1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. */
3 
4 #include "eswitch.h"
5 #include "esw/qos.h"
6 #include "en/port.h"
7 #define CREATE_TRACE_POINTS
8 #include "diag/qos_tracepoint.h"
9 
10 /* Minimum supported BW share value by the HW is 1 Mbit/sec */
11 #define MLX5_MIN_BW_SHARE 1
12 
13 #define MLX5_RATE_TO_BW_SHARE(rate, divider, limit) \
14 	min_t(u32, max_t(u32, DIV_ROUND_UP(rate, divider), MLX5_MIN_BW_SHARE), limit)
15 
16 struct mlx5_esw_rate_group {
17 	u32 tsar_ix;
18 	u32 max_rate;
19 	u32 min_rate;
20 	u32 bw_share;
21 	struct list_head list;
22 };
23 
esw_qos_tsar_config(struct mlx5_core_dev * dev,u32 * sched_ctx,u32 tsar_ix,u32 max_rate,u32 bw_share)24 static int esw_qos_tsar_config(struct mlx5_core_dev *dev, u32 *sched_ctx,
25 			       u32 tsar_ix, u32 max_rate, u32 bw_share)
26 {
27 	u32 bitmask = 0;
28 
29 	if (!MLX5_CAP_GEN(dev, qos) || !MLX5_CAP_QOS(dev, esw_scheduling))
30 		return -EOPNOTSUPP;
31 
32 	MLX5_SET(scheduling_context, sched_ctx, max_average_bw, max_rate);
33 	MLX5_SET(scheduling_context, sched_ctx, bw_share, bw_share);
34 	bitmask |= MODIFY_SCHEDULING_ELEMENT_IN_MODIFY_BITMASK_MAX_AVERAGE_BW;
35 	bitmask |= MODIFY_SCHEDULING_ELEMENT_IN_MODIFY_BITMASK_BW_SHARE;
36 
37 	return mlx5_modify_scheduling_element_cmd(dev,
38 						  SCHEDULING_HIERARCHY_E_SWITCH,
39 						  sched_ctx,
40 						  tsar_ix,
41 						  bitmask);
42 }
43 
esw_qos_group_config(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,u32 max_rate,u32 bw_share,struct netlink_ext_ack * extack)44 static int esw_qos_group_config(struct mlx5_eswitch *esw, struct mlx5_esw_rate_group *group,
45 				u32 max_rate, u32 bw_share, struct netlink_ext_ack *extack)
46 {
47 	u32 sched_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
48 	struct mlx5_core_dev *dev = esw->dev;
49 	int err;
50 
51 	err = esw_qos_tsar_config(dev, sched_ctx,
52 				  group->tsar_ix,
53 				  max_rate, bw_share);
54 	if (err)
55 		NL_SET_ERR_MSG_MOD(extack, "E-Switch modify group TSAR element failed");
56 
57 	trace_mlx5_esw_group_qos_config(dev, group, group->tsar_ix, bw_share, max_rate);
58 
59 	return err;
60 }
61 
esw_qos_vport_config(struct mlx5_eswitch * esw,struct mlx5_vport * vport,u32 max_rate,u32 bw_share,struct netlink_ext_ack * extack)62 static int esw_qos_vport_config(struct mlx5_eswitch *esw,
63 				struct mlx5_vport *vport,
64 				u32 max_rate, u32 bw_share,
65 				struct netlink_ext_ack *extack)
66 {
67 	u32 sched_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
68 	struct mlx5_core_dev *dev = esw->dev;
69 	int err;
70 
71 	if (!vport->qos.enabled)
72 		return -EIO;
73 
74 	err = esw_qos_tsar_config(dev, sched_ctx, vport->qos.esw_tsar_ix,
75 				  max_rate, bw_share);
76 	if (err) {
77 		esw_warn(esw->dev,
78 			 "E-Switch modify TSAR vport element failed (vport=%d,err=%d)\n",
79 			 vport->vport, err);
80 		NL_SET_ERR_MSG_MOD(extack, "E-Switch modify TSAR vport element failed");
81 		return err;
82 	}
83 
84 	trace_mlx5_esw_vport_qos_config(vport, bw_share, max_rate);
85 
86 	return 0;
87 }
88 
esw_qos_calculate_min_rate_divider(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,bool group_level)89 static u32 esw_qos_calculate_min_rate_divider(struct mlx5_eswitch *esw,
90 					      struct mlx5_esw_rate_group *group,
91 					      bool group_level)
92 {
93 	u32 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
94 	struct mlx5_vport *evport;
95 	u32 max_guarantee = 0;
96 	unsigned long i;
97 
98 	if (group_level) {
99 		struct mlx5_esw_rate_group *group;
100 
101 		list_for_each_entry(group, &esw->qos.groups, list) {
102 			if (group->min_rate < max_guarantee)
103 				continue;
104 			max_guarantee = group->min_rate;
105 		}
106 	} else {
107 		mlx5_esw_for_each_vport(esw, i, evport) {
108 			if (!evport->enabled || !evport->qos.enabled ||
109 			    evport->qos.group != group || evport->qos.min_rate < max_guarantee)
110 				continue;
111 			max_guarantee = evport->qos.min_rate;
112 		}
113 	}
114 
115 	if (max_guarantee)
116 		return max_t(u32, max_guarantee / fw_max_bw_share, 1);
117 
118 	/* If vports min rate divider is 0 but their group has bw_share configured, then
119 	 * need to set bw_share for vports to minimal value.
