2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
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>
39 #include "mlx5_core.h"
42 #include "fs_ft_pool.h"
43 #include "diag/fs_tracepoint.h"
46 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
47 sizeof(struct init_tree_node))
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,\
55 .children = (struct init_tree_node[]) {__VA_ARGS__},\
56 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
59 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
60 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
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__) \
69 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
72 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
74 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
75 .caps = (long[]) {__VA_ARGS__} }
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))
82 #define FS_CHAINING_CAPS_EGRESS \
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))
90 #define FS_CHAINING_CAPS_RDMA_TX \
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 \
99 #define LEFTOVERS_NUM_LEVELS 1
100 #define LEFTOVERS_NUM_PRIOS 1
102 #define RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1
103 #define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1
105 #define BY_PASS_PRIO_NUM_LEVELS 1
106 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
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)
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 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}}, IPsec policy,
117 * {IPsec RoCE MPV,Alias table},IPsec RoCE policy
119 #define KERNEL_NIC_PRIO_NUM_LEVELS 11
120 #define KERNEL_NIC_NUM_PRIOS 1
121 /* One more level for tc */
122 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
124 #define KERNEL_NIC_TC_NUM_PRIOS 1
125 #define KERNEL_NIC_TC_NUM_LEVELS 3
127 #define ANCHOR_NUM_LEVELS 1
128 #define ANCHOR_NUM_PRIOS 1
129 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
131 #define OFFLOADS_MAX_FT 2
132 #define OFFLOADS_NUM_PRIOS 2
133 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
135 #define LAG_PRIO_NUM_LEVELS 1
136 #define LAG_NUM_PRIOS 1
137 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + KERNEL_RX_MACSEC_MIN_LEVEL + 1)
139 #define KERNEL_TX_IPSEC_NUM_PRIOS 1
140 #define KERNEL_TX_IPSEC_NUM_LEVELS 4
141 #define KERNEL_TX_IPSEC_MIN_LEVEL (KERNEL_TX_IPSEC_NUM_LEVELS)
143 #define KERNEL_TX_MACSEC_NUM_PRIOS 1
144 #define KERNEL_TX_MACSEC_NUM_LEVELS 2
145 #define KERNEL_TX_MACSEC_MIN_LEVEL (KERNEL_TX_IPSEC_MIN_LEVEL + KERNEL_TX_MACSEC_NUM_PRIOS)
152 static struct init_tree_node {
153 enum fs_node_type type;
154 struct init_tree_node *children;
156 struct node_caps caps;
161 enum mlx5_flow_table_miss_action def_miss_action;
163 .type = FS_TYPE_NAMESPACE,
165 .children = (struct init_tree_node[]){
166 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
167 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
168 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
169 BY_PASS_PRIO_NUM_LEVELS))),
170 ADD_PRIO(0, KERNEL_RX_MACSEC_MIN_LEVEL, 0, FS_CHAINING_CAPS,
171 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
172 ADD_MULTIPLE_PRIO(KERNEL_RX_MACSEC_NUM_PRIOS,
173 KERNEL_RX_MACSEC_NUM_LEVELS))),
174 ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
175 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
176 ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
177 LAG_PRIO_NUM_LEVELS))),
178 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
179 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
180 ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
182 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
183 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
184 ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
185 ETHTOOL_PRIO_NUM_LEVELS))),
186 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
187 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
188 ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
189 KERNEL_NIC_TC_NUM_LEVELS),
190 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
191 KERNEL_NIC_PRIO_NUM_LEVELS))),
192 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
193 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
194 ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
195 LEFTOVERS_NUM_LEVELS))),
196 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
197 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
198 ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
199 ANCHOR_NUM_LEVELS))),
203 static struct init_tree_node egress_root_fs = {
204 .type = FS_TYPE_NAMESPACE,
206 .children = (struct init_tree_node[]) {
207 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
208 FS_CHAINING_CAPS_EGRESS,
209 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
210 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
211 BY_PASS_PRIO_NUM_LEVELS))),
212 ADD_PRIO(0, KERNEL_TX_IPSEC_MIN_LEVEL, 0,
213 FS_CHAINING_CAPS_EGRESS,
214 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
215 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
216 KERNEL_TX_IPSEC_NUM_LEVELS))),
217 ADD_PRIO(0, KERNEL_TX_MACSEC_MIN_LEVEL, 0,
218 FS_CHAINING_CAPS_EGRESS,
219 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
220 ADD_MULTIPLE_PRIO(KERNEL_TX_MACSEC_NUM_PRIOS,
221 KERNEL_TX_MACSEC_NUM_LEVELS))),
228 RDMA_RX_COUNTERS_PRIO,
233 #define RDMA_RX_IPSEC_NUM_PRIOS 1
234 #define RDMA_RX_IPSEC_NUM_LEVELS 4
235 #define RDMA_RX_IPSEC_MIN_LEVEL (RDMA_RX_IPSEC_NUM_LEVELS)
237 #define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS
238 #define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1)
239 #define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2)
241 #define RDMA_RX_MACSEC_NUM_PRIOS 1
242 #define RDMA_RX_MACSEC_PRIO_NUM_LEVELS 2
243 #define RDMA_RX_MACSEC_MIN_LEVEL (RDMA_RX_COUNTERS_MIN_LEVEL + RDMA_RX_MACSEC_NUM_PRIOS)
245 static struct init_tree_node rdma_rx_root_fs = {
246 .type = FS_TYPE_NAMESPACE,
248 .children = (struct init_tree_node[]) {
249 [RDMA_RX_IPSEC_PRIO] =
250 ADD_PRIO(0, RDMA_RX_IPSEC_MIN_LEVEL, 0,
252 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
253 ADD_MULTIPLE_PRIO(RDMA_RX_IPSEC_NUM_PRIOS,
254 RDMA_RX_IPSEC_NUM_LEVELS))),
255 [RDMA_RX_MACSEC_PRIO] =
256 ADD_PRIO(0, RDMA_RX_MACSEC_MIN_LEVEL, 0,
258 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
259 ADD_MULTIPLE_PRIO(RDMA_RX_MACSEC_NUM_PRIOS,
260 RDMA_RX_MACSEC_PRIO_NUM_LEVELS))),
261 [RDMA_RX_COUNTERS_PRIO] =
262 ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0,
264 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
265 ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS,
266 RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))),
267 [RDMA_RX_BYPASS_PRIO] =
268 ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0,
270 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
271 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
272 BY_PASS_PRIO_NUM_LEVELS))),
273 [RDMA_RX_KERNEL_PRIO] =
274 ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0,
276 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
277 ADD_MULTIPLE_PRIO(1, 1))),
282 RDMA_TX_COUNTERS_PRIO,
288 #define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS
289 #define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1)
291 #define RDMA_TX_IPSEC_NUM_PRIOS 2
292 #define RDMA_TX_IPSEC_PRIO_NUM_LEVELS 1
293 #define RDMA_TX_IPSEC_MIN_LEVEL (RDMA_TX_COUNTERS_MIN_LEVEL + RDMA_TX_IPSEC_NUM_PRIOS)
295 #define RDMA_TX_MACSEC_NUM_PRIOS 1
296 #define RDMA_TX_MACESC_PRIO_NUM_LEVELS 1
297 #define RDMA_TX_MACSEC_MIN_LEVEL (RDMA_TX_COUNTERS_MIN_LEVEL + RDMA_TX_MACSEC_NUM_PRIOS)
299 static struct init_tree_node rdma_tx_root_fs = {
300 .type = FS_TYPE_NAMESPACE,
302 .children = (struct init_tree_node[]) {
303 [RDMA_TX_COUNTERS_PRIO] =
304 ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0,
306 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
307 ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS,
308 RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))),
309 [RDMA_TX_IPSEC_PRIO] =
310 ADD_PRIO(0, RDMA_TX_IPSEC_MIN_LEVEL, 0,
312 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
313 ADD_MULTIPLE_PRIO(RDMA_TX_IPSEC_NUM_PRIOS,
314 RDMA_TX_IPSEC_PRIO_NUM_LEVELS))),
315 [RDMA_TX_MACSEC_PRIO] =
316 ADD_PRIO(0, RDMA_TX_MACSEC_MIN_LEVEL, 0,
318 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
319 ADD_MULTIPLE_PRIO(RDMA_TX_MACSEC_NUM_PRIOS,
320 RDMA_TX_MACESC_PRIO_NUM_LEVELS))),
321 [RDMA_TX_BYPASS_PRIO] =
322 ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0,
323 FS_CHAINING_CAPS_RDMA_TX,
324 ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
325 ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL,
326 BY_PASS_PRIO_NUM_LEVELS))),
330 enum fs_i_lock_class {
336 static const struct rhashtable_params rhash_fte = {
337 .key_len = sizeof_field(struct fs_fte, val),
338 .key_offset = offsetof(struct fs_fte, val),
339 .head_offset = offsetof(struct fs_fte, hash),
340 .automatic_shrinking = true,
344 static const struct rhashtable_params rhash_fg = {
345 .key_len = sizeof_field(struct mlx5_flow_group, mask),
346 .key_offset = offsetof(struct mlx5_flow_group, mask),
347 .head_offset = offsetof(struct mlx5_flow_group, hash),
348 .automatic_shrinking = true,
353 static void del_hw_flow_table(struct fs_node *node);
354 static void del_hw_flow_group(struct fs_node *node);
355 static void del_hw_fte(struct fs_node *node);
356 static void del_sw_flow_table(struct fs_node *node);
357 static void del_sw_flow_group(struct fs_node *node);
358 static void del_sw_fte(struct fs_node *node);
359 static void del_sw_prio(struct fs_node *node);
360 static void del_sw_ns(struct fs_node *node);
361 /* Delete rule (destination) is special case that
362 * requires to lock the FTE for all the deletion process.
