xref: /linux/drivers/net/ethernet/intel/ice/ice_dcb_lib.c (revision fd7d598270724cc787982ea48bbe17ad383a8b7f)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2019, Intel Corporation. */
3 
4 #include "ice_dcb_lib.h"
5 #include "ice_dcb_nl.h"
6 #include "ice_devlink.h"
7 
8 /**
9  * ice_dcb_get_ena_tc - return bitmap of enabled TCs
10  * @dcbcfg: DCB config to evaluate for enabled TCs
11  */
12 static u8 ice_dcb_get_ena_tc(struct ice_dcbx_cfg *dcbcfg)
13 {
14 	u8 i, num_tc, ena_tc = 1;
15 
16 	num_tc = ice_dcb_get_num_tc(dcbcfg);
17 
18 	for (i = 0; i < num_tc; i++)
19 		ena_tc |= BIT(i);
20 
21 	return ena_tc;
22 }
23 
24 /**
25  * ice_is_pfc_causing_hung_q
26  * @pf: pointer to PF structure
27  * @txqueue: Tx queue which is supposedly hung queue
28  *
29  * find if PFC is causing the hung queue, if yes return true else false
30  */
31 bool ice_is_pfc_causing_hung_q(struct ice_pf *pf, unsigned int txqueue)
32 {
33 	u8 num_tcs = 0, i, tc, up_mapped_tc, up_in_tc = 0;
34 	u64 ref_prio_xoff[ICE_MAX_UP];
35 	struct ice_vsi *vsi;
36 	u32 up2tc;
37 
38 	vsi = ice_get_main_vsi(pf);
39 	if (!vsi)
40 		return false;
41 
42 	ice_for_each_traffic_class(i)
43 		if (vsi->tc_cfg.ena_tc & BIT(i))
44 			num_tcs++;
45 
46 	/* first find out the TC to which the hung queue belongs to */
47 	for (tc = 0; tc < num_tcs - 1; tc++)
48 		if (ice_find_q_in_range(vsi->tc_cfg.tc_info[tc].qoffset,
49 					vsi->tc_cfg.tc_info[tc + 1].qoffset,
50 					txqueue))
51 			break;
52 
53 	/* Build a bit map of all UPs associated to the suspect hung queue TC,
54 	 * so that we check for its counter increment.
55 	 */
56 	up2tc = rd32(&pf->hw, PRTDCB_TUP2TC);
57 	for (i = 0; i < ICE_MAX_UP; i++) {
58 		up_mapped_tc = (up2tc >> (i * 3)) & 0x7;
59 		if (up_mapped_tc == tc)
60 			up_in_tc |= BIT(i);
61 	}
62 
63 	/* Now that we figured out that hung queue is PFC enabled, still the
64 	 * Tx timeout can be legitimate. So to make sure Tx timeout is
65 	 * absolutely caused by PFC storm, check if the counters are
66 	 * incrementing.
67 	 */
68 	for (i = 0; i < ICE_MAX_UP; i++)
69 		if (up_in_tc & BIT(i))
70 			ref_prio_xoff[i] = pf->stats.priority_xoff_rx[i];
71 
72 	ice_update_dcb_stats(pf);
73 
74 	for (i = 0; i < ICE_MAX_UP; i++)
75 		if (up_in_tc & BIT(i))
76 			if (pf->stats.priority_xoff_rx[i] > ref_prio_xoff[i])
77 				return true;
78 
79 	return false;
80 }
81 
82 /**
83  * ice_dcb_get_mode - gets the DCB mode
84  * @port_info: pointer to port info structure
85  * @host: if set it's HOST if not it's MANAGED
86  */
87 static u8 ice_dcb_get_mode(struct ice_port_info *port_info, bool host)
88 {
89 	u8 mode;
90 
91 	if (host)
92 		mode = DCB_CAP_DCBX_HOST;
93 	else
94 		mode = DCB_CAP_DCBX_LLD_MANAGED;
95 
96 	if (port_info->qos_cfg.local_dcbx_cfg.dcbx_mode & ICE_DCBX_MODE_CEE)
97 		return mode | DCB_CAP_DCBX_VER_CEE;
98 	else
99 		return mode | DCB_CAP_DCBX_VER_IEEE;
100 }
101 
102 /**
103  * ice_dcb_get_num_tc - Get the number of TCs from DCBX config
104  * @dcbcfg: config to retrieve number of TCs from
105  */
106 u8 ice_dcb_get_num_tc(struct ice_dcbx_cfg *dcbcfg)
107 {
108 	bool tc_unused = false;
109 	u8 num_tc = 0;
110 	u8 ret = 0;
111 	int i;
112 
113 	/* Scan the ETS Config Priority Table to find traffic classes
114 	 * enabled and create a bitmask of enabled TCs
115 	 */
116 	for (i = 0; i < CEE_DCBX_MAX_PRIO; i++)
117 		num_tc |= BIT(dcbcfg->etscfg.prio_table[i]);
118 
119 	/* Scan bitmask for contiguous TCs starting with TC0 */
120 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
121 		if (num_tc & BIT(i)) {
122 			if (!tc_unused) {
123 				ret++;
124 			} else {
125 				pr_err("Non-contiguous TCs - Disabling DCB\n");
126 				return 1;
127 			}
128 		} else {
129 			tc_unused = true;
130 		}
131 	}
132 
133 	/* There is always at least 1 TC */
134 	if (!ret)
135 		ret = 1;
136 
137 	return ret;
138 }
139 
140 /**
141  * ice_get_first_droptc - returns number of first droptc
142  * @vsi: used to find the first droptc
143  *
144  * This function returns the value of first_droptc.
