xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c (revision bf5802238dc181b1f7375d358af1d01cd72d1c11)
1 /*
2  * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/mlx5/driver.h>
34 #include <linux/mlx5/fs.h>
35 #include <linux/rbtree.h>
36 #include "mlx5_core.h"
37 #include "fs_core.h"
38 #include "fs_cmd.h"
39 
40 #define MLX5_FC_STATS_PERIOD msecs_to_jiffies(1000)
41 #define MLX5_FC_BULK_QUERY_ALLOC_PERIOD msecs_to_jiffies(180 * 1000)
42 /* Max number of counters to query in bulk read is 32K */
43 #define MLX5_SW_MAX_COUNTERS_BULK BIT(15)
44 #define MLX5_INIT_COUNTERS_BULK 8
45 #define MLX5_FC_POOL_MAX_THRESHOLD BIT(18)
46 #define MLX5_FC_POOL_USED_BUFF_RATIO 10
47 
48 struct mlx5_fc_cache {
49 	u64 packets;
50 	u64 bytes;
51 	u64 lastuse;
52 };
53 
54 struct mlx5_fc {
55 	struct list_head list;
56 	struct llist_node addlist;
57 	struct llist_node dellist;
58 
59 	/* last{packets,bytes} members are used when calculating the delta since
60 	 * last reading
61 	 */
62 	u64 lastpackets;
63 	u64 lastbytes;
64 
65 	struct mlx5_fc_bulk *bulk;
66 	u32 id;
67 	bool aging;
68 
69 	struct mlx5_fc_cache cache ____cacheline_aligned_in_smp;
70 };
71 
72 static void mlx5_fc_pool_init(struct mlx5_fc_pool *fc_pool, struct mlx5_core_dev *dev);
73 static void mlx5_fc_pool_cleanup(struct mlx5_fc_pool *fc_pool);
74 static struct mlx5_fc *mlx5_fc_pool_acquire_counter(struct mlx5_fc_pool *fc_pool);
75 static void mlx5_fc_pool_release_counter(struct mlx5_fc_pool *fc_pool, struct mlx5_fc *fc);
76 
77 /* locking scheme:
78  *
79  * It is the responsibility of the user to prevent concurrent calls or bad
80  * ordering to mlx5_fc_create(), mlx5_fc_destroy() and accessing a reference
81  * to struct mlx5_fc.
82  * e.g en_tc.c is protected by RTNL lock of its caller, and will never call a
83  * dump (access to struct mlx5_fc) after a counter is destroyed.
84  *
85  * access to counter list:
86  * - create (user context)
87  *   - mlx5_fc_create() only adds to an addlist to be used by
88  *     mlx5_fc_stats_work(). addlist is a lockless single linked list
89  *     that doesn't require any additional synchronization when adding single
90  *     node.
91  *   - spawn thread to do the actual destroy
92  *
93  * - destroy (user context)
94  *   - add a counter to lockless dellist
95  *   - spawn thread to do the actual del
96  *
97  * - dump (user context)
98  *   user should not call dump after destroy
99  *
100  * - query (single thread workqueue context)
101  *   destroy/dump - no conflict (see destroy)
102  *   query/dump - packets and bytes might be inconsistent (since update is not
103  *                atomic)
104  *   query/create - no conflict (see create)
105  *   since every create/destroy spawn the work, only after necessary time has
106  *   elapsed, the thread will actually query the hardware.