364 static void del_sw_hw_rule(struct fs_node *node);
365 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
366 struct mlx5_flow_destination *d2);
367 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
368 static struct mlx5_flow_rule *
369 find_flow_rule(struct fs_fte *fte,
370 struct mlx5_flow_destination *dest);
372 static void tree_init_node(struct fs_node *node,
373 void (*del_hw_func)(struct fs_node *),
374 void (*del_sw_func)(struct fs_node *))
376 refcount_set(&node->refcount, 1);
377 INIT_LIST_HEAD(&node->list);
378 INIT_LIST_HEAD(&node->children);
379 init_rwsem(&node->lock);
380 node->del_hw_func = del_hw_func;
381 node->del_sw_func = del_sw_func;
382 node->active = false;
385 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
388 refcount_inc(&parent->refcount);
389 node->parent = parent;
391 /* Parent is the root */
395 node->root = parent->root;
398 static int tree_get_node(struct fs_node *node)
400 return refcount_inc_not_zero(&node->refcount);
403 static void nested_down_read_ref_node(struct fs_node *node,
404 enum fs_i_lock_class class)
407 down_read_nested(&node->lock, class);
408 refcount_inc(&node->refcount);
412 static void nested_down_write_ref_node(struct fs_node *node,
413 enum fs_i_lock_class class)
416 down_write_nested(&node->lock, class);
417 refcount_inc(&node->refcount);
421 static void down_write_ref_node(struct fs_node *node, bool locked)
425 down_write(&node->lock);
426 refcount_inc(&node->refcount);
430 static void up_read_ref_node(struct fs_node *node)
432 refcount_dec(&node->refcount);
433 up_read(&node->lock);
436 static void up_write_ref_node(struct fs_node *node, bool locked)
438 refcount_dec(&node->refcount);
440 up_write(&node->lock);
443 static void tree_put_node(struct fs_node *node, bool locked)
445 struct fs_node *parent_node = node->parent;
447 if (refcount_dec_and_test(&node->refcount)) {
448 if (node->del_hw_func)
449 node->del_hw_func(node);
451 down_write_ref_node(parent_node, locked);
452 list_del_init(&node->list);
454 node->del_sw_func(node);
456 up_write_ref_node(parent_node, locked);
459 if (!node && parent_node)
460 tree_put_node(parent_node, locked);
463 static int tree_remove_node(struct fs_node *node, bool locked)
465 if (refcount_read(&node->refcount) > 1) {
466 refcount_dec(&node->refcount);
469 tree_put_node(node, locked);
473 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
476 struct fs_prio *iter_prio;
478 fs_for_each_prio(iter_prio, ns) {
479 if (iter_prio->prio == prio)
486 static bool is_fwd_next_action(u32 action)
488 return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
489 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
492 static bool is_fwd_dest_type(enum mlx5_flow_destination_type type)
494 return type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM ||
495 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE ||
496 type == MLX5_FLOW_DESTINATION_TYPE_UPLINK ||
497 type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
498 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER ||
499 type == MLX5_FLOW_DESTINATION_TYPE_TIR ||
500 type == MLX5_FLOW_DESTINATION_TYPE_RANGE ||
501 type == MLX5_FLOW_DESTINATION_TYPE_TABLE_TYPE;
504 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
508 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
509 if (spec->match_value[i] & ~spec->match_criteria[i]) {
510 pr_warn("mlx5_core: match_value differs from match_criteria\n");
517 struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
519 struct fs_node *root;
520 struct mlx5_flow_namespace *ns;
524 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
525 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
529 ns = container_of(root, struct mlx5_flow_namespace, node);
530 return container_of(ns, struct mlx5_flow_root_namespace, ns);
533 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
535 struct mlx5_flow_root_namespace *root = find_root(node);
538 return root->dev->priv.steering;
542 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
544 struct mlx5_flow_root_namespace *root = find_root(node);
551 static void del_sw_ns(struct fs_node *node)
556 static void del_sw_prio(struct fs_node *node)
561 static void del_hw_flow_table(struct fs_node *node)
563 struct mlx5_flow_root_namespace *root;
564 struct mlx5_flow_table *ft;
565 struct mlx5_core_dev *dev;
568 fs_get_obj(ft, node);
569 dev = get_dev(&ft->node);
570 root = find_root(&ft->node);
571 trace_mlx5_fs_del_ft(ft);
574 err = root->cmds->destroy_flow_table(root, ft);
576 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
580 static void del_sw_flow_table(struct fs_node *node)
582 struct mlx5_flow_table *ft;
583 struct fs_prio *prio;
585 fs_get_obj(ft, node);
587 rhltable_destroy(&ft->fgs_hash);
588 if (ft->node.parent) {
589 fs_get_obj(prio, ft->node.parent);
595 static void modify_fte(struct fs_fte *fte)
597 struct mlx5_flow_root_namespace *root;
598 struct mlx5_flow_table *ft;
599 struct mlx5_flow_group *fg;
600 struct mlx5_core_dev *dev;
603 fs_get_obj(fg, fte->node.parent);
604 fs_get_obj(ft, fg->node.parent);
605 dev = get_dev(&fte->node);
607 root = find_root(&ft->node);
608 err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
611 "%s can't del rule fg id=%d fte_index=%d\n",
612 __func__, fg->id, fte->index);
613 fte->modify_mask = 0;
616 static void del_sw_hw_rule(struct fs_node *node)
618 struct mlx5_flow_rule *rule;
621 fs_get_obj(rule, node);
622 fs_get_obj(fte, rule->node.parent);
623 trace_mlx5_fs_del_rule(rule);
624 if (is_fwd_next_action(rule->sw_action)) {
625 mutex_lock(&rule->dest_attr.ft->lock);
626 list_del(&rule->next_ft);
627 mutex_unlock(&rule->dest_attr.ft->lock);
630 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER) {
633 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
634 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
635 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
639 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT) {
641 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
642 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
646 if (is_fwd_dest_type(rule->dest_attr.type)) {
651 fte->action.action &=
652 ~MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
654 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
661 static void del_hw_fte(struct fs_node *node)
663 struct mlx5_flow_root_namespace *root;
664 struct mlx5_flow_table *ft;
665 struct mlx5_flow_group *fg;
666 struct mlx5_core_dev *dev;
670 fs_get_obj(fte, node);
671 fs_get_obj(fg, fte->node.parent);
672 fs_get_obj(ft, fg->node.parent);
674 trace_mlx5_fs_del_fte(fte);
675 WARN_ON(fte->dests_size);
676 dev = get_dev(&ft->node);
677 root = find_root(&ft->node);
679 err = root->cmds->delete_fte(root, ft, fte);
682 "flow steering can't delete fte in index %d of flow group id %d\n",
684 node->active = false;
688 static void del_sw_fte(struct fs_node *node)
690 struct mlx5_flow_steering *steering = get_steering(node);
691 struct mlx5_flow_group *fg;
695 fs_get_obj(fte, node);
696 fs_get_obj(fg, fte->node.parent);
698 err = rhashtable_remove_fast(&fg->ftes_hash,
702 ida_free(&fg->fte_allocator, fte->index - fg->start_index);
703 kmem_cache_free(steering->ftes_cache, fte);
706 static void del_hw_flow_group(struct fs_node *node)
708 struct mlx5_flow_root_namespace *root;
709 struct mlx5_flow_group *fg;
710 struct mlx5_flow_table *ft;
711 struct mlx5_core_dev *dev;
713 fs_get_obj(fg, node);
714 fs_get_obj(ft, fg->node.parent);
715 dev = get_dev(&ft->node);
716 trace_mlx5_fs_del_fg(fg);
718 root = find_root(&ft->node);
719 if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
720 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
724 static void del_sw_flow_group(struct fs_node *node)
726 struct mlx5_flow_steering *steering = get_steering(node);
727 struct mlx5_flow_group *fg;
728 struct mlx5_flow_table *ft;
731 fs_get_obj(fg, node);
732 fs_get_obj(ft, fg->node.parent);
734 rhashtable_destroy(&fg->ftes_hash);
735 ida_destroy(&fg->fte_allocator);
736 if (ft->autogroup.active &&
737 fg->max_ftes == ft->autogroup.group_size &&
738 fg->start_index < ft->autogroup.max_fte)
739 ft->autogroup.num_groups--;
740 err = rhltable_remove(&ft->fgs_hash,
744 kmem_cache_free(steering->fgs_cache, fg);
747 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
752 index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL);
756 fte->index = index + fg->start_index;
757 ret = rhashtable_insert_fast(&fg->ftes_hash,
763 tree_add_node(&fte->node, &fg->node);
764 list_add_tail(&fte->node.list, &fg->node.children);
768 ida_free(&fg->fte_allocator, index);
772 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
773 const struct mlx5_flow_spec *spec,
774 struct mlx5_flow_act *flow_act)
776 struct mlx5_flow_steering *steering = get_steering(&ft->node);
779 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
781 return ERR_PTR(-ENOMEM);
783 memcpy(fte->val, &spec->match_value, sizeof(fte->val));
784 fte->node.type = FS_TYPE_FLOW_ENTRY;
785 fte->action = *flow_act;
786 fte->flow_context = spec->flow_context;
788 tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
793 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
794 struct mlx5_flow_group *fg)
796 rhashtable_destroy(&fg->ftes_hash);
797 kmem_cache_free(steering->fgs_cache, fg);
800 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
801 u8 match_criteria_enable,
802 const void *match_criteria,
806 struct mlx5_flow_group *fg;
809 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
811 return ERR_PTR(-ENOMEM);
813 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
815 kmem_cache_free(steering->fgs_cache, fg);
819 ida_init(&fg->fte_allocator);
820 fg->mask.match_criteria_enable = match_criteria_enable;
821 memcpy(&fg->mask.match_criteria, match_criteria,
822 sizeof(fg->mask.match_criteria));
823 fg->node.type = FS_TYPE_FLOW_GROUP;
824 fg->start_index = start_index;
825 fg->max_ftes = end_index - start_index + 1;
830 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
831 u8 match_criteria_enable,
832 const void *match_criteria,
835 struct list_head *prev)
837 struct mlx5_flow_steering *steering = get_steering(&ft->node);
838 struct mlx5_flow_group *fg;
841 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
842 start_index, end_index);
846 /* initialize refcnt, add to parent list */
847 ret = rhltable_insert(&ft->fgs_hash,
851 dealloc_flow_group(steering, fg);
855 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
856 tree_add_node(&fg->node, &ft->node);
857 /* Add node to group list */
858 list_add(&fg->node.list, prev);
859 atomic_inc(&ft->node.version);
864 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport,
865 enum fs_flow_table_type table_type,
866 enum fs_flow_table_op_mod op_mod,
869 struct mlx5_flow_table *ft;
872 ft = kzalloc(sizeof(*ft), GFP_KERNEL);
874 return ERR_PTR(-ENOMEM);
876 ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
883 ft->node.type = FS_TYPE_FLOW_TABLE;
885 ft->type = table_type;
888 INIT_LIST_HEAD(&ft->fwd_rules);
889 mutex_init(&ft->lock);
894 /* If reverse is false, then we search for the first flow table in the
895 * root sub-tree from start(closest from right), else we search for the
896 * last flow table in the root sub-tree till start(closest from left).
