xref: /linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c (revision 56d95b0adfa224bb1c67733dbcad30dd8debd39e)
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 #include <net/xfrm.h>
14 
15 #include "otx2_reg.h"
16 #include "otx2_common.h"
17 #include "otx2_struct.h"
18 #include "cn10k.h"
19 
20 static bool otx2_is_pfc_enabled(struct otx2_nic *pfvf)
21 {
22 	return IS_ENABLED(CONFIG_DCB) && !!pfvf->pfc_en;
23 }
24 
25 static void otx2_nix_rq_op_stats(struct queue_stats *stats,
26 				 struct otx2_nic *pfvf, int qidx)
27 {
28 	u64 incr = (u64)qidx << 32;
29 	u64 *ptr;
30 
31 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_OCTS);
32 	stats->bytes = otx2_atomic64_add(incr, ptr);
33 
34 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_PKTS);
35 	stats->pkts = otx2_atomic64_add(incr, ptr);
36 }
37 
38 static void otx2_nix_sq_op_stats(struct queue_stats *stats,
39 				 struct otx2_nic *pfvf, int qidx)
40 {
41 	u64 incr = (u64)qidx << 32;
42 	u64 *ptr;
43 
44 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_OCTS);
45 	stats->bytes = otx2_atomic64_add(incr, ptr);
46 
47 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_PKTS);
48 	stats->pkts = otx2_atomic64_add(incr, ptr);
49 }
50 
51 void otx2_update_lmac_stats(struct otx2_nic *pfvf)
52 {
53 	struct msg_req *req;
54 
55 	if (!netif_running(pfvf->netdev))
56 		return;
57 
58 	mutex_lock(&pfvf->mbox.lock);
59 	req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
60 	if (!req) {
61 		mutex_unlock(&pfvf->mbox.lock);
62 		return;
63 	}
64 
65 	otx2_sync_mbox_msg(&pfvf->mbox);
66 	mutex_unlock(&pfvf->mbox.lock);
67 }
68 
69 void otx2_update_lmac_fec_stats(struct otx2_nic *pfvf)
70 {
71 	struct msg_req *req;
72 
73 	if (!netif_running(pfvf->netdev))
74 		return;
75 	mutex_lock(&pfvf->mbox.lock);
76 	req = otx2_mbox_alloc_msg_cgx_fec_stats(&pfvf->mbox);
77 	if (req)
78 		otx2_sync_mbox_msg(&pfvf->mbox);
79 	mutex_unlock(&pfvf->mbox.lock);
80 }
81 
82 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
83 {
84 	struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
85 
86 	if (!pfvf->qset.rq)
87 		return 0;
88 
89 	otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
90 	return 1;
91 }
92 EXPORT_SYMBOL(otx2_update_rq_stats);
93 
94 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
95 {
96 	struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
97 
98 	if (!pfvf->qset.sq)
99 		return 0;
100 
101 	if (qidx >= pfvf->hw.non_qos_queues) {
102 		if (!test_bit(qidx - pfvf->hw.non_qos_queues, pfvf->qos.qos_sq_bmap))
103 			return 0;
104 	}
105 
106 	otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
107 	return 1;
108 }
109 EXPORT_SYMBOL(otx2_update_sq_stats);
110 
111 void otx2_get_dev_stats(struct otx2_nic *pfvf)
112 {
113 	struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
114 
115 	dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
116 	dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
117 	dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
118 	dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
119 	dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
120 	dev_stats->rx_frames = dev_stats->rx_bcast_frames +
121 			       dev_stats->rx_mcast_frames +
122 			       dev_stats->rx_ucast_frames;
123 
124 	dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
125 	dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
126 	dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
127 	dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
128 	dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
129 	dev_stats->tx_frames = dev_stats->tx_bcast_frames +
130 			       dev_stats->tx_mcast_frames +
131 			       dev_stats->tx_ucast_frames;
132 }
133 
134 void otx2_get_stats64(struct net_device *netdev,
135 		      struct rtnl_link_stats64 *stats)
136 {
137 	struct otx2_nic *pfvf = netdev_priv(netdev);
138 	struct otx2_dev_stats *dev_stats;
139 
140 	otx2_get_dev_stats(pfvf);
141 
142 	dev_stats = &pfvf->hw.dev_stats;
143 	stats->rx_bytes = dev_stats->rx_bytes;
144 	stats->rx_packets = dev_stats->rx_frames;
145 	stats->rx_dropped = dev_stats->rx_drops;
146 	stats->multicast = dev_stats->rx_mcast_frames;
147 
148 	stats->tx_bytes = dev_stats->tx_bytes;
149 	stats->tx_packets = dev_stats->tx_frames;
150 	stats->tx_dropped = dev_stats->tx_drops;
151 }
152 EXPORT_SYMBOL(otx2_get_stats64);
153 
154 /* Sync MAC address with RVU AF */
155 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
156 {
157 	struct nix_set_mac_addr *req;
158 	int err;
159 
160 	mutex_lock(&pfvf->mbox.lock);
161 	req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
162 	if (!req) {
163 		mutex_unlock(&pfvf->mbox.lock);
164 		return -ENOMEM;
165 	}
166 
167 	ether_addr_copy(req->mac_addr, mac);
168 
169 	err = otx2_sync_mbox_msg(&pfvf->mbox);
170 	mutex_unlock(&pfvf->mbox.lock);
171 	return err;
172 }
173 
174 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
175 				struct net_device *netdev)
176 {
177 	struct nix_get_mac_addr_rsp *rsp;
178 	struct mbox_msghdr *msghdr;
179 	struct msg_req *req;
180 	int err;
181 
182 	mutex_lock(&pfvf->mbox.lock);
183 	req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
184 	if (!req) {
185 		mutex_unlock(&pfvf->mbox.lock);
186 		return -ENOMEM;
187 	}
188 
189 	err = otx2_sync_mbox_msg(&pfvf->mbox);
190 	if (err) {
191 		mutex_unlock(&pfvf->mbox.lock);
192 		return err;
193 	}
194 
195 	msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
196 	if (IS_ERR(msghdr)) {
197 		mutex_unlock(&pfvf->mbox.lock);
198 		return PTR_ERR(msghdr);
199 	}
200 	rsp = (struct nix_get_mac_addr_rsp *)msghdr;
201 	eth_hw_addr_set(netdev, rsp->mac_addr);
202 	mutex_unlock(&pfvf->mbox.lock);
203 
204 	return 0;
205 }
206 
207 int otx2_set_mac_address(struct net_device *netdev, void *p)
208 {
209 	struct otx2_nic *pfvf = netdev_priv(netdev);
210 	struct sockaddr *addr = p;
211 
212 	if (!is_valid_ether_addr(addr->sa_data))
213 		return -EADDRNOTAVAIL;
214 
215 	if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data)) {
216 		eth_hw_addr_set(netdev, addr->sa_data);
217 		/* update dmac field in vlan offload rule */
218 		if (netif_running(netdev) &&
219 		    pfvf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
220 			otx2_install_rxvlan_offload_flow(pfvf);
221 		/* update dmac address in ntuple and DMAC filter list */
222 		if (pfvf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
223 			otx2_dmacflt_update_pfmac_flow(pfvf);
224 	} else {
225 		return -EPERM;
226 	}
227 
228 	return 0;
229 }
230 EXPORT_SYMBOL(otx2_set_mac_address);
231 
232 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
233 {
234 	struct nix_frs_cfg *req;
235 	u16 maxlen;
236 	int err;
237 
238 	maxlen = pfvf->hw.max_mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
239 
240 	mutex_lock(&pfvf->mbox.lock);
241 	req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
242 	if (!req) {
243 		mutex_unlock(&pfvf->mbox.lock);
244 		return -ENOMEM;
245 	}
246 
247 	req->maxlen = mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
248 
249 	/* Use max receive length supported by hardware for loopback devices */
250 	if (is_otx2_lbkvf(pfvf->pdev))
251 		req->maxlen = maxlen;
252 
253 	err = otx2_sync_mbox_msg(&pfvf->mbox);
254 	mutex_unlock(&pfvf->mbox.lock);
255 	return err;
256 }
257 EXPORT_SYMBOL(otx2_hw_set_mtu);
258 
259 int otx2_config_pause_frm(struct otx2_nic *pfvf)
260 {
261 	struct cgx_pause_frm_cfg *req;
262 	int err;
263 
264 	if (is_otx2_lbkvf(pfvf->pdev) || is_otx2_sdp_rep(pfvf->pdev))
265 		return 0;
266 
267 	mutex_lock(&pfvf->mbox.lock);
268 	req = otx2_mbox_alloc_msg_cgx_cfg_pause_frm(&pfvf->mbox);
269 	if (!req) {
270 		err = -ENOMEM;
271 		goto unlock;
272 	}
273 
274 	req->rx_pause = !!(pfvf->flags & OTX2_FLAG_RX_PAUSE_ENABLED);
275 	req->tx_pause = !!(pfvf->flags & OTX2_FLAG_TX_PAUSE_ENABLED);
276 	req->set = 1;
277 
278 	err = otx2_sync_mbox_msg(&pfvf->mbox);
279 unlock:
280 	mutex_unlock(&pfvf->mbox.lock);
281 	return err;
282 }
283 EXPORT_SYMBOL(otx2_config_pause_frm);
284 
285 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
286 {
287 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
288 	struct nix_rss_flowkey_cfg_rsp *rsp;
289 	struct nix_rss_flowkey_cfg *req;
290 	int err;
291 
292 	mutex_lock(&pfvf->mbox.lock);
293 	req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
294 	if (!req) {
295 		mutex_unlock(&pfvf->mbox.lock);
296 		return -ENOMEM;
297 	}
298 	req->mcam_index = -1; /* Default or reserved index */
299 	req->flowkey_cfg = rss->flowkey_cfg;
300 	req->group = DEFAULT_RSS_CONTEXT_GROUP;
301 
302 	err = otx2_sync_mbox_msg(&pfvf->mbox);
303 	if (err)
304 		goto fail;
305 
306 	rsp = (struct nix_rss_flowkey_cfg_rsp *)
307 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
308 	if (IS_ERR(rsp)) {
309 		err = PTR_ERR(rsp);
310 		goto fail;
311 	}
312 
313 	pfvf->hw.flowkey_alg_idx = rsp->alg_idx;
314 fail:
315 	mutex_unlock(&pfvf->mbox.lock);
316 	return err;
317 }
318 
319 int otx2_set_rss_table(struct otx2_nic *pfvf, int ctx_id)
320 {
321 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
322 	const int index = rss->rss_size * ctx_id;
323 	struct mbox *mbox = &pfvf->mbox;
324 	struct otx2_rss_ctx *rss_ctx;
325 	struct nix_aq_enq_req *aq;
326 	int idx, err;
327 
328 	mutex_lock(&mbox->lock);
329 	rss_ctx = rss->rss_ctx[ctx_id];
330 	/* Get memory to put this msg */
331 	for (idx = 0; idx < rss->rss_size; idx++) {
332 		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
333 		if (!aq) {
334 			/* The shared memory buffer can be full.