107  */
108 
109 static struct list_head *mlx5_fc_counters_lookup_next(struct mlx5_core_dev *dev,
110 						      u32 id)
111 {
112 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
113 	unsigned long next_id = (unsigned long)id + 1;
114 	struct mlx5_fc *counter;
115 	unsigned long tmp;
116 
117 	rcu_read_lock();
118 	/* skip counters that are in idr, but not yet in counters list */
119 	idr_for_each_entry_continue_ul(&fc_stats->counters_idr,
120 				       counter, tmp, next_id) {
121 		if (!list_empty(&counter->list))
122 			break;
123 	}
124 	rcu_read_unlock();
125 
126 	return counter ? &counter->list : &fc_stats->counters;
127 }
128 
129 static void mlx5_fc_stats_insert(struct mlx5_core_dev *dev,
130 				 struct mlx5_fc *counter)
131 {
132 	struct list_head *next = mlx5_fc_counters_lookup_next(dev, counter->id);
133 
134 	list_add_tail(&counter->list, next);
135 }
136 
137 static void mlx5_fc_stats_remove(struct mlx5_core_dev *dev,
138 				 struct mlx5_fc *counter)
139 {
140 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
141 
142 	list_del(&counter->list);
143 
144 	spin_lock(&fc_stats->counters_idr_lock);
145 	WARN_ON(!idr_remove(&fc_stats->counters_idr, counter->id));
146 	spin_unlock(&fc_stats->counters_idr_lock);
147 }
148 
149 static int get_init_bulk_query_len(struct mlx5_core_dev *dev)
150 {
151 	return min_t(int, MLX5_INIT_COUNTERS_BULK,
152 		     (1 << MLX5_CAP_GEN(dev, log_max_flow_counter_bulk)));
153 }
154 
155 static int get_max_bulk_query_len(struct mlx5_core_dev *dev)
156 {
157 	return min_t(int, MLX5_SW_MAX_COUNTERS_BULK,
158 		     (1 << MLX5_CAP_GEN(dev, log_max_flow_counter_bulk)));
159 }
160 
161 static void update_counter_cache(int index, u32 *bulk_raw_data,
162 				 struct mlx5_fc_cache *cache)
163 {
164 	void *stats = MLX5_ADDR_OF(query_flow_counter_out, bulk_raw_data,
165 			     flow_statistics[index]);
166 	u64 packets = MLX5_GET64(traffic_counter, stats, packets);
167 	u64 bytes = MLX5_GET64(traffic_counter, stats, octets);
168 
169 	if (cache->packets == packets)
170 		return;
171 
172 	cache->packets = packets;
173 	cache->bytes = bytes;
174 	cache->lastuse = jiffies;
175 }
176 
177 static void mlx5_fc_stats_query_counter_range(struct mlx5_core_dev *dev,
178 					      struct mlx5_fc *first,
179 					      u32 last_id)
180 {
181 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
182 	bool query_more_counters = (first->id <= last_id);
183 	int cur_bulk_len = fc_stats->bulk_query_len;
184 	u32 *data = fc_stats->bulk_query_out;
185 	struct mlx5_fc *counter = first;
186 	u32 bulk_base_id;
187 	int bulk_len;
188 	int err;
189 
190 	while (query_more_counters) {
191 		/* first id must be aligned to 4 when using bulk query */
192 		bulk_base_id = counter->id & ~0x3;
193 
194 		/* number of counters to query inc. the last counter */
195 		bulk_len = min_t(int, cur_bulk_len,
196 				 ALIGN(last_id - bulk_base_id + 1, 4));
197 
198 		err = mlx5_cmd_fc_bulk_query(dev, bulk_base_id, bulk_len,
199 					     data);
200 		if (err) {
201 			mlx5_core_err(dev, "Error doing bulk query: %d\n", err);
202 			return;
203 		}
204 		query_more_counters = false;
205 
206 		list_for_each_entry_from(counter, &fc_stats->counters, list) {
207 			int counter_index = counter->id - bulk_base_id;
208 			struct mlx5_fc_cache *cache = &counter->cache;
209 
