2 * Copyright (c) 2016, 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/mlx5/driver.h>
34 #include <linux/mlx5/fs.h>
35 #include <linux/rbtree.h>
36 #include "mlx5_core.h"
40 #define MLX5_FC_STATS_PERIOD msecs_to_jiffies(1000)
41 /* Max number of counters to query in bulk read is 32K */
42 #define MLX5_SW_MAX_COUNTERS_BULK BIT(15)
43 #define MLX5_FC_POOL_MAX_THRESHOLD BIT(18)
44 #define MLX5_FC_POOL_USED_BUFF_RATIO 10
46 struct mlx5_fc_cache {
53 struct list_head list;
54 struct llist_node addlist;
55 struct llist_node dellist;
57 /* last{packets,bytes} members are used when calculating the delta since
63 struct mlx5_fc_bulk *bulk;
67 struct mlx5_fc_cache cache ____cacheline_aligned_in_smp;
70 static void mlx5_fc_pool_init(struct mlx5_fc_pool *fc_pool, struct mlx5_core_dev *dev);
71 static void mlx5_fc_pool_cleanup(struct mlx5_fc_pool *fc_pool);
72 static struct mlx5_fc *mlx5_fc_pool_acquire_counter(struct mlx5_fc_pool *fc_pool);
73 static void mlx5_fc_pool_release_counter(struct mlx5_fc_pool *fc_pool, struct mlx5_fc *fc);
77 * It is the responsibility of the user to prevent concurrent calls or bad
78 * ordering to mlx5_fc_create(), mlx5_fc_destroy() and accessing a reference
80 * e.g en_tc.c is protected by RTNL lock of its caller, and will never call a
81 * dump (access to struct mlx5_fc) after a counter is destroyed.
83 * access to counter list:
84 * - create (user context)
85 * - mlx5_fc_create() only adds to an addlist to be used by
86 * mlx5_fc_stats_work(). addlist is a lockless single linked list
87 * that doesn't require any additional synchronization when adding single
89 * - spawn thread to do the actual destroy
91 * - destroy (user context)
92 * - add a counter to lockless dellist
93 * - spawn thread to do the actual del
95 * - dump (user context)
96 * user should not call dump after destroy
98 * - query (single thread workqueue context)
99 * destroy/dump - no conflict (see destroy)
100 * query/dump - packets and bytes might be inconsistent (since update is not
102 * query/create - no conflict (see create)
103 * since every create/destroy spawn the work, only after necessary time has
104 * elapsed, the thread will actually query the hardware.
107 static struct list_head *mlx5_fc_counters_lookup_next(struct mlx5_core_dev *dev,
110 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
111 unsigned long next_id = (unsigned long)id + 1;
112 struct mlx5_fc *counter;
116 /* skip counters that are in idr, but not yet in counters list */
117 idr_for_each_entry_continue_ul(&fc_stats->counters_idr,
118 counter, tmp, next_id) {
119 if (!list_empty(&counter->list))
124 return counter ? &counter->list : &fc_stats->counters;
127 static void mlx5_fc_stats_insert(struct mlx5_core_dev *dev,
128 struct mlx5_fc *counter)
130 struct list_head *next = mlx5_fc_counters_lookup_next(dev, counter->id);
132 list_add_tail(&counter->list, next);
135 static void mlx5_fc_stats_remove(struct mlx5_core_dev *dev,
136 struct mlx5_fc *counter)
138 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
140 list_del(&counter->list);
142 spin_lock(&fc_stats->counters_idr_lock);
143 WARN_ON(!idr_remove(&fc_stats->counters_idr, counter->id));
144 spin_unlock(&fc_stats->counters_idr_lock);
147 static int get_max_bulk_query_len(struct mlx5_core_dev *dev)