120 	 */
121 	if (!group_level && !max_guarantee && group && group->bw_share)
122 		return 1;
123 	return 0;
124 }
125 
esw_qos_calc_bw_share(u32 min_rate,u32 divider,u32 fw_max)126 static u32 esw_qos_calc_bw_share(u32 min_rate, u32 divider, u32 fw_max)
127 {
128 	if (divider)
129 		return MLX5_RATE_TO_BW_SHARE(min_rate, divider, fw_max);
130 
131 	return 0;
132 }
133 
esw_qos_normalize_vports_min_rate(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,struct netlink_ext_ack * extack)134 static int esw_qos_normalize_vports_min_rate(struct mlx5_eswitch *esw,
135 					     struct mlx5_esw_rate_group *group,
136 					     struct netlink_ext_ack *extack)
137 {
138 	u32 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
139 	u32 divider = esw_qos_calculate_min_rate_divider(esw, group, false);
140 	struct mlx5_vport *evport;
141 	unsigned long i;
142 	u32 bw_share;
143 	int err;
144 
145 	mlx5_esw_for_each_vport(esw, i, evport) {
146 		if (!evport->enabled || !evport->qos.enabled || evport->qos.group != group)
147 			continue;
148 		bw_share = esw_qos_calc_bw_share(evport->qos.min_rate, divider, fw_max_bw_share);
149 
150 		if (bw_share == evport->qos.bw_share)
151 			continue;
152 
153 		err = esw_qos_vport_config(esw, evport, evport->qos.max_rate, bw_share, extack);
154 		if (err)
155 			return err;
156 
157 		evport->qos.bw_share = bw_share;
158 	}
159 
160 	return 0;
161 }
162 
esw_qos_normalize_groups_min_rate(struct mlx5_eswitch * esw,u32 divider,struct netlink_ext_ack * extack)163 static int esw_qos_normalize_groups_min_rate(struct mlx5_eswitch *esw, u32 divider,
164 					     struct netlink_ext_ack *extack)
165 {
166 	u32 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
167 	struct mlx5_esw_rate_group *group;
168 	u32 bw_share;
169 	int err;
170 
171 	list_for_each_entry(group, &esw->qos.groups, list) {
172 		bw_share = esw_qos_calc_bw_share(group->min_rate, divider, fw_max_bw_share);
173 
174 		if (bw_share == group->bw_share)
175 			continue;
176 
177 		err = esw_qos_group_config(esw, group, group->max_rate, bw_share, extack);
178 		if (err)
179 			return err;
180 
181 		group->bw_share = bw_share;
182 
183 		/* All the group's vports need to be set with default bw_share
184 		 * to enable them with QOS
185 		 */
186 		err = esw_qos_normalize_vports_min_rate(esw, group, extack);
187 
188 		if (err)
189 			return err;
190 	}
191 
192 	return 0;
193 }
194 
esw_qos_set_vport_min_rate(struct mlx5_eswitch * esw,struct mlx5_vport * evport,u32 min_rate,struct netlink_ext_ack * extack)195 static int esw_qos_set_vport_min_rate(struct mlx5_eswitch *esw, struct mlx5_vport *evport,
196 				      u32 min_rate, struct netlink_ext_ack *extack)
197 {
198 	u32 fw_max_bw_share, previous_min_rate;
199 	bool min_rate_supported;
200 	int err;
201 
202 	lockdep_assert_held(&esw->state_lock);
203 	fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
204 	min_rate_supported = MLX5_CAP_QOS(esw->dev, esw_bw_share) &&
205 				fw_max_bw_share >= MLX5_MIN_BW_SHARE;
206 	if (min_rate && !min_rate_supported)
207 		return -EOPNOTSUPP;
208 	if (min_rate == evport->qos.min_rate)
209 		return 0;
210 
211 	previous_min_rate = evport->qos.min_rate;
212 	evport->qos.min_rate = min_rate;
213 	err = esw_qos_normalize_vports_min_rate(esw, evport->qos.group, extack);
214 	if (err)
215 		evport->qos.min_rate = previous_min_rate;
216 
217 	return err;
218 }
219 
esw_qos_set_vport_max_rate(struct mlx5_eswitch * esw,struct mlx5_vport * evport,u32 max_rate,struct netlink_ext_ack * extack)220 static int esw_qos_set_vport_max_rate(struct mlx5_eswitch *esw, struct mlx5_vport *evport,
221 				      u32 max_rate, struct netlink_ext_ack *extack)
222 {
223 	u32 act_max_rate = max_rate;
224 	bool max_rate_supported;
225 	int err;
226 
227 	lockdep_assert_held(&esw->state_lock);
228 	max_rate_supported = MLX5_CAP_QOS(esw->dev, esw_rate_limit);
229 
230 	if (max_rate && !max_rate_supported)
231 		return -EOPNOTSUPP;
232 	if (max_rate == evport->qos.max_rate)
233 		return 0;
234 
235 	/* If parent group has rate limit need to set to group
236 	 * value when new max rate is 0.
