_each_prio(prio, ns) {
3049		 /* This updates prio start_level and num_levels */
3050		set_prio_attrs_in_prio(prio, acc_level);
3051		acc_level += prio->num_levels;
3052	}
3053	return acc_level;
3054}
3055
3056static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
3057{
3058	struct mlx5_flow_namespace *ns;
3059	int acc_level_ns = acc_level;
3060
3061	prio->start_level = acc_level;
3062	fs_for_each_ns(ns, prio) {
3063		/* This updates start_level and num_levels of ns's priority descendants */
3064		acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
3065
3066		/* If this a prio with chains, and we can jump from one chain
3067		 * (namespace) to another, so we accumulate the levels
3068		 */
3069		if (prio->node.type == FS_TYPE_PRIO_CHAINS)
3070			acc_level = acc_level_ns;
3071	}
3072
3073	if (!prio->num_levels)
3074		prio->num_levels = acc_level_ns - prio->start_level;
3075	WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
3076}
3077
3078static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
3079{
3080	struct mlx5_flow_namespace *ns = &root_ns->ns;
3081	struct fs_prio *prio;
3082	int start_level = 0;
3083
3084	fs_for_each_prio(prio, ns) {
3085		set_prio_attrs_in_prio(prio, start_level);
3086		start_level += prio->num_levels;
3087	}
3088}
3089
3090#define ANCHOR_PRIO 0
3091#define ANCHOR_SIZE 1
3092#define ANCHOR_LEVEL 0
3093static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
3094{
3095	struct mlx5_flow_namespace *ns = NULL;
3096	struct mlx5_flow_table_attr ft_attr = {};
3097	struct mlx5_flow_table *ft;
3098
3099	ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
3100	if (WARN_ON(!ns))
3101		return -EINVAL;
3102
3103	ft_attr.max_fte = ANCHOR_SIZE;
3104	ft_attr.level   = ANCHOR_LEVEL;
3105	ft_attr.prio    = ANCHOR_PRIO;
3106
3107	ft = mlx5_create_flow_table(ns, &ft_attr);
3108	if (IS_ERR(ft)) {
3109		mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
3110		return PTR_ERR(ft);
3111	}
3112	return 0;
3113}
3114
3115static int init_root_ns(struct mlx5_flow_steering *steering)
3116{
3117	int err;
3118
3119	steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
3120	if (!steering->root_ns)
3121		return -ENOMEM;
3122
3123	err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
3124	if (err)
3125		goto out_err;
3126
3127	set_prio_attrs(steering->root_ns);
3128	err = create_anchor_flow_table(steering);
3129	if (err)
3130		goto out_err;
3131
3132	return 0;
3133
3134out_err:
3135	cleanup_root_ns(steering->root_ns);
3136	steering->root_ns = NULL;
3137	return err;
3138}
3139
3140static void clean_tree(struct fs_node *node)
3141{
3142	if (node) {
3143		struct fs_node *iter;
3144		struct fs_node *temp;
3145
3146		tree_get_node(node);
3147		list_for_each_entry_safe(iter, temp, &node->children, list)
3148			clean_tree(iter);
3149		tree_put_node(node, false);
3150		tree_remove_node(node, false);
3151	}
3152}
3153
3154static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
3155{
3156	if (!root_ns)
3157		return;
3158
3159	clean_tree(&root_ns->ns.node);
3160}
3161
3162static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
3163{
3164	struct fs_prio *prio;
3165
3166	steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
3167	if (!steering->sniffer_tx_root_ns)
3168		return -ENOMEM;
3169
3170	/* Create single prio */
3171	prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
3172	return PTR_ERR_OR_ZERO(prio);
3173}
3174
3175static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
3176{
3177	struct fs_prio *prio;
3178
3179	steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
3180	if (!steering->sniffer_rx_root_ns)
3181		return -ENOMEM;
3182
3183	/* Create single prio */
3184	prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
3185	return PTR_ERR_OR_ZERO(prio);
3186}
3187
3188#define PORT_SEL_NUM_LEVELS 3
3189static int init_port_sel_root_ns(struct mlx5_flow_steering *steering)
3190{
3191	struct fs_prio *prio;
3192
3193	steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL);
3194	if (!steering->port_sel_root_ns)
3195		return -ENOMEM;
3196
3197	/* Create single prio */
3198	prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0,
3199			      PORT_SEL_NUM_LEVELS);
3200	return PTR_ERR_OR_ZERO(prio);
3201}
3202
3203static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
3204{
3205	int err;
3206
3207	steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
3208	if (!steering->rdma_rx_root_ns)
3209		return -ENOMEM;
3210