149 return min_t(int, MLX5_SW_MAX_COUNTERS_BULK,
150 (1 << MLX5_CAP_GEN(dev, log_max_flow_counter_bulk)));
153 static void update_counter_cache(int index, u32 *bulk_raw_data,
154 struct mlx5_fc_cache *cache)
156 void *stats = MLX5_ADDR_OF(query_flow_counter_out, bulk_raw_data,
157 flow_statistics[index]);
158 u64 packets = MLX5_GET64(traffic_counter, stats, packets);
159 u64 bytes = MLX5_GET64(traffic_counter, stats, octets);
161 if (cache->packets == packets)
164 cache->packets = packets;
165 cache->bytes = bytes;
166 cache->lastuse = jiffies;
169 static void mlx5_fc_stats_query_counter_range(struct mlx5_core_dev *dev,
170 struct mlx5_fc *first,
173 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
174 bool query_more_counters = (first->id <= last_id);
175 int max_bulk_len = get_max_bulk_query_len(dev);
176 u32 *data = fc_stats->bulk_query_out;
177 struct mlx5_fc *counter = first;
182 while (query_more_counters) {
183 /* first id must be aligned to 4 when using bulk query */
184 bulk_base_id = counter->id & ~0x3;
186 /* number of counters to query inc. the last counter */
187 bulk_len = min_t(int, max_bulk_len,
188 ALIGN(last_id - bulk_base_id + 1, 4));
190 err = mlx5_cmd_fc_bulk_query(dev, bulk_base_id, bulk_len,
193 mlx5_core_err(dev, "Error doing bulk query: %d\n", err);
196 query_more_counters = false;
198 list_for_each_entry_from(counter, &fc_stats->counters, list) {
199 int counter_index = counter->id - bulk_base_id;
200 struct mlx5_fc_cache *cache = &counter->cache;
202 if (counter->id >= bulk_base_id + bulk_len) {
203 query_more_counters = true;
207 update_counter_cache(counter_index, data, cache);
212 static void mlx5_fc_free(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
214 mlx5_cmd_fc_free(dev, counter->id);
218 static void mlx5_fc_release(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
220 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
223 mlx5_fc_pool_release_counter(&fc_stats->fc_pool, counter);
225 mlx5_fc_free(dev, counter);
228 static void mlx5_fc_stats_work(struct work_struct *work)
230 struct mlx5_core_dev *dev = container_of(work, struct mlx5_core_dev,
231 priv.fc_stats.work.work);
232 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
233 /* Take dellist first to ensure that counters cannot be deleted before
236 struct llist_node *dellist = llist_del_all(&fc_stats->dellist);
237 struct llist_node *addlist = llist_del_all(&fc_stats->addlist);
238 struct mlx5_fc *counter = NULL, *last = NULL, *tmp;
239 unsigned long now = jiffies;
241 if (addlist || !list_empty(&fc_stats->counters))
242 queue_delayed_work(fc_stats->wq, &fc_stats->work,
243 fc_stats->sampling_interval);
245 llist_for_each_entry(counter, addlist, addlist)
246 mlx5_fc_stats_insert(dev, counter);
248 llist_for_each_entry_safe(counter, tmp, dellist, dellist) {
249 mlx5_fc_stats_remove(dev, counter);
251 mlx5_fc_release(dev, counter);
254 if (time_before(now, fc_stats->next_query) ||
255 list_empty(&fc_stats->counters))
257 last = list_last_entry(&fc_stats->counters, struct mlx5_fc, list);
259 counter = list_first_entry(&fc_stats->counters, struct mlx5_fc,
262 mlx5_fc_stats_query_counter_range(dev, counter, last->id);
264 fc_stats->next_query = now + fc_stats->sampling_interval;
267 static struct mlx5_fc *mlx5_fc_single_alloc(struct mlx5_core_dev *dev)
