1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell RVU Ethernet driver
3  *
4  * Copyright (C) 2020 Marvell.
5  *
6  */
7 
8 #include <linux/interrupt.h>
9 #include <linux/pci.h>
10 #include <net/page_pool/helpers.h>
11 #include <net/tso.h>
12 #include <linux/bitfield.h>
13 
14 #include "otx2_reg.h"
15 #include "otx2_common.h"
16 #include "otx2_struct.h"
17 #include "cn10k.h"
18 
19 static void otx2_nix_rq_op_stats(struct queue_stats *stats,
20 				 struct otx2_nic *pfvf, int qidx)
21 {
22 	u64 incr = (u64)qidx << 32;
23 	u64 *ptr;
24 
25 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_OCTS);
26 	stats->bytes = otx2_atomic64_add(incr, ptr);
27 
28 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_PKTS);
29 	stats->pkts = otx2_atomic64_add(incr, ptr);
30 }
31 
32 static void otx2_nix_sq_op_stats(struct queue_stats *stats,
33 				 struct otx2_nic *pfvf, int qidx)
34 {
35 	u64 incr = (u64)qidx << 32;
36 	u64 *ptr;
37 
38 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_OCTS);
39 	stats->bytes = otx2_atomic64_add(incr, ptr);
40 
41 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_PKTS);
42 	stats->pkts = otx2_atomic64_add(incr, ptr);
43 }
44 
45 void otx2_update_lmac_stats(struct otx2_nic *pfvf)
46 {
47 	struct msg_req *req;
48 
49 	if (!netif_running(pfvf->netdev))
50 		return;
51 
52 	mutex_lock(&pfvf->mbox.lock);
53 	req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
54 	if (!req) {
55 		mutex_unlock(&pfvf->mbox.lock);
56 		return;
57 	}
58 
59 	otx2_sync_mbox_msg(&pfvf->mbox);
60 	mutex_unlock(&pfvf->mbox.lock);
61 }
62 
63 void otx2_update_lmac_fec_stats(struct otx2_nic *pfvf)
64 {
65 	struct msg_req *req;
66 
67 	if (!netif_running(pfvf->netdev))
68 		return;
69 	mutex_lock(&pfvf->mbox.lock);
70 	req = otx2_mbox_alloc_msg_cgx_fec_stats(&pfvf->mbox);
71 	if (req)
72 		otx2_sync_mbox_msg(&pfvf->mbox);
73 	mutex_unlock(&pfvf->mbox.lock);
74 }
75 
76 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
77 {
78 	struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
79 
80 	if (!pfvf->qset.rq)
81 		return 0;
82 
83 	otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
84 	return 1;
85 }
86 
87 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
88 {
89 	struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
90 
91 	if (!pfvf->qset.sq)
92 		return 0;
93 
94 	if (qidx >= pfvf->hw.non_qos_queues) {
95 		if (!test_bit(qidx - pfvf->hw.non_qos_queues, pfvf->qos.qos_sq_bmap))
96 			return 0;
97 	}
98 
99 	otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
100 	return 1;
101 }
102 
103 void otx2_get_dev_stats(struct otx2_nic *pfvf)
104 {
105 	struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
106 
107 	dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
108 	dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
109 	dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
110 	dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
111 	dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
112 	dev_stats->rx_frames = dev_stats->rx_bcast_frames +
113 			       dev_stats->rx_mcast_frames +
114 			       dev_stats->rx_ucast_frames;
115 
116 	dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
117 	dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
118 	dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
119 	dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
120 	dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
121 	dev_stats->tx_frames = dev_stats->tx_bcast_frames +
122 			       dev_stats->tx_mcast_frames +
123 			       dev_stats->tx_ucast_frames;
124 }
125 
126 void otx2_get_stats64(struct net_device *netdev,
127 		      struct rtnl_link_stats64 *stats)
128 {
129 	struct otx2_nic *pfvf = netdev_priv(netdev);
130 	struct otx2_dev_stats *dev_stats;
131 
132 	otx2_get_dev_stats(pfvf);
133 
134 	dev_stats = &pfvf->hw.dev_stats;
135 	stats->rx_bytes = dev_stats->rx_bytes;
136 	stats->rx_packets = dev_stats->rx_frames;
137 	stats->rx_dropped = dev_stats->rx_drops;
138 	stats->multicast = dev_stats->rx_mcast_frames;
139 
140 	stats->tx_bytes = dev_stats->tx_bytes;
141 	stats->tx_packets = dev_stats->tx_frames;
142 	stats->tx_dropped = dev_stats->tx_drops;
143 }
144 EXPORT_SYMBOL(otx2_get_stats64);
145 
146 /* Sync MAC address with RVU AF */
147 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
148 {
149 	struct nix_set_mac_addr *req;
150 	int err;
151 
152 	mutex_lock(&pfvf->mbox.lock);
153 	req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
154 	if (!req) {
155 		mutex_unlock(&pfvf->mbox.lock);
156 		return -ENOMEM;
157 	}
158 
159 	ether_addr_copy(req->mac_addr, mac);
160 
161 	err = otx2_sync_mbox_msg(&pfvf->mbox);
162 	mutex_unlock(&pfvf->mbox.lock);
163 	return err;
164 }
165 
166 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
167 				struct net_device *netdev)
168 {
169 	struct nix_get_mac_addr_rsp *rsp;
170 	struct mbox_msghdr *msghdr;
171 	struct msg_req *req;
172 	int err;
173 
174 	mutex_lock(&pfvf->mbox.lock);
175 	req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
176 	if (!req) {
177 		mutex_unlock(&pfvf->mbox.lock);
178 		return -ENOMEM;
179 	}
180 
181 	err = otx2_sync_mbox_msg(&pfvf->mbox);
182 	if (err) {
183 		mutex_unlock(&pfvf->mbox.lock);
184 		return err;
185 	}
186 
187 	msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
188 	if (IS_ERR(msghdr)) {
189 		mutex_unlock(&pfvf->mbox.lock);
190 		return PTR_ERR(msghdr);
191 	}
192 	rsp = (struct nix_get_mac_addr_rsp *)msghdr;
193 	eth_hw_addr_set(netdev, rsp->mac_addr);
194 	mutex_unlock(&pfvf->mbox.lock);
195 
196 	return 0;
197 }
198 
199 int otx2_set_mac_address(struct net_device *netdev, void *p)
200 {
201 	struct otx2_nic *pfvf = netdev_priv(netdev);
202 	struct sockaddr *addr = p;
203 
204 	if (!is_valid_ether_addr(addr->sa_data))
205 		return -EADDRNOTAVAIL;
206 
207 	if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data)) {
208 		eth_hw_addr_set(netdev, addr->sa_data);
209 		/* update dmac field in vlan offload rule */
210 		if (netif_running(netdev) &&
211 		    pfvf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
212 			otx2_install_rxvlan_offload_flow(pfvf);
213 		/* update dmac address in ntuple and DMAC filter list */
214 		if (pfvf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
215 			otx2_dmacflt_update_pfmac_flow(pfvf);
216 	} else {
217 		return -EPERM;
218 	}
219 
220 	return 0;
221 }
222 EXPORT_SYMBOL(otx2_set_mac_address);
223 
224 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
225 {
226 	struct nix_frs_cfg *req;
227 	u16 maxlen;
228 	int err;
229 
230 	maxlen = otx2_get_max_mtu(pfvf) + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
231 
232 	mutex_lock(&pfvf->mbox.lock);
233 	req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
234 	if (!req) {
235 		mutex_unlock(&pfvf->mbox.lock);
236 		return -ENOMEM;
237 	}
238 
239 	req->maxlen = pfvf->netdev->mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
240 
241 	/* Use max receive length supported by hardware for loopback devices */
242 	if (is_otx2_lbkvf(pfvf->pdev))
243 		req->maxlen = maxlen;
244 
245 	err = otx2_sync_mbox_msg(&pfvf->mbox);
246 	mutex_unlock(&pfvf->mbox.lock);
247 	return err;
248 }
249 
250 int otx2_config_pause_frm(struct otx2_nic *pfvf)
251 {
252 	struct cgx_pause_frm_cfg *req;
253 	int err;
254 
255 	if (is_otx2_lbkvf(pfvf->pdev))
256 		return 0;
257 
258 	mutex_lock(&pfvf->mbox.lock);
259 	req = otx2_mbox_alloc_msg_cgx_cfg_pause_frm(&pfvf->mbox);
260 	if (!req) {
261 		err = -ENOMEM;
262 		goto unlock;
263 	}
264 
265 	req->rx_pause = !!(pfvf->flags & OTX2_FLAG_RX_PAUSE_ENABLED);
266 	req->tx_pause = !!(pfvf->flags & OTX2_FLAG_TX_PAUSE_ENABLED);
267 	req->set = 1;
268 
269 	err = otx2_sync_mbox_msg(&pfvf->mbox);
270 unlock:
271 	mutex_unlock(&pfvf->mbox.lock);
272 	return err;
273 }
274 EXPORT_SYMBOL(otx2_config_pause_frm);
275 
276 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
277 {
278 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
279 	struct nix_rss_flowkey_cfg_rsp *rsp;
280 	struct nix_rss_flowkey_cfg *req;
281 	int err;
282 
283 	mutex_lock(&pfvf->mbox.lock);
284 	req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
285 	if (!req) {
286 		mutex_unlock(&pfvf->mbox.lock);
287 		return -ENOMEM;
288 	}
289 	req->mcam_index = -1; /* Default or reserved index */
290 	req->flowkey_cfg = rss->flowkey_cfg;
291 	req->group = DEFAULT_RSS_CONTEXT_GROUP;
292 
293 	err = otx2_sync_mbox_msg(&pfvf->mbox);
294 	if (err)
295 		goto fail;
296 
297 	rsp = (struct nix_rss_flowkey_cfg_rsp *)
298 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
299 	if (IS_ERR(rsp)) {
300 		err = PTR_ERR(rsp);
301 		goto fail;
302 	}
303 
304 	pfvf->hw.flowkey_alg_idx = rsp->alg_idx;
305 fail:
306 	mutex_unlock(&pfvf->mbox.lock);
307 	return err;
308 }
309 
310 int otx2_set_rss_table(struct otx2_nic *pfvf, int ctx_id)
311 {
312 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
313 	const int index = rss->rss_size * ctx_id;
314 	struct mbox *mbox = &pfvf->mbox;
315 	struct otx2_rss_ctx *rss_ctx;
316 	struct nix_aq_enq_req *aq;
317 	int idx, err;
318 
319 	mutex_lock(&mbox->lock);
320 	rss_ctx = rss->rss_ctx[ctx_id];
321 	/* Get memory to put this msg */
322 	for (idx = 0; idx < rss->rss_size; idx++) {
323 		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
324 		if (!aq) {
325 			/* The shared memory buffer can be full.