3211	err = init_root_tree(steering, &rdma_rx_root_fs,
3212			     &steering->rdma_rx_root_ns->ns.node);
3213	if (err)
3214		goto out_err;
3215
3216	set_prio_attrs(steering->rdma_rx_root_ns);
3217
3218	return 0;
3219
3220out_err:
3221	cleanup_root_ns(steering->rdma_rx_root_ns);
3222	steering->rdma_rx_root_ns = NULL;
3223	return err;
3224}
3225
3226static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
3227{
3228	int err;
3229
3230	steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
3231	if (!steering->rdma_tx_root_ns)
3232		return -ENOMEM;
3233
3234	err = init_root_tree(steering, &rdma_tx_root_fs,
3235			     &steering->rdma_tx_root_ns->ns.node);
3236	if (err)
3237		goto out_err;
3238
3239	set_prio_attrs(steering->rdma_tx_root_ns);
3240
3241	return 0;
3242
3243out_err:
3244	cleanup_root_ns(steering->rdma_tx_root_ns);
3245	steering->rdma_tx_root_ns = NULL;
3246	return err;
3247}
3248
3249static int
3250init_rdma_transport_rx_root_ns_one(struct mlx5_flow_steering *steering,
3251				   int vport_idx)
3252{
3253	struct mlx5_flow_root_namespace *root_ns;
3254	struct fs_prio *prio;
3255	int ret;
3256	int i;
3257
3258	steering->rdma_transport_rx_root_ns[vport_idx] =
3259		create_root_ns(steering, FS_FT_RDMA_TRANSPORT_RX);
3260	if (!steering->rdma_transport_rx_root_ns[vport_idx])
3261		return -ENOMEM;
3262
3263	root_ns = steering->rdma_transport_rx_root_ns[vport_idx];
3264
3265	for (i = 0; i < MLX5_RDMA_TRANSPORT_BYPASS_PRIO; i++) {
3266		prio = fs_create_prio(&root_ns->ns, i, 1);
3267		if (IS_ERR(prio)) {
3268			ret = PTR_ERR(prio);
3269			goto err;
3270		}
3271	}
3272	set_prio_attrs(root_ns);
3273	return 0;
3274
3275err:
3276	cleanup_root_ns(root_ns);
3277	return ret;
3278}
3279
3280static int
3281init_rdma_transport_tx_root_ns_one(struct mlx5_flow_steering *steering,
3282				   int vport_idx)
3283{
3284	struct mlx5_flow_root_namespace *root_ns;
3285	struct fs_prio *prio;
3286	int ret;
3287	int i;
3288
3289	steering->rdma_transport_tx_root_ns[vport_idx] =
3290		create_root_ns(steering, FS_FT_RDMA_TRANSPORT_TX);
3291	if (!steering->rdma_transport_tx_root_ns[vport_idx])
3292		return -ENOMEM;
3293
3294	root_ns = steering->rdma_transport_tx_root_ns[vport_idx];
3295
3296	for (i = 0; i < MLX5_RDMA_TRANSPORT_BYPASS_PRIO; i++) {
3297		prio = fs_create_prio(&root_ns->ns, i, 1);
3298		if (IS_ERR(prio)) {
3299			ret = PTR_ERR(prio);
3300			goto err;
3301		}
3302	}
3303	set_prio_attrs(root_ns);
3304	return 0;
3305
3306err:
3307	cleanup_root_ns(root_ns);
3308	return ret;
3309}
3310
3311static bool mlx5_fs_ns_is_empty(struct mlx5_flow_namespace *ns)
3312{
3313	struct fs_prio *iter_prio;
3314
3315	fs_for_each_prio(iter_prio, ns) {
3316		if (iter_prio->num_ft)
3317			return false;
3318	}
3319
3320	return true;
3321}
3322
3323int mlx5_fs_set_root_dev(struct mlx5_core_dev *dev,
3324			 struct mlx5_core_dev *new_dev,
3325			 enum fs_flow_table_type table_type)
3326{
3327	struct mlx5_flow_root_namespace	**root;
3328	int total_vports;
3329	int i;
3330
3331	switch (table_type) {
3332	case FS_FT_RDMA_TRANSPORT_TX:
3333		root = dev->priv.steering->rdma_transport_tx_root_ns;
3334		total_vports = dev->priv.steering->rdma_transport_tx_vports;
3335		break;
3336	case FS_FT_RDMA_TRANSPORT_RX:
3337		root = dev->priv.steering->rdma_transport_rx_root_ns;
3338		total_vports = dev->priv.steering->rdma_transport_rx_vports;
3339		break;
3340	default:
3341		WARN_ON_ONCE(true);
3342		return -EINVAL;
3343	}
3344
3345	for (i = 0; i < total_vports; i++) {
3346		mutex_lock(&root[i]->chain_lock);
3347		if (!mlx5_fs_ns_is_empty(&root[i]->ns)) {
3348			mutex_unlock(&root[i]->chain_lock);
3349			goto err;
3350		}
3351		root[i]->dev = new_dev;
3352		mutex_unlock(&root[i]->chain_lock);
3353	}
3354	return 0;
3355err:
3356	while (i--) {
3357		mutex_lock(&root[i]->chain_lock);
3358		root[i]->dev = dev;
3359		mutex_unlock(&root[i]->chain_lock);
3360	}
3361	/* If you hit this error try destroying all flow tables and try again */
3362	mlx5_core_err(dev, "Failed to set root device for RDMA TRANSPORT\n");
3363	return -EINVAL;
3364}
3365EXPORT_SYMBOL(mlx5_fs_set_root_dev);
3366
3367static int init_rdma_transport_rx_root_ns(struct mlx5_flow_steering *steering)
3368{
3369	struct mlx5_core_dev *dev = steering->dev;
3370	int total_vports;
3371	int err;
3372	int i;
3373
3374	/* In case eswitch not supported and working in legacy mode */
3375	total_vports = mlx5_eswitch_get_total_vports(dev) ?: 1;
3376
3377	steering->rdma_transport_rx_root_ns =