210 			if (counter->id >= bulk_base_id + bulk_len) {
211 				query_more_counters = true;
212 				break;
213 			}
214 
215 			update_counter_cache(counter_index, data, cache);
216 		}
217 	}
218 }
219 
220 static void mlx5_fc_free(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
221 {
222 	mlx5_cmd_fc_free(dev, counter->id);
223 	kfree(counter);
224 }
225 
226 static void mlx5_fc_release(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
227 {
228 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
229 
230 	if (counter->bulk)
231 		mlx5_fc_pool_release_counter(&fc_stats->fc_pool, counter);
232 	else
233 		mlx5_fc_free(dev, counter);
234 }
235 
236 static void mlx5_fc_stats_bulk_query_size_increase(struct mlx5_core_dev *dev)
237 {
238 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
239 	int max_bulk_len = get_max_bulk_query_len(dev);
240 	unsigned long now = jiffies;
241 	u32 *bulk_query_out_tmp;
242 	int max_out_len;
243 
244 	if (fc_stats->bulk_query_alloc_failed &&
245 	    time_before(now, fc_stats->next_bulk_query_alloc))
246 		return;
247 
248 	max_out_len = mlx5_cmd_fc_get_bulk_query_out_len(max_bulk_len);
249 	bulk_query_out_tmp = kzalloc(max_out_len, GFP_KERNEL);
250 	if (!bulk_query_out_tmp) {
251 		mlx5_core_warn_once(dev,
252 				    "Can't increase flow counters bulk query buffer size, insufficient memory, bulk_size(%d)\n",
253 				    max_bulk_len);
254 		fc_stats->bulk_query_alloc_failed = true;
255 		fc_stats->next_bulk_query_alloc =
256 			now + MLX5_FC_BULK_QUERY_ALLOC_PERIOD;
257 		return;
258 	}
259 
260 	kfree(fc_stats->bulk_query_out);
261 	fc_stats->bulk_query_out = bulk_query_out_tmp;
262 	fc_stats->bulk_query_len = max_bulk_len;
263 	if (fc_stats->bulk_query_alloc_failed) {
264 		mlx5_core_info(dev,
265 			       "Flow counters bulk query buffer size increased, bulk_size(%d)\n",
266 			       max_bulk_len);
267 		fc_stats->bulk_query_alloc_failed = false;
268 	}
269 }
270 
271 static void mlx5_fc_stats_work(struct work_struct *work)
272 {
273 	struct mlx5_core_dev *dev = container_of(work, struct mlx5_core_dev,
274 						 priv.fc_stats.work.work);
275 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
276 	/* Take dellist first to ensure that counters cannot be deleted before
277 	 * they are inserted.
278 	 */
279 	struct llist_node *dellist = llist_del_all(&fc_stats->dellist);
280 	struct llist_node *addlist = llist_del_all(&fc_stats->addlist);
281 	struct mlx5_fc *counter = NULL, *last = NULL, *tmp;
282 	unsigned long now = jiffies;
283 
284 	if (addlist || !list_empty(&fc_stats->counters))
285 		queue_delayed_work(fc_stats->wq, &fc_stats->work,
286 				   fc_stats->sampling_interval);
287 
288 	llist_for_each_entry(counter, addlist, addlist) {
289 		mlx5_fc_stats_insert(dev, counter);
290 		fc_stats->num_counters++;
291 	}
292 
293 	llist_for_each_entry_safe(counter, tmp, dellist, dellist) {
294 		mlx5_fc_stats_remove(dev, counter);
295 
296 		mlx5_fc_release(dev, counter);
297 		fc_stats->num_counters--;
298 	}
299 
300 	if (fc_stats->bulk_query_len < get_max_bulk_query_len(dev) &&
301 	    fc_stats->num_counters > get_init_bulk_query_len(dev))
302 		mlx5_fc_stats_bulk_query_size_increase(dev);
303 
304 	if (time_before(now, fc_stats->next_query) ||
305 	    list_empty(&fc_stats->counters))
306 		return;
307 	last = list_last_entry(&fc_stats->counters, struct mlx5_fc, list);