335 			 * Flush it and retry
336 			 */
337 			err = otx2_sync_mbox_msg(mbox);
338 			if (err) {
339 				mutex_unlock(&mbox->lock);
340 				return err;
341 			}
342 			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
343 			if (!aq) {
344 				mutex_unlock(&mbox->lock);
345 				return -ENOMEM;
346 			}
347 		}
348 
349 		aq->rss.rq = rss_ctx->ind_tbl[idx];
350 
351 		/* Fill AQ info */
352 		aq->qidx = index + idx;
353 		aq->ctype = NIX_AQ_CTYPE_RSS;
354 		aq->op = NIX_AQ_INSTOP_INIT;
355 	}
356 	err = otx2_sync_mbox_msg(mbox);
357 	mutex_unlock(&mbox->lock);
358 	return err;
359 }
360 
361 void otx2_set_rss_key(struct otx2_nic *pfvf)
362 {
363 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
364 	u64 *key = (u64 *)&rss->key[4];
365 	int idx;
366 
367 	/* 352bit or 44byte key needs to be configured as below
368 	 * NIX_LF_RX_SECRETX0 = key<351:288>
369 	 * NIX_LF_RX_SECRETX1 = key<287:224>
370 	 * NIX_LF_RX_SECRETX2 = key<223:160>
371 	 * NIX_LF_RX_SECRETX3 = key<159:96>
372 	 * NIX_LF_RX_SECRETX4 = key<95:32>
373 	 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
374 	 */
375 	otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
376 		     (u64)(*((u32 *)&rss->key)) << 32);
377 	idx = sizeof(rss->key) / sizeof(u64);
378 	while (idx > 0) {
379 		idx--;
380 		otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
381 	}
382 }
383 
384 int otx2_rss_init(struct otx2_nic *pfvf)
385 {
386 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
387 	struct otx2_rss_ctx *rss_ctx;
388 	int idx, ret = 0;
389 
390 	rss->rss_size = sizeof(*rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
391 
392 	/* Init RSS key if it is not setup already */
393 	if (!rss->enable)
394 		netdev_rss_key_fill(rss->key, sizeof(rss->key));
395 	otx2_set_rss_key(pfvf);
396 
397 	if (!netif_is_rxfh_configured(pfvf->netdev)) {
398 		/* Set RSS group 0 as default indirection table */
399 		rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP] = kzalloc(rss->rss_size,
400 								  GFP_KERNEL);
401 		if (!rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP])
402 			return -ENOMEM;
403 
404 		rss_ctx = rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP];
405 		for (idx = 0; idx < rss->rss_size; idx++)
406 			rss_ctx->ind_tbl[idx] =
407 				ethtool_rxfh_indir_default(idx,
408 							   pfvf->hw.rx_queues);
409 	}
410 	ret = otx2_set_rss_table(pfvf, DEFAULT_RSS_CONTEXT_GROUP);
411 	if (ret)
412 		return ret;
413 
414 	/* Flowkey or hash config to be used for generating flow tag */
415 	rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
416 			   NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
417 			   NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
418 			   NIX_FLOW_KEY_TYPE_SCTP | NIX_FLOW_KEY_TYPE_VLAN |
419 			   NIX_FLOW_KEY_TYPE_IPV4_PROTO;
420 
421 	ret = otx2_set_flowkey_cfg(pfvf);
422 	if (ret)
423 		return ret;
424 
425 	rss->enable = true;
426 	return 0;
427 }
428 
429 /* Setup UDP segmentation algorithm in HW */
430 static void otx2_setup_udp_segmentation(struct nix_lso_format_cfg *lso, bool v4)
431 {
432 	struct nix_lso_format *field;
433 
434 	field = (struct nix_lso_format *)&lso->fields[0];
435 	lso->field_mask = GENMASK(18, 0);
436 
437 	/* IP's Length field */
438 	field->layer = NIX_TXLAYER_OL3;
439 	/* In ipv4, length field is at offset 2 bytes, for ipv6 it's 4 */
440 	field->offset = v4 ? 2 : 4;
441 	field->sizem1 = 1; /* i.e 2 bytes */
442 	field->alg = NIX_LSOALG_ADD_PAYLEN;
443 	field++;
444 
445 	/* No ID field in IPv6 header */
446 	if (v4) {
447 		/* Increment IPID */
448 		field->layer = NIX_TXLAYER_OL3;
449 		field->offset = 4;
450 		field->sizem1 = 1; /* i.e 2 bytes */
451 		field->alg = NIX_LSOALG_ADD_SEGNUM;
452 		field++;
453 	}
454 
455 	/* Update length in UDP header */
456 	field->layer = NIX_TXLAYER_OL4;
457 	field->offset = 4;
458 	field->sizem1 = 1;
459 	field->alg = NIX_LSOALG_ADD_PAYLEN;
460 }
461 
462 /* Setup segmentation algorithms in HW and retrieve algorithm index */
463 void otx2_setup_segmentation(struct otx2_nic *pfvf)
464 {
465 	struct nix_lso_format_cfg_rsp *rsp;
466 	struct nix_lso_format_cfg *lso;
467 	struct otx2_hw *hw = &pfvf->hw;
468 	int err;
469 
470 	mutex_lock(&pfvf->mbox.lock);
471 
472 	/* UDPv4 segmentation */
473 	lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
474 	if (!lso)
475 		goto fail;
476 
477 	/* Setup UDP/IP header fields that HW should update per segment */
478 	otx2_setup_udp_segmentation(lso, true);
479 
480 	err = otx2_sync_mbox_msg(&pfvf->mbox);
481 	if (err)
482 		goto fail;
483 
484 	rsp = (struct nix_lso_format_cfg_rsp *)
485 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
486 	if (IS_ERR(rsp))
487 		goto fail;
488 
489 	hw->lso_udpv4_idx = rsp->lso_format_idx;
490 
491 	/* UDPv6 segmentation */
492 	lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
493 	if (!lso)
494 		goto fail;
495 
496 	/* Setup UDP/IP header fields that HW should update per segment */
497 	otx2_setup_udp_segmentation(lso, false);
498 
499 	err = otx2_sync_mbox_msg(&pfvf->mbox);
500 	if (err)
501 		goto fail;
502 
503 	rsp = (struct nix_lso_format_cfg_rsp *)
504 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
505 	if (IS_ERR(rsp))
506 		goto fail;
507 
508 	hw->lso_udpv6_idx = rsp->lso_format_idx;
509 	mutex_unlock(&pfvf->mbox.lock);
510 	return;
511 fail:
512 	mutex_unlock(&pfvf->mbox.lock);
513 	netdev_info(pfvf->netdev,
514 		    "Failed to get LSO index for UDP GSO offload, disabling\n");
515 	pfvf->netdev->hw_features &= ~NETIF_F_GSO_UDP_L4;
516 }
517 
518 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
519 {
520 	/* Configure CQE interrupt coalescing parameters
521 	 *
522 	 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
523 	 * set 1 less than cq_ecount_wait. And cq_time_wait is in
524 	 * usecs, convert that to 100ns count.