3378			kcalloc(total_vports,
3379				sizeof(*steering->rdma_transport_rx_root_ns),
3380				GFP_KERNEL);
3381	if (!steering->rdma_transport_rx_root_ns)
3382		return -ENOMEM;
3383
3384	for (i = 0; i < total_vports; i++) {
3385		err = init_rdma_transport_rx_root_ns_one(steering, i);
3386		if (err)
3387			goto cleanup_root_ns;
3388	}
3389	steering->rdma_transport_rx_vports = total_vports;
3390	return 0;
3391
3392cleanup_root_ns:
3393	while (i--)
3394		cleanup_root_ns(steering->rdma_transport_rx_root_ns[i]);
3395	kfree(steering->rdma_transport_rx_root_ns);
3396	steering->rdma_transport_rx_root_ns = NULL;
3397	return err;
3398}
3399
3400static int init_rdma_transport_tx_root_ns(struct mlx5_flow_steering *steering)
3401{
3402	struct mlx5_core_dev *dev = steering->dev;
3403	int total_vports;
3404	int err;
3405	int i;
3406
3407	/* In case eswitch not supported and working in legacy mode */
3408	total_vports = mlx5_eswitch_get_total_vports(dev) ?: 1;
3409
3410	steering->rdma_transport_tx_root_ns =
3411			kcalloc(total_vports,
3412				sizeof(*steering->rdma_transport_tx_root_ns),
3413				GFP_KERNEL);
3414	if (!steering->rdma_transport_tx_root_ns)
3415		return -ENOMEM;
3416
3417	for (i = 0; i < total_vports; i++) {
3418		err = init_rdma_transport_tx_root_ns_one(steering, i);
3419		if (err)
3420			goto cleanup_root_ns;
3421	}
3422	steering->rdma_transport_tx_vports = total_vports;
3423	return 0;
3424
3425cleanup_root_ns:
3426	while (i--)
3427		cleanup_root_ns(steering->rdma_transport_tx_root_ns[i]);
3428	kfree(steering->rdma_transport_tx_root_ns);
3429	steering->rdma_transport_tx_root_ns = NULL;
3430	return err;
3431}
3432
3433static void cleanup_rdma_transport_roots_ns(struct mlx5_flow_steering *steering)
3434{
3435	int i;
3436
3437	if (steering->rdma_transport_rx_root_ns) {
3438		for (i = 0; i < steering->rdma_transport_rx_vports; i++)
3439			cleanup_root_ns(steering->rdma_transport_rx_root_ns[i]);
3440
3441		kfree(steering->rdma_transport_rx_root_ns);
3442		steering->rdma_transport_rx_root_ns = NULL;
3443	}
3444
3445	if (steering->rdma_transport_tx_root_ns) {
3446		for (i = 0; i < steering->rdma_transport_tx_vports; i++)
3447			cleanup_root_ns(steering->rdma_transport_tx_root_ns[i]);
3448
3449		kfree(steering->rdma_transport_tx_root_ns);
3450		steering->rdma_transport_tx_root_ns = NULL;
3451	}
3452}
3453
3454/* FT and tc chains are stored in the same array so we can re-use the
3455 * mlx5_get_fdb_sub_ns() and tc api for FT chains.
3456 * When creating a new ns for each chain store it in the first available slot.
3457 * Assume tc chains are created and stored first and only then the FT chain.
3458 */
3459static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
3460					struct mlx5_flow_namespace *ns)
3461{
3462	int chain = 0;
3463
3464	while (steering->fdb_sub_ns[chain])
3465		++chain;
3466
3467	steering->fdb_sub_ns[chain] = ns;
3468}
3469
3470static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
3471					struct fs_prio *maj_prio)
3472{
3473	struct mlx5_flow_namespace *ns;
3474	struct fs_prio *min_prio;
3475	int prio;
3476
3477	ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
3478	if (IS_ERR(ns))
3479		return PTR_ERR(ns);
3480
3481	for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
3482		min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
3483		if (IS_ERR(min_prio))
3484			return PTR_ERR(min_prio);
3485	}
3486
3487	store_fdb_sub_ns_prio_chain(steering, ns);
3488
3489	return 0;
3490}
3491
3492static int create_fdb_chains(struct mlx5_flow_steering *steering,
3493			     int fs_prio,
3494			     int chains)
3495{
3496	struct fs_prio *maj_prio;
3497	int levels;
3498	int chain;
3499	int err;
3500
3501	levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
3502	maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
3503					  fs_prio,
3504					  levels);
3505	if (IS_ERR(maj_prio))
3506		return PTR_ERR(maj_prio);
3507
3508	for (chain = 0; chain < chains; chain++) {
3509		err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
3510		if (err)
3511			return err;
3512	}
3513
3514	return 0;
3515}
3516
3517static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
3518{
3519	int err;
3520
3521	steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
3522				       sizeof(*steering->fdb_sub_ns),
3523				       GFP_KERNEL);
3524	if (!steering->fdb_sub_ns)
3525		return -ENOMEM;
3526
3527	err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