237 	 */
238 	if (evport->qos.group && !max_rate)
239 		act_max_rate = evport->qos.group->max_rate;
240 
241 	err = esw_qos_vport_config(esw, evport, act_max_rate, evport->qos.bw_share, extack);
242 
243 	if (!err)
244 		evport->qos.max_rate = max_rate;
245 
246 	return err;
247 }
248 
esw_qos_set_group_min_rate(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,u32 min_rate,struct netlink_ext_ack * extack)249 static int esw_qos_set_group_min_rate(struct mlx5_eswitch *esw, struct mlx5_esw_rate_group *group,
250 				      u32 min_rate, struct netlink_ext_ack *extack)
251 {
252 	u32 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
253 	struct mlx5_core_dev *dev = esw->dev;
254 	u32 previous_min_rate, divider;
255 	int err;
256 
257 	if (!(MLX5_CAP_QOS(dev, esw_bw_share) && fw_max_bw_share >= MLX5_MIN_BW_SHARE))
258 		return -EOPNOTSUPP;
259 
260 	if (min_rate == group->min_rate)
261 		return 0;
262 
263 	previous_min_rate = group->min_rate;
264 	group->min_rate = min_rate;
265 	divider = esw_qos_calculate_min_rate_divider(esw, group, true);
266 	err = esw_qos_normalize_groups_min_rate(esw, divider, extack);
267 	if (err) {
268 		group->min_rate = previous_min_rate;
269 		NL_SET_ERR_MSG_MOD(extack, "E-Switch group min rate setting failed");
270 
271 		/* Attempt restoring previous configuration */
272 		divider = esw_qos_calculate_min_rate_divider(esw, group, true);
273 		if (esw_qos_normalize_groups_min_rate(esw, divider, extack))
274 			NL_SET_ERR_MSG_MOD(extack, "E-Switch BW share restore failed");
275 	}
276 
277 	return err;
278 }
279 
esw_qos_set_group_max_rate(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,u32 max_rate,struct netlink_ext_ack * extack)280 static int esw_qos_set_group_max_rate(struct mlx5_eswitch *esw,
281 				      struct mlx5_esw_rate_group *group,
282 				      u32 max_rate, struct netlink_ext_ack *extack)
283 {
284 	struct mlx5_vport *vport;
285 	unsigned long i;
286 	int err;
287 
288 	if (group->max_rate == max_rate)
289 		return 0;
290 
291 	err = esw_qos_group_config(esw, group, max_rate, group->bw_share, extack);
292 	if (err)
293 		return err;
294 
295 	group->max_rate = max_rate;
296 
297 	/* Any unlimited vports in the group should be set
298 	 * with the value of the group.
299 	 */
300 	mlx5_esw_for_each_vport(esw, i, vport) {
301 		if (!vport->enabled || !vport->qos.enabled ||
302 		    vport->qos.group != group || vport->qos.max_rate)
303 			continue;
304 
305 		err = esw_qos_vport_config(esw, vport, max_rate, vport->qos.bw_share, extack);
306 		if (err)
307 			NL_SET_ERR_MSG_MOD(extack,
308 					   "E-Switch vport implicit rate limit setting failed");
309 	}
310 
311 	return err;
312 }
313 
esw_qos_vport_create_sched_element(struct mlx5_eswitch * esw,struct mlx5_vport * vport,u32 max_rate,u32 bw_share)314 static int esw_qos_vport_create_sched_element(struct mlx5_eswitch *esw,
315 					      struct mlx5_vport *vport,
316 					      u32 max_rate, u32 bw_share)
317 {
318 	u32 sched_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
319 	struct mlx5_esw_rate_group *group = vport->qos.group;
320 	struct mlx5_core_dev *dev = esw->dev;
321 	u32 parent_tsar_ix;
322 	void *vport_elem;
323 	int err;
324 
325 	parent_tsar_ix = group ? group->tsar_ix : esw->qos.root_tsar_ix;
326 	MLX5_SET(scheduling_context, sched_ctx, element_type,
327 		 SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT);
328 	vport_elem = MLX5_ADDR_OF(scheduling_context, sched_ctx, element_attributes);
329 	MLX5_SET(vport_element, vport_elem, vport_number, vport->vport);
330 	MLX5_SET(scheduling_context, sched_ctx, parent_element_id, parent_tsar_ix);
331 	MLX5_SET(scheduling_context, sched_ctx, max_average_bw, max_rate);
332 	MLX5_SET(scheduling_context, sched_ctx, bw_share, bw_share);
333 
334 	err = mlx5_create_scheduling_element_cmd(dev,
335 						 SCHEDULING_HIERARCHY_E_SWITCH,
336 						 sched_ctx,
337 						 &vport->qos.esw_tsar_ix);
338 	if (err) {
339 		esw_warn(esw->dev, "E-Switch create TSAR vport element failed (vport=%d,err=%d)\n",
340 			 vport->vport, err);
341 		return err;
342 	}
343 
344 	return 0;
345 }
346 
esw_qos_update_group_scheduling_element(struct mlx5_eswitch * esw,struct mlx5_vport * vport,struct mlx5_esw_rate_group * curr_group,struct mlx5_esw_rate_group * new_group,struct netlink_ext_ack * extack)347 static int esw_qos_update_group_scheduling_element(struct mlx5_eswitch *esw,
348 						   struct mlx5_vport *vport,
349 						   struct mlx5_esw_rate_group *curr_group,
350 						   struct mlx5_esw_rate_group *new_group,
351 						   struct netlink_ext_ack *extack)
352 {
353 	u32 max_rate;
354 	int err;
355 
356 	err = mlx5_destroy_scheduling_element_cmd(esw->dev,
357 						  SCHEDULING_HIERARCHY_E_SWITCH,
358 						  vport->qos.esw_tsar_ix);
359 	if (err) {
360 		NL_SET_ERR_MSG_MOD(extack, "E-Switch destroy TSAR vport element failed");
361 		return err;
362 	}
363 
364 	vport->qos.group = new_group;
365 	max_rate = vport->qos.max_rate ? vport->qos.max_rate : new_group->max_rate;
366 
367 	/* If vport is unlimited, we set the group's value.