308 
309 	counter = list_first_entry(&fc_stats->counters, struct mlx5_fc,
310 				   list);
311 	if (counter)
312 		mlx5_fc_stats_query_counter_range(dev, counter, last->id);
313 
314 	fc_stats->next_query = now + fc_stats->sampling_interval;
315 }
316 
317 static struct mlx5_fc *mlx5_fc_single_alloc(struct mlx5_core_dev *dev)
318 {
319 	struct mlx5_fc *counter;
320 	int err;
321 
322 	counter = kzalloc(sizeof(*counter), GFP_KERNEL);
323 	if (!counter)
324 		return ERR_PTR(-ENOMEM);
325 
326 	err = mlx5_cmd_fc_alloc(dev, &counter->id);
327 	if (err) {
328 		kfree(counter);
329 		return ERR_PTR(err);
330 	}
331 
332 	return counter;
333 }
334 
335 static struct mlx5_fc *mlx5_fc_acquire(struct mlx5_core_dev *dev, bool aging)
336 {
337 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
338 	struct mlx5_fc *counter;
339 
340 	if (aging && MLX5_CAP_GEN(dev, flow_counter_bulk_alloc) != 0) {
341 		counter = mlx5_fc_pool_acquire_counter(&fc_stats->fc_pool);
342 		if (!IS_ERR(counter))
343 			return counter;
344 	}
345 
346 	return mlx5_fc_single_alloc(dev);
347 }
348 
349 struct mlx5_fc *mlx5_fc_create_ex(struct mlx5_core_dev *dev, bool aging)
350 {
351 	struct mlx5_fc *counter = mlx5_fc_acquire(dev, aging);
352 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
353 	int err;
354 
355 	if (IS_ERR(counter))
356 		return counter;
357 
358 	INIT_LIST_HEAD(&counter->list);
359 	counter->aging = aging;
360 
361 	if (aging) {
362 		u32 id = counter->id;
363 
364 		counter->cache.lastuse = jiffies;
365 		counter->lastbytes = counter->cache.bytes;
366 		counter->lastpackets = counter->cache.packets;
367 
368 		idr_preload(GFP_KERNEL);
369 		spin_lock(&fc_stats->counters_idr_lock);
370 
371 		err = idr_alloc_u32(&fc_stats->counters_idr, counter, &id, id,
372 				    GFP_NOWAIT);
373 
374 		spin_unlock(&fc_stats->counters_idr_lock);
375 		idr_preload_end();
376 		if (err)
377 			goto err_out_alloc;
378 
379 		llist_add(&counter->addlist, &fc_stats->addlist);
380 	}
381 
382 	return counter;
383 
384 err_out_alloc:
385 	mlx5_fc_release(dev, counter);
386 	return ERR_PTR(err);
387 }
388 
389 struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging)
390 {
391 	struct mlx5_fc *counter = mlx5_fc_create_ex(dev, aging);
392 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
393 
394 	if (aging)
395 		mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
396 	return counter;
397 }
398 EXPORT_SYMBOL(mlx5_fc_create);
399 
400 u32 mlx5_fc_id(struct mlx5_fc *counter)
401 {
402 	return counter->id;
403 }
404 EXPORT_SYMBOL(mlx5_fc_id);
405 
406 void mlx5_fc_destroy(struct mlx5_core_dev *dev, struct mlx5_fc *counter)
407 {
408 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
409 
410 	if (!counter)
411 		return;
412 
413 	if (counter->aging) {
414 		llist_add(&counter->dellist, &fc_stats->dellist);
415 		mod_delayed_work(fc_stats->wq, &fc_stats->work, 0);
416 		return;
417 	}
418 
419 	mlx5_fc_release(dev, counter);
420 }
421 EXPORT_SYMBOL(mlx5_fc_destroy);
422 
423 int mlx5_init_fc_stats(struct mlx5_core_dev *dev)
424 {
425 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
426 	int init_bulk_len;
427 	int init_out_len;
428 
429 	spin_lock_init(&fc_stats->counters_idr_lock);
430 	idr_init(&fc_stats->counters_idr);
431 	INIT_LIST_HEAD(&fc_stats->counters);
432 	init_llist_head(&fc_stats->addlist);