3528	if (err)
3529		return err;
3530
3531	err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
3532	if (err)
3533		return err;
3534
3535	return 0;
3536}
3537
3538static int create_fdb_bypass(struct mlx5_flow_steering *steering)
3539{
3540	struct mlx5_flow_namespace *ns;
3541	struct fs_prio *prio;
3542	int i;
3543
3544	prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0);
3545	if (IS_ERR(prio))
3546		return PTR_ERR(prio);
3547
3548	ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
3549	if (IS_ERR(ns))
3550		return PTR_ERR(ns);
3551
3552	for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) {
3553		prio = fs_create_prio(ns, i, 1);
3554		if (IS_ERR(prio))
3555			return PTR_ERR(prio);
3556	}
3557	return 0;
3558}
3559
3560static void cleanup_fdb_root_ns(struct mlx5_flow_steering *steering)
3561{
3562	cleanup_root_ns(steering->fdb_root_ns);
3563	steering->fdb_root_ns = NULL;
3564	kfree(steering->fdb_sub_ns);
3565	steering->fdb_sub_ns = NULL;
3566}
3567
3568static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
3569{
3570	struct fs_prio *maj_prio;
3571	int err;
3572
3573	steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
3574	if (!steering->fdb_root_ns)
3575		return -ENOMEM;
3576
3577	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_DROP_ROOT, 1);
3578	err = PTR_ERR_OR_ZERO(maj_prio);
3579	if (err)
3580		goto out_err;
3581
3582	err = create_fdb_bypass(steering);
3583	if (err)
3584		goto out_err;
3585
3586	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_CRYPTO_INGRESS, 3);
3587	if (IS_ERR(maj_prio)) {
3588		err = PTR_ERR(maj_prio);
3589		goto out_err;
3590	}
3591
3592	err = create_fdb_fast_path(steering);
3593	if (err)
3594		goto out_err;
3595
3596	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
3597	if (IS_ERR(maj_prio)) {
3598		err = PTR_ERR(maj_prio);
3599		goto out_err;
3600	}
3601
3602	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 4);
3603	if (IS_ERR(maj_prio)) {
3604		err = PTR_ERR(maj_prio);
3605		goto out_err;
3606	}
3607
3608	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
3609	if (IS_ERR(maj_prio)) {
3610		err = PTR_ERR(maj_prio);
3611		goto out_err;
3612	}
3613
3614	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_CRYPTO_EGRESS, 3);
3615	if (IS_ERR(maj_prio)) {
3616		err = PTR_ERR(maj_prio);
3617		goto out_err;
3618	}
3619
3620	/* We put this priority last, knowing that nothing will get here
3621	 * unless explicitly forwarded to. This is possible because the
3622	 * slow path tables have catch all rules and nothing gets passed
3623	 * those tables.
3624	 */
3625	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
3626	if (IS_ERR(maj_prio)) {
3627		err = PTR_ERR(maj_prio);
3628		goto out_err;
3629	}
3630
3631	set_prio_attrs(steering->fdb_root_ns);
3632	return 0;
3633
3634out_err:
3635	cleanup_fdb_root_ns(steering);
3636	return err;
3637}
3638
3639static void
3640mlx5_fs_remove_vport_acl_root_ns(struct xarray *esw_acl_root_ns, u16 vport_idx)
3641{
3642	struct mlx5_vport_acl_root_ns *vport_ns;
3643
3644	vport_ns = xa_erase(esw_acl_root_ns, vport_idx);
3645	if (vport_ns) {
3646		cleanup_root_ns(vport_ns->root_ns);
3647		kfree(vport_ns);
3648	}
3649}
3650
3651static int
3652mlx5_fs_add_vport_acl_root_ns(struct mlx5_flow_steering *steering,
3653			      struct xarray *esw_acl_root_ns,
3654			      enum fs_flow_table_type table_type,
3655			      u16 vport_idx)
3656{
3657	struct mlx5_vport_acl_root_ns *vport_ns;
3658	struct fs_prio *prio;
3659	int err;
3660
3661	/* sanity check, intended xarrays are used */
3662	if (WARN_ON(esw_acl_root_ns != &steering->esw_egress_root_ns &&
3663		    esw_acl_root_ns != &steering->esw_ingress_root_ns))
3664		return -EINVAL;
3665
3666	if (table_type != FS_FT_ESW_EGRESS_ACL &&
3667	    table_type != FS_FT_ESW_INGRESS_ACL) {
3668		mlx5_core_err(steering->dev,
3669			      "Invalid table type %d for egress/ingress ACLs\n",
3670			      table_type);
3671		return -EINVAL;
3672	}
3673
3674	if (xa_load(esw_acl_root_ns, vport_idx))
3675		return -EEXIST;
3676
3677	vport_ns = kzalloc(sizeof(*vport_ns), GFP_KERNEL);
3678	if (!vport_ns)
3679		return -ENOMEM;
3680
3681	vport_ns->root_ns = create_root_ns(steering, table_type);
3682	if (!vport_ns->root_ns) {
3683		err = -ENOMEM;
3684		goto kfree_vport_ns;
3685	}
3686
3687	/* create 1 prio*/
3688	prio = fs_create_prio(&vport_ns->root_ns->ns, 0, 1);
3689	if (IS_ERR(prio)) {
3690		err = PTR_ERR(prio);
3691		goto cleanup_root_ns;
3692	}