368 	 * Therefore, if the group is limited it will apply to
369 	 * the vport as well and if not, vport will remain unlimited.
370 	 */
371 	err = esw_qos_vport_create_sched_element(esw, vport, max_rate, vport->qos.bw_share);
372 	if (err) {
373 		NL_SET_ERR_MSG_MOD(extack, "E-Switch vport group set failed.");
374 		goto err_sched;
375 	}
376 
377 	return 0;
378 
379 err_sched:
380 	vport->qos.group = curr_group;
381 	max_rate = vport->qos.max_rate ? vport->qos.max_rate : curr_group->max_rate;
382 	if (esw_qos_vport_create_sched_element(esw, vport, max_rate, vport->qos.bw_share))
383 		esw_warn(esw->dev, "E-Switch vport group restore failed (vport=%d)\n",
384 			 vport->vport);
385 
386 	return err;
387 }
388 
esw_qos_vport_update_group(struct mlx5_eswitch * esw,struct mlx5_vport * vport,struct mlx5_esw_rate_group * group,struct netlink_ext_ack * extack)389 static int esw_qos_vport_update_group(struct mlx5_eswitch *esw,
390 				      struct mlx5_vport *vport,
391 				      struct mlx5_esw_rate_group *group,
392 				      struct netlink_ext_ack *extack)
393 {
394 	struct mlx5_esw_rate_group *new_group, *curr_group;
395 	int err;
396 
397 	if (!vport->enabled)
398 		return -EINVAL;
399 
400 	curr_group = vport->qos.group;
401 	new_group = group ?: esw->qos.group0;
402 	if (curr_group == new_group)
403 		return 0;
404 
405 	err = esw_qos_update_group_scheduling_element(esw, vport, curr_group, new_group, extack);
406 	if (err)
407 		return err;
408 
409 	/* Recalculate bw share weights of old and new groups */
410 	if (vport->qos.bw_share || new_group->bw_share) {
411 		esw_qos_normalize_vports_min_rate(esw, curr_group, extack);
412 		esw_qos_normalize_vports_min_rate(esw, new_group, extack);
413 	}
414 
415 	return 0;
416 }
417 
418 static struct mlx5_esw_rate_group *
__esw_qos_create_rate_group(struct mlx5_eswitch * esw,struct netlink_ext_ack * extack)419 __esw_qos_create_rate_group(struct mlx5_eswitch *esw, struct netlink_ext_ack *extack)
420 {
421 	u32 tsar_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
422 	struct mlx5_esw_rate_group *group;
423 	u32 divider;
424 	int err;
425 
426 	group = kzalloc(sizeof(*group), GFP_KERNEL);
427 	if (!group)
428 		return ERR_PTR(-ENOMEM);
429 
430 	MLX5_SET(scheduling_context, tsar_ctx, parent_element_id,
431 		 esw->qos.root_tsar_ix);
432 	err = mlx5_create_scheduling_element_cmd(esw->dev,
433 						 SCHEDULING_HIERARCHY_E_SWITCH,
434 						 tsar_ctx,
435 						 &group->tsar_ix);
436 	if (err) {
437 		NL_SET_ERR_MSG_MOD(extack, "E-Switch create TSAR for group failed");
438 		goto err_sched_elem;
439 	}
440 
441 	list_add_tail(&group->list, &esw->qos.groups);
442 
443 	divider = esw_qos_calculate_min_rate_divider(esw, group, true);
444 	if (divider) {
445 		err = esw_qos_normalize_groups_min_rate(esw, divider, extack);
446 		if (err) {
447 			NL_SET_ERR_MSG_MOD(extack, "E-Switch groups normalization failed");
448 			goto err_min_rate;
449 		}
450 	}
451 	trace_mlx5_esw_group_qos_create(esw->dev, group, group->tsar_ix);
452 
453 	return group;
454 
455 err_min_rate:
456 	list_del(&group->list);
457 	if (mlx5_destroy_scheduling_element_cmd(esw->dev,
458 						SCHEDULING_HIERARCHY_E_SWITCH,
459 						group->tsar_ix))
460 		NL_SET_ERR_MSG_MOD(extack, "E-Switch destroy TSAR for group failed");
461 err_sched_elem:
462 	kfree(group);
463 	return ERR_PTR(err);
464 }
465 
466 static int esw_qos_get(struct mlx5_eswitch *esw, struct netlink_ext_ack *extack);
467 static void esw_qos_put(struct mlx5_eswitch *esw);
468 
469 static struct mlx5_esw_rate_group *
esw_qos_create_rate_group(struct mlx5_eswitch * esw,struct netlink_ext_ack * extack)470 esw_qos_create_rate_group(struct mlx5_eswitch *esw, struct netlink_ext_ack *extack)
471 {
472 	struct mlx5_esw_rate_group *group;
473 	int err;
474 
475 	if (!MLX5_CAP_QOS(esw->dev, log_esw_max_sched_depth))
476 		return ERR_PTR(-EOPNOTSUPP);
477 
478 	err = esw_qos_get(esw, extack);
479 	if (err)
480 		return ERR_PTR(err);
481 
482 	group = __esw_qos_create_rate_group(esw, extack);
483 	if (IS_ERR(group))
484 		esw_qos_put(esw);
485 
486 	return group;
487 }
488 
__esw_qos_destroy_rate_group(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,struct netlink_ext_ack * extack)489 static int __esw_qos_destroy_rate_group(struct mlx5_eswitch *esw,
490 					struct mlx5_esw_rate_group *group,
491 					struct netlink_ext_ack *extack)
492 {
493 	u32 divider;
494 	int err;