326 			 * Flush it and retry
327 			 */
328 			err = otx2_sync_mbox_msg(mbox);
329 			if (err) {
330 				mutex_unlock(&mbox->lock);
331 				return err;
332 			}
333 			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
334 			if (!aq) {
335 				mutex_unlock(&mbox->lock);
336 				return -ENOMEM;
337 			}
338 		}
339 
340 		aq->rss.rq = rss_ctx->ind_tbl[idx];
341 
342 		/* Fill AQ info */
343 		aq->qidx = index + idx;
344 		aq->ctype = NIX_AQ_CTYPE_RSS;
345 		aq->op = NIX_AQ_INSTOP_INIT;
346 	}
347 	err = otx2_sync_mbox_msg(mbox);
348 	mutex_unlock(&mbox->lock);
349 	return err;
350 }
351 
352 void otx2_set_rss_key(struct otx2_nic *pfvf)
353 {
354 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
355 	u64 *key = (u64 *)&rss->key[4];
356 	int idx;
357 
358 	/* 352bit or 44byte key needs to be configured as below
359 	 * NIX_LF_RX_SECRETX0 = key<351:288>
360 	 * NIX_LF_RX_SECRETX1 = key<287:224>
361 	 * NIX_LF_RX_SECRETX2 = key<223:160>
362 	 * NIX_LF_RX_SECRETX3 = key<159:96>
363 	 * NIX_LF_RX_SECRETX4 = key<95:32>
364 	 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
365 	 */
366 	otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
367 		     (u64)(*((u32 *)&rss->key)) << 32);
368 	idx = sizeof(rss->key) / sizeof(u64);
369 	while (idx > 0) {
370 		idx--;
371 		otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
372 	}
373 }
374 
375 int otx2_rss_init(struct otx2_nic *pfvf)
376 {
377 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
378 	struct otx2_rss_ctx *rss_ctx;
379 	int idx, ret = 0;
380 
381 	rss->rss_size = sizeof(*rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
382 
383 	/* Init RSS key if it is not setup already */
384 	if (!rss->enable)
385 		netdev_rss_key_fill(rss->key, sizeof(rss->key));
386 	otx2_set_rss_key(pfvf);
387 
388 	if (!netif_is_rxfh_configured(pfvf->netdev)) {
389 		/* Set RSS group 0 as default indirection table */
390 		rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP] = kzalloc(rss->rss_size,
391 								  GFP_KERNEL);
392 		if (!rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP])
393 			return -ENOMEM;
394 
395 		rss_ctx = rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP];
396 		for (idx = 0; idx < rss->rss_size; idx++)
397 			rss_ctx->ind_tbl[idx] =
398 				ethtool_rxfh_indir_default(idx,
399 							   pfvf->hw.rx_queues);
400 	}
401 	ret = otx2_set_rss_table(pfvf, DEFAULT_RSS_CONTEXT_GROUP);
402 	if (ret)
403 		return ret;
404 
405 	/* Flowkey or hash config to be used for generating flow tag */
406 	rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
407 			   NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
408 			   NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
409 			   NIX_FLOW_KEY_TYPE_SCTP | NIX_FLOW_KEY_TYPE_VLAN |
410 			   NIX_FLOW_KEY_TYPE_IPV4_PROTO;
411 
412 	ret = otx2_set_flowkey_cfg(pfvf);
413 	if (ret)
414 		return ret;
415 
416 	rss->enable = true;
417 	return 0;
418 }
419 
420 /* Setup UDP segmentation algorithm in HW */
421 static void otx2_setup_udp_segmentation(struct nix_lso_format_cfg *lso, bool v4)
422 {
423 	struct nix_lso_format *field;
424 
425 	field = (struct nix_lso_format *)&lso->fields[0];
426 	lso->field_mask = GENMASK(18, 0);
427 
428 	/* IP's Length field */
429 	field->layer = NIX_TXLAYER_OL3;
430 	/* In ipv4, length field is at offset 2 bytes, for ipv6 it's 4 */
431 	field->offset = v4 ? 2 : 4;
432 	field->sizem1 = 1; /* i.e 2 bytes */
433 	field->alg = NIX_LSOALG_ADD_PAYLEN;
434 	field++;
435 
436 	/* No ID field in IPv6 header */
437 	if (v4) {
438 		/* Increment IPID */
439 		field->layer = NIX_TXLAYER_OL3;
440 		field->offset = 4;
441 		field->sizem1 = 1; /* i.e 2 bytes */
442 		field->alg = NIX_LSOALG_ADD_SEGNUM;
443 		field++;
444 	}
445 
446 	/* Update length in UDP header */
447 	field->layer = NIX_TXLAYER_OL4;
448 	field->offset = 4;
449 	field->sizem1 = 1;
450 	field->alg = NIX_LSOALG_ADD_PAYLEN;
451 }
452 
453 /* Setup segmentation algorithms in HW and retrieve algorithm index */
454 void otx2_setup_segmentation(struct otx2_nic *pfvf)
455 {
456 	struct nix_lso_format_cfg_rsp *rsp;
457 	struct nix_lso_format_cfg *lso;
458 	struct otx2_hw *hw = &pfvf->hw;
459 	int err;
460 
461 	mutex_lock(&pfvf->mbox.lock);
462 
463 	/* UDPv4 segmentation */
464 	lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
465 	if (!lso)
466 		goto fail;
467 
468 	/* Setup UDP/IP header fields that HW should update per segment */
469 	otx2_setup_udp_segmentation(lso, true);
470 
471 	err = otx2_sync_mbox_msg(&pfvf->mbox);
472 	if (err)
473 		goto fail;
474 
475 	rsp = (struct nix_lso_format_cfg_rsp *)
476 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
477 	if (IS_ERR(rsp))
478 		goto fail;
479 
480 	hw->lso_udpv4_idx = rsp->lso_format_idx;
481 
482 	/* UDPv6 segmentation */
483 	lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
484 	if (!lso)
485 		goto fail;
486 
487 	/* Setup UDP/IP header fields that HW should update per segment */
488 	otx2_setup_udp_segmentation(lso, false);
489 
490 	err = otx2_sync_mbox_msg(&pfvf->mbox);
491 	if (err)
492 		goto fail;
493 
494 	rsp = (struct nix_lso_format_cfg_rsp *)
495 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
496 	if (IS_ERR(rsp))
497 		goto fail;
498 
499 	hw->lso_udpv6_idx = rsp->lso_format_idx;
500 	mutex_unlock(&pfvf->mbox.lock);
501 	return;
502 fail:
503 	mutex_unlock(&pfvf->mbox.lock);
504 	netdev_info(pfvf->netdev,
505 		    "Failed to get LSO index for UDP GSO offload, disabling\n");
506 	pfvf->netdev->hw_features &= ~NETIF_F_GSO_UDP_L4;
507 }
508 
509 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
510 {
511 	/* Configure CQE interrupt coalescing parameters
512 	 *
513 	 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
514 	 * set 1 less than cq_ecount_wait. And cq_time_wait is in
515 	 * usecs, convert that to 100ns count.