433 	init_llist_head(&fc_stats->dellist);
434 
435 	init_bulk_len = get_init_bulk_query_len(dev);
436 	init_out_len = mlx5_cmd_fc_get_bulk_query_out_len(init_bulk_len);
437 	fc_stats->bulk_query_out = kzalloc(init_out_len, GFP_KERNEL);
438 	if (!fc_stats->bulk_query_out)
439 		return -ENOMEM;
440 	fc_stats->bulk_query_len = init_bulk_len;
441 
442 	fc_stats->wq = create_singlethread_workqueue("mlx5_fc");
443 	if (!fc_stats->wq)
444 		goto err_wq_create;
445 
446 	fc_stats->sampling_interval = MLX5_FC_STATS_PERIOD;
447 	INIT_DELAYED_WORK(&fc_stats->work, mlx5_fc_stats_work);
448 
449 	mlx5_fc_pool_init(&fc_stats->fc_pool, dev);
450 	return 0;
451 
452 err_wq_create:
453 	kfree(fc_stats->bulk_query_out);
454 	return -ENOMEM;
455 }
456 
457 void mlx5_cleanup_fc_stats(struct mlx5_core_dev *dev)
458 {
459 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
460 	struct llist_node *tmplist;
461 	struct mlx5_fc *counter;
462 	struct mlx5_fc *tmp;
463 
464 	cancel_delayed_work_sync(&dev->priv.fc_stats.work);
465 	destroy_workqueue(dev->priv.fc_stats.wq);
466 	dev->priv.fc_stats.wq = NULL;
467 
468 	tmplist = llist_del_all(&fc_stats->addlist);
469 	llist_for_each_entry_safe(counter, tmp, tmplist, addlist)
470 		mlx5_fc_release(dev, counter);
471 
472 	list_for_each_entry_safe(counter, tmp, &fc_stats->counters, list)
473 		mlx5_fc_release(dev, counter);
474 
475 	mlx5_fc_pool_cleanup(&fc_stats->fc_pool);
476 	idr_destroy(&fc_stats->counters_idr);
477 	kfree(fc_stats->bulk_query_out);
478 }
479 
480 int mlx5_fc_query(struct mlx5_core_dev *dev, struct mlx5_fc *counter,
481 		  u64 *packets, u64 *bytes)
482 {
483 	return mlx5_cmd_fc_query(dev, counter->id, packets, bytes);
484 }
485 EXPORT_SYMBOL(mlx5_fc_query);
486 
487 u64 mlx5_fc_query_lastuse(struct mlx5_fc *counter)
488 {
489 	return counter->cache.lastuse;
490 }
491 
492 void mlx5_fc_query_cached(struct mlx5_fc *counter,
493 			  u64 *bytes, u64 *packets, u64 *lastuse)
494 {
495 	struct mlx5_fc_cache c;
496 
497 	c = counter->cache;
498 
499 	*bytes = c.bytes - counter->lastbytes;
500 	*packets = c.packets - counter->lastpackets;
501 	*lastuse = c.lastuse;
502 
503 	counter->lastbytes = c.bytes;
504 	counter->lastpackets = c.packets;
505 }
506 
507 void mlx5_fc_query_cached_raw(struct mlx5_fc *counter,
508 			      u64 *bytes, u64 *packets, u64 *lastuse)
509 {
510 	struct mlx5_fc_cache c = counter->cache;
511 
512 	*bytes = c.bytes;
513 	*packets = c.packets;
514 	*lastuse = c.lastuse;
515 }
516 
517 void mlx5_fc_queue_stats_work(struct mlx5_core_dev *dev,
518 			      struct delayed_work *dwork,
519 			      unsigned long delay)
520 {
521 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
522 
523 	queue_delayed_work(fc_stats->wq, dwork, delay);
524 }
525 
526 void mlx5_fc_update_sampling_interval(struct mlx5_core_dev *dev,
527 				      unsigned long interval)
528 {
529 	struct mlx5_fc_stats *fc_stats = &dev->priv.fc_stats;
530 
531 	fc_stats->sampling_interval = min_t(unsigned long, interval,
532 					    fc_stats->sampling_interval);
533 }
534 
535 /* Flow counter bluks */
536 
537 struct mlx5_fc_bulk {
538 	struct list_head pool_list;
539 	u32 base_id;
540 	int bulk_len;
541 	unsigned long *bitmask;
542 	struct mlx5_fc fcs[] __counted_by(bulk_len);
543 };
544 
545 static void mlx5_fc_init(struct mlx5_fc *counter, struct mlx5_fc_bulk *bulk,
546 			 u32 id)