525 	 */
526 	otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
527 		     ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
528 		     ((u64)pfvf->hw.cq_qcount_wait << 32) |
529 		     (pfvf->hw.cq_ecount_wait - 1));
530 }
531 
532 static int otx2_alloc_pool_buf(struct otx2_nic *pfvf, struct otx2_pool *pool,
533 			       dma_addr_t *dma)
534 {
535 	unsigned int offset = 0;
536 	struct page *page;
537 	size_t sz;
538 
539 	sz = SKB_DATA_ALIGN(pool->rbsize);
540 	sz = ALIGN(sz, OTX2_ALIGN);
541 
542 	page = page_pool_alloc_frag(pool->page_pool, &offset, sz, GFP_ATOMIC);
543 	if (unlikely(!page))
544 		return -ENOMEM;
545 
546 	*dma = page_pool_get_dma_addr(page) + offset;
547 	return 0;
548 }
549 
550 static int __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
551 			     dma_addr_t *dma)
552 {
553 	u8 *buf;
554 
555 	if (pool->page_pool)
556 		return otx2_alloc_pool_buf(pfvf, pool, dma);
557 
558 	buf = napi_alloc_frag_align(pool->rbsize, OTX2_ALIGN);
559 	if (unlikely(!buf))
560 		return -ENOMEM;
561 
562 	*dma = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
563 				    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
564 	if (unlikely(dma_mapping_error(pfvf->dev, *dma))) {
565 		page_frag_free(buf);
566 		return -ENOMEM;
567 	}
568 
569 	return 0;
570 }
571 
572 int otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
573 		    dma_addr_t *dma)
574 {
575 	int ret;
576 
577 	local_bh_disable();
578 	ret = __otx2_alloc_rbuf(pfvf, pool, dma);
579 	local_bh_enable();
580 	return ret;
581 }
582 
583 int otx2_alloc_buffer(struct otx2_nic *pfvf, struct otx2_cq_queue *cq,
584 		      dma_addr_t *dma)
585 {
586 	if (unlikely(__otx2_alloc_rbuf(pfvf, cq->rbpool, dma)))
587 		return -ENOMEM;
588 	return 0;
589 }
590 
591 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
592 {
593 	struct otx2_nic *pfvf = netdev_priv(netdev);
594 
595 	schedule_work(&pfvf->reset_task);
596 }
597 EXPORT_SYMBOL(otx2_tx_timeout);
598 
599 void otx2_get_mac_from_af(struct net_device *netdev)
600 {
601 	struct otx2_nic *pfvf = netdev_priv(netdev);
602 	int err;
603 
604 	err = otx2_hw_get_mac_addr(pfvf, netdev);
605 	if (err)
606 		dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
607 
608 	/* If AF doesn't provide a valid MAC, generate a random one */
609 	if (!is_valid_ether_addr(netdev->dev_addr))
610 		eth_hw_addr_random(netdev);
611 }
612 EXPORT_SYMBOL(otx2_get_mac_from_af);
613 
614 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl, int prio, bool txschq_for_pfc)
615 {
616 	u16 (*schq_list)[MAX_TXSCHQ_PER_FUNC];
617 	struct otx2_hw *hw = &pfvf->hw;
618 	struct nix_txschq_config *req;
619 	u64 schq, parent;
620 	u64 dwrr_val;
621 
622 	dwrr_val = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
623 
624 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
625 	if (!req)
626 		return -ENOMEM;
627 
628 	req->lvl = lvl;
629 	req->num_regs = 1;
630 
631 	schq_list = hw->txschq_list;
632 #ifdef CONFIG_DCB
633 	if (txschq_for_pfc)
634 		schq_list = pfvf->pfc_schq_list;
635 #endif
636 
637 	schq = schq_list[lvl][prio];
638 	/* Set topology e.t.c configuration */
639 	if (lvl == NIX_TXSCH_LVL_SMQ) {
640 		req->reg[0] = NIX_AF_SMQX_CFG(schq);
641 		req->regval[0] = ((u64)pfvf->tx_max_pktlen << 8) | OTX2_MIN_MTU;
642 		req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
643 				  (0x2ULL << 36);
644 		/* Set link type for DWRR MTU selection on CN10K silicons */
645 		if (!is_dev_otx2(pfvf->pdev))
646 			req->regval[0] |= FIELD_PREP(GENMASK_ULL(58, 57),
647 						(u64)hw->smq_link_type);
648 		req->num_regs++;
649 		/* MDQ config */
650 		parent = schq_list[NIX_TXSCH_LVL_TL4][prio];
651 		req->reg[1] = NIX_AF_MDQX_PARENT(schq);
652 		req->regval[1] = parent << 16;
653 		req->num_regs++;
654 		/* Set DWRR quantum */
655 		req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
656 		req->regval[2] =  dwrr_val;
657 	} else if (lvl == NIX_TXSCH_LVL_TL4) {
658 		int sdp_chan =  hw->tx_chan_base + prio;
659 
660 		if (is_otx2_sdp_rep(pfvf->pdev))
661 			prio = 0;
662 		parent = schq_list[NIX_TXSCH_LVL_TL3][prio];
663 		req->reg[0] = NIX_AF_TL4X_PARENT(schq);
664 		req->regval[0] = (u64)parent << 16;
665 		req->num_regs++;
666 		req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
667 		req->regval[1] = dwrr_val;
668 		if (is_otx2_sdp_rep(pfvf->pdev)) {
669 			req->num_regs++;
670 			req->reg[2] = NIX_AF_TL4X_SDP_LINK_CFG(schq);
671 			req->regval[2] = BIT_ULL(12) | BIT_ULL(13) |
672 					 (sdp_chan & 0xff);
673 		}
674 	} else if (lvl == NIX_TXSCH_LVL_TL3) {
675 		parent = schq_list[NIX_TXSCH_LVL_TL2][prio];
676 		req->reg[0] = NIX_AF_TL3X_PARENT(schq);
677 		req->regval[0] = (u64)parent << 16;
678 		req->num_regs++;
679 		req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
680 		req->regval[1] = dwrr_val;
681 		if (lvl == hw->txschq_link_cfg_lvl &&
682 		    !is_otx2_sdp_rep(pfvf->pdev)) {
683 			req->num_regs++;
684 			req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
685 			/* Enable this queue and backpressure
686 			 * and set relative channel
687 			 */
688 			req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
689 		}
690 	} else if (lvl == NIX_TXSCH_LVL_TL2) {
691 		parent = schq_list[NIX_TXSCH_LVL_TL1][prio];
692 		req->reg[0] = NIX_AF_TL2X_PARENT(schq);
693 		req->regval[0] = (u64)parent << 16;
694 
695 		req->num_regs++;
696 		req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
697 		req->regval[1] = (u64)hw->txschq_aggr_lvl_rr_prio << 24 | dwrr_val;
698 
699 		if (lvl == hw->txschq_link_cfg_lvl &&
700 		    !is_otx2_sdp_rep(pfvf->pdev)) {
701 			req->num_regs++;
702 			req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
703 			/* Enable this queue and backpressure
704 			 * and set relative channel
705 			 */
706 			req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
707 		}
708 	} else if (lvl == NIX_TXSCH_LVL_TL1) {
709 		/* Default config for TL1.
710 		 * For VF this is always ignored.
711 		 */
712 
713 		/* On CN10K, if RR_WEIGHT is greater than 16384, HW will
714 		 * clip it to 16384, so configuring a 24bit max value
715 		 * will work on both OTx2 and CN10K.
