xref: /linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c (revision c5fbdf0ba7c1a6ed52dc3650bee73ce00c86cf7f)
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 <linux/dcbnl.h>
14 #include <net/xfrm.h>
15 
16 #include "otx2_reg.h"
17 #include "otx2_common.h"
18 #include "otx2_struct.h"
19 #include "cn10k.h"
20 #include "otx2_xsk.h"
21 
22 static bool otx2_is_pfc_enabled(struct otx2_nic *pfvf)
23 {
24 	return IS_ENABLED(CONFIG_DCB) && !!pfvf->pfc_en;
25 }
26 
27 static void otx2_nix_rq_op_stats(struct queue_stats *stats,
28 				 struct otx2_nic *pfvf, int qidx)
29 {
30 	u64 incr = (u64)qidx << 32;
31 	void __iomem *ptr;
32 
33 	ptr = otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_OCTS);
34 	stats->bytes = otx2_atomic64_add(incr, ptr);
35 
36 	ptr = otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_PKTS);
37 	stats->pkts = otx2_atomic64_add(incr, ptr);
38 }
39 
40 static void otx2_nix_sq_op_stats(struct queue_stats *stats,
41 				 struct otx2_nic *pfvf, int qidx)
42 {
43 	u64 incr = (u64)qidx << 32;
44 	void __iomem *ptr;
45 
46 	ptr = otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_OCTS);
47 	stats->bytes = otx2_atomic64_add(incr, ptr);
48 
49 	ptr = otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_PKTS);
50 	stats->pkts = otx2_atomic64_add(incr, ptr);
51 }
52 
53 void otx2_update_lmac_stats(struct otx2_nic *pfvf)
54 {
55 	struct msg_req *req;
56 
57 	if (!netif_running(pfvf->netdev))
58 		return;
59 
60 	mutex_lock(&pfvf->mbox.lock);
61 	req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
62 	if (!req) {
63 		mutex_unlock(&pfvf->mbox.lock);
64 		return;
65 	}
66 
67 	otx2_sync_mbox_msg(&pfvf->mbox);
68 	mutex_unlock(&pfvf->mbox.lock);
69 }
70 
71 void otx2_update_lmac_fec_stats(struct otx2_nic *pfvf)
72 {
73 	struct msg_req *req;
74 
75 	if (!netif_running(pfvf->netdev))
76 		return;
77 	mutex_lock(&pfvf->mbox.lock);
78 	req = otx2_mbox_alloc_msg_cgx_fec_stats(&pfvf->mbox);
79 	if (req)
80 		otx2_sync_mbox_msg(&pfvf->mbox);
81 	mutex_unlock(&pfvf->mbox.lock);
82 }
83 
84 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
85 {
86 	struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
87 
88 	if (!pfvf->qset.rq)
89 		return 0;
90 
91 	otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
92 	return 1;
93 }
94 EXPORT_SYMBOL(otx2_update_rq_stats);
95 
96 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
97 {
98 	struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
99 
100 	if (!pfvf->qset.sq)
101 		return 0;
102 
103 	if (qidx >= pfvf->hw.non_qos_queues) {
104 		if (!test_bit(qidx - pfvf->hw.non_qos_queues, pfvf->qos.qos_sq_bmap))
105 			return 0;
106 	}
107 
108 	otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
109 	return 1;
110 }
111 EXPORT_SYMBOL(otx2_update_sq_stats);
112 
113 void otx2_get_dev_stats(struct otx2_nic *pfvf)
114 {
115 	struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
116 
117 	dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
118 	dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
119 	dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
120 	dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
121 	dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
122 	dev_stats->rx_frames = dev_stats->rx_bcast_frames +
123 			       dev_stats->rx_mcast_frames +
124 			       dev_stats->rx_ucast_frames;
125 
126 	dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
127 	dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
128 	dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
129 	dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
130 	dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
131 	dev_stats->tx_frames = dev_stats->tx_bcast_frames +
132 			       dev_stats->tx_mcast_frames +
133 			       dev_stats->tx_ucast_frames;
134 }
135 
136 void otx2_get_stats64(struct net_device *netdev,
137 		      struct rtnl_link_stats64 *stats)
138 {
139 	struct otx2_nic *pfvf = netdev_priv(netdev);
140 	struct otx2_dev_stats *dev_stats;
141 
142 	otx2_get_dev_stats(pfvf);
143 
144 	dev_stats = &pfvf->hw.dev_stats;
145 	stats->rx_bytes = dev_stats->rx_bytes;
146 	stats->rx_packets = dev_stats->rx_frames;
147 	stats->rx_dropped = dev_stats->rx_drops;
148 	stats->multicast = dev_stats->rx_mcast_frames;
149 
150 	stats->tx_bytes = dev_stats->tx_bytes;
151 	stats->tx_packets = dev_stats->tx_frames;
152 	stats->tx_dropped = dev_stats->tx_drops;
153 }
154 EXPORT_SYMBOL(otx2_get_stats64);
155 
156 /* Sync MAC address with RVU AF */
157 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
158 {
159 	struct nix_set_mac_addr *req;
160 	int err;
161 
162 	mutex_lock(&pfvf->mbox.lock);
163 	req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
164 	if (!req) {
165 		mutex_unlock(&pfvf->mbox.lock);
166 		return -ENOMEM;
167 	}
168 
169 	ether_addr_copy(req->mac_addr, mac);
170 
171 	err = otx2_sync_mbox_msg(&pfvf->mbox);
172 	mutex_unlock(&pfvf->mbox.lock);
173 	return err;
174 }
175 
176 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
177 				struct net_device *netdev)
178 {
179 	struct nix_get_mac_addr_rsp *rsp;
180 	struct mbox_msghdr *msghdr;
181 	struct msg_req *req;
182 	int err;
183 
184 	mutex_lock(&pfvf->mbox.lock);
185 	req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
186 	if (!req) {
187 		mutex_unlock(&pfvf->mbox.lock);
188 		return -ENOMEM;
189 	}
190 
191 	err = otx2_sync_mbox_msg(&pfvf->mbox);
192 	if (err) {
193 		mutex_unlock(&pfvf->mbox.lock);
194 		return err;
195 	}
196 
197 	msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
198 	if (IS_ERR(msghdr)) {
199 		mutex_unlock(&pfvf->mbox.lock);
200 		return PTR_ERR(msghdr);
201 	}
202 	rsp = (struct nix_get_mac_addr_rsp *)msghdr;
203 	eth_hw_addr_set(netdev, rsp->mac_addr);
204 	mutex_unlock(&pfvf->mbox.lock);
205 
206 	return 0;
207 }
208 
209 int otx2_set_mac_address(struct net_device *netdev, void *p)
210 {
211 	struct otx2_nic *pfvf = netdev_priv(netdev);
212 	struct sockaddr *addr = p;
213 
214 	if (!is_valid_ether_addr(addr->sa_data))
215 		return -EADDRNOTAVAIL;
216 
217 	if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data)) {
218 		eth_hw_addr_set(netdev, addr->sa_data);
219 		/* update dmac field in vlan offload rule */
220 		if (netif_running(netdev) &&
221 		    pfvf->flags & OTX2_FLAG_RX_VLAN_SUPPORT)
222 			otx2_install_rxvlan_offload_flow(pfvf);
223 		/* update dmac address in ntuple and DMAC filter list */
224 		if (pfvf->flags & OTX2_FLAG_DMACFLTR_SUPPORT)
225 			otx2_dmacflt_update_pfmac_flow(pfvf);
226 	} else {
227 		return -EPERM;
228 	}
229 
230 	return 0;
231 }
232 EXPORT_SYMBOL(otx2_set_mac_address);
233 
234 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
235 {
236 	struct nix_frs_cfg *req;
237 	u16 maxlen;
238 	int err;
239 
240 	maxlen = pfvf->hw.max_mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
241 
242 	mutex_lock(&pfvf->mbox.lock);
243 	req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
244 	if (!req) {
245 		mutex_unlock(&pfvf->mbox.lock);
246 		return -ENOMEM;
247 	}
248 
249 	req->maxlen = mtu + OTX2_ETH_HLEN + OTX2_HW_TIMESTAMP_LEN;
250 
251 	/* Use max receive length supported by hardware for loopback devices */
252 	if (is_otx2_lbkvf(pfvf->pdev))
253 		req->maxlen = maxlen;
254 
255 	err = otx2_sync_mbox_msg(&pfvf->mbox);
256 	mutex_unlock(&pfvf->mbox.lock);
257 	return err;
258 }
259 EXPORT_SYMBOL(otx2_hw_set_mtu);
260 
261 int otx2_config_pause_frm(struct otx2_nic *pfvf)
262 {
263 	struct cgx_pause_frm_cfg *req;
264 	int err;
265 
266 	if (is_otx2_lbkvf(pfvf->pdev) || is_otx2_sdp_rep(pfvf->pdev))
267 		return 0;
268 
269 	mutex_lock(&pfvf->mbox.lock);
270 	req = otx2_mbox_alloc_msg_cgx_cfg_pause_frm(&pfvf->mbox);
271 	if (!req) {
272 		err = -ENOMEM;
273 		goto unlock;
274 	}
275 
276 	req->rx_pause = !!(pfvf->flags & OTX2_FLAG_RX_PAUSE_ENABLED);
277 	req->tx_pause = !!(pfvf->flags & OTX2_FLAG_TX_PAUSE_ENABLED);
278 	req->set = 1;
279 
280 	err = otx2_sync_mbox_msg(&pfvf->mbox);
281 unlock:
282 	mutex_unlock(&pfvf->mbox.lock);
283 	return err;
284 }
285 EXPORT_SYMBOL(otx2_config_pause_frm);
286 
287 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
288 {
289 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
290 	struct nix_rss_flowkey_cfg_rsp *rsp;
291 	struct nix_rss_flowkey_cfg *req;
292 	int err;
293 
294 	mutex_lock(&pfvf->mbox.lock);
295 	req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
296 	if (!req) {
297 		mutex_unlock(&pfvf->mbox.lock);
298 		return -ENOMEM;
299 	}
300 	req->mcam_index = -1; /* Default or reserved index */
301 	req->flowkey_cfg = rss->flowkey_cfg;
302 	req->group = DEFAULT_RSS_CONTEXT_GROUP;
303 
304 	err = otx2_sync_mbox_msg(&pfvf->mbox);
305 	if (err)
306 		goto fail;
307 
308 	rsp = (struct nix_rss_flowkey_cfg_rsp *)
309 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
310 	if (IS_ERR(rsp)) {
311 		err = PTR_ERR(rsp);
312 		goto fail;
313 	}
314 
315 	pfvf->hw.flowkey_alg_idx = rsp->alg_idx;
316 fail:
317 	mutex_unlock(&pfvf->mbox.lock);
318 	return err;
319 }
320 
321 int otx2_set_rss_table(struct otx2_nic *pfvf, int ctx_id)
322 {
323 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
324 	const int index = rss->rss_size * ctx_id;
325 	struct mbox *mbox = &pfvf->mbox;
326 	struct otx2_rss_ctx *rss_ctx;
327 	struct nix_aq_enq_req *aq;
328 	int idx, err;
329 
330 	mutex_lock(&mbox->lock);
331 	rss_ctx = rss->rss_ctx[ctx_id];
332 	/* Get memory to put this msg */
333 	for (idx = 0; idx < rss->rss_size; idx++) {
334 		/* Ignore the queue if AF_XDP zero copy is enabled */
335 		if (test_bit(rss_ctx->ind_tbl[idx], pfvf->af_xdp_zc_qidx))
336 			continue;
337 
338 		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
339 		if (!aq) {
340 			/* The shared memory buffer can be full.