547 {
548 	counter->bulk = bulk;
549 	counter->id = id;
550 }
551 
552 static int mlx5_fc_bulk_get_free_fcs_amount(struct mlx5_fc_bulk *bulk)
553 {
554 	return bitmap_weight(bulk->bitmask, bulk->bulk_len);
555 }
556 
557 static struct mlx5_fc_bulk *mlx5_fc_bulk_create(struct mlx5_core_dev *dev)
558 {
559 	enum mlx5_fc_bulk_alloc_bitmask alloc_bitmask;
560 	struct mlx5_fc_bulk *bulk;
561 	int err = -ENOMEM;
562 	int bulk_len;
563 	u32 base_id;
564 	int i;
565 
566 	alloc_bitmask = MLX5_CAP_GEN(dev, flow_counter_bulk_alloc);
567 	bulk_len = alloc_bitmask > 0 ? MLX5_FC_BULK_NUM_FCS(alloc_bitmask) : 1;
568 
569 	bulk = kvzalloc(struct_size(bulk, fcs, bulk_len), GFP_KERNEL);
570 	if (!bulk)
571 		goto err_alloc_bulk;
572 
573 	bulk->bitmask = kvcalloc(BITS_TO_LONGS(bulk_len), sizeof(unsigned long),
574 				 GFP_KERNEL);
575 	if (!bulk->bitmask)
576 		goto err_alloc_bitmask;
577 
578 	err = mlx5_cmd_fc_bulk_alloc(dev, alloc_bitmask, &base_id);
579 	if (err)
580 		goto err_mlx5_cmd_bulk_alloc;
581 
582 	bulk->base_id = base_id;
583 	bulk->bulk_len = bulk_len;
584 	for (i = 0; i < bulk_len; i++) {
585 		mlx5_fc_init(&bulk->fcs[i], bulk, base_id + i);
586 		set_bit(i, bulk->bitmask);
587 	}
588 
589 	return bulk;
590 
591 err_mlx5_cmd_bulk_alloc:
592 	kvfree(bulk->bitmask);
593 err_alloc_bitmask:
594 	kvfree(bulk);
595 err_alloc_bulk:
596 	return ERR_PTR(err);
597 }
598 
599 static int
600 mlx5_fc_bulk_destroy(struct mlx5_core_dev *dev, struct mlx5_fc_bulk *bulk)
601 {
602 	if (mlx5_fc_bulk_get_free_fcs_amount(bulk) < bulk->bulk_len) {
603 		mlx5_core_err(dev, "Freeing bulk before all counters were released\n");
604 		return -EBUSY;
605 	}
606 
607 	mlx5_cmd_fc_free(dev, bulk->base_id);
608 	kvfree(bulk->bitmask);
609 	kvfree(bulk);
610 
611 	return 0;
612 }
613 
614 static struct mlx5_fc *mlx5_fc_bulk_acquire_fc(struct mlx5_fc_bulk *bulk)
615 {
616 	int free_fc_index = find_first_bit(bulk->bitmask, bulk->bulk_len);
617 
618 	if (free_fc_index >= bulk->bulk_len)
619 		return ERR_PTR(-ENOSPC);
620 
621 	clear_bit(free_fc_index, bulk->bitmask);
622 	return &bulk->fcs[free_fc_index];
623 }
624 
625 static int mlx5_fc_bulk_release_fc(struct mlx5_fc_bulk *bulk, struct mlx5_fc *fc)
626 {
627 	int fc_index = fc->id - bulk->base_id;
628 
629 	if (test_bit(fc_index, bulk->bitmask))
630 		return -EINVAL;
631 
632 	set_bit(fc_index, bulk->bitmask);
633 	return 0;
634 }
635 
636 /* Flow counters pool API */
637 
638 static void mlx5_fc_pool_init(struct mlx5_fc_pool *fc_pool, struct mlx5_core_dev *dev)
639 {
640 	fc_pool->dev = dev;
641 	mutex_init(&fc_pool->pool_lock);
642 	INIT_LIST_HEAD(&fc_pool->fully_used);
643 	INIT_LIST_HEAD(&fc_pool->partially_used);
644 	INIT_LIST_HEAD(&fc_pool->unused);
645 	fc_pool->available_fcs = 0;
646 	fc_pool->used_fcs = 0;
647 	fc_pool->threshold = 0;
648 }
649 
650 static void mlx5_fc_pool_cleanup(struct mlx5_fc_pool *fc_pool)
651 {
652 	struct mlx5_core_dev *dev = fc_pool->dev;
653 	struct mlx5_fc_bulk *bulk;
654 	struct mlx5_fc_bulk *tmp;
655 
656 	list_for_each_entry_safe(bulk, tmp, &fc_pool->fully_used, pool_list)
657 		mlx5_fc_bulk_destroy(dev, bulk);
658 	list_for_each_entry_safe(bulk, tmp, &fc_pool->partially_used, pool_list)
659 		mlx5_fc_bulk_destroy(dev, bulk);
660 	list_for_each_entry_safe(bulk, tmp, &fc_pool->unused, pool_list)
661 		mlx5_fc_bulk_destroy(dev, bulk);