145  * When DCB is enabled, first droptc information is derived from enabled_tc
146  * and PFC enabled bits. otherwise this function returns 0 as there is one
147  * TC without DCB (tc0)
148  */
149 static u8 ice_get_first_droptc(struct ice_vsi *vsi)
150 {
151 	struct ice_dcbx_cfg *cfg = &vsi->port_info->qos_cfg.local_dcbx_cfg;
152 	struct device *dev = ice_pf_to_dev(vsi->back);
153 	u8 num_tc, ena_tc_map, pfc_ena_map;
154 	u8 i;
155 
156 	num_tc = ice_dcb_get_num_tc(cfg);
157 
158 	/* get bitmap of enabled TCs */
159 	ena_tc_map = ice_dcb_get_ena_tc(cfg);
160 
161 	/* get bitmap of PFC enabled TCs */
162 	pfc_ena_map = cfg->pfc.pfcena;
163 
164 	/* get first TC that is not PFC enabled */
165 	for (i = 0; i < num_tc; i++) {
166 		if ((ena_tc_map & BIT(i)) && (!(pfc_ena_map & BIT(i)))) {
167 			dev_dbg(dev, "first drop tc = %d\n", i);
168 			return i;
169 		}
170 	}
171 
172 	dev_dbg(dev, "first drop tc = 0\n");
173 	return 0;
174 }
175 
176 /**
177  * ice_vsi_set_dcb_tc_cfg - Set VSI's TC based on DCB configuration
178  * @vsi: pointer to the VSI instance
179  */
180 void ice_vsi_set_dcb_tc_cfg(struct ice_vsi *vsi)
181 {
182 	struct ice_dcbx_cfg *cfg = &vsi->port_info->qos_cfg.local_dcbx_cfg;
183 
184 	switch (vsi->type) {
185 	case ICE_VSI_PF:
186 		vsi->tc_cfg.ena_tc = ice_dcb_get_ena_tc(cfg);
187 		vsi->tc_cfg.numtc = ice_dcb_get_num_tc(cfg);
188 		break;
189 	case ICE_VSI_CHNL:
190 		vsi->tc_cfg.ena_tc = BIT(ice_get_first_droptc(vsi));
191 		vsi->tc_cfg.numtc = 1;
192 		break;
193 	case ICE_VSI_CTRL:
194 	case ICE_VSI_LB:
195 	default:
196 		vsi->tc_cfg.ena_tc = ICE_DFLT_TRAFFIC_CLASS;
197 		vsi->tc_cfg.numtc = 1;
198 	}
199 }
200 
201 /**
202  * ice_dcb_get_tc - Get the TC associated with the queue
203  * @vsi: ptr to the VSI
204  * @queue_index: queue number associated with VSI
205  */
206 u8 ice_dcb_get_tc(struct ice_vsi *vsi, int queue_index)
207 {
208 	return vsi->tx_rings[queue_index]->dcb_tc;
209 }
210 
211 /**
212  * ice_vsi_cfg_dcb_rings - Update rings to reflect DCB TC
213  * @vsi: VSI owner of rings being updated
214  */
215 void ice_vsi_cfg_dcb_rings(struct ice_vsi *vsi)
216 {
217 	struct ice_tx_ring *tx_ring;
218 	struct ice_rx_ring *rx_ring;
219 	u16 qoffset, qcount;
220 	int i, n;
221 
222 	if (!test_bit(ICE_FLAG_DCB_ENA, vsi->back->flags)) {
223 		/* Reset the TC information */
224 		ice_for_each_txq(vsi, i) {
225 			tx_ring = vsi->tx_rings[i];
226 			tx_ring->dcb_tc = 0;
227 		}
228 		ice_for_each_rxq(vsi, i) {
229 			rx_ring = vsi->rx_rings[i];
230 			rx_ring->dcb_tc = 0;
231 		}
232 		return;
233 	}
234 
235 	ice_for_each_traffic_class(n) {
236 		if (!(vsi->tc_cfg.ena_tc & BIT(n)))
237 			break;
238 
239 		qoffset = vsi->tc_cfg.tc_info[n].qoffset;
240 		qcount = vsi->tc_cfg.tc_info[n].qcount_tx;
241 		for (i = qoffset; i < (qoffset + qcount); i++)
242 			vsi->tx_rings[i]->dcb_tc = n;
243 
244 		qcount = vsi->tc_cfg.tc_info[n].qcount_rx;
245 		for (i = qoffset; i < (qoffset + qcount); i++)
246 			vsi->rx_rings[i]->dcb_tc = n;
247 	}
248 	/* applicable only if "all_enatc" is set, which will be set from
249 	 * setup_tc method as part of configuring channels
250 	 */
251 	if (vsi->all_enatc) {
252 		u8 first_droptc = ice_get_first_droptc(vsi);
253 
254 		/* When DCB is configured, TC for ADQ queues (which are really
255 		 * PF queues) should be the first drop TC of the main VSI
256 		 */
257 		ice_for_each_chnl_tc(n) {
258 			if (!(vsi->all_enatc & BIT(n)))
259 				break;
260 
261 			qoffset = vsi->mqprio_qopt.qopt.offset[n];
262 			qcount = vsi->mqprio_qopt.qopt.count[n];
263 			for (i = qoffset; i < (qoffset + qcount); i++) {
264 				vsi->tx_rings[i]->dcb_tc = first_droptc;
265 				vsi->rx_rings[i]->dcb_tc = first_droptc;
266 			}
267 		}
268 	}
269 }
270 
271 /**
272  * ice_dcb_ena_dis_vsi - disable certain VSIs for DCB config/reconfig
273  * @pf: pointer to the PF instance
274  * @ena: true to enable VSIs, false to disable
275  * @locked: true if caller holds RTNL lock, false otherwise
276  *
277  * Before a new DCB configuration can be applied, VSIs of type PF, SWITCHDEV
278  * and CHNL need to be brought down. Following completion of DCB configuration
279  * the VSIs that were downed need to be brought up again. This helper function
280  * does both.