341 			 * Flush it and retry
342 			 */
343 			err = otx2_sync_mbox_msg(mbox);
344 			if (err) {
345 				mutex_unlock(&mbox->lock);
346 				return err;
347 			}
348 			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
349 			if (!aq) {
350 				mutex_unlock(&mbox->lock);
351 				return -ENOMEM;
352 			}
353 		}
354 
355 		aq->rss.rq = rss_ctx->ind_tbl[idx];
356 
357 		/* Fill AQ info */
358 		aq->qidx = index + idx;
359 		aq->ctype = NIX_AQ_CTYPE_RSS;
360 		aq->op = NIX_AQ_INSTOP_INIT;
361 	}
362 	err = otx2_sync_mbox_msg(mbox);
363 	mutex_unlock(&mbox->lock);
364 	return err;
365 }
366 
367 void otx2_set_rss_key(struct otx2_nic *pfvf)
368 {
369 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
370 	u64 *key = (u64 *)&rss->key[4];
371 	int idx;
372 
373 	/* 352bit or 44byte key needs to be configured as below
374 	 * NIX_LF_RX_SECRETX0 = key<351:288>
375 	 * NIX_LF_RX_SECRETX1 = key<287:224>
376 	 * NIX_LF_RX_SECRETX2 = key<223:160>
377 	 * NIX_LF_RX_SECRETX3 = key<159:96>
378 	 * NIX_LF_RX_SECRETX4 = key<95:32>
379 	 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
380 	 */
381 	otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
382 		     (u64)(*((u32 *)&rss->key)) << 32);
383 	idx = sizeof(rss->key) / sizeof(u64);
384 	while (idx > 0) {
385 		idx--;
386 		otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
387 	}
388 }
389 
390 int otx2_rss_init(struct otx2_nic *pfvf)
391 {
392 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
393 	struct otx2_rss_ctx *rss_ctx;
394 	int idx, ret = 0;
395 
396 	rss->rss_size = sizeof(*rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP]);
397 
398 	/* Init RSS key if it is not setup already */
399 	if (!rss->enable)
400 		netdev_rss_key_fill(rss->key, sizeof(rss->key));
401 	otx2_set_rss_key(pfvf);
402 
403 	if (!netif_is_rxfh_configured(pfvf->netdev)) {
404 		/* Set RSS group 0 as default indirection table */
405 		rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP] = kzalloc(rss->rss_size,
406 								  GFP_KERNEL);
407 		if (!rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP])
408 			return -ENOMEM;
409 
410 		rss_ctx = rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP];
411 		for (idx = 0; idx < rss->rss_size; idx++)
412 			rss_ctx->ind_tbl[idx] =
413 				ethtool_rxfh_indir_default(idx,
414 							   pfvf->hw.rx_queues);
415 	}
416 	ret = otx2_set_rss_table(pfvf, DEFAULT_RSS_CONTEXT_GROUP);
417 	if (ret)
418 		return ret;
419 
420 	/* Flowkey or hash config to be used for generating flow tag */
421 	rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
422 			   NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
423 			   NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
424 			   NIX_FLOW_KEY_TYPE_SCTP | NIX_FLOW_KEY_TYPE_VLAN |
425 			   NIX_FLOW_KEY_TYPE_IPV4_PROTO;
426 
427 	ret = otx2_set_flowkey_cfg(pfvf);
428 	if (ret)
429 		return ret;
430 
431 	rss->enable = true;
432 	return 0;
433 }
434 
435 /* Setup UDP segmentation algorithm in HW */
436 static void otx2_setup_udp_segmentation(struct nix_lso_format_cfg *lso, bool v4)
437 {
438 	struct nix_lso_format *field;
439 
440 	field = (struct nix_lso_format *)&lso->fields[0];
441 	lso->field_mask = GENMASK(18, 0);
442 
443 	/* IP's Length field */
444 	field->layer = NIX_TXLAYER_OL3;
445 	/* In ipv4, length field is at offset 2 bytes, for ipv6 it's 4 */
446 	field->offset = v4 ? 2 : 4;
447 	field->sizem1 = 1; /* i.e 2 bytes */
448 	field->alg = NIX_LSOALG_ADD_PAYLEN;
449 	field++;
450 
451 	/* No ID field in IPv6 header */
452 	if (v4) {
453 		/* Increment IPID */
454 		field->layer = NIX_TXLAYER_OL3;
455 		field->offset = 4;
456 		field->sizem1 = 1; /* i.e 2 bytes */
457 		field->alg = NIX_LSOALG_ADD_SEGNUM;
458 		field++;
459 	}
460 
461 	/* Update length in UDP header */
462 	field->layer = NIX_TXLAYER_OL4;
463 	field->offset = 4;
464 	field->sizem1 = 1;
465 	field->alg = NIX_LSOALG_ADD_PAYLEN;
466 }
467 
468 /* Setup segmentation algorithms in HW and retrieve algorithm index */
469 void otx2_setup_segmentation(struct otx2_nic *pfvf)
470 {
471 	struct nix_lso_format_cfg_rsp *rsp;
472 	struct nix_lso_format_cfg *lso;
473 	struct otx2_hw *hw = &pfvf->hw;
474 	int err;
475 
476 	mutex_lock(&pfvf->mbox.lock);
477 
478 	/* UDPv4 segmentation */
479 	lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
480 	if (!lso)
481 		goto fail;
482 
483 	/* Setup UDP/IP header fields that HW should update per segment */
484 	otx2_setup_udp_segmentation(lso, true);
485 
486 	err = otx2_sync_mbox_msg(&pfvf->mbox);
487 	if (err)
488 		goto fail;
489 
490 	rsp = (struct nix_lso_format_cfg_rsp *)
491 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
492 	if (IS_ERR(rsp))
493 		goto fail;
494 
495 	hw->lso_udpv4_idx = rsp->lso_format_idx;
496 
497 	/* UDPv6 segmentation */
498 	lso = otx2_mbox_alloc_msg_nix_lso_format_cfg(&pfvf->mbox);
499 	if (!lso)
500 		goto fail;
501 
502 	/* Setup UDP/IP header fields that HW should update per segment */
503 	otx2_setup_udp_segmentation(lso, false);
504 
505 	err = otx2_sync_mbox_msg(&pfvf->mbox);
506 	if (err)
507 		goto fail;
508 
509 	rsp = (struct nix_lso_format_cfg_rsp *)
510 			otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &lso->hdr);
511 	if (IS_ERR(rsp))
512 		goto fail;
513 
514 	hw->lso_udpv6_idx = rsp->lso_format_idx;
515 	mutex_unlock(&pfvf->mbox.lock);
516 	return;
517 fail:
518 	mutex_unlock(&pfvf->mbox.lock);
519 	netdev_info(pfvf->netdev,
520 		    "Failed to get LSO index for UDP GSO offload, disabling\n");
521 	pfvf->netdev->hw_features &= ~NETIF_F_GSO_UDP_L4;
522 }
523 
524 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
525 {
526 	/* Configure CQE interrupt coalescing parameters
527 	 *
528 	 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
529 	 * set 1 less than cq_ecount_wait. And cq_time_wait is in
530 	 * usecs, convert that to 100ns count.