3693
3694	vport_ns->vport_idx = vport_idx;
3695	err = xa_insert(esw_acl_root_ns, vport_idx, vport_ns, GFP_KERNEL);
3696	if (err)
3697		goto cleanup_root_ns;
3698	return 0;
3699
3700cleanup_root_ns:
3701	cleanup_root_ns(vport_ns->root_ns);
3702kfree_vport_ns:
3703	kfree(vport_ns);
3704	return err;
3705}
3706
3707int mlx5_fs_vport_egress_acl_ns_add(struct mlx5_flow_steering *steering,
3708				    u16 vport_idx)
3709{
3710	return mlx5_fs_add_vport_acl_root_ns(steering,
3711					     &steering->esw_egress_root_ns,
3712					     FS_FT_ESW_EGRESS_ACL, vport_idx);
3713}
3714
3715int mlx5_fs_vport_ingress_acl_ns_add(struct mlx5_flow_steering *steering,
3716				     u16 vport_idx)
3717{
3718	return mlx5_fs_add_vport_acl_root_ns(steering,
3719					     &steering->esw_ingress_root_ns,
3720					     FS_FT_ESW_INGRESS_ACL, vport_idx);
3721}
3722
3723void mlx5_fs_vport_egress_acl_ns_remove(struct mlx5_flow_steering *steering,
3724					int vport_idx)
3725{
3726	mlx5_fs_remove_vport_acl_root_ns(&steering->esw_egress_root_ns,
3727					 vport_idx);
3728}
3729
3730void mlx5_fs_vport_ingress_acl_ns_remove(struct mlx5_flow_steering *steering,
3731					 int vport_idx)
3732{
3733	mlx5_fs_remove_vport_acl_root_ns(&steering->esw_ingress_root_ns,
3734					 vport_idx);
3735}
3736
3737u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type)
3738{
3739	struct mlx5_flow_root_namespace *root;
3740	struct mlx5_flow_namespace *ns;
3741
3742	ns = mlx5_get_flow_namespace(dev, type);
3743	if (!ns)
3744		return 0;
3745
3746	root = find_root(&ns->node);
3747	if (!root)
3748		return 0;
3749
3750	return root->cmds->get_capabilities(root, root->table_type);
3751}
3752
3753static int init_egress_root_ns(struct mlx5_flow_steering *steering)
3754{
3755	int err;
3756
3757	steering->egress_root_ns = create_root_ns(steering,
3758						  FS_FT_NIC_TX);
3759	if (!steering->egress_root_ns)
3760		return -ENOMEM;
3761
3762	err = init_root_tree(steering, &egress_root_fs,
3763			     &steering->egress_root_ns->ns.node);
3764	if (err)
3765		goto cleanup;
3766	set_prio_attrs(steering->egress_root_ns);
3767	return 0;
3768cleanup:
3769	cleanup_root_ns(steering->egress_root_ns);
3770	steering->egress_root_ns = NULL;
3771	return err;
3772}
3773
3774static int mlx5_fs_mode_validate(struct devlink *devlink, u32 id,
3775				 union devlink_param_value val,
3776				 struct netlink_ext_ack *extack)
3777{
3778	struct mlx5_core_dev *dev = devlink_priv(devlink);
3779	char *value = val.vstr;
3780	u8 eswitch_mode;
3781
3782	eswitch_mode = mlx5_eswitch_mode(dev);
3783	if (eswitch_mode == MLX5_ESWITCH_OFFLOADS) {
3784		NL_SET_ERR_MSG_FMT_MOD(extack,
3785				       "Changing fs mode is not supported when eswitch offloads enabled.");
3786		return -EOPNOTSUPP;
3787	}
3788
3789	if (!strcmp(value, "dmfs"))
3790		return 0;
3791
3792	if (!strcmp(value, "smfs")) {
3793		bool smfs_cap = mlx5_fs_dr_is_supported(dev);
3794
3795		if (!smfs_cap) {
3796			NL_SET_ERR_MSG_MOD(extack,
3797					   "Software managed steering is not supported by current device");
3798			return -EOPNOTSUPP;
3799		}
3800	} else if (!strcmp(value, "hmfs")) {
3801		bool hmfs_cap = mlx5_fs_hws_is_supported(dev);
3802
3803		if (!hmfs_cap) {
3804			NL_SET_ERR_MSG_MOD(extack,
3805					   "Hardware steering is not supported by current device");
3806			return -EOPNOTSUPP;
3807		}
3808	} else {
3809		NL_SET_ERR_MSG_MOD(extack,
3810				   "Bad parameter: supported values are [\"dmfs\", \"smfs\", \"hmfs\"]");
3811		return -EINVAL;
3812	}
3813
3814	return 0;
3815}
3816
3817static int mlx5_fs_mode_set(struct devlink *devlink, u32 id,
3818			    struct devlink_param_gset_ctx *ctx,
3819			    struct netlink_ext_ack *extack)
3820{
3821	struct mlx5_core_dev *dev = devlink_priv(devlink);
3822	enum mlx5_flow_steering_mode mode;
3823
3824	if (!strcmp(ctx->val.vstr, "smfs"))
3825		mode = MLX5_FLOW_STEERING_MODE_SMFS;
3826	else if (!strcmp(ctx->val.vstr, "hmfs"))
3827		mode = MLX5_FLOW_STEERING_MODE_HMFS;
3828	else
3829		mode = MLX5_FLOW_STEERING_MODE_DMFS;
3830	dev->priv.steering->mode = mode;
3831
3832	return 0;
3833}
3834
3835static int mlx5_fs_mode_get(struct devlink *devlink, u32 id,
3836			    struct devlink_param_gset_ctx *ctx,
3837			    struct netlink_ext_ack *extack)
3838{
3839	struct mlx5_core_dev *dev = devlink_priv(devlink);
3840
3841	switch (dev->priv.steering->mode) {
3842	case MLX5_FLOW_STEERING_MODE_SMFS:
3843		strscpy(ctx->val.vstr, "smfs", sizeof(ctx->val.vstr));