898 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root,
899 struct list_head *start,
902 #define list_advance_entry(pos, reverse) \
903 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
905 #define list_for_each_advance_continue(pos, head, reverse) \
906 for (pos = list_advance_entry(pos, reverse); \
907 &pos->list != (head); \
908 pos = list_advance_entry(pos, reverse))
910 struct fs_node *iter = list_entry(start, struct fs_node, list);
911 struct mlx5_flow_table *ft = NULL;
916 list_for_each_advance_continue(iter, &root->children, reverse) {
917 if (iter->type == FS_TYPE_FLOW_TABLE) {
918 fs_get_obj(ft, iter);
921 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
929 static struct fs_node *find_prio_chains_parent(struct fs_node *parent,
930 struct fs_node **child)
932 struct fs_node *node = NULL;
934 while (parent && parent->type != FS_TYPE_PRIO_CHAINS) {
936 parent = parent->parent;
945 /* If reverse is false then return the first flow table next to the passed node
946 * in the tree, else return the last flow table before the node in the tree.
947 * If skip is true, skip the flow tables in the same prio_chains prio.
949 static struct mlx5_flow_table *find_closest_ft(struct fs_node *node, bool reverse,
952 struct fs_node *prio_chains_parent = NULL;
953 struct mlx5_flow_table *ft = NULL;
954 struct fs_node *curr_node;
955 struct fs_node *parent;
958 prio_chains_parent = find_prio_chains_parent(node, NULL);
959 parent = node->parent;
961 while (!ft && parent) {
962 if (parent != prio_chains_parent)
963 ft = find_closest_ft_recursive(parent, &curr_node->list,
966 parent = curr_node->parent;
971 /* Assuming all the tree is locked by mutex chain lock */
972 static struct mlx5_flow_table *find_next_chained_ft(struct fs_node *node)
974 return find_closest_ft(node, false, true);
977 /* Assuming all the tree is locked by mutex chain lock */
978 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_node *node)
980 return find_closest_ft(node, true, true);
983 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
984 struct mlx5_flow_act *flow_act)
986 struct fs_prio *prio;
989 next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
990 fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
992 return find_next_chained_ft(&prio->node);
995 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
996 struct fs_prio *prio,
997 struct mlx5_flow_table *ft)
999 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
1000 struct mlx5_flow_table *iter;
1003 fs_for_each_ft(iter, prio) {
1004 err = root->cmds->modify_flow_table(root, iter, ft);
1007 "Failed to modify flow table id %d, type %d, err %d\n",
1008 iter->id, iter->type, err);
1009 /* The driver is out of sync with the FW */
1016 static struct mlx5_flow_table *find_closet_ft_prio_chains(struct fs_node *node,
1017 struct fs_node *parent,
1018 struct fs_node **child,
1021 struct mlx5_flow_table *ft;
1023 ft = find_closest_ft(node, reverse, false);
1025 if (ft && parent == find_prio_chains_parent(&ft->node, child))
1031 /* Connect flow tables from previous priority of prio to ft */
1032 static int connect_prev_fts(struct mlx5_core_dev *dev,
1033 struct mlx5_flow_table *ft,
1034 struct fs_prio *prio)
1036 struct fs_node *prio_parent, *parent = NULL, *child, *node;
1037 struct mlx5_flow_table *prev_ft;
1040 prio_parent = find_prio_chains_parent(&prio->node, &child);
1042 /* return directly if not under the first sub ns of prio_chains prio */
1043 if (prio_parent && !list_is_first(&child->list, &prio_parent->children))
1046 prev_ft = find_prev_chained_ft(&prio->node);
1048 struct fs_prio *prev_prio;
1050 fs_get_obj(prev_prio, prev_ft->node.parent);
1051 err = connect_fts_in_prio(dev, prev_prio, ft);
1056 parent = find_prio_chains_parent(&prev_prio->node, &child);
1062 prev_ft = find_closet_ft_prio_chains(node, parent, &child, true);
1067 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
1070 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
1071 struct mlx5_ft_underlay_qp *uqp;
1072 int min_level = INT_MAX;
1077 min_level = root->root_ft->level;
1079 if (ft->level >= min_level)
1082 if (list_empty(&root->underlay_qpns)) {
1083 /* Don't set any QPN (zero) in case QPN list is empty */
1085 err = root->cmds->update_root_ft(root, ft, qpn, false);
1087 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1089 err = root->cmds->update_root_ft(root, ft,
1097 mlx5_core_warn(root->dev,
1098 "Update root flow table of id(%u) qpn(%d) failed\n",
1106 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
1107 struct mlx5_flow_destination *dest)
1109 struct mlx5_flow_root_namespace *root;
1110 struct mlx5_flow_table *ft;
1111 struct mlx5_flow_group *fg;
1113 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1116 fs_get_obj(fte, rule->node.parent);
1117 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1119 down_write_ref_node(&fte->node, false);
1120 fs_get_obj(fg, fte->node.parent);
1121 fs_get_obj(ft, fg->node.parent);
1123 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1124 root = find_root(&ft->node);
1125 err = root->cmds->update_fte(root, ft, fg,
1127 up_write_ref_node(&fte->node, false);
1132 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
1133 struct mlx5_flow_destination *new_dest,
1134 struct mlx5_flow_destination *old_dest)
1139 if (handle->num_rules != 1)
1141 return _mlx5_modify_rule_destination(handle->rule[0],
1145 for (i = 0; i < handle->num_rules; i++) {
1146 if (mlx5_flow_dests_cmp(old_dest, &handle->rule[i]->dest_attr))
1147 return _mlx5_modify_rule_destination(handle->rule[i],
1154 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
1155 static int connect_fwd_rules(struct mlx5_core_dev *dev,
1156 struct mlx5_flow_table *new_next_ft,
1157 struct mlx5_flow_table *old_next_ft)
1159 struct mlx5_flow_destination dest = {};
1160 struct mlx5_flow_rule *iter;
1163 /* new_next_ft and old_next_ft could be NULL only
1164 * when we create/destroy the anchor flow table.
1166 if (!new_next_ft || !old_next_ft)
1169 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1170 dest.ft = new_next_ft;
1172 mutex_lock(&old_next_ft->lock);
1173 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1174 mutex_unlock(&old_next_ft->lock);
1175 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1176 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1177 iter->ft->ns == new_next_ft->ns)
1180 err = _mlx5_modify_rule_destination(iter, &dest);
1182 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1188 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1189 struct fs_prio *prio)
1191 struct mlx5_flow_table *next_ft, *first_ft;
1194 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1196 first_ft = list_first_entry_or_null(&prio->node.children,
1197 struct mlx5_flow_table, node.list);
1198 if (!first_ft || first_ft->level > ft->level) {
1199 err = connect_prev_fts(dev, ft, prio);
1203 next_ft = first_ft ? first_ft : find_next_chained_ft(&prio->node);
1204 err = connect_fwd_rules(dev, ft, next_ft);
1209 if (MLX5_CAP_FLOWTABLE(dev,
1210 flow_table_properties_nic_receive.modify_root))
1211 err = update_root_ft_create(ft, prio);
1215 static void list_add_flow_table(struct mlx5_flow_table *ft,
1216 struct fs_prio *prio)
1218 struct list_head *prev = &prio->node.children;
1219 struct mlx5_flow_table *iter;
1221 fs_for_each_ft(iter, prio) {
1222 if (iter->level > ft->level)
1224 prev = &iter->node.list;
1226 list_add(&ft->node.list, prev);
1229 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1230 struct mlx5_flow_table_attr *ft_attr,
1231 enum fs_flow_table_op_mod op_mod,
1234 struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1235 bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1236 struct mlx5_flow_table *next_ft;
1237 struct fs_prio *fs_prio = NULL;
1238 struct mlx5_flow_table *ft;
1242 pr_err("mlx5: flow steering failed to find root of namespace\n");
1243 return ERR_PTR(-ENODEV);
1246 mutex_lock(&root->chain_lock);
1247 fs_prio = find_prio(ns, ft_attr->prio);
1253 /* The level is related to the
1254 * priority level range.
1256 if (ft_attr->level >= fs_prio->num_levels) {
1261 ft_attr->level += fs_prio->start_level;
1264 /* The level is related to the
1265 * priority level range.