269 struct mlx5_fc *counter;
272 counter = kzalloc(sizeof(*counter), GFP_KERNEL);
274 return ERR_PTR(-ENOMEM);
276 err = mlx5_cmd_fc_alloc(dev, &counter->id);
285 static struct mlx5_fc *mlx5_fc_acquire(struct mlx5_core_dev *dev, bool aging)
287 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
288 struct mlx5_fc *counter;
290 if (aging && MLX5_CAP_GEN(dev, flow_counter_bulk_alloc) != 0) {
291 counter = mlx5_fc_pool_acquire_counter(&fc_stats->fc_pool);
292 if (!IS_ERR(counter))
296 return mlx5_fc_single_alloc(dev);
299 struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging)
301 struct mlx5_fc *counter = mlx5_fc_acquire(dev, aging);
302 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
308 INIT_LIST_HEAD(&counter->list);
309 counter->aging = aging;
312 u32 id = counter->id;
314 counter->cache.lastuse = jiffies;
315 counter->lastbytes = counter->cache.bytes;
316 counter->lastpackets = counter->cache.packets;
318 idr_preload(GFP_KERNEL);
319 spin_lock(&fc_stats->counters_idr_lock);
321 err = idr_alloc_u32(&fc_stats->counters_idr, counter, &id, id,
324 spin_unlock(&fc_stats->counters_idr_lock);
329 llist_add(&counter->addlist, &fc_stats->addlist);
331 mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
337 mlx5_fc_release(dev, counter);
340 EXPORT_SYMBOL(mlx5_fc_create);
342 u32 mlx5_fc_id(struct mlx5_fc *counter)
346 EXPORT_SYMBOL(mlx5_fc_id);
348 void mlx5_fc_destroy(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
350 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
355 if (counter->aging) {
356 llist_add(&counter->dellist, &fc_stats->dellist);
357 mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
361 mlx5_fc_release(dev, counter);
363 EXPORT_SYMBOL(mlx5_fc_destroy);
365 int mlx5_init_fc_stats(struct mlx5_core_dev *dev)
367 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
371 spin_lock_init(&fc_stats->counters_idr_lock);
372 idr_init(&fc_stats->counters_idr);
373 INIT_LIST_HEAD(&fc_stats->counters);
374 init_llist_head(&fc_stats->addlist);
375 init_llist_head(&fc_stats->dellist);
377 max_bulk_len = get_max_bulk_query_len(dev);
378 max_out_len = mlx5_cmd_fc_get_bulk_query_out_len(max_bulk_len);
379 fc_stats->bulk_query_out = kzalloc(max_out_len, GFP_KERNEL);
380 if (!fc_stats->bulk_query_out)
383 fc_stats->wq = create_singlethread_workqueue("mlx5_fc");
387 fc_stats->sampling_interval = MLX5_FC_STATS_PERIOD;
388 INIT_DELAYED_WORK(&fc_stats->work, mlx5_fc_stats_work);
390 mlx5_fc_pool_init(&fc_stats->fc_pool, dev);
394 kfree(fc_stats->bulk_query_out);
398 void mlx5_cleanup_fc_stats(struct mlx5_core_dev *dev)
400 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
401 struct llist_node *tmplist;
402 struct mlx5_fc *counter;
405 cancel_delayed_work_sync(&dev->priv.fc_stats.work);
406 destroy_workqueue(dev->priv.fc_stats.wq);
407 dev->priv.fc_stats.wq = NULL;
409 tmplist = llist_del_all(&fc_stats->addlist);
410 llist_for_each_entry_safe(counter, tmp, tmplist, addlist)
411 mlx5_fc_release(dev, counter);
413 list_for_each_entry_safe(counter, tmp, &fc_stats->counters, list)
414 mlx5_fc_release(dev, counter);
416 mlx5_fc_pool_cleanup(&fc_stats->fc_pool);
417 idr_destroy(&fc_stats->counters_idr);
418 kfree(fc_stats->bulk_query_out);
421 int mlx5_fc_query(struct mlx5_core_dev *dev, struct mlx5_fc *counter,