495 
496 	list_del(&group->list);
497 
498 	divider = esw_qos_calculate_min_rate_divider(esw, NULL, true);
499 	err = esw_qos_normalize_groups_min_rate(esw, divider, extack);
500 	if (err)
501 		NL_SET_ERR_MSG_MOD(extack, "E-Switch groups' normalization failed");
502 
503 	err = mlx5_destroy_scheduling_element_cmd(esw->dev,
504 						  SCHEDULING_HIERARCHY_E_SWITCH,
505 						  group->tsar_ix);
506 	if (err)
507 		NL_SET_ERR_MSG_MOD(extack, "E-Switch destroy TSAR_ID failed");
508 
509 	trace_mlx5_esw_group_qos_destroy(esw->dev, group, group->tsar_ix);
510 
511 	kfree(group);
512 
513 	return err;
514 }
515 
esw_qos_destroy_rate_group(struct mlx5_eswitch * esw,struct mlx5_esw_rate_group * group,struct netlink_ext_ack * extack)516 static int esw_qos_destroy_rate_group(struct mlx5_eswitch *esw,
517 				      struct mlx5_esw_rate_group *group,
518 				      struct netlink_ext_ack *extack)
519 {
520 	int err;
521 
522 	err = __esw_qos_destroy_rate_group(esw, group, extack);
523 	esw_qos_put(esw);
524 
525 	return err;
526 }
527 
esw_qos_element_type_supported(struct mlx5_core_dev * dev,int type)528 static bool esw_qos_element_type_supported(struct mlx5_core_dev *dev, int type)
529 {
530 	switch (type) {
531 	case SCHEDULING_CONTEXT_ELEMENT_TYPE_TSAR:
532 		return MLX5_CAP_QOS(dev, esw_element_type) &
533 		       ELEMENT_TYPE_CAP_MASK_TASR;
534 	case SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT:
535 		return MLX5_CAP_QOS(dev, esw_element_type) &
536 		       ELEMENT_TYPE_CAP_MASK_VPORT;
537 	case SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT_TC:
538 		return MLX5_CAP_QOS(dev, esw_element_type) &
539 		       ELEMENT_TYPE_CAP_MASK_VPORT_TC;
540 	case SCHEDULING_CONTEXT_ELEMENT_TYPE_PARA_VPORT_TC:
541 		return MLX5_CAP_QOS(dev, esw_element_type) &
542 		       ELEMENT_TYPE_CAP_MASK_PARA_VPORT_TC;
543 	}
544 	return false;
545 }
546 
esw_qos_create(struct mlx5_eswitch * esw,struct netlink_ext_ack * extack)547 static int esw_qos_create(struct mlx5_eswitch *esw, struct netlink_ext_ack *extack)
548 {
549 	u32 tsar_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
550 	struct mlx5_core_dev *dev = esw->dev;
551 	__be32 *attr;
552 	int err;
553 
554 	if (!MLX5_CAP_GEN(dev, qos) || !MLX5_CAP_QOS(dev, esw_scheduling))
555 		return -EOPNOTSUPP;
556 
557 	if (!esw_qos_element_type_supported(dev, SCHEDULING_CONTEXT_ELEMENT_TYPE_TSAR))
558 		return -EOPNOTSUPP;
559 
560 	MLX5_SET(scheduling_context, tsar_ctx, element_type,
561 		 SCHEDULING_CONTEXT_ELEMENT_TYPE_TSAR);
562 
563 	attr = MLX5_ADDR_OF(scheduling_context, tsar_ctx, element_attributes);
564 	*attr = cpu_to_be32(TSAR_ELEMENT_TSAR_TYPE_DWRR << 16);
565 
566 	err = mlx5_create_scheduling_element_cmd(dev,
567 						 SCHEDULING_HIERARCHY_E_SWITCH,
568 						 tsar_ctx,
569 						 &esw->qos.root_tsar_ix);
570 	if (err) {
571 		esw_warn(dev, "E-Switch create root TSAR failed (%d)\n", err);
572 		return err;
573 	}
574 
575 	INIT_LIST_HEAD(&esw->qos.groups);
576 	if (MLX5_CAP_QOS(dev, log_esw_max_sched_depth)) {
577 		esw->qos.group0 = __esw_qos_create_rate_group(esw, extack);
578 		if (IS_ERR(esw->qos.group0)) {
579 			esw_warn(dev, "E-Switch create rate group 0 failed (%ld)\n",
580 				 PTR_ERR(esw->qos.group0));
581 			err = PTR_ERR(esw->qos.group0);
582 			goto err_group0;
583 		}
584 	}
585 	refcount_set(&esw->qos.refcnt, 1);
586 
587 	return 0;
588 
589 err_group0:
590 	if (mlx5_destroy_scheduling_element_cmd(esw->dev, SCHEDULING_HIERARCHY_E_SWITCH,
591 						esw->qos.root_tsar_ix))
592 		esw_warn(esw->dev, "E-Switch destroy root TSAR failed.\n");
593 
594 	return err;
595 }
596 
esw_qos_destroy(struct mlx5_eswitch * esw)597 static void esw_qos_destroy(struct mlx5_eswitch *esw)
598 {
599 	int err;
600 
601 	if (esw->qos.group0)
602 		__esw_qos_destroy_rate_group(esw, esw->qos.group0, NULL);
603 
604 	err = mlx5_destroy_scheduling_element_cmd(esw->dev,
605 						  SCHEDULING_HIERARCHY_E_SWITCH,
606 						  esw->qos.root_tsar_ix);
607 	if (err)
608 		esw_warn(esw->dev, "E-Switch destroy root TSAR failed (%d)\n", err);
609 }
610 
esw_qos_get(struct mlx5_eswitch * esw,struct netlink_ext_ack * extack)611 static int esw_qos_get(struct mlx5_eswitch *esw, struct netlink_ext_ack *extack)
612 {
613 	int err = 0;
614 
615 	lockdep_assert_held(&esw->state_lock);
616 
617 	if (!refcount_inc_not_zero(&esw->qos.refcnt)) {
618 		/* esw_qos_create() set refcount to 1 only on success.