716 		 */
717 		req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
718 		req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
719 
720 		req->num_regs++;
721 		req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
722 		req->regval[1] = hw->txschq_aggr_lvl_rr_prio << 1;
723 
724 		req->num_regs++;
725 		req->reg[2] = NIX_AF_TL1X_CIR(schq);
726 		req->regval[2] = 0;
727 	}
728 
729 	return otx2_sync_mbox_msg(&pfvf->mbox);
730 }
731 EXPORT_SYMBOL(otx2_txschq_config);
732 
733 int otx2_smq_flush(struct otx2_nic *pfvf, int smq)
734 {
735 	struct nix_txschq_config *req;
736 	int rc;
737 
738 	mutex_lock(&pfvf->mbox.lock);
739 
740 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
741 	if (!req) {
742 		mutex_unlock(&pfvf->mbox.lock);
743 		return -ENOMEM;
744 	}
745 
746 	req->lvl = NIX_TXSCH_LVL_SMQ;
747 	req->reg[0] = NIX_AF_SMQX_CFG(smq);
748 	req->regval[0] |= BIT_ULL(49);
749 	req->num_regs++;
750 
751 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
752 	mutex_unlock(&pfvf->mbox.lock);
753 	return rc;
754 }
755 EXPORT_SYMBOL(otx2_smq_flush);
756 
757 int otx2_txsch_alloc(struct otx2_nic *pfvf)
758 {
759 	int chan_cnt = pfvf->hw.tx_chan_cnt;
760 	struct nix_txsch_alloc_req *req;
761 	struct nix_txsch_alloc_rsp *rsp;
762 	int lvl, schq, rc;
763 
764 	/* Get memory to put this msg */
765 	req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
766 	if (!req)
767 		return -ENOMEM;
768 
769 	/* Request one schq per level */
770 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
771 		req->schq[lvl] = 1;
772 
773 	if (is_otx2_sdp_rep(pfvf->pdev) && chan_cnt > 1) {
774 		req->schq[NIX_TXSCH_LVL_SMQ] = chan_cnt;
775 		req->schq[NIX_TXSCH_LVL_TL4] = chan_cnt;
776 	}
777 
778 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
779 	if (rc)
780 		return rc;
781 
782 	rsp = (struct nix_txsch_alloc_rsp *)
783 	      otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
784 	if (IS_ERR(rsp))
785 		return PTR_ERR(rsp);
786 
787 	/* Setup transmit scheduler list */
788 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
789 		pfvf->hw.txschq_cnt[lvl] = rsp->schq[lvl];
790 		for (schq = 0; schq < rsp->schq[lvl]; schq++)
791 			pfvf->hw.txschq_list[lvl][schq] =
792 				rsp->schq_list[lvl][schq];
793 	}
794 
795 	pfvf->hw.txschq_link_cfg_lvl = rsp->link_cfg_lvl;
796 	pfvf->hw.txschq_aggr_lvl_rr_prio = rsp->aggr_lvl_rr_prio;
797 
798 	return 0;
799 }
800 
801 void otx2_txschq_free_one(struct otx2_nic *pfvf, u16 lvl, u16 schq)
802 {
803 	struct nix_txsch_free_req *free_req;
804 	int err;
805 
806 	mutex_lock(&pfvf->mbox.lock);
807 
808 	free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
809 	if (!free_req) {
810 		mutex_unlock(&pfvf->mbox.lock);
811 		netdev_err(pfvf->netdev,
812 			   "Failed alloc txschq free req\n");
813 		return;
814 	}
815 
816 	free_req->schq_lvl = lvl;
817 	free_req->schq = schq;
818 
819 	err = otx2_sync_mbox_msg(&pfvf->mbox);
820 	if (err) {
821 		netdev_err(pfvf->netdev,
822 			   "Failed stop txschq %d at level %d\n", schq, lvl);
823 	}
824 
825 	mutex_unlock(&pfvf->mbox.lock);
826 }
827 EXPORT_SYMBOL(otx2_txschq_free_one);
828 
829 void otx2_txschq_stop(struct otx2_nic *pfvf)
830 {
831 	int lvl, schq, idx;
832 
833 	/* free non QOS TLx nodes */
834 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
835 		for (idx = 0; idx < pfvf->hw.txschq_cnt[lvl]; idx++) {
836 			otx2_txschq_free_one(pfvf, lvl,
837 					     pfvf->hw.txschq_list[lvl][idx]);
838 		}
839 	}
840 
841 	/* Clear the txschq list */
842 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
843 		for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
844 			pfvf->hw.txschq_list[lvl][schq] = 0;
845 	}
846 
847 }
848 
849 void otx2_sqb_flush(struct otx2_nic *pfvf)
850 {
851 	int qidx, sqe_tail, sqe_head;
852 	struct otx2_snd_queue *sq;
853 	u64 incr, *ptr, val;
854 
855 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
856 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
857 		sq = &pfvf->qset.sq[qidx];
858 		if (!sq->sqb_ptrs)
859 			continue;
860 
861 		incr = (u64)qidx << 32;
862 		val = otx2_atomic64_add(incr, ptr);
863 		sqe_head = (val >> 20) & 0x3F;
864 		sqe_tail = (val >> 28) & 0x3F;
865 		if (sqe_head != sqe_tail)
866 			usleep_range(50, 60);
867 	}
868 }
869 
870 /* RED and drop levels of CQ on packet reception.
871  * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
872  */
873 #define RQ_PASS_LVL_CQ(skid, qsize)	((((skid) + 16) * 256) / (qsize))
874 #define RQ_DROP_LVL_CQ(skid, qsize)	(((skid) * 256) / (qsize))
875 
876 /* RED and drop levels of AURA for packet reception.
877  * For AURA level is measure of fullness (0x0 = empty, 255 = full).
878  * Eg: For RQ length 1K, for pass/drop level 204/230.
879  * RED accepts pkts if free pointers > 102 & <= 205.
880  * Drops pkts if free pointers < 102.
881  */
882 #define RQ_BP_LVL_AURA   (255 - ((85 * 256) / 100)) /* BP when 85% is full */
883 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
884 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
885 
886 static int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
887 {
888 	struct otx2_qset *qset = &pfvf->qset;
889 	struct nix_aq_enq_req *aq;
890 
891 	/* Get memory to put this msg */
892 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
893 	if (!aq)
894 		return -ENOMEM;
895 
896 	aq->rq.cq = qidx;
897 	aq->rq.ena = 1;
898 	aq->rq.pb_caching = 1;
899 	aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
900 	aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
901 	aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
902 	aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
903 	aq->rq.qint_idx = 0;
904 	aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
905 	aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
906 	aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
907 	aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
908 	aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
909 	aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
910 
911 	/* Fill AQ info */
912 	aq->qidx = qidx;
913 	aq->ctype = NIX_AQ_CTYPE_RQ;
914 	aq->op = NIX_AQ_INSTOP_INIT;
915 
916 	return otx2_sync_mbox_msg(&pfvf->mbox);
917 }
918 
919 int otx2_sq_aq_init(void *dev, u16 qidx, u8 chan_offset, u16 sqb_aura)
920 {
921 	struct otx2_nic *pfvf = dev;
922 	struct otx2_snd_queue *sq;
923 	struct nix_aq_enq_req *aq;
924 
925 	sq = &pfvf->qset.sq[qidx];
926 	sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
927 	/* Get memory to put this msg */
928 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
929 	if (!aq)
930 		return -ENOMEM;
931 
932 	aq->sq.cq = pfvf->hw.rx_queues + qidx;
933 	aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
934 	aq->sq.cq_ena = 1;
935 	aq->sq.ena = 1;
936 	aq->sq.smq = otx2_get_smq_idx(pfvf, qidx);
937 	aq->sq.smq_rr_quantum = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
938 	aq->sq.default_chan = pfvf->hw.tx_chan_base + chan_offset;
939 	aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
940 	aq->sq.sqb_aura = sqb_aura;
941 	aq->sq.sq_int_ena = NIX_SQINT_BITS;
942 	aq->sq.qint_idx = 0;
943 	/* Due pipelining impact minimum 2000 unused SQ CQE's
944 	 * need to maintain to avoid CQ overflow.
945 	 */
946 	aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (pfvf->qset.sqe_cnt));
947 
948 	/* Fill AQ info */
949 	aq->qidx = qidx;
950 	aq->ctype = NIX_AQ_CTYPE_SQ;
951 	aq->op = NIX_AQ_INSTOP_INIT;
952 
953 	return otx2_sync_mbox_msg(&pfvf->mbox);
954 }
955 
956 int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
957 {
958 	struct otx2_qset *qset = &pfvf->qset;
959 	struct otx2_snd_queue *sq;
960 	struct otx2_pool *pool;
961 	u8 chan_offset;
962 	int err;
963 
964 	pool = &pfvf->qset.pool[sqb_aura];
965 	sq = &qset->sq[qidx];
966 	sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
967 	sq->sqe_cnt = qset->sqe_cnt;
968 
969 	err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
970 	if (err)
971 		return err;
972 
973 	/* Allocate memory for NIX SQE (which includes NIX SG) and CPT SG.
974 	 * SG of NIX and CPT are same in size. Allocate memory for CPT SG
975 	 * same as NIX SQE for base address alignment.
976 	 * Layout of a NIX SQE and CPT SG entry:
977 	 *      -----------------------------
978 	 *     |     CPT Scatter Gather      |
979 	 *     |       (SQE SIZE)            |
980 	 *     |                             |
981 	 *      -----------------------------
982 	 *     |       NIX SQE               |
983 	 *     |       (SQE SIZE)            |
984 	 *     |                             |
985 	 *      -----------------------------
986 	 */
987 	err = qmem_alloc(pfvf->dev, &sq->sqe_ring, qset->sqe_cnt,
988 			 sq->sqe_size * 2);
989 	if (err)
990 		return err;
991 
992 	err = qmem_alloc(pfvf->dev, &sq->cpt_resp, qset->sqe_cnt, 64);
993 	if (err)
994 		return err;
995 
996 	if (qidx < pfvf->hw.tx_queues) {
997 		err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
998 				 TSO_HEADER_SIZE);
999 		if (err)
1000 			return err;
1001 	}
1002 
1003 	sq->sqe_base = sq->sqe->base;
1004 	sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
1005 	if (!sq->sg)
1006 		return -ENOMEM;
1007 
1008 	if (pfvf->ptp && qidx < pfvf->hw.tx_queues) {
1009 		err = qmem_alloc(pfvf->dev, &sq->timestamps, qset->sqe_cnt,
1010 				 sizeof(*sq->timestamps));
1011 		if (err) {
1012 			kfree(sq->sg);
1013 			sq->sg = NULL;
1014 			return err;
1015 		}
1016 	}
1017 
1018 	sq->head = 0;
1019 	sq->cons_head = 0;
1020 	sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
1021 	sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
1022 	/* Set SQE threshold to 10% of total SQEs */
1023 	sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
1024 	sq->aura_id = sqb_aura;
1025 	sq->aura_fc_addr = pool->fc_addr->base;
1026 	sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
1027 
1028 	sq->stats.bytes = 0;
1029 	sq->stats.pkts = 0;
1030 
1031 	chan_offset = qidx % pfvf->hw.tx_chan_cnt;
1032 	err = pfvf->hw_ops->sq_aq_init(pfvf, qidx, chan_offset, sqb_aura);
1033 	if (err) {
1034 		kfree(sq->sg);
1035 		sq->sg = NULL;
1036 		return err;
1037 	}
1038 
1039 	return 0;
1040 
1041 }
1042 
1043 static int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
1044 {
1045 	struct otx2_qset *qset = &pfvf->qset;
1046 	int err, pool_id, non_xdp_queues;
1047 	struct nix_aq_enq_req *aq;
1048 	struct otx2_cq_queue *cq;
1049 
1050 	cq = &qset->cq[qidx];
1051 	cq->cq_idx = qidx;
1052 	non_xdp_queues = pfvf->hw.rx_queues + pfvf->hw.tx_queues;
1053 	if (qidx < pfvf->hw.rx_queues) {
1054 		cq->cq_type = CQ_RX;
1055 		cq->cint_idx = qidx;
1056 		cq->cqe_cnt = qset->rqe_cnt;
1057 		if (pfvf->xdp_prog)
1058 			xdp_rxq_info_reg(&cq->xdp_rxq, pfvf->netdev, qidx, 0);
1059 	} else if (qidx < non_xdp_queues) {
1060 		cq->cq_type = CQ_TX;
1061 		cq->cint_idx = qidx - pfvf->hw.rx_queues;
1062 		cq->cqe_cnt = qset->sqe_cnt;
1063 	} else {
1064 		if (pfvf->hw.xdp_queues &&
1065 		    qidx < non_xdp_queues + pfvf->hw.xdp_queues) {
1066 			cq->cq_type = CQ_XDP;
1067 			cq->cint_idx = qidx - non_xdp_queues;
1068 			cq->cqe_cnt = qset->sqe_cnt;
1069 		} else {
1070 			cq->cq_type = CQ_QOS;
1071 			cq->cint_idx = qidx - non_xdp_queues -
1072 				       pfvf->hw.xdp_queues;
1073 			cq->cqe_cnt = qset->sqe_cnt;
1074 		}
1075 	}
1076 	cq->cqe_size = pfvf->qset.xqe_size;
1077 
1078 	/* Allocate memory for CQEs */
1079 	err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
1080 	if (err)
1081 		return err;
1082 
1083 	/* Save CQE CPU base for faster reference */
1084 	cq->cqe_base = cq->cqe->base;
1085 	/* In case where all RQs auras point to single pool,
1086 	 * all CQs receive buffer pool also point to same pool.