531 	 */
532 	otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
533 		     ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
534 		     ((u64)pfvf->hw.cq_qcount_wait << 32) |
535 		     (pfvf->hw.cq_ecount_wait - 1));
536 }
537 
538 static int otx2_alloc_pool_buf(struct otx2_nic *pfvf, struct otx2_pool *pool,
539 			       dma_addr_t *dma)
540 {
541 	unsigned int offset = 0;
542 	struct page *page;
543 	size_t sz;
544 
545 	sz = SKB_DATA_ALIGN(pool->rbsize);
546 	sz = ALIGN(sz, OTX2_ALIGN);
547 
548 	page = page_pool_alloc_frag(pool->page_pool, &offset, sz, GFP_ATOMIC);
549 	if (unlikely(!page))
550 		return -ENOMEM;
551 
552 	*dma = page_pool_get_dma_addr(page) + offset;
553 	return 0;
554 }
555 
556 static int __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
557 			     dma_addr_t *dma, int qidx, int idx)
558 {
559 	u8 *buf;
560 
561 	if (pool->xsk_pool)
562 		return otx2_xsk_pool_alloc_buf(pfvf, pool, dma, idx);
563 
564 	if (pool->page_pool)
565 		return otx2_alloc_pool_buf(pfvf, pool, dma);
566 
567 	buf = napi_alloc_frag_align(pool->rbsize, OTX2_ALIGN);
568 	if (unlikely(!buf))
569 		return -ENOMEM;
570 
571 	*dma = dma_map_single_attrs(pfvf->dev, buf, pool->rbsize,
572 				    DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
573 	if (unlikely(dma_mapping_error(pfvf->dev, *dma))) {
574 		page_frag_free(buf);
575 		return -ENOMEM;
576 	}
577 
578 	return 0;
579 }
580 
581 int otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
582 		    dma_addr_t *dma, int qidx, int idx)
583 {
584 	int ret;
585 
586 	local_bh_disable();
587 	ret = __otx2_alloc_rbuf(pfvf, pool, dma, qidx, idx);
588 	local_bh_enable();
589 	return ret;
590 }
591 
592 int otx2_alloc_buffer(struct otx2_nic *pfvf, struct otx2_cq_queue *cq,
593 		      dma_addr_t *dma)
594 {
595 	if (unlikely(__otx2_alloc_rbuf(pfvf, cq->rbpool, dma,
596 				       cq->cq_idx, cq->pool_ptrs - 1)))
597 		return -ENOMEM;
598 	return 0;
599 }
600 
601 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
602 {
603 	struct otx2_nic *pfvf = netdev_priv(netdev);
604 
605 	schedule_work(&pfvf->reset_task);
606 }
607 EXPORT_SYMBOL(otx2_tx_timeout);
608 
609 void otx2_get_mac_from_af(struct net_device *netdev)
610 {
611 	struct otx2_nic *pfvf = netdev_priv(netdev);
612 	int err;
613 
614 	err = otx2_hw_get_mac_addr(pfvf, netdev);
615 	if (err)
616 		dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
617 
618 	/* If AF doesn't provide a valid MAC, generate a random one */
619 	if (!is_valid_ether_addr(netdev->dev_addr))
620 		eth_hw_addr_random(netdev);
621 }
622 EXPORT_SYMBOL(otx2_get_mac_from_af);
623 
624 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl, int prio, bool txschq_for_pfc)
625 {
626 	u16 (*schq_list)[MAX_TXSCHQ_PER_FUNC];
627 	struct otx2_hw *hw = &pfvf->hw;
628 	struct nix_txschq_config *req;
629 	u64 schq, parent;
630 	u64 dwrr_val;
631 
632 	dwrr_val = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
633 
634 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
635 	if (!req)
636 		return -ENOMEM;
637 
638 	req->lvl = lvl;
639 	req->num_regs = 1;
640 
641 	schq_list = hw->txschq_list;
642 #ifdef CONFIG_DCB
643 	if (txschq_for_pfc)
644 		schq_list = pfvf->pfc_schq_list;
645 #endif
646 
647 	schq = schq_list[lvl][prio];
648 	/* Set topology e.t.c configuration */
649 	if (lvl == NIX_TXSCH_LVL_SMQ) {
650 		req->reg[0] = NIX_AF_SMQX_CFG(schq);
651 		req->regval[0] = ((u64)pfvf->tx_max_pktlen << 8) | OTX2_MIN_MTU;
652 		req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
653 				  (0x2ULL << 36);
654 		/* Set link type for DWRR MTU selection on CN10K silicons */
655 		if (!is_dev_otx2(pfvf->pdev))
656 			req->regval[0] |= FIELD_PREP(GENMASK_ULL(58, 57),
657 						(u64)hw->smq_link_type);
658 		req->num_regs++;
659 		/* MDQ config */
660 		parent = schq_list[NIX_TXSCH_LVL_TL4][prio];
661 		req->reg[1] = NIX_AF_MDQX_PARENT(schq);
662 		req->regval[1] = parent << 16;
663 		req->num_regs++;
664 		/* Set DWRR quantum */
665 		req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
666 		req->regval[2] =  dwrr_val;
667 	} else if (lvl == NIX_TXSCH_LVL_TL4) {
668 		int sdp_chan =  hw->tx_chan_base + prio;
669 
670 		if (is_otx2_sdp_rep(pfvf->pdev))
671 			prio = 0;
672 		parent = schq_list[NIX_TXSCH_LVL_TL3][prio];
673 		req->reg[0] = NIX_AF_TL4X_PARENT(schq);
674 		req->regval[0] = (u64)parent << 16;
675 		req->num_regs++;
676 		req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
677 		req->regval[1] = dwrr_val;
678 		if (is_otx2_sdp_rep(pfvf->pdev)) {
679 			req->num_regs++;
680 			req->reg[2] = NIX_AF_TL4X_SDP_LINK_CFG(schq);
681 			req->regval[2] = BIT_ULL(12) | BIT_ULL(13) |
682 					 (sdp_chan & 0xff);
683 		}
684 	} else if (lvl == NIX_TXSCH_LVL_TL3) {
685 		parent = schq_list[NIX_TXSCH_LVL_TL2][prio];
686 		req->reg[0] = NIX_AF_TL3X_PARENT(schq);
687 		req->regval[0] = (u64)parent << 16;
688 		req->num_regs++;
689 		req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
690 		req->regval[1] = dwrr_val;
691 		if (lvl == hw->txschq_link_cfg_lvl &&
692 		    !is_otx2_sdp_rep(pfvf->pdev)) {
693 			req->num_regs++;
694 			req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
695 			/* Enable this queue and backpressure
696 			 * and set relative channel
697 			 */
698 			req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
699 		}
700 	} else if (lvl == NIX_TXSCH_LVL_TL2) {
701 		parent = schq_list[NIX_TXSCH_LVL_TL1][prio];
702 		req->reg[0] = NIX_AF_TL2X_PARENT(schq);
703 		req->regval[0] = (u64)parent << 16;
704 
705 		req->num_regs++;
706 		req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
707 		req->regval[1] = (u64)hw->txschq_aggr_lvl_rr_prio << 24 | dwrr_val;
708 
709 		if (lvl == hw->txschq_link_cfg_lvl &&
710 		    !is_otx2_sdp_rep(pfvf->pdev)) {
711 			req->num_regs++;
712 			req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq, hw->tx_link);
713 			/* Enable this queue and backpressure
714 			 * and set relative channel
715 			 */
716 			req->regval[2] = BIT_ULL(13) | BIT_ULL(12) | prio;
717 		}
718 	} else if (lvl == NIX_TXSCH_LVL_TL1) {
719 		/* Default config for TL1.
720 		 * For VF this is always ignored.
721 		 */
722 
723 		/* On CN10K, if RR_WEIGHT is greater than 16384, HW will
724 		 * clip it to 16384, so configuring a 24bit max value
725 		 * will work on both OTx2 and CN10K.