662 }
663 
664 static void mlx5_fc_pool_update_threshold(struct mlx5_fc_pool *fc_pool)
665 {
666 	fc_pool->threshold = min_t(int, MLX5_FC_POOL_MAX_THRESHOLD,
667 				   fc_pool->used_fcs / MLX5_FC_POOL_USED_BUFF_RATIO);
668 }
669 
670 static struct mlx5_fc_bulk *
671 mlx5_fc_pool_alloc_new_bulk(struct mlx5_fc_pool *fc_pool)
672 {
673 	struct mlx5_core_dev *dev = fc_pool->dev;
674 	struct mlx5_fc_bulk *new_bulk;
675 
676 	new_bulk = mlx5_fc_bulk_create(dev);
677 	if (!IS_ERR(new_bulk))
678 		fc_pool->available_fcs += new_bulk->bulk_len;
679 	mlx5_fc_pool_update_threshold(fc_pool);
680 	return new_bulk;
681 }
682 
683 static void
684 mlx5_fc_pool_free_bulk(struct mlx5_fc_pool *fc_pool, struct mlx5_fc_bulk *bulk)
685 {
686 	struct mlx5_core_dev *dev = fc_pool->dev;
687 
688 	fc_pool->available_fcs -= bulk->bulk_len;
689 	mlx5_fc_bulk_destroy(dev, bulk);
690 	mlx5_fc_pool_update_threshold(fc_pool);
691 }
692 
693 static struct mlx5_fc *
694 mlx5_fc_pool_acquire_from_list(struct list_head *src_list,
695 			       struct list_head *next_list,
696 			       bool move_non_full_bulk)
697 {
698 	struct mlx5_fc_bulk *bulk;
699 	struct mlx5_fc *fc;
700 
701 	if (list_empty(src_list))
702 		return ERR_PTR(-ENODATA);
703 
704 	bulk = list_first_entry(src_list, struct mlx5_fc_bulk, pool_list);
705 	fc = mlx5_fc_bulk_acquire_fc(bulk);
706 	if (move_non_full_bulk || mlx5_fc_bulk_get_free_fcs_amount(bulk) == 0)
707 		list_move(&bulk->pool_list, next_list);
708 	return fc;
709 }
710 
711 static struct mlx5_fc *
712 mlx5_fc_pool_acquire_counter(struct mlx5_fc_pool *fc_pool)
713 {
714 	struct mlx5_fc_bulk *new_bulk;
715 	struct mlx5_fc *fc;
716 
717 	mutex_lock(&fc_pool->pool_lock);
718 
719 	fc = mlx5_fc_pool_acquire_from_list(&fc_pool->partially_used,
720 					    &fc_pool->fully_used, false);
721 	if (IS_ERR(fc))
722 		fc = mlx5_fc_pool_acquire_from_list(&fc_pool->unused,
723 						    &fc_pool->partially_used,
724 						    true);
725 	if (IS_ERR(fc)) {
726 		new_bulk = mlx5_fc_pool_alloc_new_bulk(fc_pool);
727 		if (IS_ERR(new_bulk)) {
728 			fc = ERR_CAST(new_bulk);
729 			goto out;
730 		}
731 		fc = mlx5_fc_bulk_acquire_fc(new_bulk);
732 		list_add(&new_bulk->pool_list, &fc_pool->partially_used);
733 	}
734 	fc_pool->available_fcs--;
735 	fc_pool->used_fcs++;
736 
737 out:
738 	mutex_unlock(&fc_pool->pool_lock);
739 	return fc;
740 }
741 
742 static void
743 mlx5_fc_pool_release_counter(struct mlx5_fc_pool *fc_pool, struct mlx5_fc *fc)
744 {
745 	struct mlx5_core_dev *dev = fc_pool->dev;
746 	struct mlx5_fc_bulk *bulk = fc->bulk;
747 	int bulk_free_fcs_amount;
748 
749 	mutex_lock(&fc_pool->pool_lock);
750 
751 	if (mlx5_fc_bulk_release_fc(bulk, fc)) {
752 		mlx5_core_warn(dev, "Attempted to release a counter which is not acquired\n");
753 		goto unlock;
754 	}
755 
756 	fc_pool->available_fcs++;
757 	fc_pool->used_fcs--;
758 
759 	bulk_free_fcs_amount = mlx5_fc_bulk_get_free_fcs_amount(bulk);
760 	if (bulk_free_fcs_amount == 1)
761 		list_move_tail(&bulk->pool_list, &fc_pool->partially_used);
762 	if (bulk_free_fcs_amount == bulk->bulk_len) {
763 		list_del(&bulk->pool_list);
764 		if (fc_pool->available_fcs > fc_pool->threshold)
765 			mlx5_fc_pool_free_bulk(fc_pool, bulk);
766 		else
767 			list_add(&bulk->pool_list, &fc_pool->unused);
768 	}
769 
770 unlock:
771 	mutex_unlock(&fc_pool->pool_lock);
772 }
773