1267 ft = alloc_flow_table(ft_attr->level,
1270 op_mod, ft_attr->flags);
1276 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1277 next_ft = unmanaged ? ft_attr->next_ft :
1278 find_next_chained_ft(&fs_prio->node);
1279 ft->def_miss_action = ns->def_miss_action;
1281 err = root->cmds->create_flow_table(root, ft, ft_attr, next_ft);
1286 err = connect_flow_table(root->dev, ft, fs_prio);
1291 ft->node.active = true;
1292 down_write_ref_node(&fs_prio->node, false);
1294 tree_add_node(&ft->node, &fs_prio->node);
1295 list_add_flow_table(ft, fs_prio);
1297 ft->node.root = fs_prio->node.root;
1300 up_write_ref_node(&fs_prio->node, false);
1301 mutex_unlock(&root->chain_lock);
1302 trace_mlx5_fs_add_ft(ft);
1305 root->cmds->destroy_flow_table(root, ft);
1307 rhltable_destroy(&ft->fgs_hash);
1310 mutex_unlock(&root->chain_lock);
1311 return ERR_PTR(err);
1314 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1315 struct mlx5_flow_table_attr *ft_attr)
1317 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1319 EXPORT_SYMBOL(mlx5_create_flow_table);
1321 u32 mlx5_flow_table_id(struct mlx5_flow_table *ft)
1325 EXPORT_SYMBOL(mlx5_flow_table_id);
1327 struct mlx5_flow_table *
1328 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1329 struct mlx5_flow_table_attr *ft_attr, u16 vport)
1331 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1334 struct mlx5_flow_table*
1335 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1336 int prio, u32 level)
1338 struct mlx5_flow_table_attr ft_attr = {};
1340 ft_attr.level = level;
1341 ft_attr.prio = prio;
1342 ft_attr.max_fte = 1;
1344 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1346 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1348 #define MAX_FLOW_GROUP_SIZE BIT(24)
1349 struct mlx5_flow_table*
1350 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1351 struct mlx5_flow_table_attr *ft_attr)
1353 int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1354 int max_num_groups = ft_attr->autogroup.max_num_groups;
1355 struct mlx5_flow_table *ft;
1356 int autogroups_max_fte;
1358 ft = mlx5_create_flow_table(ns, ft_attr);
1362 autogroups_max_fte = ft->max_fte - num_reserved_entries;
1363 if (max_num_groups > autogroups_max_fte)
1365 if (num_reserved_entries > ft->max_fte)
1368 /* Align the number of groups according to the largest group size */
1369 if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1370 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1372 ft->autogroup.active = true;
1373 ft->autogroup.required_groups = max_num_groups;
1374 ft->autogroup.max_fte = autogroups_max_fte;
1375 /* We save place for flow groups in addition to max types */
1376 ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1381 mlx5_destroy_flow_table(ft);
1382 return ERR_PTR(-ENOSPC);
1384 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1386 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1389 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1390 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1391 fg_in, match_criteria);
1392 u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1394 match_criteria_enable);
1395 int start_index = MLX5_GET(create_flow_group_in, fg_in,
1397 int end_index = MLX5_GET(create_flow_group_in, fg_in,
1399 struct mlx5_flow_group *fg;
1402 if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1403 return ERR_PTR(-EPERM);
1405 down_write_ref_node(&ft->node, false);
1406 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1407 start_index, end_index,
1408 ft->node.children.prev);
1409 up_write_ref_node(&ft->node, false);
1413 err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1415 tree_put_node(&fg->node, false);
1416 return ERR_PTR(err);
1418 trace_mlx5_fs_add_fg(fg);
1419 fg->node.active = true;
1423 EXPORT_SYMBOL(mlx5_create_flow_group);
1425 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1427 struct mlx5_flow_rule *rule;
1429 rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1433 INIT_LIST_HEAD(&rule->next_ft);
1434 rule->node.type = FS_TYPE_FLOW_DEST;
1436 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1438 rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE;
1443 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1445 struct mlx5_flow_handle *handle;
1447 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1451 handle->num_rules = num_rules;
1456 static void destroy_flow_handle(struct fs_fte *fte,
1457 struct mlx5_flow_handle *handle,
1458 struct mlx5_flow_destination *dest,
1462 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1464 list_del(&handle->rule[i]->node.list);
1465 kfree(handle->rule[i]);
1471 static struct mlx5_flow_handle *
1472 create_flow_handle(struct fs_fte *fte,
1473 struct mlx5_flow_destination *dest,
1478 struct mlx5_flow_handle *handle;
1479 struct mlx5_flow_rule *rule = NULL;
1480 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1481 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1485 handle = alloc_handle((dest_num) ? dest_num : 1);
1487 return ERR_PTR(-ENOMEM);
1491 rule = find_flow_rule(fte, dest + i);
1493 refcount_inc(&rule->node.refcount);
1499 rule = alloc_rule(dest + i);
1503 /* Add dest to dests list- we need flow tables to be in the
1504 * end of the list for forward to next prio rules.
1506 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1508 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1509 list_add(&rule->node.list, &fte->node.children);
1511 list_add_tail(&rule->node.list, &fte->node.children);
1515 if (is_fwd_dest_type(dest[i].type))
1518 type = dest[i].type ==
1519 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1520 *modify_mask |= type ? count : dst;
1523 handle->rule[i] = rule;
1524 } while (++i < dest_num);
1529 destroy_flow_handle(fte, handle, dest, i);
1530 return ERR_PTR(-ENOMEM);
1533 /* fte should not be deleted while calling this function */
1534 static struct mlx5_flow_handle *
1535 add_rule_fte(struct fs_fte *fte,
1536 struct mlx5_flow_group *fg,
1537 struct mlx5_flow_destination *dest,
1541 struct mlx5_flow_root_namespace *root;
1542 struct mlx5_flow_handle *handle;
1543 struct mlx5_flow_table *ft;
1544 int modify_mask = 0;
1546 bool new_rule = false;
1548 handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1550 if (IS_ERR(handle) || !new_rule)
1554 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1556 fs_get_obj(ft, fg->node.parent);
1557 root = find_root(&fg->node);
1558 if (!(fte->status & FS_FTE_STATUS_EXISTING))
1559 err = root->cmds->create_fte(root, ft, fg, fte);
1561 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1565 fte->node.active = true;
1566 fte->status |= FS_FTE_STATUS_EXISTING;
1567 atomic_inc(&fg->node.version);
1573 destroy_flow_handle(fte, handle, dest, handle->num_rules);
1574 return ERR_PTR(err);
1577 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft,
1578 const struct mlx5_flow_spec *spec)
1580 struct list_head *prev = &ft->node.children;
1581 u32 max_fte = ft->autogroup.max_fte;
1582 unsigned int candidate_index = 0;
1583 unsigned int group_size = 0;
1584 struct mlx5_flow_group *fg;
1586 if (!ft->autogroup.active)
1587 return ERR_PTR(-ENOENT);
1589 if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1590 group_size = ft->autogroup.group_size;
1592 /* max_fte == ft->autogroup.max_types */
1593 if (group_size == 0)
1596 /* sorted by start_index */
1597 fs_for_each_fg(fg, ft) {
1598 if (candidate_index + group_size > fg->start_index)
1599 candidate_index = fg->start_index + fg->max_ftes;
1602 prev = &fg->node.list;
1605 if (candidate_index + group_size > max_fte)
1606 return ERR_PTR(-ENOSPC);
1608 fg = alloc_insert_flow_group(ft,
1609 spec->match_criteria_enable,
1610 spec->match_criteria,
1612 candidate_index + group_size - 1,
1617 if (group_size == ft->autogroup.group_size)
1618 ft->autogroup.num_groups++;
1624 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1625 struct mlx5_flow_group *fg)
1627 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1628 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1629 void *match_criteria_addr;
1630 u8 src_esw_owner_mask_on;
1635 in = kvzalloc(inlen, GFP_KERNEL);
1639 MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1640 fg->mask.match_criteria_enable);
1641 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1642 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index +
1645 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1647 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1648 source_eswitch_owner_vhca_id);
1649 MLX5_SET(create_flow_group_in, in,
1650 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1652 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1653 in, match_criteria);
1654 memcpy(match_criteria_addr, fg->mask.match_criteria,
1655 sizeof(fg->mask.match_criteria));
1657 err = root->cmds->create_flow_group(root, ft, in, fg);
1659 fg->node.active = true;
1660 trace_mlx5_fs_add_fg(fg);
1667 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1668 struct mlx5_flow_destination *d2)
1670 if (d1->type == d2->type) {
1671 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
1672 d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) &&
1673 d1->vport.num == d2->vport.num &&
1674 d1->vport.flags == d2->vport.flags &&
1675 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1676 (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1677 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1678 (d1->vport.pkt_reformat->id ==
1679 d2->vport.pkt_reformat->id) : true)) ||
1680 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1681 d1->ft == d2->ft) ||
1682 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1683 d1->tir_num == d2->tir_num) ||
1684 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1685 d1->ft_num == d2->ft_num) ||
1686 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1687 d1->sampler_id == d2->sampler_id) ||
1688 (d1->type == MLX5_FLOW_DESTINATION_TYPE_RANGE &&
1689 d1->range.field == d2->range.field &&
1690 d1->range.hit_ft == d2->range.hit_ft &&
1691 d1->range.miss_ft == d2->range.miss_ft &&
1692 d1->range.min == d2->range.min &&
1693 d1->range.max == d2->range.max))
1700 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1701 struct mlx5_flow_destination *dest)
1703 struct mlx5_flow_rule *rule;
1705 list_for_each_entry(rule, &fte->node.children, node.list) {
1706 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1712 static bool check_conflicting_actions_vlan(const struct mlx5_fs_vlan *vlan0,
1713 const struct mlx5_fs_vlan *vlan1)
1715 return vlan0->ethtype != vlan1->ethtype ||
1716 vlan0->vid != vlan1->vid ||
1717 vlan0->prio != vlan1->prio;
1720 static bool check_conflicting_actions(const struct mlx5_flow_act *act1,
1721 const struct mlx5_flow_act *act2)
1723 u32 action1 = act1->action;
1724 u32 action2 = act2->action;
1727 xored_actions = action1 ^ action2;
1729 /* if one rule only wants to count, it's ok */
1730 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1731 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1734 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1735 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1736 MLX5_FLOW_CONTEXT_ACTION_DECAP |
1737 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1738 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1739 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1740 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1741 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1744 if (action1 & MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT &&
1745 act1->pkt_reformat != act2->pkt_reformat)
1748 if (action1 & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR &&
1749 act1->modify_hdr != act2->modify_hdr)
1752 if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH &&
1753 check_conflicting_actions_vlan(&act1->vlan[0], &act2->vlan[0]))
1756 if (action1 & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2 &&
1757 check_conflicting_actions_vlan(&act1->vlan[1], &act2->vlan[1]))
1763 static int check_conflicting_ftes(struct fs_fte *fte,
1764 const struct mlx5_flow_context *flow_context,
1765 const struct mlx5_flow_act *flow_act)
1767 if (check_conflicting_actions(flow_act, &fte->action)) {
1768 mlx5_core_warn(get_dev(&fte->node),
1769 "Found two FTEs with conflicting actions\n");
1773 if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1774 fte->flow_context.flow_tag != flow_context->flow_tag) {
1775 mlx5_core_warn(get_dev(&fte->node),
1776 "FTE flow tag %u already exists with different flow tag %u\n",
1777 fte->flow_context.flow_tag,
1778 flow_context->flow_tag);
1785 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1786 const struct mlx5_flow_spec *spec,
1787 struct mlx5_flow_act *flow_act,