3844		break;
3845	case MLX5_FLOW_STEERING_MODE_HMFS:
3846		strscpy(ctx->val.vstr, "hmfs", sizeof(ctx->val.vstr));
3847		break;
3848	default:
3849		strscpy(ctx->val.vstr, "dmfs", sizeof(ctx->val.vstr));
3850	}
3851
3852	return 0;
3853}
3854
3855static const struct devlink_param mlx5_fs_params[] = {
3856	DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_FLOW_STEERING_MODE,
3857			     "flow_steering_mode", DEVLINK_PARAM_TYPE_STRING,
3858			     BIT(DEVLINK_PARAM_CMODE_RUNTIME),
3859			     mlx5_fs_mode_get, mlx5_fs_mode_set,
3860			     mlx5_fs_mode_validate),
3861};
3862
3863void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev)
3864{
3865	struct mlx5_flow_steering *steering = dev->priv.steering;
3866
3867	WARN_ON(!xa_empty(&steering->esw_egress_root_ns));
3868	WARN_ON(!xa_empty(&steering->esw_ingress_root_ns));
3869	xa_destroy(&steering->esw_egress_root_ns);
3870	xa_destroy(&steering->esw_ingress_root_ns);
3871
3872	cleanup_root_ns(steering->root_ns);
3873	cleanup_fdb_root_ns(steering);
3874	cleanup_root_ns(steering->port_sel_root_ns);
3875	cleanup_root_ns(steering->sniffer_rx_root_ns);
3876	cleanup_root_ns(steering->sniffer_tx_root_ns);
3877	cleanup_root_ns(steering->rdma_rx_root_ns);
3878	cleanup_root_ns(steering->rdma_tx_root_ns);
3879	cleanup_root_ns(steering->egress_root_ns);
3880	cleanup_rdma_transport_roots_ns(steering);
3881
3882	devl_params_unregister(priv_to_devlink(dev), mlx5_fs_params,
3883			       ARRAY_SIZE(mlx5_fs_params));
3884}
3885
3886int mlx5_fs_core_init(struct mlx5_core_dev *dev)
3887{
3888	struct mlx5_flow_steering *steering = dev->priv.steering;
3889	int err;
3890
3891	err = devl_params_register(priv_to_devlink(dev), mlx5_fs_params,
3892				   ARRAY_SIZE(mlx5_fs_params));
3893	if (err)
3894		return err;
3895
3896	if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
3897	      (MLX5_CAP_GEN(dev, nic_flow_table))) ||
3898	     ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
3899	      MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
3900	    MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
3901		err = init_root_ns(steering);
3902		if (err)
3903			goto err;
3904	}
3905
3906	if (MLX5_ESWITCH_MANAGER(dev)) {
3907		if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3908			err = init_fdb_root_ns(steering);
3909			if (err)
3910				goto err;
3911		}
3912	}
3913
3914	if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3915		err = init_sniffer_rx_root_ns(steering);
3916		if (err)
3917			goto err;
3918	}
3919
3920	if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3921		err = init_sniffer_tx_root_ns(steering);
3922		if (err)
3923			goto err;
3924	}
3925
3926	if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) {
3927		err = init_port_sel_root_ns(steering);
3928		if (err)
3929			goto err;
3930	}
3931
3932	if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support)) {
3933		err = init_rdma_rx_root_ns(steering);
3934		if (err)
3935			goto err;
3936	}
3937
3938	if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3939		err = init_rdma_tx_root_ns(steering);
3940		if (err)
3941			goto err;
3942	}
3943
3944	if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3945		err = init_egress_root_ns(steering);
3946		if (err)
3947			goto err;
3948	}
3949
3950	if (MLX5_CAP_FLOWTABLE_RDMA_TRANSPORT_RX(dev, ft_support)) {
3951		err = init_rdma_transport_rx_root_ns(steering);
3952		if (err)
3953			goto err;
3954	}
3955
3956	if (MLX5_CAP_FLOWTABLE_RDMA_TRANSPORT_TX(dev, ft_support)) {
3957		err = init_rdma_transport_tx_root_ns(steering);
3958		if (err)
3959			goto err;
3960	}
3961
3962	xa_init(&steering->esw_egress_root_ns);
3963	xa_init(&steering->esw_ingress_root_ns);
3964	return 0;
3965
3966err:
3967	mlx5_fs_core_cleanup(dev);
3968	return err;
3969}
3970
3971void mlx5_fs_core_free(struct mlx5_core_dev *dev)
3972{
3973	struct mlx5_flow_steering *steering = dev->priv.steering;
3974
3975	kmem_cache_destroy(steering->ftes_cache);
3976	kmem_cache_destroy(steering->fgs_cache);
3977	kfree(steering);
3978	mlx5_ft_pool_destroy(dev);
3979	mlx5_cleanup_fc_stats(dev);
3980}
3981
3982int mlx5_fs_core_alloc(struct mlx5_core_dev *dev)
3983{
3984	struct mlx5_flow_steering *steering;
3985	char name[80];
3986	int err = 0;
3987
3988	err = mlx5_init_fc_stats(dev);
3989	if (err)
3990		return err;
3991
3992	err = mlx5_ft_pool_init(dev);
3993	if (err)
3994		goto err;
3995
3996	steering = kzalloc(sizeof(*steering), GFP_KERNEL);
3997	if (!steering) {