1087 	 */
1088 	pool_id = ((cq->cq_type == CQ_RX) &&
1089 		   (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
1090 	cq->rbpool = &qset->pool[pool_id];
1091 	cq->refill_task_sched = false;
1092 
1093 	/* Get memory to put this msg */
1094 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
1095 	if (!aq)
1096 		return -ENOMEM;
1097 
1098 	aq->cq.ena = 1;
1099 	aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
1100 	aq->cq.caching = 1;
1101 	aq->cq.base = cq->cqe->iova;
1102 	aq->cq.cint_idx = cq->cint_idx;
1103 	aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
1104 	aq->cq.qint_idx = 0;
1105 	aq->cq.avg_level = 255;
1106 
1107 	if (qidx < pfvf->hw.rx_queues) {
1108 		aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
1109 		aq->cq.drop_ena = 1;
1110 
1111 		if (!is_otx2_lbkvf(pfvf->pdev)) {
1112 			/* Enable receive CQ backpressure */
1113 			aq->cq.bp_ena = 1;
1114 #ifdef CONFIG_DCB
1115 			aq->cq.bpid = pfvf->bpid[pfvf->queue_to_pfc_map[qidx]];
1116 #else
1117 			aq->cq.bpid = pfvf->bpid[0];
1118 #endif
1119 
1120 			/* Set backpressure level is same as cq pass level */
1121 			aq->cq.bp = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
1122 		}
1123 	}
1124 
1125 	/* Fill AQ info */
1126 	aq->qidx = qidx;
1127 	aq->ctype = NIX_AQ_CTYPE_CQ;
1128 	aq->op = NIX_AQ_INSTOP_INIT;
1129 
1130 	return otx2_sync_mbox_msg(&pfvf->mbox);
1131 }
1132 
1133 static void otx2_pool_refill_task(struct work_struct *work)
1134 {
1135 	struct otx2_cq_queue *cq;
1136 	struct refill_work *wrk;
1137 	struct otx2_nic *pfvf;
1138 	int qidx;
1139 
1140 	wrk = container_of(work, struct refill_work, pool_refill_work.work);
1141 	pfvf = wrk->pf;
1142 	qidx = wrk - pfvf->refill_wrk;
1143 	cq = &pfvf->qset.cq[qidx];
1144 
1145 	cq->refill_task_sched = false;
1146 
1147 	local_bh_disable();
1148 	napi_schedule(wrk->napi);
1149 	local_bh_enable();
1150 }
1151 
1152 int otx2_config_nix_queues(struct otx2_nic *pfvf)
1153 {
1154 	int qidx, err;
1155 
1156 	/* Initialize RX queues */
1157 	for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
1158 		u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
1159 
1160 		err = otx2_rq_init(pfvf, qidx, lpb_aura);
1161 		if (err)
1162 			return err;
1163 	}
1164 
1165 	/* Initialize TX queues */
1166 	for (qidx = 0; qidx < pfvf->hw.non_qos_queues; qidx++) {
1167 		u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1168 
1169 		err = otx2_sq_init(pfvf, qidx, sqb_aura);
1170 		if (err)
1171 			return err;
1172 	}
1173 
1174 	/* Initialize completion queues */
1175 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1176 		err = otx2_cq_init(pfvf, qidx);
1177 		if (err)
1178 			return err;
1179 	}
1180 
1181 	pfvf->cq_op_addr = (__force u64 *)otx2_get_regaddr(pfvf,
1182 							   NIX_LF_CQ_OP_STATUS);
1183 
1184 	/* Initialize work queue for receive buffer refill */
1185 	pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
1186 					sizeof(struct refill_work), GFP_KERNEL);
1187 	if (!pfvf->refill_wrk)
1188 		return -ENOMEM;
1189 
1190 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1191 		pfvf->refill_wrk[qidx].pf = pfvf;
1192 		INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
1193 				  otx2_pool_refill_task);
1194 	}
1195 	return 0;
1196 }
1197 
1198 int otx2_config_nix(struct otx2_nic *pfvf)
1199 {
1200 	struct nix_lf_alloc_req  *nixlf;
1201 	struct nix_lf_alloc_rsp *rsp;
1202 	int err;
1203 
1204 	pfvf->qset.xqe_size = pfvf->hw.xqe_size;
1205 
1206 	/* Get memory to put this msg */
1207 	nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
1208 	if (!nixlf)
1209 		return -ENOMEM;
1210 
1211 	/* Set RQ/SQ/CQ counts */
1212 	nixlf->rq_cnt = pfvf->hw.rx_queues;
1213 	nixlf->sq_cnt = otx2_get_total_tx_queues(pfvf);
1214 	nixlf->cq_cnt = pfvf->qset.cq_cnt;
1215 	nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
1216 	nixlf->rss_grps = MAX_RSS_GROUPS;
1217 	nixlf->xqe_sz = pfvf->hw.xqe_size == 128 ? NIX_XQESZ_W16 : NIX_XQESZ_W64;
1218 	/* We don't know absolute NPA LF idx attached.
1219 	 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
1220 	 * NPA LF attached to this RVU PF/VF.
1221 	 */
1222 	nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
1223 	/* Disable alignment pad, enable L2 length check,
1224 	 * enable L4 TCP/UDP checksum verification.