1788 struct mlx5_flow_destination *dest,
1792 struct mlx5_flow_handle *handle;
1797 ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1799 return ERR_PTR(ret);
1801 old_action = fte->action.action;
1802 fte->action.action |= flow_act->action;
1803 handle = add_rule_fte(fte, fg, dest, dest_num,
1804 old_action != flow_act->action);
1805 if (IS_ERR(handle)) {
1806 fte->action.action = old_action;
1809 trace_mlx5_fs_set_fte(fte, false);
1811 for (i = 0; i < handle->num_rules; i++) {
1812 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1813 tree_add_node(&handle->rule[i]->node, &fte->node);
1814 trace_mlx5_fs_add_rule(handle->rule[i]);
1820 static bool counter_is_valid(u32 action)
1822 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1823 MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1824 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1827 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1828 struct mlx5_flow_act *flow_act,
1829 struct mlx5_flow_table *ft)
1831 bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1832 u32 action = flow_act->action;
1834 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1835 return counter_is_valid(action);
1837 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1841 if (ft->type != FS_FT_FDB &&
1842 ft->type != FS_FT_NIC_RX &&
1843 ft->type != FS_FT_NIC_TX)
1846 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1847 ft->type != dest->ft->type)
1851 if (!dest || ((dest->type ==
1852 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1853 (dest->ft->level <= ft->level && !ignore_level)))
1859 struct list_head list;
1860 struct mlx5_flow_group *g;
1863 static void free_match_list(struct match_list *head, bool ft_locked)
1865 struct match_list *iter, *match_tmp;
1867 list_for_each_entry_safe(iter, match_tmp, &head->list,
1869 tree_put_node(&iter->g->node, ft_locked);
1870 list_del(&iter->list);
1875 static int build_match_list(struct match_list *match_head,
1876 struct mlx5_flow_table *ft,
1877 const struct mlx5_flow_spec *spec,
1878 struct mlx5_flow_group *fg,
1881 struct rhlist_head *tmp, *list;
1882 struct mlx5_flow_group *g;
1885 INIT_LIST_HEAD(&match_head->list);
1886 /* Collect all fgs which has a matching match_criteria */
1887 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1888 /* RCU is atomic, we can't execute FW commands here */
1889 rhl_for_each_entry_rcu(g, tmp, list, hash) {
1890 struct match_list *curr_match;
1895 if (unlikely(!tree_get_node(&g->node)))
1898 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1901 free_match_list(match_head, ft_locked);
1905 list_add_tail(&curr_match->list, &match_head->list);
1911 static u64 matched_fgs_get_version(struct list_head *match_head)
1913 struct match_list *iter;
1916 list_for_each_entry(iter, match_head, list)
1917 version += (u64)atomic_read(&iter->g->node.version);
1921 static struct fs_fte *
1922 lookup_fte_locked(struct mlx5_flow_group *g,
1923 const u32 *match_value,
1926 struct fs_fte *fte_tmp;
1929 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1931 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1932 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1934 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1938 if (!fte_tmp->node.active) {
1939 tree_put_node(&fte_tmp->node, false);
1944 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1947 up_write_ref_node(&g->node, false);
1949 up_read_ref_node(&g->node);
1953 static struct mlx5_flow_handle *
1954 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1955 struct list_head *match_head,
1956 const struct mlx5_flow_spec *spec,
1957 struct mlx5_flow_act *flow_act,
1958 struct mlx5_flow_destination *dest,
1962 struct mlx5_flow_steering *steering = get_steering(&ft->node);
1963 struct mlx5_flow_group *g;
1964 struct mlx5_flow_handle *rule;
1965 struct match_list *iter;
1966 bool take_write = false;
1971 fte = alloc_fte(ft, spec, flow_act);
1973 return ERR_PTR(-ENOMEM);
1975 search_again_locked:
1976 if (flow_act->flags & FLOW_ACT_NO_APPEND)
1978 version = matched_fgs_get_version(match_head);
1979 /* Try to find an fte with identical match value and attempt update its
1982 list_for_each_entry(iter, match_head, list) {
1983 struct fs_fte *fte_tmp;
1986 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1989 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1990 /* No error check needed here, because insert_fte() is not called */
1991 up_write_ref_node(&fte_tmp->node, false);
1992 tree_put_node(&fte_tmp->node, false);
1993 kmem_cache_free(steering->ftes_cache, fte);
1998 /* No group with matching fte found, or we skipped the search.
1999 * Try to add a new fte to any matching fg.
2002 /* Check the ft version, for case that new flow group
2003 * was added while the fgs weren't locked
2005 if (atomic_read(&ft->node.version) != ft_version) {
2006 rule = ERR_PTR(-EAGAIN);
2010 /* Check the fgs version. If version have changed it could be that an
2011 * FTE with the same match value was added while the fgs weren't
2014 if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
2015 version != matched_fgs_get_version(match_head)) {
2017 goto search_again_locked;
2020 list_for_each_entry(iter, match_head, list) {
2023 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2025 if (!g->node.active) {
2026 up_write_ref_node(&g->node, false);
2030 err = insert_fte(g, fte);
2032 up_write_ref_node(&g->node, false);
2035 kmem_cache_free(steering->ftes_cache, fte);
2036 return ERR_PTR(err);
2039 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
2040 up_write_ref_node(&g->node, false);
2041 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
2042 up_write_ref_node(&fte->node, false);
2044 tree_put_node(&fte->node, false);
2047 rule = ERR_PTR(-ENOENT);
2049 kmem_cache_free(steering->ftes_cache, fte);
2053 static struct mlx5_flow_handle *
2054 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2055 const struct mlx5_flow_spec *spec,
2056 struct mlx5_flow_act *flow_act,
2057 struct mlx5_flow_destination *dest,
2061 struct mlx5_flow_steering *steering = get_steering(&ft->node);
2062 struct mlx5_flow_handle *rule;
2063 struct match_list match_head;
2064 struct mlx5_flow_group *g;
2065 bool take_write = false;
2071 if (!check_valid_spec(spec))
2072 return ERR_PTR(-EINVAL);
2074 if (flow_act->fg && ft->autogroup.active)
2075 return ERR_PTR(-EINVAL);
2077 if (dest && dest_num <= 0)
2078 return ERR_PTR(-EINVAL);
2080 for (i = 0; i < dest_num; i++) {
2081 if (!dest_is_valid(&dest[i], flow_act, ft))
2082 return ERR_PTR(-EINVAL);
2084 nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
2085 search_again_locked:
2086 version = atomic_read(&ft->node.version);
2088 /* Collect all fgs which has a matching match_criteria */
2089 err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write);
2092 up_write_ref_node(&ft->node, false);
2094 up_read_ref_node(&ft->node);
2095 return ERR_PTR(err);
2099 up_read_ref_node(&ft->node);
2101 rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
2103 free_match_list(&match_head, take_write);
2104 if (!IS_ERR(rule) ||
2105 (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
2107 up_write_ref_node(&ft->node, false);
2112 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
2116 if (PTR_ERR(rule) == -EAGAIN ||
2117 version != atomic_read(&ft->node.version))
2118 goto search_again_locked;
2120 g = alloc_auto_flow_group(ft, spec);
2123 up_write_ref_node(&ft->node, false);
2127 fte = alloc_fte(ft, spec, flow_act);
2129 up_write_ref_node(&ft->node, false);
2134 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
2135 up_write_ref_node(&ft->node, false);
2137 err = create_auto_flow_group(ft, g);
2139 goto err_release_fg;
2141 err = insert_fte(g, fte);
2143 goto err_release_fg;
2145 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
2146 up_write_ref_node(&g->node, false);
2147 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
2148 up_write_ref_node(&fte->node, false);
2150 tree_put_node(&fte->node, false);
2151 tree_put_node(&g->node, false);
2155 up_write_ref_node(&g->node, false);
2156 kmem_cache_free(steering->ftes_cache, fte);
2158 tree_put_node(&g->node, false);
2159 return ERR_PTR(err);
2162 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
2164 return ((ft->type == FS_FT_NIC_RX) &&
2165 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
2168 struct mlx5_flow_handle *
2169 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2170 const struct mlx5_flow_spec *spec,
2171 struct mlx5_flow_act *flow_act,
2172 struct mlx5_flow_destination *dest,
2175 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2176 static const struct mlx5_flow_spec zero_spec = {};
2177 struct mlx5_flow_destination *gen_dest = NULL;
2178 struct mlx5_flow_table *next_ft = NULL;
2179 struct mlx5_flow_handle *handle = NULL;
2180 u32 sw_action = flow_act->action;
2186 if (!is_fwd_next_action(sw_action))
2187 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2189 if (!fwd_next_prio_supported(ft))
2190 return ERR_PTR(-EOPNOTSUPP);
2192 mutex_lock(&root->chain_lock);
2193 next_ft = find_next_fwd_ft(ft, flow_act);
2195 handle = ERR_PTR(-EOPNOTSUPP);
2199 gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
2202 handle = ERR_PTR(-ENOMEM);
2205 for (i = 0; i < num_dest; i++)
2206 gen_dest[i] = dest[i];
2208 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2209 gen_dest[i].ft = next_ft;
2212 flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
2213 MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
2214 flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2215 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2219 if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
2220 mutex_lock(&next_ft->lock);
2221 list_add(&handle->rule[num_dest - 1]->next_ft,
2222 &next_ft->fwd_rules);
2223 mutex_unlock(&next_ft->lock);
2224 handle->rule[num_dest - 1]->sw_action = sw_action;
2225 handle->rule[num_dest - 1]->ft = ft;
2228 mutex_unlock(&root->chain_lock);
2232 EXPORT_SYMBOL(mlx5_add_flow_rules);
2234 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2239 /* In order to consolidate the HW changes we lock the FTE for other
2240 * changes, and increase its refcount, in order not to perform the
2241 * "del" functions of the FTE. Will handle them here.
2242 * The removal of the rules is done under locked FTE.
2243 * After removing all the handle's rules, if there are remaining
2244 * rules, it means we just need to modify the FTE in FW, and
2245 * unlock/decrease the refcount we increased before.
2246 * Otherwise, it means the FTE should be deleted. First delete the
2247 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2248 * the FTE, which will handle the last decrease of the refcount, as
2249 * well as required handling of its parent.
2251 fs_get_obj(fte, handle->rule[0]->node.parent);
2252 down_write_ref_node(&fte->node, false);
2253 for (i = handle->num_rules - 1; i >= 0; i--)
2254 tree_remove_node(&handle->rule[i]->node, true);
2255 if (list_empty(&fte->node.children)) {
2256 fte->node.del_hw_func(&fte->node);
2257 /* Avoid double call to del_hw_fte */
2258 fte->node.del_hw_func = NULL;
2259 up_write_ref_node(&fte->node, false);
2260 tree_put_node(&fte->node, false);
2261 } else if (fte->dests_size) {
2262 if (fte->modify_mask)
2264 up_write_ref_node(&fte->node, false);
2266 up_write_ref_node(&fte->node, false);
2270 EXPORT_SYMBOL(mlx5_del_flow_rules);
2272 /* Assuming prio->node.children(flow tables) is sorted by level */
2273 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2275 struct fs_node *prio_parent, *child;
2276 struct fs_prio *prio;
2278 fs_get_obj(prio, ft->node.parent);
2280 if (!list_is_last(&ft->node.list, &prio->node.children))
2281 return list_next_entry(ft, node.list);
2283 prio_parent = find_prio_chains_parent(&prio->node, &child);
2285 if (prio_parent && list_is_first(&child->list, &prio_parent->children))
2286 return find_closest_ft(&prio->node, false, false);
2288 return find_next_chained_ft(&prio->node);
2291 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2293 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2294 struct mlx5_ft_underlay_qp *uqp;
2295 struct mlx5_flow_table *new_root_ft = NULL;
2299 if (root->root_ft != ft)
2302 new_root_ft = find_next_ft(ft);
2304 root->root_ft = NULL;
2308 if (list_empty(&root->underlay_qpns)) {
2309 /* Don't set any QPN (zero) in case QPN list is empty */
2311 err = root->cmds->update_root_ft(root, new_root_ft,
2314 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2316 err = root->cmds->update_root_ft(root,
2325 mlx5_core_warn(root->dev,
2326 "Update root flow table of id(%u) qpn(%d) failed\n",
2329 root->root_ft = new_root_ft;
2334 /* Connect flow table from previous priority to
2335 * the next flow table.