3998		err = -ENOMEM;
3999		goto err;
4000	}
4001
4002	steering->dev = dev;
4003	dev->priv.steering = steering;
4004
4005	if (mlx5_fs_dr_is_supported(dev))
4006		steering->mode = MLX5_FLOW_STEERING_MODE_SMFS;
4007	else if (mlx5_fs_hws_is_supported(dev))
4008		steering->mode = MLX5_FLOW_STEERING_MODE_HMFS;
4009	else
4010		steering->mode = MLX5_FLOW_STEERING_MODE_DMFS;
4011
4012	snprintf(name, sizeof(name), "%s-mlx5_fs_fgs", dev_name(dev->device));
4013	steering->fgs_cache = kmem_cache_create(name,
4014						sizeof(struct mlx5_flow_group), 0,
4015						0, NULL);
4016	snprintf(name, sizeof(name), "%s-mlx5_fs_ftes", dev_name(dev->device));
4017	steering->ftes_cache = kmem_cache_create(name, sizeof(struct fs_fte), 0,
4018						 0, NULL);
4019	if (!steering->ftes_cache || !steering->fgs_cache) {
4020		err = -ENOMEM;
4021		goto err;
4022	}
4023
4024	return 0;
4025
4026err:
4027	mlx5_fs_core_free(dev);
4028	return err;
4029}
4030
4031int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
4032{
4033	struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
4034	struct mlx5_ft_underlay_qp *new_uqp;
4035	int err = 0;
4036
4037	new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
4038	if (!new_uqp)
4039		return -ENOMEM;
4040
4041	mutex_lock(&root->chain_lock);
4042
4043	if (!root->root_ft) {
4044		err = -EINVAL;
4045		goto update_ft_fail;
4046	}
4047
4048	err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
4049					 false);
4050	if (err) {
4051		mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
4052			       underlay_qpn, err);
4053		goto update_ft_fail;
4054	}
4055
4056	new_uqp->qpn = underlay_qpn;
4057	list_add_tail(&new_uqp->list, &root->underlay_qpns);
4058
4059	mutex_unlock(&root->chain_lock);
4060
4061	return 0;
4062
4063update_ft_fail:
4064	mutex_unlock(&root->chain_lock);
4065	kfree(new_uqp);
4066	return err;
4067}
4068EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
4069
4070int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
4071{
4072	struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
4073	struct mlx5_ft_underlay_qp *uqp;
4074	bool found = false;
4075	int err = 0;
4076
4077	mutex_lock(&root->chain_lock);
4078	list_for_each_entry(uqp, &root->underlay_qpns, list) {
4079		if (uqp->qpn == underlay_qpn) {
4080			found = true;
4081			break;
4082		}
4083	}
4084
4085	if (!found) {
4086		mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
4087			       underlay_qpn);
4088		err = -EINVAL;
4089		goto out;
4090	}
4091
4092	err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
4093					 true);
4094	if (err)
4095		mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
4096			       underlay_qpn, err);
4097
4098	list_del(&uqp->list);
4099	mutex_unlock(&root->chain_lock);
4100	kfree(uqp);
4101
4102	return 0;
4103
4104out:
4105	mutex_unlock(&root->chain_lock);
4106	return err;
4107}
4108EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
4109
4110struct mlx5_flow_root_namespace *
4111mlx5_get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
4112{
4113	struct mlx5_flow_namespace *ns;
4114
4115	if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
4116	    ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS ||
4117	    ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_TX ||
4118	    ns_type == MLX5_FLOW_NAMESPACE_RDMA_TRANSPORT_RX)
4119		ns = mlx5_get_flow_vport_namespace(dev, ns_type, 0);
4120	else
4121		ns = mlx5_get_flow_namespace(dev, ns_type);
4122	if (!ns)
4123		return NULL;
4124
4125	return find_root(&ns->node);
4126}
4127
4128struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
4129						 u8 ns_type, u8 num_actions,
4130						 void *modify_actions)
4131{
4132	struct mlx5_flow_root_namespace *root;
4133	struct mlx5_modify_hdr *modify_hdr;
4134	int err;
4135
4136	root = mlx5_get_root_namespace(dev, ns_type);
4137	if (!root)
4138		return ERR_PTR(-EOPNOTSUPP);
4139
4140	modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
4141	if (!modify_hdr)
4142		return ERR_PTR(-ENOMEM);
4143
4144	modify_hdr->ns_type = ns_type;
4145	err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
4146					      modify_actions, modify_hdr);
4147	if (err) {
4148		kfree(modify_hdr);
4149		return ERR_PTR(err);
4150	}
4151
4152	return modify_hdr;
4153}
4154EXPORT_SYMBOL(mlx5_modify_header_alloc);
4155
4156void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
4157				struct mlx5_modify_hdr *modify_hdr)
4158{
4159	struct mlx5_flow_root_namespace *root;