619 		 * No need to decrement on failure.
620 		 */
621 		err = esw_qos_create(esw, extack);
622 	}
623 
624 	return err;
625 }
626 
esw_qos_put(struct mlx5_eswitch * esw)627 static void esw_qos_put(struct mlx5_eswitch *esw)
628 {
629 	lockdep_assert_held(&esw->state_lock);
630 	if (refcount_dec_and_test(&esw->qos.refcnt))
631 		esw_qos_destroy(esw);
632 }
633 
esw_qos_vport_enable(struct mlx5_eswitch * esw,struct mlx5_vport * vport,u32 max_rate,u32 bw_share,struct netlink_ext_ack * extack)634 static int esw_qos_vport_enable(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
635 				u32 max_rate, u32 bw_share, struct netlink_ext_ack *extack)
636 {
637 	int err;
638 
639 	lockdep_assert_held(&esw->state_lock);
640 	if (vport->qos.enabled)
641 		return 0;
642 
643 	err = esw_qos_get(esw, extack);
644 	if (err)
645 		return err;
646 
647 	vport->qos.group = esw->qos.group0;
648 
649 	err = esw_qos_vport_create_sched_element(esw, vport, max_rate, bw_share);
650 	if (err)
651 		goto err_out;
652 
653 	vport->qos.enabled = true;
654 	trace_mlx5_esw_vport_qos_create(vport, bw_share, max_rate);
655 
656 	return 0;
657 
658 err_out:
659 	esw_qos_put(esw);
660 
661 	return err;
662 }
663 
mlx5_esw_qos_vport_disable(struct mlx5_eswitch * esw,struct mlx5_vport * vport)664 void mlx5_esw_qos_vport_disable(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
665 {
666 	int err;
667 
668 	lockdep_assert_held(&esw->state_lock);
669 	if (!vport->qos.enabled)
670 		return;
671 	WARN(vport->qos.group && vport->qos.group != esw->qos.group0,
672 	     "Disabling QoS on port before detaching it from group");
673 
674 	err = mlx5_destroy_scheduling_element_cmd(esw->dev,
675 						  SCHEDULING_HIERARCHY_E_SWITCH,
676 						  vport->qos.esw_tsar_ix);
677 	if (err)
678 		esw_warn(esw->dev, "E-Switch destroy TSAR vport element failed (vport=%d,err=%d)\n",
679 			 vport->vport, err);
680 
681 	memset(&vport->qos, 0, sizeof(vport->qos));
682 	trace_mlx5_esw_vport_qos_destroy(vport);
683 
684 	esw_qos_put(esw);
685 }
686 
mlx5_esw_qos_set_vport_rate(struct mlx5_eswitch * esw,struct mlx5_vport * vport,u32 max_rate,u32 min_rate)687 int mlx5_esw_qos_set_vport_rate(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
688 				u32 max_rate, u32 min_rate)
689 {
690 	int err;
691 
692 	lockdep_assert_held(&esw->state_lock);
693 	err = esw_qos_vport_enable(esw, vport, 0, 0, NULL);
694 	if (err)
695 		return err;
696 
697 	err = esw_qos_set_vport_min_rate(esw, vport, min_rate, NULL);
698 	if (!err)
699 		err = esw_qos_set_vport_max_rate(esw, vport, max_rate, NULL);
700 
701 	return err;
702 }
703 
mlx5_esw_qos_modify_vport_rate(struct mlx5_eswitch * esw,u16 vport_num,u32 rate_mbps)704 int mlx5_esw_qos_modify_vport_rate(struct mlx5_eswitch *esw, u16 vport_num, u32 rate_mbps)
705 {
706 	u32 ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
707 	struct mlx5_vport *vport;
708 	u32 bitmask;
709 	int err;
710 
711 	vport = mlx5_eswitch_get_vport(esw, vport_num);
712 	if (IS_ERR(vport))
713 		return PTR_ERR(vport);
714 
715 	mutex_lock(&esw->state_lock);
716 	if (!vport->qos.enabled) {
717 		/* Eswitch QoS wasn't enabled yet. Enable it and vport QoS. */
718 		err = esw_qos_vport_enable(esw, vport, rate_mbps, vport->qos.bw_share, NULL);
719 	} else {
720 		MLX5_SET(scheduling_context, ctx, max_average_bw, rate_mbps);
721 
722 		bitmask = MODIFY_SCHEDULING_ELEMENT_IN_MODIFY_BITMASK_MAX_AVERAGE_BW;
723 		err = mlx5_modify_scheduling_element_cmd(esw->dev,
724 							 SCHEDULING_HIERARCHY_E_SWITCH,
725 							 ctx,
726 							 vport->qos.esw_tsar_ix,
727 							 bitmask);
728 	}
729 	mutex_unlock(&esw->state_lock);
730 
731 	return err;
732 }
733 
734 #define MLX5_LINKSPEED_UNIT 125000 /* 1Mbps in Bps */
735 
736 /* Converts bytes per second value passed in a pointer into megabits per
737  * second, rewriting last. If converted rate exceed link speed or is not a
738  * fraction of Mbps - returns error.