516 	 */
517 	otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
518 		     ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
519 		     ((u64)pfvf->hw.cq_qcount_wait << 32) |
520 		     (pfvf->hw.cq_ecount_wait - 1));
521 }
522 
523 static int otx2_alloc_pool_buf(struct otx2_nic *pfvf, struct otx2_pool *pool,
524 			       dma_addr_t *dma)
525 {
526 	unsigned int offset = 0;
527 	struct page *page;
528 	size_t sz;
529 
530 	sz = SKB_DATA_ALIGN(pool->rbsize);
531 	sz = ALIGN(sz, OTX2_ALIGN);
532 
533 	page = page_pool_alloc_frag(pool->page_pool, &offset, sz, GFP_ATOMIC);
534 	if (unlikely(!page))
535 		return -ENOMEM;
536 
537 	*dma = page_pool_get_dma_addr(page) + offset;
538 	return 0;
539 }
540 
541 static int __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
542 			     dma_addr_t *dma)
543 {
544 	u8 *buf;
545 
546 	if (pool->page_pool)
547 		return otx2_alloc_pool_buf(pfvf, pool, dma);
548 
549 	buf = napi_alloc_frag_align(pool->rbsize, OTX2_ALIGN);
550 	if (unlikely(!buf))
551 		return -ENOMEM;
552 
553 	*dma = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
554 				    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
555 	if (unlikely(dma_mapping_error(pfvf->dev, *dma))) {
556 		page_frag_free(buf);
557 		return -ENOMEM;
558 	}
559 
560 	return 0;
561 }
562 
563 int otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
564 		    dma_addr_t *dma)
565 {
566 	int ret;
567 
568 	local_bh_disable();
569 	ret = __otx2_alloc_rbuf(pfvf, pool, dma);
570 	local_bh_enable();
571 	return ret;
572 }
573 
574 int otx2_alloc_buffer(struct otx2_nic *pfvf, struct otx2_cq_queue *cq,
575 		      dma_addr_t *dma)
576 {
577 	if (unlikely(__otx2_alloc_rbuf(pfvf, cq->rbpool, dma))) {
578 		struct refill_work *work;
579 		struct delayed_work *dwork;
580 
581 		work = &pfvf->refill_wrk[cq->cq_idx];
582 		dwork = &work->pool_refill_work;
583 		/* Schedule a task if no other task is running */
584 		if (!cq->refill_task_sched) {
585 			cq->refill_task_sched = true;
586 			schedule_delayed_work(dwork,
587 					      msecs_to_jiffies(100));
588 		}
589 		return -ENOMEM;
590 	}
591 	return 0;
592 }
593 
594 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
595 {
596 	struct otx2_nic *pfvf = netdev_priv(netdev);
597 
598 	schedule_work(&pfvf->reset_task);
599 }
600 EXPORT_SYMBOL(otx2_tx_timeout);
601 
602 void otx2_get_mac_from_af(struct net_device *netdev)
603 {
604 	struct otx2_nic *pfvf = netdev_priv(netdev);
605 	int err;
606 
607 	err = otx2_hw_get_mac_addr(pfvf, netdev);
608 	if (err)
609 		dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
610 
611 	/* If AF doesn't provide a valid MAC, generate a random one */
612 	if (!is_valid_ether_addr(netdev->dev_addr))
613 		eth_hw_addr_random(netdev);
614 }
615 EXPORT_SYMBOL(otx2_get_mac_from_af);
616 
617 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl, int prio, bool txschq_for_pfc)
618 {
619 	u16 (*schq_list)[MAX_TXSCHQ_PER_FUNC];
620 	struct otx2_hw *hw = &pfvf->hw;
621 	struct nix_txschq_config *req;
622 	u64 schq, parent;
623 	u64 dwrr_val;
624 
625 	dwrr_val = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
626 
627 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
628 	if (!req)
629 		return -ENOMEM;
630 
631 	req->lvl = lvl;
632 	req->num_regs = 1;
633 
634 	schq_list = hw->txschq_list;
635 #ifdef CONFIG_DCB
636 	if (txschq_for_pfc)
637 		schq_list = pfvf->pfc_schq_list;
638 #endif
639 
640 	schq = schq_list[lvl][prio];
641 	/* Set topology e.t.c configuration */
642 	if (lvl == NIX_TXSCH_LVL_SMQ) {
643 		req->reg[0] = NIX_AF_SMQX_CFG(schq);
644 		req->regval[0] = ((u64)pfvf->tx_max_pktlen << 8) | OTX2_MIN_MTU;
645 		req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
646 				  (0x2ULL << 36);
647 		/* Set link type for DWRR MTU selection on CN10K silicons */
648 		if (!is_dev_otx2(pfvf->pdev))
649 			req->regval[0] |= FIELD_PREP(GENMASK_ULL(58, 57),
650 						(u64)hw->smq_link_type);
651 		req->num_regs++;
652 		/* MDQ config */
653 		parent = schq_list[NIX_TXSCH_LVL_TL4][prio];
654 		req->reg[1] = NIX_AF_MDQX_PARENT(schq);
655 		req->regval[1] = parent << 16;
656 		req->num_regs++;
657 		/* Set DWRR quantum */
658 		req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
659 		req->regval[2] =  dwrr_val;
660 	} else if (lvl == NIX_TXSCH_LVL_TL4) {
661 		parent = schq_list[NIX_TXSCH_LVL_TL3][prio];
662 		req->reg[0] = NIX_AF_TL4X_PARENT(schq);
663 		req->regval[0] = parent << 16;
664 		req->num_regs++;
665 		req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
666 		req->regval[1] = dwrr_val;
667 	} else if (lvl == NIX_TXSCH_LVL_TL3) {
668 		parent = schq_list[NIX_TXSCH_LVL_TL2][prio];
669 		req->reg[0] = NIX_AF_TL3X_PARENT(schq);
670 		req->regval[0] = parent << 16;
671 		req->num_regs++;
672 		req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
673 		req->regval[1] = dwrr_val;
674 		if (lvl == hw->txschq_link_cfg_lvl) {
675 			req->num_regs++;
676 			req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
677 			/* Enable this queue and backpressure
678 			 * and set relative channel
679 			 */
680 			req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
681 		}
682 	} else if (lvl == NIX_TXSCH_LVL_TL2) {
683 		parent = schq_list[NIX_TXSCH_LVL_TL1][prio];
684 		req->reg[0] = NIX_AF_TL2X_PARENT(schq);
685 		req->regval[0] = parent << 16;
686 
687 		req->num_regs++;
688 		req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
689 		req->regval[1] = TXSCH_TL1_DFLT_RR_PRIO << 24 | dwrr_val;
690 
691 		if (lvl == hw->txschq_link_cfg_lvl) {
692 			req->num_regs++;
693 			req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
694 			/* Enable this queue and backpressure
695 			 * and set relative channel
696 			 */
697 			req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
698 		}
699 	} else if (lvl == NIX_TXSCH_LVL_TL1) {
700 		/* Default config for TL1.
701 		 * For VF this is always ignored.
702 		 */
703 
704 		/* On CN10K, if RR_WEIGHT is greater than 16384, HW will
705 		 * clip it to 16384, so configuring a 24bit max value
706 		 * will work on both OTx2 and CN10K.