1225 	 */
1226 	nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
1227 
1228 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1229 	if (err)
1230 		return err;
1231 
1232 	rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
1233 							   &nixlf->hdr);
1234 	if (IS_ERR(rsp))
1235 		return PTR_ERR(rsp);
1236 
1237 	if (rsp->qints < 1)
1238 		return -ENXIO;
1239 
1240 	return rsp->hdr.rc;
1241 }
1242 
1243 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
1244 {
1245 	struct otx2_qset *qset = &pfvf->qset;
1246 	struct otx2_hw *hw = &pfvf->hw;
1247 	struct otx2_snd_queue *sq;
1248 	int sqb, qidx;
1249 	u64 iova, pa;
1250 
1251 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
1252 		sq = &qset->sq[qidx];
1253 		if (!sq->sqb_ptrs)
1254 			continue;
1255 		for (sqb = 0; sqb < sq->sqb_count; sqb++) {
1256 			if (!sq->sqb_ptrs[sqb])
1257 				continue;
1258 			iova = sq->sqb_ptrs[sqb];
1259 			pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1260 			dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
1261 					     DMA_FROM_DEVICE,
1262 					     DMA_ATTR_SKIP_CPU_SYNC);
1263 			put_page(virt_to_page(phys_to_virt(pa)));
1264 		}
1265 		sq->sqb_count = 0;
1266 	}
1267 }
1268 
1269 void otx2_free_bufs(struct otx2_nic *pfvf, struct otx2_pool *pool,
1270 		    u64 iova, int size)
1271 {
1272 	struct page *page;
1273 	u64 pa;
1274 
1275 	pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1276 	page = virt_to_head_page(phys_to_virt(pa));
1277 
1278 	if (pool->page_pool) {
1279 		page_pool_put_full_page(pool->page_pool, page, true);
1280 	} else {
1281 		dma_unmap_page_attrs(pfvf->dev, iova, size,
1282 				     DMA_FROM_DEVICE,
1283 				     DMA_ATTR_SKIP_CPU_SYNC);
1284 
1285 		put_page(page);
1286 	}
1287 }
1288 
1289 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
1290 {
1291 	int pool_id, pool_start = 0, pool_end = 0, size = 0;
1292 	struct otx2_pool *pool;
1293 	u64 iova;
1294 
1295 	if (type == AURA_NIX_SQ) {
1296 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
1297 		pool_end =  pool_start + pfvf->hw.sqpool_cnt;
1298 		size = pfvf->hw.sqb_size;
1299 	}
1300 	if (type == AURA_NIX_RQ) {
1301 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
1302 		pool_end = pfvf->hw.rqpool_cnt;
1303 		size = pfvf->rbsize;
1304 	}
1305 
1306 	/* Free SQB and RQB pointers from the aura pool */
1307 	for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
1308 		iova = otx2_aura_allocptr(pfvf, pool_id);
1309 		pool = &pfvf->qset.pool[pool_id];
1310 		while (iova) {
1311 			if (type == AURA_NIX_RQ)
1312 				iova -= OTX2_HEAD_ROOM;
1313 
1314 			otx2_free_bufs(pfvf, pool, iova, size);
1315 
1316 			iova = otx2_aura_allocptr(pfvf, pool_id);
1317 		}
1318 	}
1319 }
1320 
1321 void otx2_aura_pool_free(struct otx2_nic *pfvf)
1322 {
1323 	struct otx2_pool *pool;
1324 	int pool_id;
1325 
1326 	if (!pfvf->qset.pool)
1327 		return;
1328 
1329 	for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
1330 		pool = &pfvf->qset.pool[pool_id];
1331 		qmem_free(pfvf->dev, pool->stack);
1332 		qmem_free(pfvf->dev, pool->fc_addr);
1333 		page_pool_destroy(pool->page_pool);
1334 		pool->page_pool = NULL;
1335 	}
1336 	devm_kfree(pfvf->dev, pfvf->qset.pool);
1337 	pfvf->qset.pool = NULL;
1338 }
1339 
1340 int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
1341 		   int pool_id, int numptrs)
1342 {
1343 	struct npa_aq_enq_req *aq;
1344 	struct otx2_pool *pool;
1345 	int err;
1346 
1347 	pool = &pfvf->qset.pool[pool_id];
1348 
1349 	/* Allocate memory for HW to update Aura count.
1350 	 * Alloc one cache line, so that it fits all FC_STYPE modes.
1351 	 */
1352 	if (!pool->fc_addr) {
1353 		err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
1354 		if (err)
1355 			return err;
1356 	}
1357 
1358 	/* Initialize this aura's context via AF */
1359 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1360 	if (!aq) {
1361 		/* Shared mbox memory buffer is full, flush it and retry */
1362 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1363 		if (err)
1364 			return err;
1365 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1366 		if (!aq)
1367 			return -ENOMEM;
1368 	}
1369 
1370 	aq->aura_id = aura_id;
1371 	/* Will be filled by AF with correct pool context address */
1372 	aq->aura.pool_addr = pool_id;
1373 	aq->aura.pool_caching = 1;
1374 	aq->aura.shift = ilog2(numptrs) - 8;
1375 	aq->aura.count = numptrs;
1376 	aq->aura.limit = numptrs;
1377 	aq->aura.avg_level = 255;
1378 	aq->aura.ena = 1;
1379 	aq->aura.fc_ena = 1;
1380 	aq->aura.fc_addr = pool->fc_addr->iova;
1381 	aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1382 
1383 	/* Enable backpressure for RQ aura */
1384 	if (aura_id < pfvf->hw.rqpool_cnt && !is_otx2_lbkvf(pfvf->pdev)) {
1385 		aq->aura.bp_ena = 0;
1386 		/* If NIX1 LF is attached then specify NIX1_RX.
1387 		 *
1388 		 * Below NPA_AURA_S[BP_ENA] is set according to the
1389 		 * NPA_BPINTF_E enumeration given as:
1390 		 * 0x0 + a*0x1 where 'a' is 0 for NIX0_RX and 1 for NIX1_RX so
1391 		 * NIX0_RX is 0x0 + 0*0x1 = 0
1392 		 * NIX1_RX is 0x0 + 1*0x1 = 1
1393 		 * But in HRM it is given that
1394 		 * "NPA_AURA_S[BP_ENA](w1[33:32]) - Enable aura backpressure to
1395 		 * NIX-RX based on [BP] level. One bit per NIX-RX; index
1396 		 * enumerated by NPA_BPINTF_E."
1397 		 */
1398 		if (pfvf->nix_blkaddr == BLKADDR_NIX1)
1399 			aq->aura.bp_ena = 1;
1400 #ifdef CONFIG_DCB
1401 		aq->aura.nix0_bpid = pfvf->bpid[pfvf->queue_to_pfc_map[aura_id]];
1402 #else
1403 		aq->aura.nix0_bpid = pfvf->bpid[0];
1404 #endif
1405 
1406 		/* Set backpressure level for RQ's Aura */
1407 		aq->aura.bp = RQ_BP_LVL_AURA;
1408 	}
1409 
1410 	/* Fill AQ info */
1411 	aq->ctype = NPA_AQ_CTYPE_AURA;
1412 	aq->op = NPA_AQ_INSTOP_INIT;
1413 
1414 	return 0;
1415 }
1416 
1417 int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1418 		   int stack_pages, int numptrs, int buf_size, int type)
1419 {
1420 	struct page_pool_params pp_params = { 0 };
1421 	struct npa_aq_enq_req *aq;
1422 	struct otx2_pool *pool;
1423 	int err;
1424 
1425 	pool = &pfvf->qset.pool[pool_id];
1426 	/* Alloc memory for stack which is used to store buffer pointers */
1427 	err = qmem_alloc(pfvf->dev, &pool->stack,
1428 			 stack_pages, pfvf->hw.stack_pg_bytes);
1429 	if (err)
1430 		return err;
1431 
1432 	pool->rbsize = buf_size;
1433 
1434 	/* Initialize this pool's context via AF */
1435 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1436 	if (!aq) {
1437 		/* Shared mbox memory buffer is full, flush it and retry */
1438 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1439 		if (err) {
1440 			qmem_free(pfvf->dev, pool->stack);
1441 			return err;
1442 		}
1443 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1444 		if (!aq) {
1445 			qmem_free(pfvf->dev, pool->stack);
1446 			return -ENOMEM;
1447 		}
1448 	}
1449 
1450 	aq->aura_id = pool_id;
1451 	aq->pool.stack_base = pool->stack->iova;
1452 	aq->pool.stack_caching = 1;
1453 	aq->pool.ena = 1;
1454 	aq->pool.buf_size = buf_size / 128;
1455 	aq->pool.stack_max_pages = stack_pages;
1456 	aq->pool.shift = ilog2(numptrs) - 8;
1457 	aq->pool.ptr_start = 0;
1458 	aq->pool.ptr_end = ~0ULL;
1459 
1460 	/* Fill AQ info */
1461 	aq->ctype = NPA_AQ_CTYPE_POOL;
1462 	aq->op = NPA_AQ_INSTOP_INIT;
1463 
1464 	if (type != AURA_NIX_RQ) {
1465 		pool->page_pool = NULL;
1466 		return 0;
1467 	}
1468 
1469 	pp_params.order = get_order(buf_size);
1470 	pp_params.flags = PP_FLAG_DMA_MAP;
1471 	pp_params.pool_size = min(OTX2_PAGE_POOL_SZ, numptrs);
1472 	pp_params.nid = NUMA_NO_NODE;
1473 	pp_params.dev = pfvf->dev;
1474 	pp_params.dma_dir = DMA_FROM_DEVICE;
1475 	pool->page_pool = page_pool_create(&pp_params);
1476 	if (IS_ERR(pool->page_pool)) {
1477 		netdev_err(pfvf->netdev, "Creation of page pool failed\n");
1478 		return PTR_ERR(pool->page_pool);
1479 	}
1480 
1481 	return 0;
1482 }
1483 
1484 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1485 {
1486 	int qidx, pool_id, stack_pages, num_sqbs;
1487 	struct otx2_qset *qset = &pfvf->qset;
1488 	struct otx2_hw *hw = &pfvf->hw;
1489 	struct otx2_snd_queue *sq;
1490 	struct otx2_pool *pool;
1491 	dma_addr_t bufptr;
1492 	int err, ptr;
1493 
1494 	/* Calculate number of SQBs needed.
1495 	 *
1496 	 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1497 	 * Last SQE is used for pointing to next SQB.