281  */
282 static void ice_dcb_ena_dis_vsi(struct ice_pf *pf, bool ena, bool locked)
283 {
284 	int i;
285 
286 	ice_for_each_vsi(pf, i) {
287 		struct ice_vsi *vsi = pf->vsi[i];
288 
289 		if (!vsi)
290 			continue;
291 
292 		switch (vsi->type) {
293 		case ICE_VSI_CHNL:
294 		case ICE_VSI_SWITCHDEV_CTRL:
295 		case ICE_VSI_PF:
296 			if (ena)
297 				ice_ena_vsi(vsi, locked);
298 			else
299 				ice_dis_vsi(vsi, locked);
300 			break;
301 		default:
302 			continue;
303 		}
304 	}
305 }
306 
307 /**
308  * ice_dcb_bwchk - check if ETS bandwidth input parameters are correct
309  * @pf: pointer to the PF struct
310  * @dcbcfg: pointer to DCB config structure
311  */
312 int ice_dcb_bwchk(struct ice_pf *pf, struct ice_dcbx_cfg *dcbcfg)
313 {
314 	struct ice_dcb_ets_cfg *etscfg = &dcbcfg->etscfg;
315 	u8 num_tc, total_bw = 0;
316 	int i;
317 
318 	/* returns number of contigous TCs and 1 TC for non-contigous TCs,
319 	 * since at least 1 TC has to be configured
320 	 */
321 	num_tc = ice_dcb_get_num_tc(dcbcfg);
322 
323 	/* no bandwidth checks required if there's only one TC, so assign
324 	 * all bandwidth to TC0 and return
325 	 */
326 	if (num_tc == 1) {
327 		etscfg->tcbwtable[0] = ICE_TC_MAX_BW;
328 		return 0;
329 	}
330 
331 	for (i = 0; i < num_tc; i++)
332 		total_bw += etscfg->tcbwtable[i];
333 
334 	if (!total_bw) {
335 		etscfg->tcbwtable[0] = ICE_TC_MAX_BW;
336 	} else if (total_bw != ICE_TC_MAX_BW) {
337 		dev_err(ice_pf_to_dev(pf), "Invalid config, total bandwidth must equal 100\n");
338 		return -EINVAL;
339 	}
340 
341 	return 0;
342 }
343 
344 /**
345  * ice_pf_dcb_cfg - Apply new DCB configuration
346  * @pf: pointer to the PF struct
347  * @new_cfg: DCBX config to apply
348  * @locked: is the RTNL held
349  */
350 int ice_pf_dcb_cfg(struct ice_pf *pf, struct ice_dcbx_cfg *new_cfg, bool locked)
351 {
352 	struct ice_aqc_port_ets_elem buf = { 0 };
353 	struct ice_dcbx_cfg *old_cfg, *curr_cfg;
354 	struct device *dev = ice_pf_to_dev(pf);
355 	int ret = ICE_DCB_NO_HW_CHG;
356 	struct iidc_event *event;
357 	struct ice_vsi *pf_vsi;
358 
359 	curr_cfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg;
360 
361 	/* FW does not care if change happened */
362 	if (!pf->hw.port_info->qos_cfg.is_sw_lldp)
363 		ret = ICE_DCB_HW_CHG_RST;
364 
365 	/* Enable DCB tagging only when more than one TC */
366 	if (ice_dcb_get_num_tc(new_cfg) > 1) {
367 		dev_dbg(dev, "DCB tagging enabled (num TC > 1)\n");
368 		if (pf->hw.port_info->is_custom_tx_enabled) {
369 			dev_err(dev, "Custom Tx scheduler feature enabled, can't configure DCB\n");
370 			return -EBUSY;
371 		}
372 		ice_tear_down_devlink_rate_tree(pf);
373 
374 		set_bit(ICE_FLAG_DCB_ENA, pf->flags);
375 	} else {
376 		dev_dbg(dev, "DCB tagging disabled (num TC = 1)\n");
377 		clear_bit(ICE_FLAG_DCB_ENA, pf->flags);
378 	}
379 
380 	if (!memcmp(new_cfg, curr_cfg, sizeof(*new_cfg))) {
381 		dev_dbg(dev, "No change in DCB config required\n");
382 		return ret;
383 	}
384 
385 	if (ice_dcb_bwchk(pf, new_cfg))
386 		return -EINVAL;
387 
388 	/* Store old config in case FW config fails */
389 	old_cfg = kmemdup(curr_cfg, sizeof(*old_cfg), GFP_KERNEL);
390 	if (!old_cfg)
391 		return -ENOMEM;
392 
393 	dev_info(dev, "Commit DCB Configuration to the hardware\n");
394 	pf_vsi = ice_get_main_vsi(pf);
395 	if (!pf_vsi) {
396 		dev_dbg(dev, "PF VSI doesn't exist\n");
397 		ret = -EINVAL;
398 		goto free_cfg;
399 	}
400 
401 	/* Notify AUX drivers about impending change to TCs */
402 	event = kzalloc(sizeof(*event), GFP_KERNEL);
403 	if (!event) {
404 		ret = -ENOMEM;
405 		goto free_cfg;
406 	}
407 
408 	set_bit(IIDC_EVENT_BEFORE_TC_CHANGE, event->type);
409 	ice_send_event_to_aux(pf, event);
410 	kfree(event);
411 
412 	/* avoid race conditions by holding the lock while disabling and
413 	 * re-enabling the VSI
414 	 */
415 	if (!locked)
416 		rtnl_lock();
417 
418 	/* disable VSIs affected by DCB changes */
419 	ice_dcb_ena_dis_vsi(pf, false, true);
420 
421 	memcpy(curr_cfg, new_cfg, sizeof(*curr_cfg));
422 	memcpy(&curr_cfg->etsrec, &curr_cfg->etscfg, sizeof(curr_cfg->etsrec));
423 	memcpy(&new_cfg->etsrec, &curr_cfg->etscfg, sizeof(curr_cfg->etsrec));
424 
425 	/* Only send new config to HW if we are in SW LLDP mode. Otherwise,
426 	 * the new config came from the HW in the first place.