739  */
esw_qos_devlink_rate_to_mbps(struct mlx5_core_dev * mdev,const char * name,u64 * rate,struct netlink_ext_ack * extack)740 static int esw_qos_devlink_rate_to_mbps(struct mlx5_core_dev *mdev, const char *name,
741 					u64 *rate, struct netlink_ext_ack *extack)
742 {
743 	u32 link_speed_max, reminder;
744 	u64 value;
745 	int err;
746 
747 	err = mlx5e_port_max_linkspeed(mdev, &link_speed_max);
748 	if (err) {
749 		NL_SET_ERR_MSG_MOD(extack, "Failed to get link maximum speed");
750 		return err;
751 	}
752 
753 	value = div_u64_rem(*rate, MLX5_LINKSPEED_UNIT, &reminder);
754 	if (reminder) {
755 		pr_err("%s rate value %lluBps not in link speed units of 1Mbps.\n",
756 		       name, *rate);
757 		NL_SET_ERR_MSG_MOD(extack, "TX rate value not in link speed units of 1Mbps");
758 		return -EINVAL;
759 	}
760 
761 	if (value > link_speed_max) {
762 		pr_err("%s rate value %lluMbps exceed link maximum speed %u.\n",
763 		       name, value, link_speed_max);
764 		NL_SET_ERR_MSG_MOD(extack, "TX rate value exceed link maximum speed");
765 		return -EINVAL;
766 	}
767 
768 	*rate = value;
769 	return 0;
770 }
771 
772 /* Eswitch devlink rate API */
773 
mlx5_esw_devlink_rate_leaf_tx_share_set(struct devlink_rate * rate_leaf,void * priv,u64 tx_share,struct netlink_ext_ack * extack)774 int mlx5_esw_devlink_rate_leaf_tx_share_set(struct devlink_rate *rate_leaf, void *priv,
775 					    u64 tx_share, struct netlink_ext_ack *extack)
776 {
777 	struct mlx5_vport *vport = priv;
778 	struct mlx5_eswitch *esw;
779 	int err;
780 
781 	esw = vport->dev->priv.eswitch;
782 	if (!mlx5_esw_allowed(esw))
783 		return -EPERM;
784 
785 	err = esw_qos_devlink_rate_to_mbps(vport->dev, "tx_share", &tx_share, extack);
786 	if (err)
787 		return err;
788 
789 	mutex_lock(&esw->state_lock);
790 	err = esw_qos_vport_enable(esw, vport, 0, 0, extack);
791 	if (err)
792 		goto unlock;
793 
794 	err = esw_qos_set_vport_min_rate(esw, vport, tx_share, extack);
795 unlock:
796 	mutex_unlock(&esw->state_lock);
797 	return err;
798 }
799 
mlx5_esw_devlink_rate_leaf_tx_max_set(struct devlink_rate * rate_leaf,void * priv,u64 tx_max,struct netlink_ext_ack * extack)800 int mlx5_esw_devlink_rate_leaf_tx_max_set(struct devlink_rate *rate_leaf, void *priv,
801 					  u64 tx_max, struct netlink_ext_ack *extack)
802 {
803 	struct mlx5_vport *vport = priv;
804 	struct mlx5_eswitch *esw;
805 	int err;
806 
807 	esw = vport->dev->priv.eswitch;
808 	if (!mlx5_esw_allowed(esw))
809 		return -EPERM;
810 
811 	err = esw_qos_devlink_rate_to_mbps(vport->dev, "tx_max", &tx_max, extack);
812 	if (err)
813 		return err;
814 
815 	mutex_lock(&esw->state_lock);
816 	err = esw_qos_vport_enable(esw, vport, 0, 0, extack);
817 	if (err)
818 		goto unlock;
819 
820 	err = esw_qos_set_vport_max_rate(esw, vport, tx_max, extack);
821 unlock:
822 	mutex_unlock(&esw->state_lock);
823 	return err;
824 }
825 
mlx5_esw_devlink_rate_node_tx_share_set(struct devlink_rate * rate_node,void * priv,u64 tx_share,struct netlink_ext_ack * extack)826 int mlx5_esw_devlink_rate_node_tx_share_set(struct devlink_rate *rate_node, void *priv,
827 					    u64 tx_share, struct netlink_ext_ack *extack)
828 {
829 	struct mlx5_core_dev *dev = devlink_priv(rate_node->devlink);
830 	struct mlx5_eswitch *esw = dev->priv.eswitch;
831 	struct mlx5_esw_rate_group *group = priv;
832 	int err;
833 
834 	err = esw_qos_devlink_rate_to_mbps(dev, "tx_share", &tx_share, extack);
835 	if (err)
836 		return err;
837 
838 	mutex_lock(&esw->state_lock);
839 	err = esw_qos_set_group_min_rate(esw, group, tx_share, extack);
840 	mutex_unlock(&esw->state_lock);
841 	return err;
842 }
843 