726 		 */
727 		req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
728 		req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
729 
730 		req->num_regs++;
731 		req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
732 		req->regval[1] = hw->txschq_aggr_lvl_rr_prio << 1;
733 
734 		req->num_regs++;
735 		req->reg[2] = NIX_AF_TL1X_CIR(schq);
736 		req->regval[2] = 0;
737 	}
738 
739 	return otx2_sync_mbox_msg(&pfvf->mbox);
740 }
741 EXPORT_SYMBOL(otx2_txschq_config);
742 
743 int otx2_smq_flush(struct otx2_nic *pfvf, int smq)
744 {
745 	struct nix_txschq_config *req;
746 	int rc;
747 
748 	mutex_lock(&pfvf->mbox.lock);
749 
750 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
751 	if (!req) {
752 		mutex_unlock(&pfvf->mbox.lock);
753 		return -ENOMEM;
754 	}
755 
756 	req->lvl = NIX_TXSCH_LVL_SMQ;
757 	req->reg[0] = NIX_AF_SMQX_CFG(smq);
758 	req->regval[0] |= BIT_ULL(49);
759 	req->num_regs++;
760 
761 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
762 	mutex_unlock(&pfvf->mbox.lock);
763 	return rc;
764 }
765 EXPORT_SYMBOL(otx2_smq_flush);
766 
767 int otx2_txsch_alloc(struct otx2_nic *pfvf)
768 {
769 	int chan_cnt = pfvf->hw.tx_chan_cnt;
770 	struct nix_txsch_alloc_req *req;
771 	struct nix_txsch_alloc_rsp *rsp;
772 	int lvl, schq, rc;
773 
774 	/* Get memory to put this msg */
775 	req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
776 	if (!req)
777 		return -ENOMEM;
778 
779 	/* Request one schq per level */
780 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
781 		req->schq[lvl] = 1;
782 
783 	if (is_otx2_sdp_rep(pfvf->pdev) && chan_cnt > 1) {
784 		req->schq[NIX_TXSCH_LVL_SMQ] = chan_cnt;
785 		req->schq[NIX_TXSCH_LVL_TL4] = chan_cnt;
786 	}
787 
788 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
789 	if (rc)
790 		return rc;
791 
792 	rsp = (struct nix_txsch_alloc_rsp *)
793 	      otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
794 	if (IS_ERR(rsp))
795 		return PTR_ERR(rsp);
796 
797 	/* Setup transmit scheduler list */
798 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
799 		pfvf->hw.txschq_cnt[lvl] = rsp->schq[lvl];
800 		for (schq = 0; schq < rsp->schq[lvl]; schq++)
801 			pfvf->hw.txschq_list[lvl][schq] =
802 				rsp->schq_list[lvl][schq];
803 	}
804 
805 	pfvf->hw.txschq_link_cfg_lvl = rsp->link_cfg_lvl;
806 	pfvf->hw.txschq_aggr_lvl_rr_prio = rsp->aggr_lvl_rr_prio;
807 
808 	return 0;
809 }
810 
811 void otx2_txschq_free_one(struct otx2_nic *pfvf, u16 lvl, u16 schq)
812 {
813 	struct nix_txsch_free_req *free_req;
814 	int err;
815 
816 	mutex_lock(&pfvf->mbox.lock);
817 
818 	free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
819 	if (!free_req) {
820 		mutex_unlock(&pfvf->mbox.lock);
821 		netdev_err(pfvf->netdev,
822 			   "Failed alloc txschq free req\n");
823 		return;
824 	}
825 
826 	free_req->schq_lvl = lvl;
827 	free_req->schq = schq;
828 
829 	err = otx2_sync_mbox_msg(&pfvf->mbox);
830 	if (err) {
831 		netdev_err(pfvf->netdev,
832 			   "Failed stop txschq %d at level %d\n", schq, lvl);
833 	}
834 
835 	mutex_unlock(&pfvf->mbox.lock);
836 }
837 EXPORT_SYMBOL(otx2_txschq_free_one);
838 
839 void otx2_txschq_stop(struct otx2_nic *pfvf)
840 {
841 	int lvl, schq, idx;
842 
843 	/* free non QOS TLx nodes */
844 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
845 		for (idx = 0; idx < pfvf->hw.txschq_cnt[lvl]; idx++) {
846 			otx2_txschq_free_one(pfvf, lvl,
847 					     pfvf->hw.txschq_list[lvl][idx]);
848 		}
849 	}
850 
851 	/* Clear the txschq list */
852 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
853 		for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
854 			pfvf->hw.txschq_list[lvl][schq] = 0;
855 	}
856 
857 }
858 
859 void otx2_sqb_flush(struct otx2_nic *pfvf)
860 {
861 	int qidx, sqe_tail, sqe_head;
862 	struct otx2_snd_queue *sq;
863 	void __iomem *ptr;
864 	u64 incr, val;
865 
866 	ptr = otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
867 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
868 		sq = &pfvf->qset.sq[qidx];
869 		if (!sq->sqb_ptrs)
870 			continue;
871 
872 		incr = (u64)qidx << 32;
873 		val = otx2_atomic64_add(incr, ptr);
874 		sqe_head = (val >> 20) & 0x3F;
875 		sqe_tail = (val >> 28) & 0x3F;
876 		if (sqe_head != sqe_tail)
877 			usleep_range(50, 60);
878 	}
879 }
880 
881 /* RED and drop levels of CQ on packet reception.
882  * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
883  */
884 #define RQ_PASS_LVL_CQ(skid, qsize)	((((skid) + 16) * 256) / (qsize))
885 #define RQ_DROP_LVL_CQ(skid, qsize)	(((skid) * 256) / (qsize))
886 
887 /* RED and drop levels of AURA for packet reception.
888  * For AURA level is measure of fullness (0x0 = empty, 255 = full).
889  * Eg: For RQ length 1K, for pass/drop level 204/230.
890  * RED accepts pkts if free pointers > 102 & <= 205.
891  * Drops pkts if free pointers < 102.
892  */
893 #define RQ_BP_LVL_AURA   (255 - ((85 * 256) / 100)) /* BP when 85% is full */
894 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
895 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
896 
897 int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
898 {
899 	struct otx2_qset *qset = &pfvf->qset;
900 	struct nix_aq_enq_req *aq;
901 
902 	/* Get memory to put this msg */
903 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
904 	if (!aq)
905 		return -ENOMEM;
906 
907 	aq->rq.cq = qidx;
908 	aq->rq.ena = 1;
909 	aq->rq.pb_caching = 1;
910 	aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
911 	aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
912 	aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
913 	aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
914 	aq->rq.qint_idx = 0;
915 	aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
916 	aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
917 	aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
918 	aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
919 	aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
920 	aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
921 
922 	/* Fill AQ info */
923 	aq->qidx = qidx;
924 	aq->ctype = NIX_AQ_CTYPE_RQ;
925 	aq->op = NIX_AQ_INSTOP_INIT;
926 
927 	return otx2_sync_mbox_msg(&pfvf->mbox);
928 }
929 
930 int otx2_sq_aq_init(void *dev, u16 qidx, u8 chan_offset, u16 sqb_aura)
931 {
932 	struct otx2_nic *pfvf = dev;
933 	struct otx2_snd_queue *sq;
934 	struct nix_aq_enq_req *aq;
935 
936 	sq = &pfvf->qset.sq[qidx];
937 	sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
938 	/* Get memory to put this msg */
939 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
940 	if (!aq)
941 		return -ENOMEM;
942 
943 	aq->sq.cq = pfvf->hw.rx_queues + qidx;
944 	aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
945 	aq->sq.cq_ena = 1;
946 	aq->sq.ena = 1;
947 	aq->sq.smq = otx2_get_smq_idx(pfvf, qidx);
948 	aq->sq.smq_rr_quantum = mtu_to_dwrr_weight(pfvf, pfvf->tx_max_pktlen);
949 	aq->sq.default_chan = pfvf->hw.tx_chan_base + chan_offset;
950 	aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
951 	aq->sq.sqb_aura = sqb_aura;
952 	aq->sq.sq_int_ena = NIX_SQINT_BITS;
953 	aq->sq.qint_idx = 0;
954 	/* Due pipelining impact minimum 2000 unused SQ CQE's
955 	 * need to maintain to avoid CQ overflow.
956 	 */
957 	aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (pfvf->qset.sqe_cnt));
958 
959 	/* Fill AQ info */
960 	aq->qidx = qidx;
961 	aq->ctype = NIX_AQ_CTYPE_SQ;
962 	aq->op = NIX_AQ_INSTOP_INIT;
963 
964 	return otx2_sync_mbox_msg(&pfvf->mbox);
965 }
966 
967 int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
968 {
969 	struct otx2_qset *qset = &pfvf->qset;
970 	struct otx2_snd_queue *sq;
971 	struct otx2_pool *pool;
972 	u8 chan_offset;
973 	int err;
974 
975 	pool = &pfvf->qset.pool[sqb_aura];
976 	sq = &qset->sq[qidx];
977 	sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
978 	sq->sqe_cnt = qset->sqe_cnt;
979 
980 	err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
981 	if (err)
982 		return err;
983 
984 	/* Allocate memory for NIX SQE (which includes NIX SG) and CPT SG.
985 	 * SG of NIX and CPT are same in size. Allocate memory for CPT SG
986 	 * same as NIX SQE for base address alignment.