707 		 */
708 		req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
709 		req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
710 
711 		req->num_regs++;
712 		req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
713 		req->regval[1] = (TXSCH_TL1_DFLT_RR_PRIO << 1);
714 
715 		req->num_regs++;
716 		req->reg[2] = NIX_AF_TL1X_CIR(schq);
717 		req->regval[2] = 0;
718 	}
719 
720 	return otx2_sync_mbox_msg(&pfvf->mbox);
721 }
722 EXPORT_SYMBOL(otx2_txschq_config);
723 
724 int otx2_smq_flush(struct otx2_nic *pfvf, int smq)
725 {
726 	struct nix_txschq_config *req;
727 	int rc;
728 
729 	mutex_lock(&pfvf->mbox.lock);
730 
731 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
732 	if (!req) {
733 		mutex_unlock(&pfvf->mbox.lock);
734 		return -ENOMEM;
735 	}
736 
737 	req->lvl = NIX_TXSCH_LVL_SMQ;
738 	req->reg[0] = NIX_AF_SMQX_CFG(smq);
739 	req->regval[0] |= BIT_ULL(49);
740 	req->num_regs++;
741 
742 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
743 	mutex_unlock(&pfvf->mbox.lock);
744 	return rc;
745 }
746 EXPORT_SYMBOL(otx2_smq_flush);
747 
748 int otx2_txsch_alloc(struct otx2_nic *pfvf)
749 {
750 	struct nix_txsch_alloc_req *req;
751 	struct nix_txsch_alloc_rsp *rsp;
752 	int lvl, schq, rc;
753 
754 	/* Get memory to put this msg */
755 	req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
756 	if (!req)
757 		return -ENOMEM;
758 
759 	/* Request one schq per level */
760 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
761 		req->schq[lvl] = 1;
762 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
763 	if (rc)
764 		return rc;
765 
766 	rsp = (struct nix_txsch_alloc_rsp *)
767 	      otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
768 	if (IS_ERR(rsp))
769 		return PTR_ERR(rsp);
770 
771 	/* Setup transmit scheduler list */
772 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
773 		for (schq = 0; schq < rsp->schq[lvl]; schq++)
774 			pfvf->hw.txschq_list[lvl][schq] =
775 				rsp->schq_list[lvl][schq];
776 
777 	pfvf->hw.txschq_link_cfg_lvl = rsp->link_cfg_lvl;
778 	pfvf->hw.txschq_aggr_lvl_rr_prio = rsp->aggr_lvl_rr_prio;
779 
780 	return 0;
781 }
782 
783 void otx2_txschq_free_one(struct otx2_nic *pfvf, u16 lvl, u16 schq)
784 {
785 	struct nix_txsch_free_req *free_req;
786 	int err;
787 
788 	mutex_lock(&pfvf->mbox.lock);
789 
790 	free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
791 	if (!free_req) {
792 		mutex_unlock(&pfvf->mbox.lock);
793 		netdev_err(pfvf->netdev,
794 			   "Failed alloc txschq free req\n");
795 		return;
796 	}
797 
798 	free_req->schq_lvl = lvl;
799 	free_req->schq = schq;
800 
801 	err = otx2_sync_mbox_msg(&pfvf->mbox);
802 	if (err) {
803 		netdev_err(pfvf->netdev,
804 			   "Failed stop txschq %d at level %d\n", schq, lvl);
805 	}
806 
807 	mutex_unlock(&pfvf->mbox.lock);
808 }
809 
810 void otx2_txschq_stop(struct otx2_nic *pfvf)
811 {
812 	int lvl, schq;
813 
814 	/* free non QOS TLx nodes */
815 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
816 		otx2_txschq_free_one(pfvf, lvl,
817 				     pfvf->hw.txschq_list[lvl][0]);
818 
819 	/* Clear the txschq list */
820 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
821 		for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
822 			pfvf->hw.txschq_list[lvl][schq] = 0;
823 	}
824 
825 }
826 
827 void otx2_sqb_flush(struct otx2_nic *pfvf)
828 {
829 	int qidx, sqe_tail, sqe_head;
830 	struct otx2_snd_queue *sq;
831 	u64 incr, *ptr, val;
832 	int timeout = 1000;
833 
834 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
835 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
836 		sq = &pfvf->qset.sq[qidx];
837 		if (!sq->sqb_ptrs)
838 			continue;
839 
840 		incr = (u64)qidx << 32;
841 		while (timeout) {
842 			val = otx2_atomic64_add(incr, ptr);
843 			sqe_head = (val >> 20) & 0x3F;
844 			sqe_tail = (val >> 28) & 0x3F;
845 			if (sqe_head == sqe_tail)
846 				break;
847 			usleep_range(1, 3);
848 			timeout--;
849 		}
850 	}
851 }
852 
853 /* RED and drop levels of CQ on packet reception.
854  * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
855  */
856 #define RQ_PASS_LVL_CQ(skid, qsize)	((((skid) + 16) * 256) / (qsize))
857 #define RQ_DROP_LVL_CQ(skid, qsize)	(((skid) * 256) / (qsize))
858 
859 /* RED and drop levels of AURA for packet reception.
860  * For AURA level is measure of fullness (0x0 = empty, 255 = full).
861  * Eg: For RQ length 1K, for pass/drop level 204/230.
862  * RED accepts pkts if free pointers > 102 & <= 205.
863  * Drops pkts if free pointers < 102.
864  */
865 #define RQ_BP_LVL_AURA   (255 - ((85 * 256) / 100)) /* BP when 85% is full */
866 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
867 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
868 
869 static int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
870 {
871 	struct otx2_qset *qset = &pfvf->qset;
872 	struct nix_aq_enq_req *aq;
873 
874 	/* Get memory to put this msg */
875 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
876 	if (!aq)
877 		return -ENOMEM;
878 
879 	aq->rq.cq = qidx;
880 	aq->rq.ena = 1;
881 	aq->rq.pb_caching = 1;
882 	aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
883 	aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
884 	aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
885 	aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
886 	aq->rq.qint_idx = 0;
887 	aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
888 	aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
889 	aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
890 	aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
891 	aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
892 	aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
893 
894 	/* Fill AQ info */
895 	aq->qidx = qidx;
896 	aq->ctype = NIX_AQ_CTYPE_RQ;
897 	aq->op = NIX_AQ_INSTOP_INIT;
898 
899 	return otx2_sync_mbox_msg(&pfvf->mbox);
900 }
901 
902 int otx2_sq_aq_init(void *dev, u16 qidx, u16 sqb_aura)
903 {
904 	struct otx2_nic *pfvf = dev;
905 	struct otx2_snd_queue *sq;
906 	struct nix_aq_enq_req *aq;
907 
908 	sq = &pfvf->qset.sq[qidx];
909 	sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
910 	/* Get memory to put this msg */
911 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
912 	if (!aq)
913 		return -ENOMEM;
914 
915 	aq->sq.cq = pfvf->hw.rx_queues + qidx;
916 	aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
917 	aq->sq.cq_ena = 1;
918 	aq->sq.ena = 1;
919 	aq->sq.smq = otx2_get_smq_idx(pfvf, qidx);
920 	aq->sq.smq_rr_quantum = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
921 	aq->sq.default_chan = pfvf->hw.tx_chan_base;
922 	aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
923 	aq->sq.sqb_aura = sqb_aura;
924 	aq->sq.sq_int_ena = NIX_SQINT_BITS;
925 	aq->sq.qint_idx = 0;
926 	/* Due pipelining impact minimum 2000 unused SQ CQE's
927 	 * need to maintain to avoid CQ overflow.
928 	 */
929 	aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (pfvf->qset.sqe_cnt));
930 
931 	/* Fill AQ info */
932 	aq->qidx = qidx;
933 	aq->ctype = NIX_AQ_CTYPE_SQ;
934 	aq->op = NIX_AQ_INSTOP_INIT;
935 
936 	return otx2_sync_mbox_msg(&pfvf->mbox);
937 }
938 
939 int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
940 {
941 	struct otx2_qset *qset = &pfvf->qset;
942 	struct otx2_snd_queue *sq;
943 	struct otx2_pool *pool;
944 	int err;
945 
946 	pool = &pfvf->qset.pool[sqb_aura];
947 	sq = &qset->sq[qidx];
948 	sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
949 	sq->sqe_cnt = qset->sqe_cnt;
950 
951 	err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
952 	if (err)
953 		return err;
954 
955 	if (qidx < pfvf->hw.tx_queues) {
956 		err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
957 				 TSO_HEADER_SIZE);
958 		if (err)
959 			return err;
960 	}
961 
962 	sq->sqe_base = sq->sqe->base;
963 	sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
964 	if (!sq->sg)
965 		return -ENOMEM;
966 
967 	if (pfvf->ptp && qidx < pfvf->hw.tx_queues) {
968 		err = qmem_alloc(pfvf->dev, &sq->timestamps, qset->sqe_cnt,
969 				 sizeof(*sq->timestamps));
970 		if (err)
971 			return err;
972 	}
973 
974 	sq->head = 0;
975 	sq->cons_head = 0;
976 	sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
977 	sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
978 	/* Set SQE threshold to 10% of total SQEs */
979 	sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
980 	sq->aura_id = sqb_aura;
981 	sq->aura_fc_addr = pool->fc_addr->base;
982 	sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
983 
984 	sq->stats.bytes = 0;
985 	sq->stats.pkts = 0;
986 
987 	return pfvf->hw_ops->sq_aq_init(pfvf, qidx, sqb_aura);
988 
989 }
990 
991 static int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
992 {
993 	struct otx2_qset *qset = &pfvf->qset;
994 	int err, pool_id, non_xdp_queues;
995 	struct nix_aq_enq_req *aq;
996 	struct otx2_cq_queue *cq;
997 
998 	cq = &qset->cq[qidx];
999 	cq->cq_idx = qidx;
1000 	non_xdp_queues = pfvf->hw.rx_queues + pfvf->hw.tx_queues;
1001 	if (qidx < pfvf->hw.rx_queues) {
1002 		cq->cq_type = CQ_RX;
1003 		cq->cint_idx = qidx;
1004 		cq->cqe_cnt = qset->rqe_cnt;
1005 		if (pfvf->xdp_prog)
1006 			xdp_rxq_info_reg(&cq->xdp_rxq, pfvf->netdev, qidx, 0);
1007 	} else if (qidx < non_xdp_queues) {
1008 		cq->cq_type = CQ_TX;
1009 		cq->cint_idx = qidx - pfvf->hw.rx_queues;
1010 		cq->cqe_cnt = qset->sqe_cnt;
1011 	} else {
1012 		if (pfvf->hw.xdp_queues &&
1013 		    qidx < non_xdp_queues + pfvf->hw.xdp_queues) {
1014 			cq->cq_type = CQ_XDP;
1015 			cq->cint_idx = qidx - non_xdp_queues;
1016 			cq->cqe_cnt = qset->sqe_cnt;
1017 		} else {
1018 			cq->cq_type = CQ_QOS;
1019 			cq->cint_idx = qidx - non_xdp_queues -
1020 				       pfvf->hw.xdp_queues;
1021 			cq->cqe_cnt = qset->sqe_cnt;
1022 		}
1023 	}
1024 	cq->cqe_size = pfvf->qset.xqe_size;
1025 
1026 	/* Allocate memory for CQEs */
1027 	err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
1028 	if (err)
1029 		return err;
1030 
1031 	/* Save CQE CPU base for faster reference */
1032 	cq->cqe_base = cq->cqe->base;
1033 	/* In case where all RQs auras point to single pool,
1034 	 * all CQs receive buffer pool also point to same pool.