422 u64 *packets, u64 *bytes)
424 return mlx5_cmd_fc_query(dev, counter->id, packets, bytes);
426 EXPORT_SYMBOL(mlx5_fc_query);
428 u64 mlx5_fc_query_lastuse(struct mlx5_fc *counter)
430 return counter->cache.lastuse;
433 void mlx5_fc_query_cached(struct mlx5_fc *counter,
434 u64 *bytes, u64 *packets, u64 *lastuse)
436 struct mlx5_fc_cache c;
440 *bytes = c.bytes - counter->lastbytes;
441 *packets = c.packets - counter->lastpackets;
442 *lastuse = c.lastuse;
444 counter->lastbytes = c.bytes;
445 counter->lastpackets = c.packets;
448 void mlx5_fc_queue_stats_work(struct mlx5_core_dev *dev,
449 struct delayed_work *dwork,
452 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
454 queue_delayed_work(fc_stats->wq, dwork, delay);
457 void mlx5_fc_update_sampling_interval(struct mlx5_core_dev *dev,
458 unsigned long interval)
460 struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
462 fc_stats->sampling_interval = min_t(unsigned long, interval,
463 fc_stats->sampling_interval);
466 /* Flow counter bluks */
468 struct mlx5_fc_bulk {
469 struct list_head pool_list;
472 unsigned long *bitmask;
473 struct mlx5_fc fcs[];
476 static void mlx5_fc_init(struct mlx5_fc *counter, struct mlx5_fc_bulk *bulk,
479 counter->bulk = bulk;
483 static int mlx5_fc_bulk_get_free_fcs_amount(struct mlx5_fc_bulk *bulk)
485 return bitmap_weight(bulk->bitmask, bulk->bulk_len);
488 static struct mlx5_fc_bulk *mlx5_fc_bulk_create(struct mlx5_core_dev *dev)
490 enum mlx5_fc_bulk_alloc_bitmask alloc_bitmask;
491 struct mlx5_fc_bulk *bulk;
497 alloc_bitmask = MLX5_CAP_GEN(dev, flow_counter_bulk_alloc);
498 bulk_len = alloc_bitmask > 0 ? MLX5_FC_BULK_NUM_FCS(alloc_bitmask) : 1;
500 bulk = kzalloc(sizeof(*bulk) + bulk_len * sizeof(struct mlx5_fc),
505 bulk->bitmask = kcalloc(BITS_TO_LONGS(bulk_len), sizeof(unsigned long),
508 goto err_alloc_bitmask;
510 err = mlx5_cmd_fc_bulk_alloc(dev, alloc_bitmask, &base_id);
512 goto err_mlx5_cmd_bulk_alloc;
514 bulk->base_id = base_id;
515 bulk->bulk_len = bulk_len;
516 for (i = 0; i < bulk_len; i++) {
517 mlx5_fc_init(&bulk->fcs[i], bulk, base_id + i);
518 set_bit(i, bulk->bitmask);
523 err_mlx5_cmd_bulk_alloc:
524 kfree(bulk->bitmask);
532 mlx5_fc_bulk_destroy(struct mlx5_core_dev *dev, struct mlx5_fc_bulk *bulk)
534 if (mlx5_fc_bulk_get_free_fcs_amount(bulk) < bulk->bulk_len) {
535 mlx5_core_err(dev, "Freeing bulk before all counters were released\n");
539 mlx5_cmd_fc_free(dev, bulk->base_id);
540 kfree(bulk->bitmask);
546 static struct mlx5_fc *mlx5_fc_bulk_acquire_fc(struct mlx5_fc_bulk *bulk)
548 int free_fc_index = find_first_bit(bulk->bitmask, bulk->bulk_len);
550 if (free_fc_index >= bulk->bulk_len)
551 return ERR_PTR(-ENOSPC);
553 clear_bit(free_fc_index, bulk->bitmask);
554 return &bulk->fcs[free_fc_index];
557 static int mlx5_fc_bulk_release_fc(struct mlx5_fc_bulk *bulk, struct mlx5_fc *fc)
559 int fc_index = fc->id - bulk->base_id;
561 if (test_bit(fc_index, bulk->bitmask))
564 set_bit(fc_index, bulk->bitmask);
568 /* Flow counters pool API */
570 static void mlx5_fc_pool_init(struct mlx5_fc_pool *fc_pool, struct mlx5_core_dev *dev)
573 mutex_init(&fc_pool->pool_lock);
574 INIT_LIST_HEAD(&fc_pool->fully_used);
575 INIT_LIST_HEAD(&fc_pool->partially_used);
576 INIT_LIST_HEAD(&fc_pool->unused);
577 fc_pool->available_fcs = 0;