987 	 * Layout of a NIX SQE and CPT SG entry:
988 	 *      -----------------------------
989 	 *     |     CPT Scatter Gather      |
990 	 *     |       (SQE SIZE)            |
991 	 *     |                             |
992 	 *      -----------------------------
993 	 *     |       NIX SQE               |
994 	 *     |       (SQE SIZE)            |
995 	 *     |                             |
996 	 *      -----------------------------
997 	 */
998 	err = qmem_alloc(pfvf->dev, &sq->sqe_ring, qset->sqe_cnt,
999 			 sq->sqe_size * 2);
1000 	if (err)
1001 		return err;
1002 
1003 	err = qmem_alloc(pfvf->dev, &sq->cpt_resp, qset->sqe_cnt, 64);
1004 	if (err)
1005 		return err;
1006 
1007 	if (qidx < pfvf->hw.tx_queues) {
1008 		err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
1009 				 TSO_HEADER_SIZE);
1010 		if (err)
1011 			return err;
1012 	}
1013 
1014 	sq->sqe_base = sq->sqe->base;
1015 	sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
1016 	if (!sq->sg)
1017 		return -ENOMEM;
1018 
1019 	if (pfvf->ptp && qidx < pfvf->hw.tx_queues) {
1020 		err = qmem_alloc(pfvf->dev, &sq->timestamps, qset->sqe_cnt,
1021 				 sizeof(*sq->timestamps));
1022 		if (err) {
1023 			kfree(sq->sg);
1024 			sq->sg = NULL;
1025 			return err;
1026 		}
1027 	}
1028 
1029 	sq->head = 0;
1030 	sq->cons_head = 0;
1031 	sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
1032 	sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
1033 	/* Set SQE threshold to 10% of total SQEs */
1034 	sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
1035 	sq->aura_id = sqb_aura;
1036 	sq->aura_fc_addr = pool->fc_addr->base;
1037 	sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
1038 
1039 	sq->stats.bytes = 0;
1040 	sq->stats.pkts = 0;
1041 	/* Attach XSK_BUFF_POOL to XDP queue */
1042 	if (qidx > pfvf->hw.xdp_queues)
1043 		otx2_attach_xsk_buff(pfvf, sq, (qidx - pfvf->hw.xdp_queues));
1044 
1045 
1046 	chan_offset = qidx % pfvf->hw.tx_chan_cnt;
1047 	err = pfvf->hw_ops->sq_aq_init(pfvf, qidx, chan_offset, sqb_aura);
1048 	if (err) {
1049 		kfree(sq->sg);
1050 		sq->sg = NULL;
1051 		return err;
1052 	}
1053 
1054 	return 0;
1055 
1056 }
1057 
1058 int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
1059 {
1060 	struct otx2_qset *qset = &pfvf->qset;
1061 	int err, pool_id, non_xdp_queues;
1062 	struct nix_aq_enq_req *aq;
1063 	struct otx2_cq_queue *cq;
1064 	struct otx2_pool *pool;
1065 
1066 	cq = &qset->cq[qidx];
1067 	cq->cq_idx = qidx;
1068 	non_xdp_queues = pfvf->hw.rx_queues + pfvf->hw.tx_queues;
1069 	if (qidx < pfvf->hw.rx_queues) {
1070 		cq->cq_type = CQ_RX;
1071 		cq->cint_idx = qidx;
1072 		cq->cqe_cnt = qset->rqe_cnt;
1073 		if (pfvf->xdp_prog) {
1074 			xdp_rxq_info_reg(&cq->xdp_rxq, pfvf->netdev, qidx, 0);
1075 			pool = &qset->pool[qidx];
1076 			if (pool->xsk_pool) {
1077 				xdp_rxq_info_reg_mem_model(&cq->xdp_rxq,
1078 							   MEM_TYPE_XSK_BUFF_POOL,
1079 							   NULL);
1080 				xsk_pool_set_rxq_info(pool->xsk_pool, &cq->xdp_rxq);
1081 			} else if (pool->page_pool) {
1082 				xdp_rxq_info_reg_mem_model(&cq->xdp_rxq,
1083 							   MEM_TYPE_PAGE_POOL,
1084 							   pool->page_pool);
1085 			}
1086 		}
1087 	} else if (qidx < non_xdp_queues) {
1088 		cq->cq_type = CQ_TX;
1089 		cq->cint_idx = qidx - pfvf->hw.rx_queues;
1090 		cq->cqe_cnt = qset->sqe_cnt;
1091 	} else {
1092 		if (pfvf->hw.xdp_queues &&
1093 		    qidx < non_xdp_queues + pfvf->hw.xdp_queues) {
1094 			cq->cq_type = CQ_XDP;
1095 			cq->cint_idx = qidx - non_xdp_queues;
1096 			cq->cqe_cnt = qset->sqe_cnt;
1097 		} else {
1098 			cq->cq_type = CQ_QOS;
1099 			cq->cint_idx = qidx - non_xdp_queues -
1100 				       pfvf->hw.xdp_queues;
1101 			cq->cqe_cnt = qset->sqe_cnt;
1102 		}
1103 	}
1104 	cq->cqe_size = pfvf->qset.xqe_size;
1105 
1106 	/* Allocate memory for CQEs */
1107 	err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
1108 	if (err)
1109 		return err;
1110 
1111 	/* Save CQE CPU base for faster reference */
1112 	cq->cqe_base = cq->cqe->base;
1113 	/* In case where all RQs auras point to single pool,
1114 	 * all CQs receive buffer pool also point to same pool.
1115 	 */
1116 	pool_id = ((cq->cq_type == CQ_RX) &&
1117 		   (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
1118 	cq->rbpool = &qset->pool[pool_id];
1119 	cq->refill_task_sched = false;
1120 
1121 	/* Get memory to put this msg */
1122 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
1123 	if (!aq)
1124 		return -ENOMEM;
1125 
1126 	aq->cq.ena = 1;
1127 	aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
1128 	aq->cq.caching = 1;
1129 	aq->cq.base = cq->cqe->iova;
1130 	aq->cq.cint_idx = cq->cint_idx;
1131 	aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
1132 	aq->cq.qint_idx = 0;
1133 	aq->cq.avg_level = 255;
1134 
1135 	if (qidx < pfvf->hw.rx_queues) {
1136 		aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
1137 		aq->cq.drop_ena = 1;
1138 
1139 		if (!is_otx2_lbkvf(pfvf->pdev)) {
1140 			/* Enable receive CQ backpressure */
1141 			aq->cq.bp_ena = 1;
1142 #ifdef CONFIG_DCB
1143 			aq->cq.bpid = pfvf->bpid[pfvf->queue_to_pfc_map[qidx]];
1144 #else
1145 			aq->cq.bpid = pfvf->bpid[0];
1146 #endif
1147 
1148 			/* Set backpressure level is same as cq pass level */
1149 			aq->cq.bp = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
1150 		}
1151 	}
1152 
1153 	/* Fill AQ info */
1154 	aq->qidx = qidx;
1155 	aq->ctype = NIX_AQ_CTYPE_CQ;
1156 	aq->op = NIX_AQ_INSTOP_INIT;
1157 
1158 	return otx2_sync_mbox_msg(&pfvf->mbox);
1159 }
1160 
1161 static void otx2_pool_refill_task(struct work_struct *work)
1162 {
1163 	struct otx2_cq_queue *cq;
1164 	struct refill_work *wrk;
1165 	struct otx2_nic *pfvf;
1166 	int qidx;
1167 
1168 	wrk = container_of(work, struct refill_work, pool_refill_work.work);
1169 	pfvf = wrk->pf;
1170 	qidx = wrk - pfvf->refill_wrk;
1171 	cq = &pfvf->qset.cq[qidx];
1172 
1173 	cq->refill_task_sched = false;
1174 
1175 	local_bh_disable();
1176 	napi_schedule(wrk->napi);
1177 	local_bh_enable();
1178 }
1179 
1180 int otx2_config_nix_queues(struct otx2_nic *pfvf)
1181 {
1182 	int qidx, err;
1183 
1184 	/* Initialize RX queues */
1185 	for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
1186 		u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
1187 
1188 		err = otx2_rq_init(pfvf, qidx, lpb_aura);
1189 		if (err)
1190 			return err;
1191 	}
1192 
1193 	/* Initialize TX queues */
1194 	for (qidx = 0; qidx < pfvf->hw.non_qos_queues; qidx++) {
1195 		u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1196 
1197 		err = otx2_sq_init(pfvf, qidx, sqb_aura);
1198 		if (err)
1199 			return err;
1200 	}
1201 
1202 	/* Initialize completion queues */
1203 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1204 		err = otx2_cq_init(pfvf, qidx);
1205 		if (err)
1206 			return err;
1207 	}
1208 
1209 	pfvf->cq_op_addr = (__force u64 *)otx2_get_regaddr(pfvf,
1210 							   NIX_LF_CQ_OP_STATUS);
1211 
1212 	/* Initialize work queue for receive buffer refill */
1213 	pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
1214 					sizeof(struct refill_work), GFP_KERNEL);
1215 	if (!pfvf->refill_wrk)
1216 		return -ENOMEM;
1217 
1218 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
1219 		pfvf->refill_wrk[qidx].pf = pfvf;
1220 		INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
1221 				  otx2_pool_refill_task);
1222 	}
1223 	return 0;
1224 }
1225 
1226 int otx2_config_nix(struct otx2_nic *pfvf)
1227 {
1228 	struct nix_lf_alloc_req  *nixlf;
1229 	struct nix_lf_alloc_rsp *rsp;
1230 	int err;
1231 
1232 	pfvf->qset.xqe_size = pfvf->hw.xqe_size;
1233 
1234 	/* Get memory to put this msg */
1235 	nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
1236 	if (!nixlf)
1237 		return -ENOMEM;
1238 
1239 	/* Set RQ/SQ/CQ counts */
1240 	nixlf->rq_cnt = pfvf->hw.rx_queues;
1241 	nixlf->sq_cnt = otx2_get_total_tx_queues(pfvf);
1242 	nixlf->cq_cnt = pfvf->qset.cq_cnt;
1243 	nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
1244 	nixlf->rss_grps = MAX_RSS_GROUPS;
1245 	nixlf->xqe_sz = pfvf->hw.xqe_size == 128 ? NIX_XQESZ_W16 : NIX_XQESZ_W64;
1246 	/* We don't know absolute NPA LF idx attached.