2337 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2339 struct mlx5_core_dev *dev = get_dev(&ft->node);
2340 struct mlx5_flow_table *next_ft;
2341 struct fs_prio *prio;
2344 err = update_root_ft_destroy(ft);
2348 fs_get_obj(prio, ft->node.parent);
2349 if (!(list_first_entry(&prio->node.children,
2350 struct mlx5_flow_table,
2354 next_ft = find_next_ft(ft);
2355 err = connect_fwd_rules(dev, next_ft, ft);
2359 err = connect_prev_fts(dev, next_ft, prio);
2361 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2366 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2368 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2371 mutex_lock(&root->chain_lock);
2372 if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2373 err = disconnect_flow_table(ft);
2375 mutex_unlock(&root->chain_lock);
2378 if (tree_remove_node(&ft->node, false))
2379 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2381 mutex_unlock(&root->chain_lock);
2385 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2387 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2389 if (tree_remove_node(&fg->node, false))
2390 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2393 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2395 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2398 struct mlx5_flow_steering *steering = dev->priv.steering;
2400 if (!steering || !steering->fdb_sub_ns)
2403 return steering->fdb_sub_ns[n];
2405 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2407 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type)
2410 case MLX5_FLOW_NAMESPACE_BYPASS:
2411 case MLX5_FLOW_NAMESPACE_KERNEL_RX_MACSEC:
2412 case MLX5_FLOW_NAMESPACE_LAG:
2413 case MLX5_FLOW_NAMESPACE_OFFLOADS:
2414 case MLX5_FLOW_NAMESPACE_ETHTOOL:
2415 case MLX5_FLOW_NAMESPACE_KERNEL:
2416 case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2417 case MLX5_FLOW_NAMESPACE_ANCHOR:
2424 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2425 enum mlx5_flow_namespace_type type)
2427 struct mlx5_flow_steering *steering = dev->priv.steering;
2428 struct mlx5_flow_root_namespace *root_ns;
2430 struct fs_prio *fs_prio;
2431 struct mlx5_flow_namespace *ns;
2437 case MLX5_FLOW_NAMESPACE_FDB:
2438 if (steering->fdb_root_ns)
2439 return &steering->fdb_root_ns->ns;
2441 case MLX5_FLOW_NAMESPACE_PORT_SEL:
2442 if (steering->port_sel_root_ns)
2443 return &steering->port_sel_root_ns->ns;
2445 case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2446 if (steering->sniffer_rx_root_ns)
2447 return &steering->sniffer_rx_root_ns->ns;
2449 case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2450 if (steering->sniffer_tx_root_ns)
2451 return &steering->sniffer_tx_root_ns->ns;
2453 case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
2454 root_ns = steering->fdb_root_ns;
2455 prio = FDB_BYPASS_PATH;
2457 case MLX5_FLOW_NAMESPACE_EGRESS:
2458 case MLX5_FLOW_NAMESPACE_EGRESS_IPSEC:
2459 case MLX5_FLOW_NAMESPACE_EGRESS_MACSEC:
2460 root_ns = steering->egress_root_ns;
2461 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2463 case MLX5_FLOW_NAMESPACE_RDMA_RX:
2464 root_ns = steering->rdma_rx_root_ns;
2465 prio = RDMA_RX_BYPASS_PRIO;
2467 case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL:
2468 root_ns = steering->rdma_rx_root_ns;
2469 prio = RDMA_RX_KERNEL_PRIO;
2471 case MLX5_FLOW_NAMESPACE_RDMA_TX:
2472 root_ns = steering->rdma_tx_root_ns;
2474 case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS:
2475 root_ns = steering->rdma_rx_root_ns;
2476 prio = RDMA_RX_COUNTERS_PRIO;
2478 case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS:
2479 root_ns = steering->rdma_tx_root_ns;
2480 prio = RDMA_TX_COUNTERS_PRIO;
2482 case MLX5_FLOW_NAMESPACE_RDMA_RX_IPSEC:
2483 root_ns = steering->rdma_rx_root_ns;
2484 prio = RDMA_RX_IPSEC_PRIO;
2486 case MLX5_FLOW_NAMESPACE_RDMA_TX_IPSEC:
2487 root_ns = steering->rdma_tx_root_ns;
2488 prio = RDMA_TX_IPSEC_PRIO;
2490 case MLX5_FLOW_NAMESPACE_RDMA_RX_MACSEC:
2491 root_ns = steering->rdma_rx_root_ns;
2492 prio = RDMA_RX_MACSEC_PRIO;
2494 case MLX5_FLOW_NAMESPACE_RDMA_TX_MACSEC:
2495 root_ns = steering->rdma_tx_root_ns;
2496 prio = RDMA_TX_MACSEC_PRIO;
2498 default: /* Must be NIC RX */
2499 WARN_ON(!is_nic_rx_ns(type));
2500 root_ns = steering->root_ns;
2508 fs_prio = find_prio(&root_ns->ns, prio);
2512 ns = list_first_entry(&fs_prio->node.children,
2518 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2520 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2521 enum mlx5_flow_namespace_type type,
2524 struct mlx5_flow_steering *steering = dev->priv.steering;
2530 case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2531 if (vport >= steering->esw_egress_acl_vports)
2533 if (steering->esw_egress_root_ns &&
2534 steering->esw_egress_root_ns[vport])
2535 return &steering->esw_egress_root_ns[vport]->ns;
2538 case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2539 if (vport >= steering->esw_ingress_acl_vports)
2541 if (steering->esw_ingress_root_ns &&
2542 steering->esw_ingress_root_ns[vport])
2543 return &steering->esw_ingress_root_ns[vport]->ns;
2551 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2554 enum fs_node_type type)
2556 struct fs_prio *fs_prio;
2558 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2560 return ERR_PTR(-ENOMEM);
2562 fs_prio->node.type = type;
2563 tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2564 tree_add_node(&fs_prio->node, &ns->node);
2565 fs_prio->num_levels = num_levels;
2566 fs_prio->prio = prio;
2567 list_add_tail(&fs_prio->node.list, &ns->node.children);
2572 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2576 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2579 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2580 unsigned int prio, int num_levels)
2582 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2585 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2588 ns->node.type = FS_TYPE_NAMESPACE;
2593 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2596 struct mlx5_flow_namespace *ns;
2598 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2600 return ERR_PTR(-ENOMEM);
2602 fs_init_namespace(ns);
2603 ns->def_miss_action = def_miss_act;
2604 tree_init_node(&ns->node, NULL, del_sw_ns);
2605 tree_add_node(&ns->node, &prio->node);
2606 list_add_tail(&ns->node.list, &prio->node.children);
2611 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2612 struct init_tree_node *prio_metadata)
2614 struct fs_prio *fs_prio;
2617 for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2618 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2619 if (IS_ERR(fs_prio))
2620 return PTR_ERR(fs_prio);
2625 #define FLOW_TABLE_BIT_SZ 1
2626 #define GET_FLOW_TABLE_CAP(dev, offset) \
2627 ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) + \
2629 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2630 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2634 for (i = 0; i < caps->arr_sz; i++) {
2635 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2641 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2642 struct init_tree_node *init_node,
2643 struct fs_node *fs_parent_node,
2644 struct init_tree_node *init_parent_node,
2647 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2648 flow_table_properties_nic_receive.