427 	 */
428 	if (pf->hw.port_info->qos_cfg.is_sw_lldp) {
429 		ret = ice_set_dcb_cfg(pf->hw.port_info);
430 		if (ret) {
431 			dev_err(dev, "Set DCB Config failed\n");
432 			/* Restore previous settings to local config */
433 			memcpy(curr_cfg, old_cfg, sizeof(*curr_cfg));
434 			goto out;
435 		}
436 	}
437 
438 	ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
439 	if (ret) {
440 		dev_err(dev, "Query Port ETS failed\n");
441 		goto out;
442 	}
443 
444 	ice_pf_dcb_recfg(pf, false);
445 
446 out:
447 	/* enable previously downed VSIs */
448 	ice_dcb_ena_dis_vsi(pf, true, true);
449 	if (!locked)
450 		rtnl_unlock();
451 free_cfg:
452 	kfree(old_cfg);
453 	return ret;
454 }
455 
456 /**
457  * ice_cfg_etsrec_defaults - Set default ETS recommended DCB config
458  * @pi: port information structure
459  */
460 static void ice_cfg_etsrec_defaults(struct ice_port_info *pi)
461 {
462 	struct ice_dcbx_cfg *dcbcfg = &pi->qos_cfg.local_dcbx_cfg;
463 	u8 i;
464 
465 	/* Ensure ETS recommended DCB configuration is not already set */
466 	if (dcbcfg->etsrec.maxtcs)
467 		return;
468 
469 	/* In CEE mode, set the default to 1 TC */
470 	dcbcfg->etsrec.maxtcs = 1;
471 	for (i = 0; i < ICE_MAX_TRAFFIC_CLASS; i++) {
472 		dcbcfg->etsrec.tcbwtable[i] = i ? 0 : 100;
473 		dcbcfg->etsrec.tsatable[i] = i ? ICE_IEEE_TSA_STRICT :
474 						 ICE_IEEE_TSA_ETS;
475 	}
476 }
477 
478 /**
479  * ice_dcb_need_recfg - Check if DCB needs reconfig
480  * @pf: board private structure
481  * @old_cfg: current DCB config
482  * @new_cfg: new DCB config
483  */
484 static bool
485 ice_dcb_need_recfg(struct ice_pf *pf, struct ice_dcbx_cfg *old_cfg,
486 		   struct ice_dcbx_cfg *new_cfg)
487 {
488 	struct device *dev = ice_pf_to_dev(pf);
489 	bool need_reconfig = false;
490 
491 	/* Check if ETS configuration has changed */
492 	if (memcmp(&new_cfg->etscfg, &old_cfg->etscfg,
493 		   sizeof(new_cfg->etscfg))) {
494 		/* If Priority Table has changed reconfig is needed */
495 		if (memcmp(&new_cfg->etscfg.prio_table,
496 			   &old_cfg->etscfg.prio_table,
497 			   sizeof(new_cfg->etscfg.prio_table))) {
498 			need_reconfig = true;
499 			dev_dbg(dev, "ETS UP2TC changed.\n");
500 		}
501 
502 		if (memcmp(&new_cfg->etscfg.tcbwtable,
503 			   &old_cfg->etscfg.tcbwtable,
504 			   sizeof(new_cfg->etscfg.tcbwtable)))
505 			dev_dbg(dev, "ETS TC BW Table changed.\n");
506 
507 		if (memcmp(&new_cfg->etscfg.tsatable,
508 			   &old_cfg->etscfg.tsatable,
509 			   sizeof(new_cfg->etscfg.tsatable)))
510 			dev_dbg(dev, "ETS TSA Table changed.\n");
511 	}
512 
513 	/* Check if PFC configuration has changed */
514 	if (memcmp(&new_cfg->pfc, &old_cfg->pfc, sizeof(new_cfg->pfc))) {
515 		need_reconfig = true;
516 		dev_dbg(dev, "PFC config change detected.\n");
517 	}
518 
519 	/* Check if APP Table has changed */
520 	if (memcmp(&new_cfg->app, &old_cfg->app, sizeof(new_cfg->app))) {
521 		need_reconfig = true;
522 		dev_dbg(dev, "APP Table change detected.\n");
523 	}
524 
525 	dev_dbg(dev, "dcb need_reconfig=%d\n", need_reconfig);
526 	return need_reconfig;
527 }
528 
529 /**
530  * ice_dcb_rebuild - rebuild DCB post reset
531  * @pf: physical function instance
532  */
533 void ice_dcb_rebuild(struct ice_pf *pf)
534 {
535 	struct ice_aqc_port_ets_elem buf = { 0 };
536 	struct device *dev = ice_pf_to_dev(pf);
537 	struct ice_dcbx_cfg *err_cfg;
538 	int ret;
539 
540 	ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
541 	if (ret) {
542 		dev_err(dev, "Query Port ETS failed\n");
543 		goto dcb_error;
544 	}
545 
546 	mutex_lock(&pf->tc_mutex);
547 
548 	if (!pf->hw.port_info->qos_cfg.is_sw_lldp)
549 		ice_cfg_etsrec_defaults(pf->hw.port_info);
550 
551 	ret = ice_set_dcb_cfg(pf->hw.port_info);
552 	if (ret) {
553 		dev_err(dev, "Failed to set DCB config in rebuild\n");
554 		goto dcb_error;
555 	}
556 
557 	if (!pf->hw.port_info->qos_cfg.is_sw_lldp) {
558 		ret = ice_cfg_lldp_mib_change(&pf->hw, true);
559 		if (ret && !pf->hw.port_info->qos_cfg.is_sw_lldp) {
560 			dev_err(dev, "Failed to register for MIB changes\n");
561 			goto dcb_error;
562 		}
563 	}
564 
565 	dev_info(dev, "DCB info restored\n");
566 	ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
567 	if (ret) {
568 		dev_err(dev, "Query Port ETS failed\n");
569 		goto dcb_error;
570 	}
571 
572 	mutex_unlock(&pf->tc_mutex);
573 
574 	return;
575 
576 dcb_error:
577 	dev_err(dev, "Disabling DCB until new settings occur\n");
578 	err_cfg = kzalloc(sizeof(*err_cfg), GFP_KERNEL);
579 	if (!err_cfg) {
580 		mutex_unlock(&pf->tc_mutex);
581 		return;
582 	}
583 
584 	err_cfg->etscfg.willing = true;
585 	err_cfg->etscfg.tcbwtable[0] = ICE_TC_MAX_BW;
586 	err_cfg->etscfg.tsatable[0] = ICE_IEEE_TSA_ETS;
587 	memcpy(&err_cfg->etsrec, &err_cfg->etscfg, sizeof(err_cfg->etsrec));
588 	/* Coverity warns the return code of ice_pf_dcb_cfg() is not checked
589 	 * here as is done for other calls to that function. That check is
590 	 * not necessary since this is in this function's error cleanup path.