1247 	 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
1248 	 * NPA LF attached to this RVU PF/VF.
1249 	 */
1250 	nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
1251 	/* Disable alignment pad, enable L2 length check,
1252 	 * enable L4 TCP/UDP checksum verification.
1253 	 */
1254 	nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
1255 
1256 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1257 	if (err)
1258 		return err;
1259 
1260 	rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
1261 							   &nixlf->hdr);
1262 	if (IS_ERR(rsp))
1263 		return PTR_ERR(rsp);
1264 
1265 	if (rsp->qints < 1)
1266 		return -ENXIO;
1267 
1268 	return rsp->hdr.rc;
1269 }
1270 
1271 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
1272 {
1273 	struct otx2_qset *qset = &pfvf->qset;
1274 	struct otx2_hw *hw = &pfvf->hw;
1275 	struct otx2_snd_queue *sq;
1276 	int sqb, qidx;
1277 	u64 iova, pa;
1278 
1279 	for (qidx = 0; qidx < otx2_get_total_tx_queues(pfvf); qidx++) {
1280 		sq = &qset->sq[qidx];
1281 		if (!sq->sqb_ptrs)
1282 			continue;
1283 		for (sqb = 0; sqb < sq->sqb_count; sqb++) {
1284 			if (!sq->sqb_ptrs[sqb])
1285 				continue;
1286 			iova = sq->sqb_ptrs[sqb];
1287 			pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1288 			dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
1289 					     DMA_FROM_DEVICE,
1290 					     DMA_ATTR_SKIP_CPU_SYNC);
1291 			put_page(virt_to_page(phys_to_virt(pa)));
1292 		}
1293 		sq->sqb_count = 0;
1294 	}
1295 }
1296 
1297 void otx2_free_bufs(struct otx2_nic *pfvf, struct otx2_pool *pool,
1298 		    u64 iova, int size)
1299 {
1300 	struct page *page;
1301 	u64 pa;
1302 
1303 	pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
1304 	page = virt_to_head_page(phys_to_virt(pa));
1305 	if (pool->page_pool) {
1306 		page_pool_put_full_page(pool->page_pool, page, true);
1307 	} else if (pool->xsk_pool) {
1308 		/* Note: No way of identifying xdp_buff */
1309 	} else {
1310 		dma_unmap_page_attrs(pfvf->dev, iova, size,
1311 				     DMA_FROM_DEVICE,
1312 				     DMA_ATTR_SKIP_CPU_SYNC);
1313 
1314 		put_page(page);
1315 	}
1316 }
1317 
1318 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
1319 {
1320 	int pool_id, pool_start = 0, pool_end = 0, size = 0;
1321 	struct otx2_pool *pool;
1322 	u64 iova;
1323 	int idx;
1324 
1325 	if (type == AURA_NIX_SQ) {
1326 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
1327 		pool_end =  pool_start + pfvf->hw.sqpool_cnt;
1328 		size = pfvf->hw.sqb_size;
1329 	}
1330 	if (type == AURA_NIX_RQ) {
1331 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
1332 		pool_end = pfvf->hw.rqpool_cnt;
1333 		size = pfvf->rbsize;
1334 	}
1335 
1336 	/* Free SQB and RQB pointers from the aura pool */
1337 	for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
1338 		pool = &pfvf->qset.pool[pool_id];
1339 		iova = otx2_aura_allocptr(pfvf, pool_id);
1340 		while (iova) {
1341 			if (type == AURA_NIX_RQ)
1342 				iova -= OTX2_HEAD_ROOM;
1343 			otx2_free_bufs(pfvf, pool, iova, size);
1344 			iova = otx2_aura_allocptr(pfvf, pool_id);
1345 		}
1346 
1347 		for (idx = 0 ; idx < pool->xdp_cnt; idx++) {
1348 			if (!pool->xdp[idx])
1349 				continue;
1350 
1351 			xsk_buff_free(pool->xdp[idx]);
1352 		}
1353 	}
1354 }
1355 
1356 void otx2_aura_pool_free(struct otx2_nic *pfvf)
1357 {
1358 	struct otx2_pool *pool;
1359 	int pool_id;
1360 
1361 	if (!pfvf->qset.pool)
1362 		return;
1363 
1364 	for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
1365 		pool = &pfvf->qset.pool[pool_id];
1366 		qmem_free(pfvf->dev, pool->stack);
1367 		qmem_free(pfvf->dev, pool->fc_addr);
1368 		page_pool_destroy(pool->page_pool);
1369 		devm_kfree(pfvf->dev, pool->xdp);
1370 		pool->xsk_pool = NULL;
1371 	}
1372 	devm_kfree(pfvf->dev, pfvf->qset.pool);
1373 	pfvf->qset.pool = NULL;
1374 }
1375 
1376 int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
1377 		   int pool_id, int numptrs)
1378 {
1379 	struct npa_aq_enq_req *aq;
1380 	struct otx2_pool *pool;
1381 	int err;
1382 
1383 	pool = &pfvf->qset.pool[pool_id];
1384 
1385 	/* Allocate memory for HW to update Aura count.
1386 	 * Alloc one cache line, so that it fits all FC_STYPE modes.
1387 	 */
1388 	if (!pool->fc_addr) {
1389 		err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
1390 		if (err)
1391 			return err;
1392 	}
1393 
1394 	/* Initialize this aura's context via AF */
1395 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1396 	if (!aq) {
1397 		/* Shared mbox memory buffer is full, flush it and retry */
1398 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1399 		if (err)
1400 			return err;
1401 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1402 		if (!aq)
1403 			return -ENOMEM;
1404 	}
1405 
1406 	aq->aura_id = aura_id;
1407 	/* Will be filled by AF with correct pool context address */
1408 	aq->aura.pool_addr = pool_id;
1409 	aq->aura.pool_caching = 1;
1410 	aq->aura.shift = ilog2(numptrs) - 8;
1411 	aq->aura.count = numptrs;
1412 	aq->aura.limit = numptrs;
1413 	aq->aura.avg_level = 255;
1414 	aq->aura.ena = 1;
1415 	aq->aura.fc_ena = 1;
1416 	aq->aura.fc_addr = pool->fc_addr->iova;
1417 	aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1418 
1419 	/* Enable backpressure for RQ aura */
1420 	if (aura_id < pfvf->hw.rqpool_cnt && !is_otx2_lbkvf(pfvf->pdev)) {
1421 		aq->aura.bp_ena = 0;
1422 		/* If NIX1 LF is attached then specify NIX1_RX.
1423 		 *
1424 		 * Below NPA_AURA_S[BP_ENA] is set according to the
1425 		 * NPA_BPINTF_E enumeration given as:
1426 		 * 0x0 + a*0x1 where 'a' is 0 for NIX0_RX and 1 for NIX1_RX so
1427 		 * NIX0_RX is 0x0 + 0*0x1 = 0
1428 		 * NIX1_RX is 0x0 + 1*0x1 = 1
1429 		 * But in HRM it is given that
1430 		 * "NPA_AURA_S[BP_ENA](w1[33:32]) - Enable aura backpressure to
1431 		 * NIX-RX based on [BP] level. One bit per NIX-RX; index
1432 		 * enumerated by NPA_BPINTF_E."