1498 	 */
1499 	num_sqbs = (hw->sqb_size / 128) - 1;
1500 	num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1501 
1502 	/* Get no of stack pages needed */
1503 	stack_pages =
1504 		(num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1505 
1506 	for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1507 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1508 		/* Initialize aura context */
1509 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1510 		if (err)
1511 			goto fail;
1512 
1513 		/* Initialize pool context */
1514 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1515 				     num_sqbs, hw->sqb_size, AURA_NIX_SQ);
1516 		if (err)
1517 			goto fail;
1518 	}
1519 
1520 	/* Flush accumulated messages */
1521 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1522 	if (err)
1523 		goto fail;
1524 
1525 	/* Allocate pointers and free them to aura/pool */
1526 	for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1527 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1528 		pool = &pfvf->qset.pool[pool_id];
1529 
1530 		sq = &qset->sq[qidx];
1531 		sq->sqb_count = 0;
1532 		sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(*sq->sqb_ptrs), GFP_KERNEL);
1533 		if (!sq->sqb_ptrs) {
1534 			err = -ENOMEM;
1535 			goto err_mem;
1536 		}
1537 
1538 		for (ptr = 0; ptr < num_sqbs; ptr++) {
1539 			err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1540 			if (err)
1541 				goto err_mem;
1542 			pfvf->hw_ops->aura_freeptr(pfvf, pool_id, bufptr);
1543 			sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1544 		}
1545 	}
1546 
1547 err_mem:
1548 	return err ? -ENOMEM : 0;
1549 
1550 fail:
1551 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1552 	otx2_aura_pool_free(pfvf);
1553 	return err;
1554 }
1555 
1556 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1557 {
1558 	struct otx2_hw *hw = &pfvf->hw;
1559 	int stack_pages, pool_id, rq;
1560 	struct otx2_pool *pool;
1561 	int err, ptr, num_ptrs;
1562 	dma_addr_t bufptr;
1563 
1564 	num_ptrs = pfvf->qset.rqe_cnt;
1565 
1566 	stack_pages =
1567 		(num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1568 
1569 	for (rq = 0; rq < hw->rx_queues; rq++) {
1570 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1571 		/* Initialize aura context */
1572 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1573 		if (err)
1574 			goto fail;
1575 	}
1576 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1577 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1578 				     num_ptrs, pfvf->rbsize, AURA_NIX_RQ);
1579 		if (err)
1580 			goto fail;
1581 	}
1582 
1583 	/* Flush accumulated messages */
1584 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1585 	if (err)
1586 		goto fail;
1587 
1588 	/* Allocate pointers and free them to aura/pool */
1589 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1590 		pool = &pfvf->qset.pool[pool_id];
1591 		for (ptr = 0; ptr < num_ptrs; ptr++) {
1592 			err = otx2_alloc_rbuf(pfvf, pool, &bufptr);
1593 			if (err)
1594 				return -ENOMEM;
1595 			pfvf->hw_ops->aura_freeptr(pfvf, pool_id,
1596 						   bufptr + OTX2_HEAD_ROOM);
1597 		}
1598 	}
1599 	return 0;
1600 fail:
1601 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1602 	otx2_aura_pool_free(pfvf);
1603 	return err;
1604 }
1605 
1606 int otx2_config_npa(struct otx2_nic *pfvf)
1607 {
1608 	struct otx2_qset *qset = &pfvf->qset;
1609 	struct npa_lf_alloc_req  *npalf;
1610 	struct otx2_hw *hw = &pfvf->hw;
1611 	int aura_cnt;
1612 
1613 	/* Pool - Stack of free buffer pointers
1614 	 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1615 	 */
1616 
1617 	if (!hw->pool_cnt)
1618 		return -EINVAL;
1619 
1620 	qset->pool = devm_kcalloc(pfvf->dev, hw->pool_cnt,
1621 				  sizeof(struct otx2_pool), GFP_KERNEL);
1622 	if (!qset->pool)
1623 		return -ENOMEM;
1624 
1625 	/* Get memory to put this msg */
1626 	npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1627 	if (!npalf)
1628 		return -ENOMEM;
1629 
1630 	/* Set aura and pool counts */
1631 	npalf->nr_pools = hw->pool_cnt;
1632 	aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1633 	npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1634 
1635 	return otx2_sync_mbox_msg(&pfvf->mbox);
1636 }
1637 
1638 int otx2_detach_resources(struct mbox *mbox)
1639 {
1640 	struct rsrc_detach *detach;
1641 
1642 	mutex_lock(&mbox->lock);
1643 	detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1644 	if (!detach) {
1645 		mutex_unlock(&mbox->lock);
1646 		return -ENOMEM;
1647 	}
1648 
1649 	/* detach all */
1650 	detach->partial = false;
1651 
1652 	/* Send detach request to AF */
1653 	otx2_sync_mbox_msg(mbox);
1654 	mutex_unlock(&mbox->lock);
1655 	return 0;
1656 }
1657 EXPORT_SYMBOL(otx2_detach_resources);
1658 
1659 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1660 {
1661 	struct rsrc_attach *attach;
1662 	struct msg_req *msix;
1663 	int err;
1664 
1665 	mutex_lock(&pfvf->mbox.lock);
1666 	/* Get memory to put this msg */
1667 	attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1668 	if (!attach) {
1669 		mutex_unlock(&pfvf->mbox.lock);
1670 		return -ENOMEM;
1671 	}
1672 
1673 	attach->npalf = true;
1674 	attach->nixlf = true;
1675 
1676 	/* Send attach request to AF */
1677 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1678 	if (err) {
1679 		mutex_unlock(&pfvf->mbox.lock);
1680 		return err;
1681 	}
1682 
1683 	pfvf->nix_blkaddr = BLKADDR_NIX0;
1684 
1685 	/* If the platform has two NIX blocks then LF may be
1686 	 * allocated from NIX1.
1687 	 */
1688 	if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1689 		pfvf->nix_blkaddr = BLKADDR_NIX1;
1690 
1691 	/* Get NPA and NIX MSIX vector offsets */
1692 	msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1693 	if (!msix) {
1694 		mutex_unlock(&pfvf->mbox.lock);
1695 		return -ENOMEM;
1696 	}
1697 
1698 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1699 	if (err) {
1700 		mutex_unlock(&pfvf->mbox.lock);
1701 		return err;
1702 	}
1703 	mutex_unlock(&pfvf->mbox.lock);
1704 
1705 	if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1706 	    pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1707 		dev_err(pfvf->dev,
1708 			"RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1709 		return -EINVAL;
1710 	}
1711 
1712 	return 0;
1713 }
1714 EXPORT_SYMBOL(otx2_attach_npa_nix);
1715 
1716 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1717 {
1718 	struct hwctx_disable_req *req;
1719 
1720 	mutex_lock(&mbox->lock);
1721 	/* Request AQ to disable this context */
1722 	if (npa)
1723 		req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1724 	else
1725 		req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1726 
1727 	if (!req) {
1728 		mutex_unlock(&mbox->lock);
1729 		return;
1730 	}
1731 
1732 	req->ctype = type;
1733 
1734 	if (otx2_sync_mbox_msg(mbox))
1735 		dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1736 			__func__);
1737 
1738 	mutex_unlock(&mbox->lock);
1739 }
1740 
1741 int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
1742 {
1743 	struct nix_bp_cfg_req *req;
1744 
1745 	if (enable)
1746 		req = otx2_mbox_alloc_msg_nix_bp_enable(&pfvf->mbox);
1747 	else
1748 		req = otx2_mbox_alloc_msg_nix_bp_disable(&pfvf->mbox);
1749 
1750 	if (!req)
1751 		return -ENOMEM;
1752 
1753 	req->chan_base = 0;
1754 	if (otx2_is_pfc_enabled(pfvf)) {
1755 		req->chan_cnt = IEEE_8021QAZ_MAX_TCS;
1756 		req->bpid_per_chan = 1;
1757 	} else {
1758 		req->chan_cnt = 1;
1759 		req->bpid_per_chan = 0;
1760 	}
1761 
1762 	return otx2_sync_mbox_msg(&pfvf->mbox);
1763 }
1764 EXPORT_SYMBOL(otx2_nix_config_bp);
1765 
1766 int otx2_nix_cpt_config_bp(struct otx2_nic *pfvf, bool enable)
1767 {
1768 	struct nix_bp_cfg_req *req;
1769 
1770 	if (enable)
1771 		req = otx2_mbox_alloc_msg_nix_cpt_bp_enable(&pfvf->mbox);
1772 	else
1773 		req = otx2_mbox_alloc_msg_nix_cpt_bp_disable(&pfvf->mbox);
1774 
1775 	if (!req)
1776 		return -ENOMEM;
1777 
1778 	req->chan_base = 0;
1779 	if (otx2_is_pfc_enabled(pfvf)) {
1780 		req->chan_cnt = IEEE_8021QAZ_MAX_TCS;
1781 		req->bpid_per_chan = 1;
1782 	} else {
1783 		req->chan_cnt = 1;
1784 		req->bpid_per_chan = 0;
1785 	}
1786 
1787 	return otx2_sync_mbox_msg(&pfvf->mbox);
1788 }
1789 EXPORT_SYMBOL(otx2_nix_cpt_config_bp);
1790 
1791 /* Mbox message handlers */