591 	 * Suppress the Coverity warning with the following comment...
592 	 */
593 	/* coverity[check_return] */
594 	ice_pf_dcb_cfg(pf, err_cfg, false);
595 	kfree(err_cfg);
596 
597 	mutex_unlock(&pf->tc_mutex);
598 }
599 
600 /**
601  * ice_dcb_init_cfg - set the initial DCB config in SW
602  * @pf: PF to apply config to
603  * @locked: Is the RTNL held
604  */
605 static int ice_dcb_init_cfg(struct ice_pf *pf, bool locked)
606 {
607 	struct ice_dcbx_cfg *newcfg;
608 	struct ice_port_info *pi;
609 	int ret = 0;
610 
611 	pi = pf->hw.port_info;
612 	newcfg = kmemdup(&pi->qos_cfg.local_dcbx_cfg, sizeof(*newcfg),
613 			 GFP_KERNEL);
614 	if (!newcfg)
615 		return -ENOMEM;
616 
617 	memset(&pi->qos_cfg.local_dcbx_cfg, 0, sizeof(*newcfg));
618 
619 	dev_info(ice_pf_to_dev(pf), "Configuring initial DCB values\n");
620 	if (ice_pf_dcb_cfg(pf, newcfg, locked))
621 		ret = -EINVAL;
622 
623 	kfree(newcfg);
624 
625 	return ret;
626 }
627 
628 /**
629  * ice_dcb_sw_dflt_cfg - Apply a default DCB config
630  * @pf: PF to apply config to
631  * @ets_willing: configure ETS willing
632  * @locked: was this function called with RTNL held
633  */
634 int ice_dcb_sw_dflt_cfg(struct ice_pf *pf, bool ets_willing, bool locked)
635 {
636 	struct ice_aqc_port_ets_elem buf = { 0 };
637 	struct ice_dcbx_cfg *dcbcfg;
638 	struct ice_port_info *pi;
639 	struct ice_hw *hw;
640 	int ret;
641 
642 	hw = &pf->hw;
643 	pi = hw->port_info;
644 	dcbcfg = kzalloc(sizeof(*dcbcfg), GFP_KERNEL);
645 	if (!dcbcfg)
646 		return -ENOMEM;
647 
648 	memset(&pi->qos_cfg.local_dcbx_cfg, 0, sizeof(*dcbcfg));
649 
650 	dcbcfg->etscfg.willing = ets_willing ? 1 : 0;
651 	dcbcfg->etscfg.maxtcs = hw->func_caps.common_cap.maxtc;
652 	dcbcfg->etscfg.tcbwtable[0] = 100;
653 	dcbcfg->etscfg.tsatable[0] = ICE_IEEE_TSA_ETS;
654 
655 	memcpy(&dcbcfg->etsrec, &dcbcfg->etscfg,
656 	       sizeof(dcbcfg->etsrec));
657 	dcbcfg->etsrec.willing = 0;
658 
659 	dcbcfg->pfc.willing = 1;
660 	dcbcfg->pfc.pfccap = hw->func_caps.common_cap.maxtc;
661 
662 	dcbcfg->numapps = 1;
663 	dcbcfg->app[0].selector = ICE_APP_SEL_ETHTYPE;
664 	dcbcfg->app[0].priority = 3;
665 	dcbcfg->app[0].prot_id = ETH_P_FCOE;
666 
667 	ret = ice_pf_dcb_cfg(pf, dcbcfg, locked);
668 	kfree(dcbcfg);
669 	if (ret)
670 		return ret;
671 
672 	return ice_query_port_ets(pi, &buf, sizeof(buf), NULL);
673 }
674 
675 /**
676  * ice_dcb_tc_contig - Check that TCs are contiguous
677  * @prio_table: pointer to priority table
678  *
679  * Check if TCs begin with TC0 and are contiguous
680  */
681 static bool ice_dcb_tc_contig(u8 *prio_table)
682 {
683 	bool found_empty = false;
684 	u8 used_tc = 0;
685 	int i;
686 
687 	/* Create a bitmap of used TCs */
688 	for (i = 0; i < CEE_DCBX_MAX_PRIO; i++)
689 		used_tc |= BIT(prio_table[i]);
690 
691 	for (i = 0; i < CEE_DCBX_MAX_PRIO; i++) {
692 		if (used_tc & BIT(i)) {
693 			if (found_empty)
694 				return false;
695 		} else {
696 			found_empty = true;
697 		}
698 	}
699 
700 	return true;
701 }
702 
703 /**
704  * ice_dcb_noncontig_cfg - Configure DCB for non-contiguous TCs
705  * @pf: pointer to the PF struct
706  *
707  * If non-contiguous TCs, then configure SW DCB with TC0 and ETS non-willing
708  */
709 static int ice_dcb_noncontig_cfg(struct ice_pf *pf)
710 {
711 	struct ice_dcbx_cfg *dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg;
712 	struct device *dev = ice_pf_to_dev(pf);
713 	int ret;
714 
715 	/* Configure SW DCB default with ETS non-willing */
716 	ret = ice_dcb_sw_dflt_cfg(pf, false, true);
717 	if (ret) {
718 		dev_err(dev, "Failed to set local DCB config %d\n", ret);
719 		return ret;
720 	}
721 
722 	/* Reconfigure with ETS willing so that FW will send LLDP MIB event */
723 	dcbcfg->etscfg.willing = 1;
724 	ret = ice_set_dcb_cfg(pf->hw.port_info);
725 	if (ret)
726 		dev_err(dev, "Failed to set DCB to unwilling\n");
727 
728 	return ret;
729 }
730 
731 /**
732  * ice_pf_dcb_recfg - Reconfigure all VEBs and VSIs
733  * @pf: pointer to the PF struct
734  * @locked: is adev device lock held
735  *
736  * Assumed caller has already disabled all VSIs before
737  * calling this function. Reconfiguring DCB based on
738  * local_dcbx_cfg.