2650 struct mlx5_flow_namespace *fs_ns;
2651 struct fs_prio *fs_prio;
2652 struct fs_node *base;
2656 if (init_node->type == FS_TYPE_PRIO) {
2657 if ((init_node->min_ft_level > max_ft_level) ||
2658 !has_required_caps(steering->dev, &init_node->caps))
2661 fs_get_obj(fs_ns, fs_parent_node);
2662 if (init_node->num_leaf_prios)
2663 return create_leaf_prios(fs_ns, prio, init_node);
2664 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2665 if (IS_ERR(fs_prio))
2666 return PTR_ERR(fs_prio);
2667 base = &fs_prio->node;
2668 } else if (init_node->type == FS_TYPE_NAMESPACE) {
2669 fs_get_obj(fs_prio, fs_parent_node);
2670 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2672 return PTR_ERR(fs_ns);
2673 base = &fs_ns->node;
2678 for (i = 0; i < init_node->ar_size; i++) {
2679 err = init_root_tree_recursive(steering, &init_node->children[i],
2680 base, init_node, prio);
2683 if (init_node->children[i].type == FS_TYPE_PRIO &&
2684 init_node->children[i].num_leaf_prios) {
2685 prio += init_node->children[i].num_leaf_prios;
2692 static int init_root_tree(struct mlx5_flow_steering *steering,
2693 struct init_tree_node *init_node,
2694 struct fs_node *fs_parent_node)
2699 for (i = 0; i < init_node->ar_size; i++) {
2700 err = init_root_tree_recursive(steering, &init_node->children[i],
2709 static void del_sw_root_ns(struct fs_node *node)
2711 struct mlx5_flow_root_namespace *root_ns;
2712 struct mlx5_flow_namespace *ns;
2714 fs_get_obj(ns, node);
2715 root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2716 mutex_destroy(&root_ns->chain_lock);
2720 static struct mlx5_flow_root_namespace
2721 *create_root_ns(struct mlx5_flow_steering *steering,
2722 enum fs_flow_table_type table_type)
2724 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2725 struct mlx5_flow_root_namespace *root_ns;
2726 struct mlx5_flow_namespace *ns;
2728 /* Create the root namespace */
2729 root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2733 root_ns->dev = steering->dev;
2734 root_ns->table_type = table_type;
2735 root_ns->cmds = cmds;
2737 INIT_LIST_HEAD(&root_ns->underlay_qpns);
2740 fs_init_namespace(ns);
2741 mutex_init(&root_ns->chain_lock);
2742 tree_init_node(&ns->node, NULL, del_sw_root_ns);
2743 tree_add_node(&ns->node, NULL);
2748 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2750 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2752 struct fs_prio *prio;
2754 fs_for_each_prio(prio, ns) {
2755 /* This updates prio start_level and num_levels */
2756 set_prio_attrs_in_prio(prio, acc_level);
2757 acc_level += prio->num_levels;
2762 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2764 struct mlx5_flow_namespace *ns;
2765 int acc_level_ns = acc_level;
2767 prio->start_level = acc_level;
2768 fs_for_each_ns(ns, prio) {
2769 /* This updates start_level and num_levels of ns's priority descendants */
2770 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2772 /* If this a prio with chains, and we can jump from one chain
2773 * (namespace) to another, so we accumulate the levels
2775 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2776 acc_level = acc_level_ns;
2779 if (!prio->num_levels)
2780 prio->num_levels = acc_level_ns - prio->start_level;
2781 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2784 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2786 struct mlx5_flow_namespace *ns = &root_ns->ns;
2787 struct fs_prio *prio;
2788 int start_level = 0;
2790 fs_for_each_prio(prio, ns) {
2791 set_prio_attrs_in_prio(prio, start_level);
2792 start_level += prio->num_levels;
2796 #define ANCHOR_PRIO 0
2797 #define ANCHOR_SIZE 1
2798 #define ANCHOR_LEVEL 0
2799 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2801 struct mlx5_flow_namespace *ns = NULL;
2802 struct mlx5_flow_table_attr ft_attr = {};
2803 struct mlx5_flow_table *ft;
2805 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2809 ft_attr.max_fte = ANCHOR_SIZE;
2810 ft_attr.level = ANCHOR_LEVEL;
2811 ft_attr.prio = ANCHOR_PRIO;
2813 ft = mlx5_create_flow_table(ns, &ft_attr);
2815 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2821 static int init_root_ns(struct mlx5_flow_steering *steering)
2825 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2826 if (!steering->root_ns)
2829 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2833 set_prio_attrs(steering->root_ns);
2834 err = create_anchor_flow_table(steering);
2841 cleanup_root_ns(steering->root_ns);
2842 steering->root_ns = NULL;
2846 static void clean_tree(struct fs_node *node)
2849 struct fs_node *iter;
2850 struct fs_node *temp;
2852 tree_get_node(node);
2853 list_for_each_entry_safe(iter, temp, &node->children, list)
2855 tree_put_node(node, false);
2856 tree_remove_node(node, false);
2860 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2865 clean_tree(&root_ns->ns.node);
2868 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2870 struct fs_prio *prio;
2872 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2873 if (!steering->sniffer_tx_root_ns)
2876 /* Create single prio */
2877 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2878 return PTR_ERR_OR_ZERO(prio);
2881 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2883 struct fs_prio *prio;
2885 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2886 if (!steering->sniffer_rx_root_ns)
2889 /* Create single prio */
2890 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2891 return PTR_ERR_OR_ZERO(prio);
2894 #define PORT_SEL_NUM_LEVELS 3
2895 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering)
2897 struct fs_prio *prio;
2899 steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL);
2900 if (!steering->port_sel_root_ns)
2903 /* Create single prio */
2904 prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0,
2905 PORT_SEL_NUM_LEVELS);
2906 return PTR_ERR_OR_ZERO(prio);
2909 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2913 steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2914 if (!steering->rdma_rx_root_ns)
2917 err = init_root_tree(steering, &rdma_rx_root_fs,
2918 &steering->rdma_rx_root_ns->ns.node);
2922 set_prio_attrs(steering->rdma_rx_root_ns);
2927 cleanup_root_ns(steering->rdma_rx_root_ns);
2928 steering->rdma_rx_root_ns = NULL;
2932 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2936 steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2937 if (!steering->rdma_tx_root_ns)
2940 err = init_root_tree(steering, &rdma_tx_root_fs,
2941 &steering->rdma_tx_root_ns->ns.node);
2945 set_prio_attrs(steering->rdma_tx_root_ns);
2950 cleanup_root_ns(steering->rdma_tx_root_ns);
2951 steering->rdma_tx_root_ns = NULL;
2955 /* FT and tc chains are stored in the same array so we can re-use the
2956 * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2957 * When creating a new ns for each chain store it in the first available slot.
2958 * Assume tc chains are created and stored first and only then the FT chain.
2960 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2961 struct mlx5_flow_namespace *ns)
2965 while (steering->fdb_sub_ns[chain])
2968 steering->fdb_sub_ns[chain] = ns;
2971 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2972 struct fs_prio *maj_prio)
2974 struct mlx5_flow_namespace *ns;
2975 struct fs_prio *min_prio;
2978 ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2982 for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2983 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2984 if (IS_ERR(min_prio))
2985 return PTR_ERR(min_prio);
2988 store_fdb_sub_ns_prio_chain(steering, ns);
2993 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2997 struct fs_prio *maj_prio;
3002 levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
3003 maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
3006 if (IS_ERR(maj_prio))
3007 return PTR_ERR(maj_prio);
3009 for (chain = 0; chain < chains; chain++) {
3010 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
3018 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
3022 steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
3023 sizeof(*steering->fdb_sub_ns),
3025 if (!steering->fdb_sub_ns)
3028 err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
3032 err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
3039 static int create_fdb_bypass(struct mlx5_flow_steering *steering)
3041 struct mlx5_flow_namespace *ns;
3042 struct fs_prio *prio;
3045 prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0);
3047 return PTR_ERR(prio);
3049 ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
3053 for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) {
3054 prio = fs_create_prio(ns, i, 1);
3056 return PTR_ERR(prio);
3061 static void cleanup_fdb_root_ns(struct mlx5_flow_steering *steering)
3063 cleanup_root_ns(steering->fdb_root_ns);
3064 steering->fdb_root_ns = NULL;
3065 kfree(steering->fdb_sub_ns);
3066 steering->fdb_sub_ns = NULL;
3069 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
3071 struct fs_prio *maj_prio;
3074 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
3075 if (!steering->fdb_root_ns)
3078 err = create_fdb_bypass(steering);
3082 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_CRYPTO_INGRESS, 3);
3083 if (IS_ERR(maj_prio)) {
3084 err = PTR_ERR(maj_prio);
3088 err = create_fdb_fast_path(steering);
3092 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
3093 if (IS_ERR(maj_prio)) {
3094 err = PTR_ERR(maj_prio);
3098 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 4);
3099 if (IS_ERR(maj_prio)) {
3100 err = PTR_ERR(maj_prio);
3104 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
3105 if (IS_ERR(maj_prio)) {
3106 err = PTR_ERR(maj_prio);
3110 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_CRYPTO_EGRESS, 3);
3111 if (IS_ERR(maj_prio)) {
3112 err = PTR_ERR(maj_prio);
3116 /* We put this priority last, knowing that nothing will get here
3117 * unless explicitly forwarded to. This is possible because the
3118 * slow path tables have catch all rules and nothing gets passed
3121 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
3122 if (IS_ERR(maj_prio)) {
3123 err = PTR_ERR(maj_prio);
3127 set_prio_attrs(steering->fdb_root_ns);
3131 cleanup_fdb_root_ns(steering);
3135 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
3137 struct fs_prio *prio;
3139 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
3140 if (!steering->esw_egress_root_ns[vport])
3144 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
3145 return PTR_ERR_OR_ZERO(prio);
3148 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
3150 struct fs_prio *prio;
3152 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
3153 if (!steering->esw_ingress_root_ns[vport])
3157 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
3158 return PTR_ERR_OR_ZERO(prio);
3161 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
3163 struct mlx5_flow_steering *steering = dev->priv.steering;
3167 steering->esw_egress_root_ns =
3168 kcalloc(total_vports,
3169 sizeof(*steering->esw_egress_root_ns),
3171 if (!steering->esw_egress_root_ns)
3174 for (i = 0; i < total_vports; i++) {
3175 err = init_egress_acl_root_ns(steering, i);
3177 goto cleanup_root_ns;
3179 steering->esw_egress_acl_vports = total_vports;
3183 for (i--; i >= 0; i--)
3184 cleanup_root_ns(steering->esw_egress_root_ns[i]);
3185 kfree(steering->esw_egress_root_ns);
3186 steering->esw_egress_root_ns = NULL;
3190 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
3192 struct mlx5_flow_steering *steering = dev->priv.steering;
3195 if (!steering->esw_egress_root_ns)
3198 for (i = 0; i < steering->esw_egress_acl_vports; i++)
3199 cleanup_root_ns(steering->esw_egress_root_ns[i]);
3201 kfree(steering->esw_egress_root_ns);
3202 steering->esw_egress_root_ns = NULL;
3205 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
3207 struct mlx5_flow_steering *steering = dev->priv.steering;
3211 steering->esw_ingress_root_ns =