1035 	 */
1036 	pool_id = ((cq->cq_type == CQ_RX) &&
1037 		   (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
1038 	cq->rbpool = &qset->pool[pool_id];
1039 	cq->refill_task_sched = false;
1040 
1041 	/* Get memory to put this msg */
1042 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
1043 	if (!aq)
1044 		return -ENOMEM;
1045 
1046 	aq->cq.ena = 1;
1047 	aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
1048 	aq->cq.caching = 1;
1049 	aq->cq.base = cq->cqe->iova;
1050 	aq->cq.cint_idx = cq->cint_idx;
1051 	aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
1052 	aq->cq.qint_idx = 0;
1053 	aq->cq.avg_level = 255;
1054 
1055 	if (qidx < pfvf->hw.rx_queues) {
1056 		aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
1057 		aq->cq.drop_ena = 1;
1058 
1059 		if (!is_otx2_lbkvf(pfvf->pdev)) {
1060 			/* Enable receive CQ backpressure */
1061 			aq->cq.bp_ena = 1;
1062 #ifdef CONFIG_DCB
1063 			aq->cq.bpid = pfvf->bpid[pfvf->queue_to_pfc_map[qidx]];
1064 #else
1065 			aq->cq.bpid = pfvf->bpid[0];
1066 #endif
1067 
1068 			/* Set backpressure level is same as cq pass level */
1069 			aq->cq.bp = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
1070 		}
1071 	}
1072 
1073 	/* Fill AQ info */
1074 	aq->qidx = qidx;
1075 	aq->ctype = NIX_AQ_CTYPE_CQ;
1076 	aq->op = NIX_AQ_INSTOP_INIT;
1077 
1078 	return otx2_sync_mbox_msg(&pfvf->mbox);
1079 }
1080 
1081 static void otx2_pool_refill_task(struct work_struct *work)
1082 {
1083 	struct otx2_cq_queue *cq;
1084 	struct otx2_pool *rbpool;
1085 	struct refill_work *wrk;
1086 	int qidx, free_ptrs = 0;
1087 	struct otx2_nic *pfvf;
1088 	dma_addr_t bufptr;
1089 
1090 	wrk = container_of(work, struct refill_work, pool_refill_work.work);
1091 	pfvf = wrk->pf;
1092 	qidx = wrk - pfvf->refill_wrk;
1093 	cq = &pfvf->qset.cq[qidx];
1094 	rbpool = cq->rbpool;
1095 	free_ptrs = cq->pool_ptrs;
1096 
1097 	while (cq->pool_ptrs) {
1098 		if (otx2_alloc_rbuf(pfvf, rbpool, &bufptr)) {
1099 			/* Schedule a WQ if we fails to free atleast half of the
1100 			 * pointers else enable napi for this RQ.
1101 			 */
1102 			if (!((free_ptrs - cq->pool_ptrs) > free_ptrs / 2)) {
1103 				struct delayed_work *dwork;
1104 
1105 				dwork = &wrk->pool_refill_work;
1106 				schedule_delayed_work(dwork,
1107 						      msecs_to_jiffies(100));
1108 			} else {
1109 				cq->refill_task_sched = false;
1110 			}
1111 			return;
1112 		}
1113 		pfvf->hw_ops->aura_freeptr(pfvf, qidx, bufptr + OTX2_HEAD_ROOM);
1114 		cq->pool_ptrs--;
1115 	}
1116 	cq->refill_task_sched = false;
1117 }
1118 
1119 int otx2_config_nix_queues(struct otx2_nic *pfvf)
1120 {
1121 	int qidx, err;
1122 
1123 	/* Initialize RX queues */
1124 	for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
1125 		u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
1126 
1127 		err = otx2_rq_init(pfvf, qidx, lpb_aura);
1128 		if (err)
1129 			return err;
1130 	}
1131 
1132 	/* Initialize TX queues */
1133 	for (qidx = 0; qidx < pfvf->hw.non_qos_queues; qidx++) {
1134 		u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1135 
1136 		err = otx2_sq_init(pfvf, qidx, sqb_aura);
1137 		if (err)
1138 			return err;
1139 	}
1140 
1141 	/* Initialize completion queues */
1142 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1143 		err = otx2_cq_init(pfvf, qidx);
1144 		if (err)
1145 			return err;
1146 	}
1147 
1148 	pfvf->cq_op_addr = (__force u64 *)otx2_get_regaddr(pfvf,
1149 							   NIX_LF_CQ_OP_STATUS);
1150 
1151 	/* Initialize work queue for receive buffer refill */
1152 	pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
1153 					sizeof(struct refill_work), GFP_KERNEL);
1154 	if (!pfvf->refill_wrk)
1155 		return -ENOMEM;
1156 
1157 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1158 		pfvf->refill_wrk[qidx].pf = pfvf;
1159 		INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
1160 				  otx2_pool_refill_task);
1161 	}
1162 	return 0;
1163 }
1164 
1165 int otx2_config_nix(struct otx2_nic *pfvf)
1166 {
1167 	struct nix_lf_alloc_req  *nixlf;
1168 	struct nix_lf_alloc_rsp *rsp;
1169 	int err;
1170 
1171 	pfvf->qset.xqe_size = pfvf->hw.xqe_size;
1172 
1173 	/* Get memory to put this msg */
1174 	nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
1175 	if (!nixlf)
1176 		return -ENOMEM;
1177 
1178 	/* Set RQ/SQ/CQ counts */
1179 	nixlf->rq_cnt = pfvf->hw.rx_queues;
1180 	nixlf->sq_cnt = otx2_get_total_tx_queues(pfvf);
1181 	nixlf->cq_cnt = pfvf->qset.cq_cnt;
1182 	nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
1183 	nixlf->rss_grps = MAX_RSS_GROUPS;
1184 	nixlf->xqe_sz = pfvf->hw.xqe_size == 128 ? NIX_XQESZ_W16 : NIX_XQESZ_W64;
1185 	/* We don't know absolute NPA LF idx attached.
1186 	 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
1187 	 * NPA LF attached to this RVU PF/VF.
1188 	 */
1189 	nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
1190 	/* Disable alignment pad, enable L2 length check,
1191 	 * enable L4 TCP/UDP checksum verification.