739  */
740 void ice_pf_dcb_recfg(struct ice_pf *pf, bool locked)
741 {
742 	struct ice_dcbx_cfg *dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg;
743 	struct iidc_event *event;
744 	u8 tc_map = 0;
745 	int v, ret;
746 
747 	/* Update each VSI */
748 	ice_for_each_vsi(pf, v) {
749 		struct ice_vsi *vsi = pf->vsi[v];
750 
751 		if (!vsi)
752 			continue;
753 
754 		if (vsi->type == ICE_VSI_PF) {
755 			tc_map = ice_dcb_get_ena_tc(dcbcfg);
756 
757 			/* If DCBX request non-contiguous TC, then configure
758 			 * default TC
759 			 */
760 			if (!ice_dcb_tc_contig(dcbcfg->etscfg.prio_table)) {
761 				tc_map = ICE_DFLT_TRAFFIC_CLASS;
762 				ice_dcb_noncontig_cfg(pf);
763 			}
764 		} else if (vsi->type == ICE_VSI_CHNL) {
765 			tc_map = BIT(ice_get_first_droptc(vsi));
766 		} else {
767 			tc_map = ICE_DFLT_TRAFFIC_CLASS;
768 		}
769 
770 		ret = ice_vsi_cfg_tc(vsi, tc_map);
771 		if (ret) {
772 			dev_err(ice_pf_to_dev(pf), "Failed to config TC for VSI index: %d\n",
773 				vsi->idx);
774 			continue;
775 		}
776 		/* no need to proceed with remaining cfg if it is CHNL
777 		 * or switchdev VSI
778 		 */
779 		if (vsi->type == ICE_VSI_CHNL ||
780 		    vsi->type == ICE_VSI_SWITCHDEV_CTRL)
781 			continue;
782 
783 		ice_vsi_map_rings_to_vectors(vsi);
784 		if (vsi->type == ICE_VSI_PF)
785 			ice_dcbnl_set_all(vsi);
786 	}
787 	if (!locked) {
788 		/* Notify the AUX drivers that TC change is finished */
789 		event = kzalloc(sizeof(*event), GFP_KERNEL);
790 		if (!event)
791 			return;
792 
793 		set_bit(IIDC_EVENT_AFTER_TC_CHANGE, event->type);
794 		ice_send_event_to_aux(pf, event);
795 		kfree(event);
796 	}
797 }
798 
799 /**
800  * ice_init_pf_dcb - initialize DCB for a PF
801  * @pf: PF to initialize DCB for
802  * @locked: Was function called with RTNL held
803  */
804 int ice_init_pf_dcb(struct ice_pf *pf, bool locked)
805 {
806 	struct device *dev = ice_pf_to_dev(pf);
807 	struct ice_port_info *port_info;
808 	struct ice_hw *hw = &pf->hw;
809 	int err;
810 
811 	port_info = hw->port_info;
812 
813 	err = ice_init_dcb(hw, false);
814 	if (err && !port_info->qos_cfg.is_sw_lldp) {
815 		dev_err(dev, "Error initializing DCB %d\n", err);
816 		goto dcb_init_err;
817 	}
818 
819 	dev_info(dev, "DCB is enabled in the hardware, max number of TCs supported on this port are %d\n",
820 		 pf->hw.func_caps.common_cap.maxtc);
821 	if (err) {
822 		struct ice_vsi *pf_vsi;
823 
824 		/* FW LLDP is disabled, activate SW DCBX/LLDP mode */
825 		dev_info(dev, "FW LLDP is disabled, DCBx/LLDP in SW mode.\n");
826 		clear_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags);
827 		err = ice_aq_set_pfc_mode(&pf->hw, ICE_AQC_PFC_VLAN_BASED_PFC,
828 					  NULL);
829 		if (err)
830 			dev_info(dev, "Failed to set VLAN PFC mode\n");
831 
832 		err = ice_dcb_sw_dflt_cfg(pf, true, locked);
833 		if (err) {
834 			dev_err(dev, "Failed to set local DCB config %d\n",
835 				err);
836 			err = -EIO;
837 			goto dcb_init_err;
838 		}
839 
840 		/* If the FW DCBX engine is not running then Rx LLDP packets
841 		 * need to be redirected up the stack.