mlx5_esw_devlink_rate_node_tx_max_set(struct devlink_rate * rate_node,void * priv,u64 tx_max,struct netlink_ext_ack * extack)844 int mlx5_esw_devlink_rate_node_tx_max_set(struct devlink_rate *rate_node, void *priv,
845 					  u64 tx_max, struct netlink_ext_ack *extack)
846 {
847 	struct mlx5_core_dev *dev = devlink_priv(rate_node->devlink);
848 	struct mlx5_eswitch *esw = dev->priv.eswitch;
849 	struct mlx5_esw_rate_group *group = priv;
850 	int err;
851 
852 	err = esw_qos_devlink_rate_to_mbps(dev, "tx_max", &tx_max, extack);
853 	if (err)
854 		return err;
855 
856 	mutex_lock(&esw->state_lock);
857 	err = esw_qos_set_group_max_rate(esw, group, tx_max, extack);
858 	mutex_unlock(&esw->state_lock);
859 	return err;
860 }
861 
mlx5_esw_devlink_rate_node_new(struct devlink_rate * rate_node,void ** priv,struct netlink_ext_ack * extack)862 int mlx5_esw_devlink_rate_node_new(struct devlink_rate *rate_node, void **priv,
863 				   struct netlink_ext_ack *extack)
864 {
865 	struct mlx5_esw_rate_group *group;
866 	struct mlx5_eswitch *esw;
867 	int err = 0;
868 
869 	esw = mlx5_devlink_eswitch_get(rate_node->devlink);
870 	if (IS_ERR(esw))
871 		return PTR_ERR(esw);
872 
873 	mutex_lock(&esw->state_lock);
874 	if (esw->mode != MLX5_ESWITCH_OFFLOADS) {
875 		NL_SET_ERR_MSG_MOD(extack,
876 				   "Rate node creation supported only in switchdev mode");
877 		err = -EOPNOTSUPP;
878 		goto unlock;
879 	}
880 
881 	group = esw_qos_create_rate_group(esw, extack);
882 	if (IS_ERR(group)) {
883 		err = PTR_ERR(group);
884 		goto unlock;
885 	}
886 
887 	*priv = group;
888 unlock:
889 	mutex_unlock(&esw->state_lock);
890 	return err;
891 }
892 
mlx5_esw_devlink_rate_node_del(struct devlink_rate * rate_node,void * priv,struct netlink_ext_ack * extack)893 int mlx5_esw_devlink_rate_node_del(struct devlink_rate *rate_node, void *priv,
894 				   struct netlink_ext_ack *extack)
895 {
896 	struct mlx5_esw_rate_group *group = priv;
897 	struct mlx5_eswitch *esw;
898 	int err;
899 
900 	esw = mlx5_devlink_eswitch_get(rate_node->devlink);
901 	if (IS_ERR(esw))
902 		return PTR_ERR(esw);
903 
904 	mutex_lock(&esw->state_lock);
905 	err = esw_qos_destroy_rate_group(esw, group, extack);
906 	mutex_unlock(&esw->state_lock);
907 	return err;
908 }
909 
mlx5_esw_qos_vport_update_group(struct mlx5_eswitch * esw,struct mlx5_vport * vport,struct mlx5_esw_rate_group * group,struct netlink_ext_ack * extack)910 int mlx5_esw_qos_vport_update_group(struct mlx5_eswitch *esw,
911 				    struct mlx5_vport *vport,
912 				    struct mlx5_esw_rate_group *group,
913 				    struct netlink_ext_ack *extack)
914 {
915 	int err = 0;
916 
917 	mutex_lock(&esw->state_lock);
918 	if (!vport->qos.enabled && !group)
919 		goto unlock;
920 
921 	err = esw_qos_vport_enable(esw, vport, 0, 0, extack);
922 	if (!err)
923 		err = esw_qos_vport_update_group(esw, vport, group, extack);
924 unlock:
925 	mutex_unlock(&esw->state_lock);
926 	return err;
927 }
928 
mlx5_esw_devlink_rate_parent_set(struct devlink_rate * devlink_rate,struct devlink_rate * parent,void * priv,void * parent_priv,struct netlink_ext_ack * extack)929 int mlx5_esw_devlink_rate_parent_set(struct devlink_rate *devlink_rate,
930 				     struct devlink_rate *parent,
931 				     void *priv, void *parent_priv,
932 				     struct netlink_ext_ack *extack)
933 {
934 	struct mlx5_esw_rate_group *group;
935 	struct mlx5_vport *vport = priv;
936 
937 	if (!parent)
938 		return mlx5_esw_qos_vport_update_group(vport->dev->priv.eswitch,
939 						       vport, NULL, extack);
940 
941 	group = parent_priv;
942 	return mlx5_esw_qos_vport_update_group(vport->dev->priv.eswitch, vport, group, extack);
943 }
944