1192 	 */
1193 	nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
1194 
1195 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1196 	if (err)
1197 		return err;
1198 
1199 	rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
1200 							   &nixlf->hdr);
1201 	if (IS_ERR(rsp))
1202 		return PTR_ERR(rsp);
1203 
1204 	if (rsp->qints < 1)
1205 		return -ENXIO;
1206 
1207 	return rsp->hdr.rc;
1208 }
1209 
1210 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
1211 {
1212 	struct otx2_qset *qset = &pfvf->qset;
1213 	struct otx2_hw *hw = &pfvf->hw;
1214 	struct otx2_snd_queue *sq;
1215 	int sqb, qidx;
1216 	u64 iova, pa;
1217 
1218 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
1219 		sq = &qset->sq[qidx];
1220 		if (!sq->sqb_ptrs)
1221 			continue;
1222 		for (sqb = 0; sqb < sq->sqb_count; sqb++) {
1223 			if (!sq->sqb_ptrs[sqb])
1224 				continue;
1225 			iova = sq->sqb_ptrs[sqb];
1226 			pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1227 			dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
1228 					     DMA_FROM_DEVICE,
1229 					     DMA_ATTR_SKIP_CPU_SYNC);
1230 			put_page(virt_to_page(phys_to_virt(pa)));
1231 		}
1232 		sq->sqb_count = 0;
1233 	}
1234 }
1235 
1236 void otx2_free_bufs(struct otx2_nic *pfvf, struct otx2_pool *pool,
1237 		    u64 iova, int size)
1238 {
1239 	struct page *page;
1240 	u64 pa;
1241 
1242 	pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1243 	page = virt_to_head_page(phys_to_virt(pa));
1244 
1245 	if (pool->page_pool) {
1246 		page_pool_put_full_page(pool->page_pool, page, true);
1247 	} else {
1248 		dma_unmap_page_attrs(pfvf->dev, iova, size,
1249 				     DMA_FROM_DEVICE,
1250 				     DMA_ATTR_SKIP_CPU_SYNC);
1251 
1252 		put_page(page);
1253 	}
1254 }
1255 
1256 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
1257 {
1258 	int pool_id, pool_start = 0, pool_end = 0, size = 0;
1259 	struct otx2_pool *pool;
1260 	u64 iova;
1261 
1262 	if (type == AURA_NIX_SQ) {
1263 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
1264 		pool_end =  pool_start + pfvf->hw.sqpool_cnt;
1265 		size = pfvf->hw.sqb_size;
1266 	}
1267 	if (type == AURA_NIX_RQ) {
1268 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
1269 		pool_end = pfvf->hw.rqpool_cnt;
1270 		size = pfvf->rbsize;
1271 	}
1272 
1273 	/* Free SQB and RQB pointers from the aura pool */
1274 	for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
1275 		iova = otx2_aura_allocptr(pfvf, pool_id);
1276 		pool = &pfvf->qset.pool[pool_id];
1277 		while (iova) {
1278 			if (type == AURA_NIX_RQ)
1279 				iova -= OTX2_HEAD_ROOM;
1280 
1281 			otx2_free_bufs(pfvf, pool, iova, size);
1282 
1283 			iova = otx2_aura_allocptr(pfvf, pool_id);
1284 		}
1285 	}
1286 }
1287 
1288 void otx2_aura_pool_free(struct otx2_nic *pfvf)
1289 {
1290 	struct otx2_pool *pool;
1291 	int pool_id;
1292 
1293 	if (!pfvf->qset.pool)
1294 		return;
1295 
1296 	for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
1297 		pool = &pfvf->qset.pool[pool_id];
1298 		qmem_free(pfvf->dev, pool->stack);
1299 		qmem_free(pfvf->dev, pool->fc_addr);
1300 		page_pool_destroy(pool->page_pool);
1301 		pool->page_pool = NULL;
1302 	}
1303 	devm_kfree(pfvf->dev, pfvf->qset.pool);
1304 	pfvf->qset.pool = NULL;
1305 }
1306 
1307 int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
1308 		   int pool_id, int numptrs)
1309 {
1310 	struct npa_aq_enq_req *aq;
1311 	struct otx2_pool *pool;
1312 	int err;
1313 
1314 	pool = &pfvf->qset.pool[pool_id];
1315 
1316 	/* Allocate memory for HW to update Aura count.
1317 	 * Alloc one cache line, so that it fits all FC_STYPE modes.
1318 	 */
1319 	if (!pool->fc_addr) {
1320 		err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
1321 		if (err)
1322 			return err;
1323 	}
1324 
1325 	/* Initialize this aura's context via AF */
1326 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1327 	if (!aq) {
1328 		/* Shared mbox memory buffer is full, flush it and retry */
1329 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1330 		if (err)
1331 			return err;
1332 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1333 		if (!aq)
1334 			return -ENOMEM;
1335 	}
1336 
1337 	aq->aura_id = aura_id;
1338 	/* Will be filled by AF with correct pool context address */
1339 	aq->aura.pool_addr = pool_id;
1340 	aq->aura.pool_caching = 1;
1341 	aq->aura.shift = ilog2(numptrs) - 8;
1342 	aq->aura.count = numptrs;
1343 	aq->aura.limit = numptrs;
1344 	aq->aura.avg_level = 255;
1345 	aq->aura.ena = 1;
1346 	aq->aura.fc_ena = 1;
1347 	aq->aura.fc_addr = pool->fc_addr->iova;
1348 	aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1349 
1350 	/* Enable backpressure for RQ aura */
1351 	if (aura_id < pfvf->hw.rqpool_cnt && !is_otx2_lbkvf(pfvf->pdev)) {
1352 		aq->aura.bp_ena = 0;
1353 		/* If NIX1 LF is attached then specify NIX1_RX.
1354 		 *
1355 		 * Below NPA_AURA_S[BP_ENA] is set according to the
1356 		 * NPA_BPINTF_E enumeration given as:
1357 		 * 0x0 + a*0x1 where 'a' is 0 for NIX0_RX and 1 for NIX1_RX so
1358 		 * NIX0_RX is 0x0 + 0*0x1 = 0
1359 		 * NIX1_RX is 0x0 + 1*0x1 = 1
1360 		 * But in HRM it is given that
1361 		 * "NPA_AURA_S[BP_ENA](w1[33:32]) - Enable aura backpressure to
1362 		 * NIX-RX based on [BP] level. One bit per NIX-RX; index
1363 		 * enumerated by NPA_BPINTF_E."
1364 		 */
1365 		if (pfvf->nix_blkaddr == BLKADDR_NIX1)
1366 			aq->aura.bp_ena = 1;
1367 #ifdef CONFIG_DCB
1368 		aq->aura.nix0_bpid = pfvf->bpid[pfvf->queue_to_pfc_map[aura_id]];
1369 #else
1370 		aq->aura.nix0_bpid = pfvf->bpid[0];
1371 #endif
1372 
1373 		/* Set backpressure level for RQ's Aura */
1374 		aq->aura.bp = RQ_BP_LVL_AURA;
1375 	}
1376 
1377 	/* Fill AQ info */
1378 	aq->ctype = NPA_AQ_CTYPE_AURA;
1379 	aq->op = NPA_AQ_INSTOP_INIT;
1380 
1381 	return 0;
1382 }
1383 
1384 int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1385 		   int stack_pages, int numptrs, int buf_size, int type)
1386 {
1387 	struct page_pool_params pp_params = { 0 };
1388 	struct npa_aq_enq_req *aq;
1389 	struct otx2_pool *pool;
1390 	int err;
1391 
1392 	pool = &pfvf->qset.pool[pool_id];
1393 	/* Alloc memory for stack which is used to store buffer pointers */
1394 	err = qmem_alloc(pfvf->dev, &pool->stack,
1395 			 stack_pages, pfvf->hw.stack_pg_bytes);
1396 	if (err)
1397 		return err;
1398 
1399 	pool->rbsize = buf_size;
1400 
1401 	/* Initialize this pool's context via AF */
1402 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1403 	if (!aq) {
1404 		/* Shared mbox memory buffer is full, flush it and retry */
1405 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1406 		if (err) {
1407 			qmem_free(pfvf->dev, pool->stack);
1408 			return err;
1409 		}
1410 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1411 		if (!aq) {
1412 			qmem_free(pfvf->dev, pool->stack);
1413 			return -ENOMEM;
1414 		}
1415 	}
1416 
1417 	aq->aura_id = pool_id;
1418 	aq->pool.stack_base = pool->stack->iova;
1419 	aq->pool.stack_caching = 1;
1420 	aq->pool.ena = 1;
1421 	aq->pool.buf_size = buf_size / 128;
1422 	aq->pool.stack_max_pages = stack_pages;
1423 	aq->pool.shift = ilog2(numptrs) - 8;
1424 	aq->pool.ptr_start = 0;
1425 	aq->pool.ptr_end = ~0ULL;
1426 
1427 	/* Fill AQ info */
1428 	aq->ctype = NPA_AQ_CTYPE_POOL;
1429 	aq->op = NPA_AQ_INSTOP_INIT;
1430 
1431 	if (type != AURA_NIX_RQ) {
1432 		pool->page_pool = NULL;
1433 		return 0;
1434 	}
1435 
1436 	pp_params.flags = PP_FLAG_PAGE_FRAG | PP_FLAG_DMA_MAP;
1437 	pp_params.pool_size = numptrs;
1438 	pp_params.nid = NUMA_NO_NODE;
1439 	pp_params.dev = pfvf->dev;
1440 	pp_params.dma_dir = DMA_FROM_DEVICE;
1441 	pool->page_pool = page_pool_create(&pp_params);
1442 	if (IS_ERR(pool->page_pool)) {
1443 		netdev_err(pfvf->netdev, "Creation of page pool failed\n");
1444 		return PTR_ERR(pool->page_pool);
1445 	}
1446 
1447 	return 0;
1448 }
1449 
1450 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1451 {
1452 	int qidx, pool_id, stack_pages, num_sqbs;
1453 	struct otx2_qset *qset = &pfvf->qset;
1454 	struct otx2_hw *hw = &pfvf->hw;
1455 	struct otx2_snd_queue *sq;
1456 	struct otx2_pool *pool;
1457 	dma_addr_t bufptr;
1458 	int err, ptr;
1459 
1460 	/* Calculate number of SQBs needed.