842 		 */
843 		pf_vsi = ice_get_main_vsi(pf);
844 		if (!pf_vsi) {
845 			dev_err(dev, "Failed to set local DCB config\n");
846 			err = -EIO;
847 			goto dcb_init_err;
848 		}
849 
850 		ice_cfg_sw_lldp(pf_vsi, false, true);
851 
852 		pf->dcbx_cap = ice_dcb_get_mode(port_info, true);
853 		return 0;
854 	}
855 
856 	set_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags);
857 
858 	/* DCBX/LLDP enabled in FW, set DCBNL mode advertisement */
859 	pf->dcbx_cap = ice_dcb_get_mode(port_info, false);
860 
861 	err = ice_dcb_init_cfg(pf, locked);
862 	if (err)
863 		goto dcb_init_err;
864 
865 	return 0;
866 
867 dcb_init_err:
868 	dev_err(dev, "DCB init failed\n");
869 	return err;
870 }
871 
872 /**
873  * ice_update_dcb_stats - Update DCB stats counters
874  * @pf: PF whose stats needs to be updated
875  */
876 void ice_update_dcb_stats(struct ice_pf *pf)
877 {
878 	struct ice_hw_port_stats *prev_ps, *cur_ps;
879 	struct ice_hw *hw = &pf->hw;
880 	u8 port;
881 	int i;
882 
883 	port = hw->port_info->lport;
884 	prev_ps = &pf->stats_prev;
885 	cur_ps = &pf->stats;
886 
887 	if (ice_is_reset_in_progress(pf->state))
888 		pf->stat_prev_loaded = false;
889 
890 	for (i = 0; i < 8; i++) {
891 		ice_stat_update32(hw, GLPRT_PXOFFRXC(port, i),
892 				  pf->stat_prev_loaded,
893 				  &prev_ps->priority_xoff_rx[i],
894 				  &cur_ps->priority_xoff_rx[i]);
895 		ice_stat_update32(hw, GLPRT_PXONRXC(port, i),
896 				  pf->stat_prev_loaded,
897 				  &prev_ps->priority_xon_rx[i],
898 				  &cur_ps->priority_xon_rx[i]);
899 		ice_stat_update32(hw, GLPRT_PXONTXC(port, i),
900 				  pf->stat_prev_loaded,
901 				  &prev_ps->priority_xon_tx[i],
902 				  &cur_ps->priority_xon_tx[i]);
903 		ice_stat_update32(hw, GLPRT_PXOFFTXC(port, i),
904 				  pf->stat_prev_loaded,
905 				  &prev_ps->priority_xoff_tx[i],
906 				  &cur_ps->priority_xoff_tx[i]);
907 		ice_stat_update32(hw, GLPRT_RXON2OFFCNT(port, i),
908 				  pf->stat_prev_loaded,
909 				  &prev_ps->priority_xon_2_xoff[i],
910 				  &cur_ps->priority_xon_2_xoff[i]);
911 	}
912 }
913 
914 /**
915  * ice_tx_prepare_vlan_flags_dcb - prepare VLAN tagging for DCB
916  * @tx_ring: ring to send buffer on
917  * @first: pointer to struct ice_tx_buf
918  *
919  * This should not be called if the outer VLAN is software offloaded as the VLAN
920  * tag will already be configured with the correct ID and priority bits
921  */
922 void
923 ice_tx_prepare_vlan_flags_dcb(struct ice_tx_ring *tx_ring,
924 			      struct ice_tx_buf *first)
925 {
926 	struct sk_buff *skb = first->skb;
927 
928 	if (!test_bit(ICE_FLAG_DCB_ENA, tx_ring->vsi->back->flags))
929 		return;
930 
931 	/* Insert 802.1p priority into VLAN header */
932 	if ((first->tx_flags & ICE_TX_FLAGS_HW_VLAN ||
933 	     first->tx_flags & ICE_TX_FLAGS_HW_OUTER_SINGLE_VLAN) ||
934 	    skb->priority != TC_PRIO_CONTROL) {
935 		first->vid &= ~VLAN_PRIO_MASK;
936 		/* Mask the lower 3 bits to set the 802.1p priority */
937 		first->vid |= (skb->priority << VLAN_PRIO_SHIFT) & VLAN_PRIO_MASK;
938 		/* if this is not already set it means a VLAN 0 + priority needs
939 		 * to be offloaded
940 		 */
941 		if (tx_ring->flags & ICE_TX_FLAGS_RING_VLAN_L2TAG2)
942 			first->tx_flags |= ICE_TX_FLAGS_HW_OUTER_SINGLE_VLAN;
943 		else
944 			first->tx_flags |= ICE_TX_FLAGS_HW_VLAN;
945 	}
946 }
947 
948 /**
949  * ice_dcb_is_mib_change_pending - Check if MIB change is pending
950  * @state: MIB change state
951  */
952 static bool ice_dcb_is_mib_change_pending(u8 state)
953 {
954 	return ICE_AQ_LLDP_MIB_CHANGE_PENDING ==
955 		FIELD_GET(ICE_AQ_LLDP_MIB_CHANGE_STATE_M, state);
956 }
957 
958 /**
959  * ice_dcb_process_lldp_set_mib_change - Process MIB change
960  * @pf: ptr to ice_pf
961  * @event: pointer to the admin queue receive event
962  */
963 void
964 ice_dcb_process_lldp_set_mib_change(struct ice_pf *pf,
965 				    struct ice_rq_event_info *event)
966 {
967 	struct ice_aqc_port_ets_elem buf = { 0 };
968 	struct device *dev = ice_pf_to_dev(pf);
969 	struct ice_aqc_lldp_get_mib *mib;
970 	struct ice_dcbx_cfg tmp_dcbx_cfg;
971 	bool pending_handled = true;
972 	bool need_reconfig = false;