4160
4161	root = mlx5_get_root_namespace(dev, modify_hdr->ns_type);
4162	if (WARN_ON(!root))
4163		return;
4164	root->cmds->modify_header_dealloc(root, modify_hdr);
4165	kfree(modify_hdr);
4166}
4167EXPORT_SYMBOL(mlx5_modify_header_dealloc);
4168
4169struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
4170						     struct mlx5_pkt_reformat_params *params,
4171						     enum mlx5_flow_namespace_type ns_type)
4172{
4173	struct mlx5_pkt_reformat *pkt_reformat;
4174	struct mlx5_flow_root_namespace *root;
4175	int err;
4176
4177	root = mlx5_get_root_namespace(dev, ns_type);
4178	if (!root)
4179		return ERR_PTR(-EOPNOTSUPP);
4180
4181	pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
4182	if (!pkt_reformat)
4183		return ERR_PTR(-ENOMEM);
4184
4185	pkt_reformat->ns_type = ns_type;
4186	pkt_reformat->reformat_type = params->type;
4187	err = root->cmds->packet_reformat_alloc(root, params, ns_type,
4188						pkt_reformat);
4189	if (err) {
4190		kfree(pkt_reformat);
4191		return ERR_PTR(err);
4192	}
4193
4194	return pkt_reformat;
4195}
4196EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
4197
4198void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
4199				  struct mlx5_pkt_reformat *pkt_reformat)
4200{
4201	struct mlx5_flow_root_namespace *root;
4202
4203	root = mlx5_get_root_namespace(dev, pkt_reformat->ns_type);
4204	if (WARN_ON(!root))
4205		return;
4206	root->cmds->packet_reformat_dealloc(root, pkt_reformat);
4207	kfree(pkt_reformat);
4208}
4209EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
4210
4211int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer)
4212{
4213	return definer->id;
4214}
4215
4216struct mlx5_flow_definer *
4217mlx5_create_match_definer(struct mlx5_core_dev *dev,
4218			  enum mlx5_flow_namespace_type ns_type, u16 format_id,
4219			  u32 *match_mask)
4220{
4221	struct mlx5_flow_root_namespace *root;
4222	struct mlx5_flow_definer *definer;
4223	int id;
4224
4225	root = mlx5_get_root_namespace(dev, ns_type);
4226	if (!root)
4227		return ERR_PTR(-EOPNOTSUPP);
4228
4229	definer = kzalloc(sizeof(*definer), GFP_KERNEL);
4230	if (!definer)
4231		return ERR_PTR(-ENOMEM);
4232
4233	definer->ns_type = ns_type;
4234	id = root->cmds->create_match_definer(root, format_id, match_mask);
4235	if (id < 0) {
4236		mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id);
4237		kfree(definer);
4238		return ERR_PTR(id);
4239	}
4240	definer->id = id;
4241	return definer;
4242}
4243
4244void mlx5_destroy_match_definer(struct mlx5_core_dev *dev,
4245				struct mlx5_flow_definer *definer)
4246{
4247	struct mlx5_flow_root_namespace *root;
4248
4249	root = mlx5_get_root_namespace(dev, definer->ns_type);
4250	if (WARN_ON(!root))
4251		return;
4252
4253	root->cmds->destroy_match_definer(root, definer->id);
4254	kfree(definer);
4255}
4256
4257int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
4258				 struct mlx5_flow_root_namespace *peer_ns,
4259				 u16 peer_vhca_id)
4260{
4261	if (peer_ns && ns->mode != peer_ns->mode) {
4262		mlx5_core_err(ns->dev,
4263			      "Can't peer namespace of different steering mode\n");
4264		return -EINVAL;
4265	}
4266
4267	return ns->cmds->set_peer(ns, peer_ns, peer_vhca_id);
4268}
4269
4270/* This function should be called only at init stage of the namespace.
4271 * It is not safe to call this function while steering operations
4272 * are executed in the namespace.
4273 */
4274int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
4275				 enum mlx5_flow_steering_mode mode)
4276{
4277	struct mlx5_flow_root_namespace *root;
4278	const struct mlx5_flow_cmds *cmds;
4279	int err;
4280
4281	root = find_root(&ns->node);
4282	if (&root->ns != ns)
4283	/* Can't set cmds to non root namespace */
4284		return -EINVAL;
4285
4286	if (root->table_type != FS_FT_FDB)
4287		return -EOPNOTSUPP;
4288
4289	if (root->mode == mode)
4290		return 0;
4291
4292	if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
4293		cmds = mlx5_fs_cmd_get_dr_cmds();
4294	else if (mode == MLX5_FLOW_STEERING_MODE_HMFS)
4295		cmds = mlx5_fs_cmd_get_hws_cmds();
4296	else
4297		cmds = mlx5_fs_cmd_get_fw_cmds();
4298	if (!cmds)
4299		return -EOPNOTSUPP;
4300
4301	err = cmds->create_ns(root);
4302	if (err) {
4303		mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
4304			      err);
4305		return err;
4306	}
4307
4308	root->cmds->destroy_ns(root);
4309	root->cmds = cmds;
4310	root->mode = mode;
4311
4312	return 0;
4313}