1433 		 */
1434 		if (pfvf->nix_blkaddr == BLKADDR_NIX1)
1435 			aq->aura.bp_ena = 1;
1436 #ifdef CONFIG_DCB
1437 		aq->aura.nix0_bpid = pfvf->bpid[pfvf->queue_to_pfc_map[aura_id]];
1438 #else
1439 		aq->aura.nix0_bpid = pfvf->bpid[0];
1440 #endif
1441 
1442 		/* Set backpressure level for RQ's Aura */
1443 		aq->aura.bp = RQ_BP_LVL_AURA;
1444 	}
1445 
1446 	/* Fill AQ info */
1447 	aq->ctype = NPA_AQ_CTYPE_AURA;
1448 	aq->op = NPA_AQ_INSTOP_INIT;
1449 
1450 	return 0;
1451 }
1452 
1453 int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1454 		   int stack_pages, int numptrs, int buf_size, int type)
1455 {
1456 	struct page_pool_params pp_params = { 0 };
1457 	struct xsk_buff_pool *xsk_pool;
1458 	struct npa_aq_enq_req *aq;
1459 	struct otx2_pool *pool;
1460 	int err;
1461 
1462 	pool = &pfvf->qset.pool[pool_id];
1463 	/* Alloc memory for stack which is used to store buffer pointers */
1464 	err = qmem_alloc(pfvf->dev, &pool->stack,
1465 			 stack_pages, pfvf->hw.stack_pg_bytes);
1466 	if (err)
1467 		return err;
1468 
1469 	pool->rbsize = buf_size;
1470 
1471 	/* Initialize this pool's context via AF */
1472 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1473 	if (!aq) {
1474 		/* Shared mbox memory buffer is full, flush it and retry */
1475 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1476 		if (err) {
1477 			qmem_free(pfvf->dev, pool->stack);
1478 			return err;
1479 		}
1480 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1481 		if (!aq) {
1482 			qmem_free(pfvf->dev, pool->stack);
1483 			return -ENOMEM;
1484 		}
1485 	}
1486 
1487 	aq->aura_id = pool_id;
1488 	aq->pool.stack_base = pool->stack->iova;
1489 	aq->pool.stack_caching = 1;
1490 	aq->pool.ena = 1;
1491 	aq->pool.buf_size = buf_size / 128;
1492 	aq->pool.stack_max_pages = stack_pages;
1493 	aq->pool.shift = ilog2(numptrs) - 8;
1494 	aq->pool.ptr_start = 0;
1495 	aq->pool.ptr_end = ~0ULL;
1496 
1497 	/* Fill AQ info */
1498 	aq->ctype = NPA_AQ_CTYPE_POOL;
1499 	aq->op = NPA_AQ_INSTOP_INIT;
1500 
1501 	if (type != AURA_NIX_RQ)
1502 		return 0;
1503 
1504 	if (!test_bit(pool_id, pfvf->af_xdp_zc_qidx)) {
1505 		pp_params.order = get_order(buf_size);
1506 		pp_params.flags = PP_FLAG_DMA_MAP;
1507 		pp_params.pool_size = min(OTX2_PAGE_POOL_SZ, numptrs);
1508 		pp_params.nid = NUMA_NO_NODE;
1509 		pp_params.dev = pfvf->dev;
1510 		pp_params.dma_dir = DMA_FROM_DEVICE;
1511 		pool->page_pool = page_pool_create(&pp_params);
1512 		if (IS_ERR(pool->page_pool)) {
1513 			netdev_err(pfvf->netdev, "Creation of page pool failed\n");
1514 			return PTR_ERR(pool->page_pool);
1515 		}
1516 		return 0;
1517 	}
1518 
1519 	/* Set XSK pool to support AF_XDP zero-copy */
1520 	xsk_pool = xsk_get_pool_from_qid(pfvf->netdev, pool_id);
1521 	if (xsk_pool) {
1522 		pool->xsk_pool = xsk_pool;
1523 		pool->xdp_cnt = numptrs;
1524 		pool->xdp = devm_kcalloc(pfvf->dev,
1525 					 numptrs, sizeof(struct xdp_buff *), GFP_KERNEL);
1526 		if (IS_ERR(pool->xdp)) {
1527 			netdev_err(pfvf->netdev, "Creation of xsk pool failed\n");
1528 			return PTR_ERR(pool->xdp);
1529 		}
1530 	}
1531 
1532 	return 0;
1533 }
1534 
1535 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1536 {
1537 	int qidx, pool_id, stack_pages, num_sqbs;
1538 	struct otx2_qset *qset = &pfvf->qset;
1539 	struct otx2_hw *hw = &pfvf->hw;
1540 	struct otx2_snd_queue *sq;
1541 	struct otx2_pool *pool;
1542 	dma_addr_t bufptr;
1543 	int err, ptr;
1544 
1545 	/* Calculate number of SQBs needed.
1546 	 *
1547 	 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1548 	 * Last SQE is used for pointing to next SQB.
1549 	 */
1550 	num_sqbs = (hw->sqb_size / 128) - 1;
1551 	num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1552 
1553 	/* Get no of stack pages needed */
1554 	stack_pages =
1555 		(num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1556 
1557 	for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1558 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1559 		/* Initialize aura context */
1560 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1561 		if (err)
1562 			goto fail;
1563 
1564 		/* Initialize pool context */
1565 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1566 				     num_sqbs, hw->sqb_size, AURA_NIX_SQ);
1567 		if (err)
1568 			goto fail;
1569 	}
1570 
1571 	/* Flush accumulated messages */
1572 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1573 	if (err)
1574 		goto fail;
1575 
1576 	/* Allocate pointers and free them to aura/pool */
1577 	for (qidx = 0; qidx < hw->non_qos_queues; qidx++) {
1578 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1579 		pool = &pfvf->qset.pool[pool_id];
1580 
1581 		sq = &qset->sq[qidx];
1582 		sq->sqb_count = 0;
1583 		sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(*sq->sqb_ptrs), GFP_KERNEL);
1584 		if (!sq->sqb_ptrs) {
1585 			err = -ENOMEM;
1586 			goto err_mem;
1587 		}
1588 
1589 		for (ptr = 0; ptr < num_sqbs; ptr++) {
1590 			err = otx2_alloc_rbuf(pfvf, pool, &bufptr, pool_id, ptr);
1591 			if (err) {
1592 				if (pool->xsk_pool) {
1593 					ptr--;
1594 					while (ptr >= 0) {
1595 						xsk_buff_free(pool->xdp[ptr]);
1596 						ptr--;
1597 					}
1598 				}
1599 				goto err_mem;
1600 			}
1601 
1602 			pfvf->hw_ops->aura_freeptr(pfvf, pool_id, bufptr);
1603 			sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1604 		}
1605 	}
1606 
1607 err_mem:
1608 	return err ? -ENOMEM : 0;
1609 
1610 fail:
1611 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1612 	otx2_aura_pool_free(pfvf);
1613 	return err;
1614 }
1615 
1616 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1617 {
1618 	struct otx2_hw *hw = &pfvf->hw;
1619 	int stack_pages, pool_id, rq;
1620 	struct otx2_pool *pool;
1621 	int err, ptr, num_ptrs;
1622 	dma_addr_t bufptr;
1623 
1624 	num_ptrs = pfvf->qset.rqe_cnt;
1625 
1626 	stack_pages =
1627 		(num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1628 
1629 	for (rq = 0; rq < hw->rx_queues; rq++) {
1630 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1631 		/* Initialize aura context */
1632 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1633 		if (err)
1634 			goto fail;
1635 	}
1636 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1637 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1638 				     num_ptrs, pfvf->rbsize, AURA_NIX_RQ);
1639 		if (err)
1640 			goto fail;
1641 	}
1642 
1643 	/* Flush accumulated messages */
1644 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1645 	if (err)
1646 		goto fail;
1647 
1648 	/* Allocate pointers and free them to aura/pool */
1649 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1650 		pool = &pfvf->qset.pool[pool_id];
1651 
1652 		for (ptr = 0; ptr < num_ptrs; ptr++) {
1653 			err = otx2_alloc_rbuf(pfvf, pool, &bufptr, pool_id, ptr);
1654 			if (err) {
1655 				if (pool->xsk_pool) {
1656 					while (ptr)
1657 						xsk_buff_free(pool->xdp[--ptr]);
1658 				}
1659 				return -ENOMEM;
1660 			}
1661 
1662 			pfvf->hw_ops->aura_freeptr(pfvf, pool_id,
1663 						   pool->xsk_pool ? bufptr :
1664 						   bufptr + OTX2_HEAD_ROOM);
1665 		}
1666 	}
1667 	return 0;
1668 fail:
1669 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1670 	otx2_aura_pool_free(pfvf);
1671 	return err;
1672 }
1673 
1674 int otx2_config_npa(struct otx2_nic *pfvf)
1675 {
1676 	struct otx2_qset *qset = &pfvf->qset;
1677 	struct npa_lf_alloc_req  *npalf;
1678 	struct otx2_hw *hw = &pfvf->hw;
1679 	int aura_cnt;
1680 
1681 	/* Pool - Stack of free buffer pointers
1682 	 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1683 	 */
1684 
1685 	if (!hw->pool_cnt)
1686 		return -EINVAL;
1687 
1688 	qset->pool = devm_kcalloc(pfvf->dev, hw->pool_cnt,
1689 				  sizeof(struct otx2_pool), GFP_KERNEL);
1690 	if (!qset->pool)
1691 		return -ENOMEM;
1692 
1693 	/* Get memory to put this msg */
1694 	npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1695 	if (!npalf)
1696 		return -ENOMEM;
1697 
1698 	/* Set aura and pool counts */
1699 	npalf->nr_pools = hw->pool_cnt;
1700 	aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1701 	npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1702 
1703 	return otx2_sync_mbox_msg(&pfvf->mbox);
1704 }
1705 
1706 int otx2_detach_resources(struct mbox *mbox)
1707 {
1708 	struct rsrc_detach *detach;
1709 
1710 	mutex_lock(&mbox->lock);
1711 	detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1712 	if (!detach) {
1713 		mutex_unlock(&mbox->lock);
1714 		return -ENOMEM;
1715 	}
1716 
1717 	/* detach all */
1718 	detach->partial = false;
1719 
1720 	/* Send detach request to AF */
1721 	otx2_sync_mbox_msg(mbox);
1722 	mutex_unlock(&mbox->lock);
1723 	return 0;
1724 }
1725 EXPORT_SYMBOL(otx2_detach_resources);
1726 
1727 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1728 {
1729 	struct rsrc_attach *attach;
1730 	struct msg_req *msix;
1731 	int err;
1732 
1733 	mutex_lock(&pfvf->mbox.lock);
1734 	/* Get memory to put this msg */
1735 	attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1736 	if (!attach) {
1737 		mutex_unlock(&pfvf->mbox.lock);
1738 		return -ENOMEM;
1739 	}
1740 
1741 	attach->npalf = true;
1742 	attach->nixlf = true;
1743 
1744 	/* Send attach request to AF */
1745 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1746 	if (err) {
1747 		mutex_unlock(&pfvf->mbox.lock);
1748 		return err;
1749 	}
1750 
1751 	pfvf->nix_blkaddr = BLKADDR_NIX0;
1752 
1753 	/* If the platform has two NIX blocks then LF may be
1754 	 * allocated from NIX1.