973 	struct ice_port_info *pi;
974 	u8 mib_type;
975 	int ret;
976 
977 	/* Not DCB capable or capability disabled */
978 	if (!(test_bit(ICE_FLAG_DCB_CAPABLE, pf->flags)))
979 		return;
980 
981 	if (pf->dcbx_cap & DCB_CAP_DCBX_HOST) {
982 		dev_dbg(dev, "MIB Change Event in HOST mode\n");
983 		return;
984 	}
985 
986 	pi = pf->hw.port_info;
987 	mib = (struct ice_aqc_lldp_get_mib *)&event->desc.params.raw;
988 
989 	/* Ignore if event is not for Nearest Bridge */
990 	mib_type = FIELD_GET(ICE_AQ_LLDP_BRID_TYPE_M, mib->type);
991 	dev_dbg(dev, "LLDP event MIB bridge type 0x%x\n", mib_type);
992 	if (mib_type != ICE_AQ_LLDP_BRID_TYPE_NEAREST_BRID)
993 		return;
994 
995 	/* A pending change event contains accurate config information, and
996 	 * the FW setting has not been updaed yet, so detect if change is
997 	 * pending to determine where to pull config information from
998 	 * (FW vs event)
999 	 */
1000 	if (ice_dcb_is_mib_change_pending(mib->state))
1001 		pending_handled = false;
1002 
1003 	/* Check MIB Type and return if event for Remote MIB update */
1004 	mib_type = FIELD_GET(ICE_AQ_LLDP_MIB_TYPE_M, mib->type);
1005 	dev_dbg(dev, "LLDP event mib type %s\n", mib_type ? "remote" : "local");
1006 	if (mib_type == ICE_AQ_LLDP_MIB_REMOTE) {
1007 		/* Update the remote cached instance and return */
1008 		if (!pending_handled) {
1009 			ice_get_dcb_cfg_from_mib_change(pi, event);
1010 		} else {
1011 			ret =
1012 			  ice_aq_get_dcb_cfg(pi->hw, ICE_AQ_LLDP_MIB_REMOTE,
1013 					     ICE_AQ_LLDP_BRID_TYPE_NEAREST_BRID,
1014 					     &pi->qos_cfg.remote_dcbx_cfg);
1015 			if (ret)
1016 				dev_dbg(dev, "Failed to get remote DCB config\n");
1017 		}
1018 		return;
1019 	}
1020 
1021 	/* That a DCB change has happened is now determined */
1022 	mutex_lock(&pf->tc_mutex);
1023 
1024 	/* store the old configuration */
1025 	tmp_dcbx_cfg = pi->qos_cfg.local_dcbx_cfg;
1026 
1027 	/* Reset the old DCBX configuration data */
1028 	memset(&pi->qos_cfg.local_dcbx_cfg, 0,
1029 	       sizeof(pi->qos_cfg.local_dcbx_cfg));
1030 
1031 	/* Get updated DCBX data from firmware */
1032 	if (!pending_handled) {
1033 		ice_get_dcb_cfg_from_mib_change(pi, event);
1034 	} else {
1035 		ret = ice_get_dcb_cfg(pi);
1036 		if (ret) {
1037 			dev_err(dev, "Failed to get DCB config\n");
1038 			goto out;
1039 		}
1040 	}
1041 
1042 	/* No change detected in DCBX configs */
1043 	if (!memcmp(&tmp_dcbx_cfg, &pi->qos_cfg.local_dcbx_cfg,
1044 		    sizeof(tmp_dcbx_cfg))) {
1045 		dev_dbg(dev, "No change detected in DCBX configuration.\n");
1046 		goto out;
1047 	}
1048 
1049 	pf->dcbx_cap = ice_dcb_get_mode(pi, false);
1050 
1051 	need_reconfig = ice_dcb_need_recfg(pf, &tmp_dcbx_cfg,
1052 					   &pi->qos_cfg.local_dcbx_cfg);
1053 	ice_dcbnl_flush_apps(pf, &tmp_dcbx_cfg, &pi->qos_cfg.local_dcbx_cfg);
1054 	if (!need_reconfig)
1055 		goto out;
1056 
1057 	/* Enable DCB tagging only when more than one TC */
1058 	if (ice_dcb_get_num_tc(&pi->qos_cfg.local_dcbx_cfg) > 1) {
1059 		dev_dbg(dev, "DCB tagging enabled (num TC > 1)\n");
1060 		set_bit(ICE_FLAG_DCB_ENA, pf->flags);
1061 	} else {
1062 		dev_dbg(dev, "DCB tagging disabled (num TC = 1)\n");
1063 		clear_bit(ICE_FLAG_DCB_ENA, pf->flags);
1064 	}
1065 
1066 	/* Send Execute Pending MIB Change event if it is a Pending event */
1067 	if (!pending_handled) {
1068 		ice_lldp_execute_pending_mib(&pf->hw);
1069 		pending_handled = true;
1070 	}
1071 
1072 	rtnl_lock();
1073 	/* disable VSIs affected by DCB changes */
1074 	ice_dcb_ena_dis_vsi(pf, false, true);
1075 
1076 	ret = ice_query_port_ets(pi, &buf, sizeof(buf), NULL);
1077 	if (ret) {
1078 		dev_err(dev, "Query Port ETS failed\n");
1079 		goto unlock_rtnl;
1080 	}
1081 
1082 	/* changes in configuration update VSI */
1083 	ice_pf_dcb_recfg(pf, false);
1084 
1085 	/* enable previously downed VSIs */
1086 	ice_dcb_ena_dis_vsi(pf, true, true);
1087 unlock_rtnl:
1088 	rtnl_unlock();
1089 out:
1090 	mutex_unlock(&pf->tc_mutex);
1091 
1092 	/* Send Execute Pending MIB Change event if it is a Pending event */
1093 	if (!pending_handled)
1094 		ice_lldp_execute_pending_mib(&pf->hw);
1095 }
1096