3212 kcalloc(total_vports,
3213 sizeof(*steering->esw_ingress_root_ns),
3215 if (!steering->esw_ingress_root_ns)
3218 for (i = 0; i < total_vports; i++) {
3219 err = init_ingress_acl_root_ns(steering, i);
3221 goto cleanup_root_ns;
3223 steering->esw_ingress_acl_vports = total_vports;
3227 for (i--; i >= 0; i--)
3228 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3229 kfree(steering->esw_ingress_root_ns);
3230 steering->esw_ingress_root_ns = NULL;
3234 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
3236 struct mlx5_flow_steering *steering = dev->priv.steering;
3239 if (!steering->esw_ingress_root_ns)
3242 for (i = 0; i < steering->esw_ingress_acl_vports; i++)
3243 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3245 kfree(steering->esw_ingress_root_ns);
3246 steering->esw_ingress_root_ns = NULL;
3249 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type)
3251 struct mlx5_flow_root_namespace *root;
3252 struct mlx5_flow_namespace *ns;
3254 ns = mlx5_get_flow_namespace(dev, type);
3258 root = find_root(&ns->node);
3262 return root->cmds->get_capabilities(root, root->table_type);
3265 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
3269 steering->egress_root_ns = create_root_ns(steering,
3271 if (!steering->egress_root_ns)
3274 err = init_root_tree(steering, &egress_root_fs,
3275 &steering->egress_root_ns->ns.node);
3278 set_prio_attrs(steering->egress_root_ns);
3281 cleanup_root_ns(steering->egress_root_ns);
3282 steering->egress_root_ns = NULL;
3286 static int mlx5_fs_mode_validate(struct devlink *devlink, u32 id,
3287 union devlink_param_value val,
3288 struct netlink_ext_ack *extack)
3290 struct mlx5_core_dev *dev = devlink_priv(devlink);
3291 char *value = val.vstr;
3294 if (!strcmp(value, "dmfs")) {
3296 } else if (!strcmp(value, "smfs")) {
3300 eswitch_mode = mlx5_eswitch_mode(dev);
3301 smfs_cap = mlx5_fs_dr_is_supported(dev);
3305 NL_SET_ERR_MSG_MOD(extack,
3306 "Software managed steering is not supported by current device");
3309 else if (eswitch_mode == MLX5_ESWITCH_OFFLOADS) {
3310 NL_SET_ERR_MSG_MOD(extack,
3311 "Software managed steering is not supported when eswitch offloads enabled.");
3315 NL_SET_ERR_MSG_MOD(extack,
3316 "Bad parameter: supported values are [\"dmfs\", \"smfs\"]");
3323 static int mlx5_fs_mode_set(struct devlink *devlink, u32 id,
3324 struct devlink_param_gset_ctx *ctx)
3326 struct mlx5_core_dev *dev = devlink_priv(devlink);
3327 enum mlx5_flow_steering_mode mode;
3329 if (!strcmp(ctx->val.vstr, "smfs"))
3330 mode = MLX5_FLOW_STEERING_MODE_SMFS;
3332 mode = MLX5_FLOW_STEERING_MODE_DMFS;
3333 dev->priv.steering->mode = mode;
3338 static int mlx5_fs_mode_get(struct devlink *devlink, u32 id,
3339 struct devlink_param_gset_ctx *ctx)
3341 struct mlx5_core_dev *dev = devlink_priv(devlink);
3343 if (dev->priv.steering->mode == MLX5_FLOW_STEERING_MODE_SMFS)
3344 strcpy(ctx->val.vstr, "smfs");
3346 strcpy(ctx->val.vstr, "dmfs");
3350 static const struct devlink_param mlx5_fs_params[] = {
3351 DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_FLOW_STEERING_MODE,
3352 "flow_steering_mode", DEVLINK_PARAM_TYPE_STRING,
3353 BIT(DEVLINK_PARAM_CMODE_RUNTIME),
3354 mlx5_fs_mode_get, mlx5_fs_mode_set,
3355 mlx5_fs_mode_validate),
3358 void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev)
3360 struct mlx5_flow_steering *steering = dev->priv.steering;
3362 cleanup_root_ns(steering->root_ns);
3363 cleanup_fdb_root_ns(steering);
3364 cleanup_root_ns(steering->port_sel_root_ns);
3365 cleanup_root_ns(steering->sniffer_rx_root_ns);
3366 cleanup_root_ns(steering->sniffer_tx_root_ns);
3367 cleanup_root_ns(steering->rdma_rx_root_ns);
3368 cleanup_root_ns(steering->rdma_tx_root_ns);
3369 cleanup_root_ns(steering->egress_root_ns);
3371 devl_params_unregister(priv_to_devlink(dev), mlx5_fs_params,
3372 ARRAY_SIZE(mlx5_fs_params));
3375 int mlx5_fs_core_init(struct mlx5_core_dev *dev)
3377 struct mlx5_flow_steering *steering = dev->priv.steering;
3380 err = devl_params_register(priv_to_devlink(dev), mlx5_fs_params,
3381 ARRAY_SIZE(mlx5_fs_params));
3385 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
3386 (MLX5_CAP_GEN(dev, nic_flow_table))) ||
3387 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
3388 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
3389 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
3390 err = init_root_ns(steering);
3395 if (MLX5_ESWITCH_MANAGER(dev)) {
3396 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3397 err = init_fdb_root_ns(steering);
3403 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3404 err = init_sniffer_rx_root_ns(steering);
3409 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3410 err = init_sniffer_tx_root_ns(steering);
3415 if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) {
3416 err = init_port_sel_root_ns(steering);
3421 if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3422 MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3423 err = init_rdma_rx_root_ns(steering);
3428 if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3429 err = init_rdma_tx_root_ns(steering);
3434 if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3435 err = init_egress_root_ns(steering);
3443 mlx5_fs_core_cleanup(dev);
3447 void mlx5_fs_core_free(struct mlx5_core_dev *dev)
3449 struct mlx5_flow_steering *steering = dev->priv.steering;
3451 kmem_cache_destroy(steering->ftes_cache);
3452 kmem_cache_destroy(steering->fgs_cache);
3454 mlx5_ft_pool_destroy(dev);
3455 mlx5_cleanup_fc_stats(dev);
3458 int mlx5_fs_core_alloc(struct mlx5_core_dev *dev)
3460 struct mlx5_flow_steering *steering;
3463 err = mlx5_init_fc_stats(dev);
3467 err = mlx5_ft_pool_init(dev);
3471 steering = kzalloc(sizeof(*steering), GFP_KERNEL);
3477 steering->dev = dev;
3478 dev->priv.steering = steering;
3480 if (mlx5_fs_dr_is_supported(dev))
3481 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS;
3483 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS;
3485 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
3486 sizeof(struct mlx5_flow_group), 0,
3488 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
3490 if (!steering->ftes_cache || !steering->fgs_cache) {
3498 mlx5_fs_core_free(dev);
3502 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3504 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3505 struct mlx5_ft_underlay_qp *new_uqp;
3508 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3512 mutex_lock(&root->chain_lock);
3514 if (!root->root_ft) {
3516 goto update_ft_fail;
3519 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3522 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3524 goto update_ft_fail;
3527 new_uqp->qpn = underlay_qpn;
3528 list_add_tail(&new_uqp->list, &root->underlay_qpns);
3530 mutex_unlock(&root->chain_lock);
3535 mutex_unlock(&root->chain_lock);
3539 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3541 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3543 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3544 struct mlx5_ft_underlay_qp *uqp;
3548 mutex_lock(&root->chain_lock);
3549 list_for_each_entry(uqp, &root->underlay_qpns, list) {
3550 if (uqp->qpn == underlay_qpn) {
3557 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3563 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3566 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3569 list_del(&uqp->list);
3570 mutex_unlock(&root->chain_lock);
3576 mutex_unlock(&root->chain_lock);
3579 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3581 static struct mlx5_flow_root_namespace
3582 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3584 struct mlx5_flow_namespace *ns;
3586 if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3587 ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3588 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3590 ns = mlx5_get_flow_namespace(dev, ns_type);
3594 return find_root(&ns->node);
3597 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3598 u8 ns_type, u8 num_actions,
3599 void *modify_actions)
3601 struct mlx5_flow_root_namespace *root;
3602 struct mlx5_modify_hdr *modify_hdr;
3605 root = get_root_namespace(dev, ns_type);
3607 return ERR_PTR(-EOPNOTSUPP);
3609 modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3611 return ERR_PTR(-ENOMEM);
3613 modify_hdr->ns_type = ns_type;
3614 err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3615 modify_actions, modify_hdr);
3618 return ERR_PTR(err);
3623 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3625 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3626 struct mlx5_modify_hdr *modify_hdr)
3628 struct mlx5_flow_root_namespace *root;
3630 root = get_root_namespace(dev, modify_hdr->ns_type);
3633 root->cmds->modify_header_dealloc(root, modify_hdr);
3636 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3638 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3639 struct mlx5_pkt_reformat_params *params,
3640 enum mlx5_flow_namespace_type ns_type)
3642 struct mlx5_pkt_reformat *pkt_reformat;
3643 struct mlx5_flow_root_namespace *root;
3646 root = get_root_namespace(dev, ns_type);
3648 return ERR_PTR(-EOPNOTSUPP);
3650 pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3652 return ERR_PTR(-ENOMEM);
3654 pkt_reformat->ns_type = ns_type;
3655 pkt_reformat->reformat_type = params->type;
3656 err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3659 kfree(pkt_reformat);
3660 return ERR_PTR(err);
3663 return pkt_reformat;
3665 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3667 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3668 struct mlx5_pkt_reformat *pkt_reformat)
3670 struct mlx5_flow_root_namespace *root;
3672 root = get_root_namespace(dev, pkt_reformat->ns_type);
3675 root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3676 kfree(pkt_reformat);
3678 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3680 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer)
3685 struct mlx5_flow_definer *
3686 mlx5_create_match_definer(struct mlx5_core_dev *dev,
3687 enum mlx5_flow_namespace_type ns_type, u16 format_id,
3690 struct mlx5_flow_root_namespace *root;
3691 struct mlx5_flow_definer *definer;
3694 root = get_root_namespace(dev, ns_type);
3696 return ERR_PTR(-EOPNOTSUPP);
3698 definer = kzalloc(sizeof(*definer), GFP_KERNEL);
3700 return ERR_PTR(-ENOMEM);
3702 definer->ns_type = ns_type;
3703 id = root->cmds->create_match_definer(root, format_id, match_mask);
3705 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id);
3713 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev,
3714 struct mlx5_flow_definer *definer)
3716 struct mlx5_flow_root_namespace *root;
3718 root = get_root_namespace(dev, definer->ns_type);
3722 root->cmds->destroy_match_definer(root, definer->id);
3726 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3727 struct mlx5_flow_root_namespace *peer_ns,
3730 if (peer_ns && ns->mode != peer_ns->mode) {
3731 mlx5_core_err(ns->dev,
3732 "Can't peer namespace of different steering mode\n");
3736 return ns->cmds->set_peer(ns, peer_ns, peer_vhca_id);
3739 /* This function should be called only at init stage of the namespace.
3740 * It is not safe to call this function while steering operations
3741 * are executed in the namespace.
3743 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3744 enum mlx5_flow_steering_mode mode)
3746 struct mlx5_flow_root_namespace *root;
3747 const struct mlx5_flow_cmds *cmds;
3750 root = find_root(&ns->node);
3751 if (&root->ns != ns)
3752 /* Can't set cmds to non root namespace */
3755 if (root->table_type != FS_FT_FDB)
3758 if (root->mode == mode)
3761 if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3762 cmds = mlx5_fs_cmd_get_dr_cmds();
3764 cmds = mlx5_fs_cmd_get_fw_cmds();
3768 err = cmds->create_ns(root);
3770 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3775 root->cmds->destroy_ns(root);