578 fc_pool->used_fcs = 0;
579 fc_pool->threshold = 0;
582 static void mlx5_fc_pool_cleanup(struct mlx5_fc_pool *fc_pool)
584 struct mlx5_core_dev *dev = fc_pool->dev;
585 struct mlx5_fc_bulk *bulk;
586 struct mlx5_fc_bulk *tmp;
588 list_for_each_entry_safe(bulk, tmp, &fc_pool->fully_used, pool_list)
589 mlx5_fc_bulk_destroy(dev, bulk);
590 list_for_each_entry_safe(bulk, tmp, &fc_pool->partially_used, pool_list)
591 mlx5_fc_bulk_destroy(dev, bulk);
592 list_for_each_entry_safe(bulk, tmp, &fc_pool->unused, pool_list)
593 mlx5_fc_bulk_destroy(dev, bulk);
596 static void mlx5_fc_pool_update_threshold(struct mlx5_fc_pool *fc_pool)
598 fc_pool->threshold = min_t(int, MLX5_FC_POOL_MAX_THRESHOLD,
599 fc_pool->used_fcs / MLX5_FC_POOL_USED_BUFF_RATIO);
602 static struct mlx5_fc_bulk *
603 mlx5_fc_pool_alloc_new_bulk(struct mlx5_fc_pool *fc_pool)
605 struct mlx5_core_dev *dev = fc_pool->dev;
606 struct mlx5_fc_bulk *new_bulk;
608 new_bulk = mlx5_fc_bulk_create(dev);
609 if (!IS_ERR(new_bulk))
610 fc_pool->available_fcs += new_bulk->bulk_len;
611 mlx5_fc_pool_update_threshold(fc_pool);
616 mlx5_fc_pool_free_bulk(struct mlx5_fc_pool *fc_pool, struct mlx5_fc_bulk *bulk)
618 struct mlx5_core_dev *dev = fc_pool->dev;
620 fc_pool->available_fcs -= bulk->bulk_len;
621 mlx5_fc_bulk_destroy(dev, bulk);
622 mlx5_fc_pool_update_threshold(fc_pool);
625 static struct mlx5_fc *
626 mlx5_fc_pool_acquire_from_list(struct list_head *src_list,
627 struct list_head *next_list,
628 bool move_non_full_bulk)
630 struct mlx5_fc_bulk *bulk;
633 if (list_empty(src_list))
634 return ERR_PTR(-ENODATA);
636 bulk = list_first_entry(src_list, struct mlx5_fc_bulk, pool_list);
637 fc = mlx5_fc_bulk_acquire_fc(bulk);
638 if (move_non_full_bulk || mlx5_fc_bulk_get_free_fcs_amount(bulk) == 0)
639 list_move(&bulk->pool_list, next_list);
643 static struct mlx5_fc *
644 mlx5_fc_pool_acquire_counter(struct mlx5_fc_pool *fc_pool)
646 struct mlx5_fc_bulk *new_bulk;
649 mutex_lock(&fc_pool->pool_lock);
651 fc = mlx5_fc_pool_acquire_from_list(&fc_pool->partially_used,
652 &fc_pool->fully_used, false);
654 fc = mlx5_fc_pool_acquire_from_list(&fc_pool->unused,
655 &fc_pool->partially_used,
658 new_bulk = mlx5_fc_pool_alloc_new_bulk(fc_pool);
659 if (IS_ERR(new_bulk)) {
660 fc = ERR_CAST(new_bulk);
663 fc = mlx5_fc_bulk_acquire_fc(new_bulk);
664 list_add(&new_bulk->pool_list, &fc_pool->partially_used);
666 fc_pool->available_fcs--;
670 mutex_unlock(&fc_pool->pool_lock);
675 mlx5_fc_pool_release_counter(struct mlx5_fc_pool *fc_pool, struct mlx5_fc *fc)
677 struct mlx5_core_dev *dev = fc_pool->dev;
678 struct mlx5_fc_bulk *bulk = fc->bulk;
679 int bulk_free_fcs_amount;
681 mutex_lock(&fc_pool->pool_lock);
683 if (mlx5_fc_bulk_release_fc(bulk, fc)) {
684 mlx5_core_warn(dev, "Attempted to release a counter which is not acquired\n");
688 fc_pool->available_fcs++;
691 bulk_free_fcs_amount = mlx5_fc_bulk_get_free_fcs_amount(bulk);
692 if (bulk_free_fcs_amount == 1)
693 list_move_tail(&bulk->pool_list, &fc_pool->partially_used);
694 if (bulk_free_fcs_amount == bulk->bulk_len) {
695 list_del(&bulk->pool_list);
696 if (fc_pool->available_fcs > fc_pool->threshold)
697 mlx5_fc_pool_free_bulk(fc_pool, bulk);
699 list_add(&bulk->pool_list, &fc_pool->unused);
703 mutex_unlock(&fc_pool->pool_lock);