1461 	 *
1462 	 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1463 	 * Last SQE is used for pointing to next SQB.
1464 	 */
1465 	num_sqbs = (hw->sqb_size / 128) - 1;
1466 	num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1467 
1468 	/* Get no of stack pages needed */
1469 	stack_pages =
1470 		(num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1471 
1472 	for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1473 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1474 		/* Initialize aura context */
1475 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1476 		if (err)
1477 			goto fail;
1478 
1479 		/* Initialize pool context */
1480 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1481 				     num_sqbs, hw->sqb_size, AURA_NIX_SQ);
1482 		if (err)
1483 			goto fail;
1484 	}
1485 
1486 	/* Flush accumulated messages */
1487 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1488 	if (err)
1489 		goto fail;
1490 
1491 	/* Allocate pointers and free them to aura/pool */
1492 	for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1493 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1494 		pool = &pfvf->qset.pool[pool_id];
1495 
1496 		sq = &qset->sq[qidx];
1497 		sq->sqb_count = 0;
1498 		sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(*sq->sqb_ptrs), GFP_KERNEL);
1499 		if (!sq->sqb_ptrs) {
1500 			err = -ENOMEM;
1501 			goto err_mem;
1502 		}
1503 
1504 		for (ptr = 0; ptr < num_sqbs; ptr++) {
1505 			err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1506 			if (err)
1507 				goto err_mem;
1508 			pfvf->hw_ops->aura_freeptr(pfvf, pool_id, bufptr);
1509 			sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1510 		}
1511 	}
1512 
1513 err_mem:
1514 	return err ? -ENOMEM : 0;
1515 
1516 fail:
1517 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1518 	otx2_aura_pool_free(pfvf);
1519 	return err;
1520 }
1521 
1522 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1523 {
1524 	struct otx2_hw *hw = &pfvf->hw;
1525 	int stack_pages, pool_id, rq;
1526 	struct otx2_pool *pool;
1527 	int err, ptr, num_ptrs;
1528 	dma_addr_t bufptr;
1529 
1530 	num_ptrs = pfvf->qset.rqe_cnt;
1531 
1532 	stack_pages =
1533 		(num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1534 
1535 	for (rq = 0; rq < hw->rx_queues; rq++) {
1536 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1537 		/* Initialize aura context */
1538 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1539 		if (err)
1540 			goto fail;
1541 	}
1542 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1543 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1544 				     num_ptrs, pfvf->rbsize, AURA_NIX_RQ);
1545 		if (err)
1546 			goto fail;
1547 	}
1548 
1549 	/* Flush accumulated messages */
1550 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1551 	if (err)
1552 		goto fail;
1553 
1554 	/* Allocate pointers and free them to aura/pool */
1555 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1556 		pool = &pfvf->qset.pool[pool_id];
1557 		for (ptr = 0; ptr < num_ptrs; ptr++) {
1558 			err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1559 			if (err)
1560 				return -ENOMEM;
1561 			pfvf->hw_ops->aura_freeptr(pfvf, pool_id,
1562 						   bufptr + OTX2_HEAD_ROOM);
1563 		}
1564 	}
1565 	return 0;
1566 fail:
1567 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1568 	otx2_aura_pool_free(pfvf);
1569 	return err;
1570 }
1571 
1572 int otx2_config_npa(struct otx2_nic *pfvf)
1573 {
1574 	struct otx2_qset *qset = &pfvf->qset;
1575 	struct npa_lf_alloc_req  *npalf;
1576 	struct otx2_hw *hw = &pfvf->hw;
1577 	int aura_cnt;
1578 
1579 	/* Pool - Stack of free buffer pointers
1580 	 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1581 	 */
1582 
1583 	if (!hw->pool_cnt)
1584 		return -EINVAL;
1585 
1586 	qset->pool = devm_kcalloc(pfvf->dev, hw->pool_cnt,
1587 				  sizeof(struct otx2_pool), GFP_KERNEL);
1588 	if (!qset->pool)
1589 		return -ENOMEM;
1590 
1591 	/* Get memory to put this msg */
1592 	npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1593 	if (!npalf)
1594 		return -ENOMEM;
1595 
1596 	/* Set aura and pool counts */
1597 	npalf->nr_pools = hw->pool_cnt;
1598 	aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1599 	npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1600 
1601 	return otx2_sync_mbox_msg(&pfvf->mbox);
1602 }
1603 
1604 int otx2_detach_resources(struct mbox *mbox)
1605 {
1606 	struct rsrc_detach *detach;
1607 
1608 	mutex_lock(&mbox->lock);
1609 	detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1610 	if (!detach) {
1611 		mutex_unlock(&mbox->lock);
1612 		return -ENOMEM;
1613 	}
1614 
1615 	/* detach all */
1616 	detach->partial = false;
1617 
1618 	/* Send detach request to AF */
1619 	otx2_mbox_msg_send(&mbox->mbox, 0);
1620 	mutex_unlock(&mbox->lock);
1621 	return 0;
1622 }
1623 EXPORT_SYMBOL(otx2_detach_resources);
1624 
1625 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1626 {
1627 	struct rsrc_attach *attach;
1628 	struct msg_req *msix;
1629 	int err;
1630 
1631 	mutex_lock(&pfvf->mbox.lock);
1632 	/* Get memory to put this msg */
1633 	attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1634 	if (!attach) {
1635 		mutex_unlock(&pfvf->mbox.lock);
1636 		return -ENOMEM;
1637 	}
1638 
1639 	attach->npalf = true;
1640 	attach->nixlf = true;
1641 
1642 	/* Send attach request to AF */
1643 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1644 	if (err) {
1645 		mutex_unlock(&pfvf->mbox.lock);
1646 		return err;
1647 	}
1648 
1649 	pfvf->nix_blkaddr = BLKADDR_NIX0;
1650 
1651 	/* If the platform has two NIX blocks then LF may be
1652 	 * allocated from NIX1.