1792 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1793 			    struct cgx_stats_rsp *rsp)
1794 {
1795 	int id;
1796 
1797 	for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1798 		pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1799 	for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1800 		pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1801 }
1802 
1803 void mbox_handler_cgx_fec_stats(struct otx2_nic *pfvf,
1804 				struct cgx_fec_stats_rsp *rsp)
1805 {
1806 	pfvf->hw.cgx_fec_corr_blks += rsp->fec_corr_blks;
1807 	pfvf->hw.cgx_fec_uncorr_blks += rsp->fec_uncorr_blks;
1808 }
1809 
1810 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1811 			       struct npa_lf_alloc_rsp *rsp)
1812 {
1813 	pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1814 	pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1815 }
1816 EXPORT_SYMBOL(mbox_handler_npa_lf_alloc);
1817 
1818 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1819 			       struct nix_lf_alloc_rsp *rsp)
1820 {
1821 	pfvf->hw.sqb_size = rsp->sqb_size;
1822 	pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1823 	pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1824 	pfvf->hw.rx_chan_cnt = rsp->rx_chan_cnt;
1825 	pfvf->hw.tx_chan_cnt = rsp->tx_chan_cnt;
1826 	pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1827 	pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1828 	pfvf->hw.cgx_links = rsp->cgx_links;
1829 	pfvf->hw.lbk_links = rsp->lbk_links;
1830 	pfvf->hw.tx_link = rsp->tx_link;
1831 }
1832 EXPORT_SYMBOL(mbox_handler_nix_lf_alloc);
1833 
1834 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1835 			      struct msix_offset_rsp *rsp)
1836 {
1837 	pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1838 	pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1839 }
1840 EXPORT_SYMBOL(mbox_handler_msix_offset);
1841 
1842 void mbox_handler_nix_bp_enable(struct otx2_nic *pfvf,
1843 				struct nix_bp_cfg_rsp *rsp)
1844 {
1845 	int chan, chan_id;
1846 
1847 	for (chan = 0; chan < rsp->chan_cnt; chan++) {
1848 		chan_id = ((rsp->chan_bpid[chan] >> 10) & 0x7F);
1849 		pfvf->bpid[chan_id] = rsp->chan_bpid[chan] & 0x3FF;
1850 	}
1851 }
1852 EXPORT_SYMBOL(mbox_handler_nix_bp_enable);
1853 
1854 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1855 {
1856 	struct otx2_qset *qset = &pfvf->qset;
1857 	struct otx2_hw *hw = &pfvf->hw;
1858 	int irq, qidx;
1859 
1860 	for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1861 	     qidx < n;
1862 	     qidx++, irq++) {
1863 		int vector = pci_irq_vector(pfvf->pdev, irq);
1864 
1865 		irq_set_affinity_hint(vector, NULL);
1866 		free_cpumask_var(hw->affinity_mask[irq]);
1867 		free_irq(vector, &qset->napi[qidx]);
1868 	}
1869 }
1870 EXPORT_SYMBOL(otx2_free_cints);
1871 
1872 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1873 {
1874 	struct otx2_hw *hw = &pfvf->hw;
1875 	int vec, cpu, irq, cint;
1876 
1877 	vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1878 	cpu = cpumask_first(cpu_online_mask);
1879 
1880 	/* CQ interrupts */
1881 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1882 		if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1883 			return;
1884 
1885 		cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1886 
1887 		irq = pci_irq_vector(pfvf->pdev, vec);
1888 		irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1889 
1890 		cpu = cpumask_next(cpu, cpu_online_mask);
1891 		if (unlikely(cpu >= nr_cpu_ids))
1892 			cpu = 0;
1893 	}
1894 }
1895 
1896 static u32 get_dwrr_mtu(struct otx2_nic *pfvf, struct nix_hw_info *hw)
1897 {
1898 	if (is_otx2_lbkvf(pfvf->pdev)) {
1899 		pfvf->hw.smq_link_type = SMQ_LINK_TYPE_LBK;
1900 		return hw->lbk_dwrr_mtu;
1901 	}
1902 
1903 	pfvf->hw.smq_link_type = SMQ_LINK_TYPE_RPM;
1904 	return hw->rpm_dwrr_mtu;
1905 }
1906 
1907 u16 otx2_get_max_mtu(struct otx2_nic *pfvf)
1908 {
1909 	struct nix_hw_info *rsp;
1910 	struct msg_req *req;
1911 	u16 max_mtu;
1912 	int rc;
1913 
1914 	mutex_lock(&pfvf->mbox.lock);
1915 
1916 	req = otx2_mbox_alloc_msg_nix_get_hw_info(&pfvf->mbox);
1917 	if (!req) {
1918 		rc =  -ENOMEM;
1919 		goto out;
1920 	}
1921 
1922 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
1923 	if (!rc) {
1924 		rsp = (struct nix_hw_info *)
1925 		       otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
1926 		if (IS_ERR(rsp)) {
1927 			rc = PTR_ERR(rsp);
1928 			goto out;
1929 		}
1930 
1931 		/* HW counts VLAN insertion bytes (8 for double tag)
1932 		 * irrespective of whether SQE is requesting to insert VLAN
1933 		 * in the packet or not. Hence these 8 bytes have to be
1934 		 * discounted from max packet size otherwise HW will throw
1935 		 * SMQ errors
1936 		 */
1937 		max_mtu = rsp->max_mtu - 8 - OTX2_ETH_HLEN;
1938 
1939 		/* Also save DWRR MTU, needed for DWRR weight calculation */
1940 		pfvf->hw.dwrr_mtu = get_dwrr_mtu(pfvf, rsp);
1941 		if (!pfvf->hw.dwrr_mtu)
1942 			pfvf->hw.dwrr_mtu = 1;
1943 	}
1944 
1945 out:
1946 	mutex_unlock(&pfvf->mbox.lock);
1947 	if (rc) {
1948 		dev_warn(pfvf->dev,
1949 			 "Failed to get MTU from hardware setting default value(1500)\n");
1950 		max_mtu = 1500;
1951 	}
1952 	return max_mtu;
1953 }
1954 EXPORT_SYMBOL(otx2_get_max_mtu);
1955 
1956 int otx2_handle_ntuple_tc_features(struct net_device *netdev, netdev_features_t features)
1957 {
1958 	netdev_features_t changed = features ^ netdev->features;
1959 	struct otx2_nic *pfvf = netdev_priv(netdev);
1960 	bool ntuple = !!(features & NETIF_F_NTUPLE);
1961 	bool tc = !!(features & NETIF_F_HW_TC);
1962 
1963 	if ((changed & NETIF_F_NTUPLE) && !ntuple)
1964 		otx2_destroy_ntuple_flows(pfvf);
1965 
1966 	if ((changed & NETIF_F_NTUPLE) && ntuple) {
1967 		if (!pfvf->flow_cfg->max_flows) {
1968 			netdev_err(netdev,
1969 				   "Can't enable NTUPLE, MCAM entries not allocated\n");
1970 			return -EINVAL;
1971 		}
1972 	}
1973 
1974 	if ((changed & NETIF_F_HW_TC) && !tc &&
1975 	    otx2_tc_flower_rule_cnt(pfvf)) {
1976 		netdev_err(netdev, "Can't disable TC hardware offload while flows are active\n");
1977 		return -EBUSY;
1978 	}
1979 
1980 	if ((changed & NETIF_F_NTUPLE) && ntuple &&
1981 	    otx2_tc_flower_rule_cnt(pfvf) && !(changed & NETIF_F_HW_TC)) {
1982 		netdev_err(netdev,
1983 			   "Can't enable NTUPLE when TC flower offload is active, disable TC rules and retry\n");
1984 		return -EINVAL;
1985 	}
1986 
1987 	return 0;
1988 }
1989 EXPORT_SYMBOL(otx2_handle_ntuple_tc_features);
1990 
1991 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
1992 int __weak								\
1993 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf,		\
1994 				struct _req_type *req,			\
1995 				struct _rsp_type *rsp)			\
1996 {									\
1997 	/* Nothing to do here */					\
1998 	return 0;							\
1999 }									\
2000 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
2001 MBOX_UP_CGX_MESSAGES
2002 MBOX_UP_MCS_MESSAGES
2003 #undef M
2004 
2005 dma_addr_t otx2_dma_map_skb_frag(struct otx2_nic *pfvf,
2006 				 struct sk_buff *skb, int seg, int *len)
2007 {
2008 	enum dma_data_direction dir = DMA_TO_DEVICE;
2009 	const skb_frag_t *frag;
2010 	struct page *page;
2011 	int offset;
2012 
2013 	/* Crypto hardware need write permission for ipsec crypto offload */
2014 	if (unlikely(xfrm_offload(skb))) {
2015 		dir = DMA_BIDIRECTIONAL;
2016 		skb = skb_unshare(skb, GFP_ATOMIC);
2017 	}
2018 
2019 	/* First segment is always skb->data */
2020 	if (!seg) {
2021 		page = virt_to_page(skb->data);
2022 		offset = offset_in_page(skb->data);
2023 		*len = skb_headlen(skb);
2024 	} else {
2025 		frag = &skb_shinfo(skb)->frags[seg - 1];
2026 		page = skb_frag_page(frag);
2027 		offset = skb_frag_off(frag);
2028 		*len = skb_frag_size(frag);
2029 	}
2030 	return otx2_dma_map_page(pfvf, page, offset, *len, dir);
2031 }
2032 
2033 void otx2_dma_unmap_skb_frags(struct otx2_nic *pfvf, struct sg_list *sg)
2034 {
2035 	enum dma_data_direction dir = DMA_TO_DEVICE;
2036 	struct sk_buff *skb = NULL;
2037 	int seg;
2038 
2039 	skb = (struct sk_buff *)sg->skb;
2040 	if (unlikely(xfrm_offload(skb)))
2041 		dir = DMA_BIDIRECTIONAL;
2042 
2043 	for (seg = 0; seg < sg->num_segs; seg++) {
2044 		otx2_dma_unmap_page(pfvf, sg->dma_addr[seg],
2045 				    sg->size[seg], dir);
2046 	}
2047 	sg->num_segs = 0;
2048 }
2049