1653 	 */
1654 	if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1655 		pfvf->nix_blkaddr = BLKADDR_NIX1;
1656 
1657 	/* Get NPA and NIX MSIX vector offsets */
1658 	msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1659 	if (!msix) {
1660 		mutex_unlock(&pfvf->mbox.lock);
1661 		return -ENOMEM;
1662 	}
1663 
1664 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1665 	if (err) {
1666 		mutex_unlock(&pfvf->mbox.lock);
1667 		return err;
1668 	}
1669 	mutex_unlock(&pfvf->mbox.lock);
1670 
1671 	if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1672 	    pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1673 		dev_err(pfvf->dev,
1674 			"RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1675 		return -EINVAL;
1676 	}
1677 
1678 	return 0;
1679 }
1680 EXPORT_SYMBOL(otx2_attach_npa_nix);
1681 
1682 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1683 {
1684 	struct hwctx_disable_req *req;
1685 
1686 	mutex_lock(&mbox->lock);
1687 	/* Request AQ to disable this context */
1688 	if (npa)
1689 		req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1690 	else
1691 		req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1692 
1693 	if (!req) {
1694 		mutex_unlock(&mbox->lock);
1695 		return;
1696 	}
1697 
1698 	req->ctype = type;
1699 
1700 	if (otx2_sync_mbox_msg(mbox))
1701 		dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1702 			__func__);
1703 
1704 	mutex_unlock(&mbox->lock);
1705 }
1706 
1707 int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
1708 {
1709 	struct nix_bp_cfg_req *req;
1710 
1711 	if (enable)
1712 		req = otx2_mbox_alloc_msg_nix_bp_enable(&pfvf->mbox);
1713 	else
1714 		req = otx2_mbox_alloc_msg_nix_bp_disable(&pfvf->mbox);
1715 
1716 	if (!req)
1717 		return -ENOMEM;
1718 
1719 	req->chan_base = 0;
1720 #ifdef CONFIG_DCB
1721 	req->chan_cnt = pfvf->pfc_en ? IEEE_8021QAZ_MAX_TCS : 1;
1722 	req->bpid_per_chan = pfvf->pfc_en ? 1 : 0;
1723 #else
1724 	req->chan_cnt =  1;
1725 	req->bpid_per_chan = 0;
1726 #endif
1727 
1728 	return otx2_sync_mbox_msg(&pfvf->mbox);
1729 }
1730 EXPORT_SYMBOL(otx2_nix_config_bp);
1731 
1732 /* Mbox message handlers */
1733 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1734 			    struct cgx_stats_rsp *rsp)
1735 {
1736 	int id;
1737 
1738 	for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1739 		pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1740 	for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1741 		pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1742 }
1743 
1744 void mbox_handler_cgx_fec_stats(struct otx2_nic *pfvf,
1745 				struct cgx_fec_stats_rsp *rsp)
1746 {
1747 	pfvf->hw.cgx_fec_corr_blks += rsp->fec_corr_blks;
1748 	pfvf->hw.cgx_fec_uncorr_blks += rsp->fec_uncorr_blks;
1749 }
1750 
1751 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1752 			       struct npa_lf_alloc_rsp *rsp)
1753 {
1754 	pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1755 	pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1756 }
1757 EXPORT_SYMBOL(mbox_handler_npa_lf_alloc);
1758 
1759 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1760 			       struct nix_lf_alloc_rsp *rsp)
1761 {
1762 	pfvf->hw.sqb_size = rsp->sqb_size;
1763 	pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1764 	pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1765 	pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1766 	pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1767 	pfvf->hw.cgx_links = rsp->cgx_links;
1768 	pfvf->hw.lbk_links = rsp->lbk_links;
1769 	pfvf->hw.tx_link = rsp->tx_link;
1770 }
1771 EXPORT_SYMBOL(mbox_handler_nix_lf_alloc);
1772 
1773 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1774 			      struct msix_offset_rsp *rsp)
1775 {
1776 	pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1777 	pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1778 }
1779 EXPORT_SYMBOL(mbox_handler_msix_offset);
1780 
1781 void mbox_handler_nix_bp_enable(struct otx2_nic *pfvf,
1782 				struct nix_bp_cfg_rsp *rsp)
1783 {
1784 	int chan, chan_id;
1785 
1786 	for (chan = 0; chan < rsp->chan_cnt; chan++) {
1787 		chan_id = ((rsp->chan_bpid[chan] >> 10) & 0x7F);
1788 		pfvf->bpid[chan_id] = rsp->chan_bpid[chan] & 0x3FF;
1789 	}
1790 }
1791 EXPORT_SYMBOL(mbox_handler_nix_bp_enable);
1792 
1793 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1794 {
1795 	struct otx2_qset *qset = &pfvf->qset;
1796 	struct otx2_hw *hw = &pfvf->hw;
1797 	int irq, qidx;
1798 
1799 	for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1800 	     qidx < n;
1801 	     qidx++, irq++) {
1802 		int vector = pci_irq_vector(pfvf->pdev, irq);
1803 
1804 		irq_set_affinity_hint(vector, NULL);
1805 		free_cpumask_var(hw->affinity_mask[irq]);
1806 		free_irq(vector, &qset->napi[qidx]);
1807 	}
1808 }
1809 
1810 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1811 {
1812 	struct otx2_hw *hw = &pfvf->hw;
1813 	int vec, cpu, irq, cint;
1814 
1815 	vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1816 	cpu = cpumask_first(cpu_online_mask);
1817 
1818 	/* CQ interrupts */
1819 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1820 		if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1821 			return;
1822 
1823 		cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1824 
1825 		irq = pci_irq_vector(pfvf->pdev, vec);
1826 		irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1827 
1828 		cpu = cpumask_next(cpu, cpu_online_mask);
1829 		if (unlikely(cpu >= nr_cpu_ids))
1830 			cpu = 0;
1831 	}
1832 }
1833 
1834 static u32 get_dwrr_mtu(struct otx2_nic *pfvf, struct nix_hw_info *hw)
1835 {
1836 	if (is_otx2_lbkvf(pfvf->pdev)) {
1837 		pfvf->hw.smq_link_type = SMQ_LINK_TYPE_LBK;
1838 		return hw->lbk_dwrr_mtu;
1839 	}
1840 
1841 	pfvf->hw.smq_link_type = SMQ_LINK_TYPE_RPM;
1842 	return hw->rpm_dwrr_mtu;
1843 }
1844 
1845 u16 otx2_get_max_mtu(struct otx2_nic *pfvf)
1846 {
1847 	struct nix_hw_info *rsp;
1848 	struct msg_req *req;
1849 	u16 max_mtu;
1850 	int rc;
1851 
1852 	mutex_lock(&pfvf->mbox.lock);
1853 
1854 	req = otx2_mbox_alloc_msg_nix_get_hw_info(&pfvf->mbox);
1855 	if (!req) {
1856 		rc =  -ENOMEM;
1857 		goto out;
1858 	}
1859 
1860 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
1861 	if (!rc) {
1862 		rsp = (struct nix_hw_info *)
1863 		       otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
1864 
1865 		/* HW counts VLAN insertion bytes (8 for double tag)
1866 		 * irrespective of whether SQE is requesting to insert VLAN
1867 		 * in the packet or not. Hence these 8 bytes have to be
1868 		 * discounted from max packet size otherwise HW will throw
1869 		 * SMQ errors
1870 		 */
1871 		max_mtu = rsp->max_mtu - 8 - OTX2_ETH_HLEN;
1872 
1873 		/* Also save DWRR MTU, needed for DWRR weight calculation */
1874 		pfvf->hw.dwrr_mtu = get_dwrr_mtu(pfvf, rsp);
1875 		if (!pfvf->hw.dwrr_mtu)
1876 			pfvf->hw.dwrr_mtu = 1;
1877 	}
1878 
1879 out:
1880 	mutex_unlock(&pfvf->mbox.lock);
1881 	if (rc) {
1882 		dev_warn(pfvf->dev,
1883 			 "Failed to get MTU from hardware setting default value(1500)\n");
1884 		max_mtu = 1500;
1885 	}
1886 	return max_mtu;
1887 }
1888 EXPORT_SYMBOL(otx2_get_max_mtu);
1889 
1890 int otx2_handle_ntuple_tc_features(struct net_device *netdev, netdev_features_t features)
1891 {
1892 	netdev_features_t changed = features ^ netdev->features;
1893 	struct otx2_nic *pfvf = netdev_priv(netdev);
1894 	bool ntuple = !!(features & NETIF_F_NTUPLE);
1895 	bool tc = !!(features & NETIF_F_HW_TC);
1896 
1897 	if ((changed & NETIF_F_NTUPLE) && !ntuple)
1898 		otx2_destroy_ntuple_flows(pfvf);
1899 
1900 	if ((changed & NETIF_F_NTUPLE) && ntuple) {
1901 		if (!pfvf->flow_cfg->max_flows) {
1902 			netdev_err(netdev,
1903 				   "Can't enable NTUPLE, MCAM entries not allocated\n");
1904 			return -EINVAL;
1905 		}
1906 	}
1907 
1908 	if ((changed & NETIF_F_HW_TC) && !tc &&
1909 	    otx2_tc_flower_rule_cnt(pfvf)) {
1910 		netdev_err(netdev, "Can't disable TC hardware offload while flows are active\n");
1911 		return -EBUSY;
1912 	}
1913 
1914 	if ((changed & NETIF_F_NTUPLE) && ntuple &&
1915 	    otx2_tc_flower_rule_cnt(pfvf) && !(changed & NETIF_F_HW_TC)) {
1916 		netdev_err(netdev,
1917 			   "Can't enable NTUPLE when TC flower offload is active, disable TC rules and retry\n");
1918 		return -EINVAL;
1919 	}
1920 
1921 	return 0;
1922 }
1923 EXPORT_SYMBOL(otx2_handle_ntuple_tc_features);
1924 
1925 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
1926 int __weak								\
1927 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf,		\
1928 				struct _req_type *req,			\
1929 				struct _rsp_type *rsp)			\
1930 {									\
1931 	/* Nothing to do here */					\
1932 	return 0;							\
1933 }									\
1934 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
1935 MBOX_UP_CGX_MESSAGES
1936 MBOX_UP_MCS_MESSAGES
1937 #undef M
1938