1755 	 */
1756 	if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1757 		pfvf->nix_blkaddr = BLKADDR_NIX1;
1758 
1759 	/* Get NPA and NIX MSIX vector offsets */
1760 	msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1761 	if (!msix) {
1762 		mutex_unlock(&pfvf->mbox.lock);
1763 		return -ENOMEM;
1764 	}
1765 
1766 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1767 	if (err) {
1768 		mutex_unlock(&pfvf->mbox.lock);
1769 		return err;
1770 	}
1771 	mutex_unlock(&pfvf->mbox.lock);
1772 
1773 	if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1774 	    pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1775 		dev_err(pfvf->dev,
1776 			"RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1777 		return -EINVAL;
1778 	}
1779 
1780 	return 0;
1781 }
1782 EXPORT_SYMBOL(otx2_attach_npa_nix);
1783 
1784 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1785 {
1786 	struct hwctx_disable_req *req;
1787 
1788 	mutex_lock(&mbox->lock);
1789 	/* Request AQ to disable this context */
1790 	if (npa)
1791 		req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1792 	else
1793 		req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1794 
1795 	if (!req) {
1796 		mutex_unlock(&mbox->lock);
1797 		return;
1798 	}
1799 
1800 	req->ctype = type;
1801 
1802 	if (otx2_sync_mbox_msg(mbox))
1803 		dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1804 			__func__);
1805 
1806 	mutex_unlock(&mbox->lock);
1807 }
1808 
1809 int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
1810 {
1811 	struct nix_bp_cfg_req *req;
1812 
1813 	if (enable)
1814 		req = otx2_mbox_alloc_msg_nix_bp_enable(&pfvf->mbox);
1815 	else
1816 		req = otx2_mbox_alloc_msg_nix_bp_disable(&pfvf->mbox);
1817 
1818 	if (!req)
1819 		return -ENOMEM;
1820 
1821 	req->chan_base = 0;
1822 	if (otx2_is_pfc_enabled(pfvf)) {
1823 		req->chan_cnt = IEEE_8021QAZ_MAX_TCS;
1824 		req->bpid_per_chan = 1;
1825 	} else {
1826 		req->chan_cnt = pfvf->hw.rx_chan_cnt;
1827 		req->bpid_per_chan = 0;
1828 	}
1829 
1830 	return otx2_sync_mbox_msg(&pfvf->mbox);
1831 }
1832 EXPORT_SYMBOL(otx2_nix_config_bp);
1833 
1834 int otx2_nix_cpt_config_bp(struct otx2_nic *pfvf, bool enable)
1835 {
1836 	struct nix_bp_cfg_req *req;
1837 
1838 	if (enable)
1839 		req = otx2_mbox_alloc_msg_nix_cpt_bp_enable(&pfvf->mbox);
1840 	else
1841 		req = otx2_mbox_alloc_msg_nix_cpt_bp_disable(&pfvf->mbox);
1842 
1843 	if (!req)
1844 		return -ENOMEM;
1845 
1846 	req->chan_base = 0;
1847 	if (otx2_is_pfc_enabled(pfvf)) {
1848 		req->chan_cnt = IEEE_8021QAZ_MAX_TCS;
1849 		req->bpid_per_chan = 1;
1850 	} else {
1851 		req->chan_cnt = pfvf->hw.rx_chan_cnt;
1852 		req->bpid_per_chan = 0;
1853 	}
1854 
1855 	return otx2_sync_mbox_msg(&pfvf->mbox);
1856 }
1857 EXPORT_SYMBOL(otx2_nix_cpt_config_bp);
1858 
1859 /* Mbox message handlers */
1860 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1861 			    struct cgx_stats_rsp *rsp)
1862 {
1863 	int id;
1864 
1865 	for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1866 		pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1867 	for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1868 		pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1869 }
1870 
1871 void mbox_handler_cgx_fec_stats(struct otx2_nic *pfvf,
1872 				struct cgx_fec_stats_rsp *rsp)
1873 {
1874 	pfvf->hw.cgx_fec_corr_blks += rsp->fec_corr_blks;
1875 	pfvf->hw.cgx_fec_uncorr_blks += rsp->fec_uncorr_blks;
1876 }
1877 
1878 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1879 			       struct npa_lf_alloc_rsp *rsp)
1880 {
1881 	pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1882 	pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1883 }
1884 EXPORT_SYMBOL(mbox_handler_npa_lf_alloc);
1885 
1886 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1887 			       struct nix_lf_alloc_rsp *rsp)
1888 {
1889 	pfvf->hw.sqb_size = rsp->sqb_size;
1890 	pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1891 	pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1892 	pfvf->hw.rx_chan_cnt = rsp->rx_chan_cnt;
1893 	pfvf->hw.tx_chan_cnt = rsp->tx_chan_cnt;
1894 	pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1895 	pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1896 	pfvf->hw.cgx_links = rsp->cgx_links;
1897 	pfvf->hw.lbk_links = rsp->lbk_links;
1898 	pfvf->hw.tx_link = rsp->tx_link;
1899 }
1900 EXPORT_SYMBOL(mbox_handler_nix_lf_alloc);
1901 
1902 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1903 			      struct msix_offset_rsp *rsp)
1904 {
1905 	pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1906 	pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1907 }
1908 EXPORT_SYMBOL(mbox_handler_msix_offset);
1909 
1910 void mbox_handler_nix_bp_enable(struct otx2_nic *pfvf,
1911 				struct nix_bp_cfg_rsp *rsp)
1912 {
1913 	int chan, chan_id;
1914 
1915 	for (chan = 0; chan < rsp->chan_cnt; chan++) {
1916 		chan_id = ((rsp->chan_bpid[chan] >> 10) & 0x7F);
1917 		pfvf->bpid[chan_id] = rsp->chan_bpid[chan] & 0x3FF;
1918 	}
1919 }
1920 EXPORT_SYMBOL(mbox_handler_nix_bp_enable);
1921 
1922 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1923 {
1924 	struct otx2_qset *qset = &pfvf->qset;
1925 	struct otx2_hw *hw = &pfvf->hw;
1926 	int irq, qidx;
1927 
1928 	for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1929 	     qidx < n;
1930 	     qidx++, irq++) {
1931 		int vector = pci_irq_vector(pfvf->pdev, irq);
1932 
1933 		irq_set_affinity_hint(vector, NULL);
1934 		free_cpumask_var(hw->affinity_mask[irq]);
1935 		free_irq(vector, &qset->napi[qidx]);
1936 	}
1937 }
1938 EXPORT_SYMBOL(otx2_free_cints);
1939 
1940 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1941 {
1942 	struct otx2_hw *hw = &pfvf->hw;
1943 	int vec, cpu, irq, cint;
1944 
1945 	vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1946 	cpu = cpumask_first(cpu_online_mask);
1947 
1948 	/* CQ interrupts */
1949 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1950 		if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1951 			return;
1952 
1953 		cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1954 
1955 		irq = pci_irq_vector(pfvf->pdev, vec);
1956 		irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1957 
1958 		cpu = cpumask_next(cpu, cpu_online_mask);
1959 		if (unlikely(cpu >= nr_cpu_ids))
1960 			cpu = 0;
1961 	}
1962 }
1963 
1964 static u32 get_dwrr_mtu(struct otx2_nic *pfvf, struct nix_hw_info *hw)
1965 {
1966 	if (is_otx2_lbkvf(pfvf->pdev)) {
1967 		pfvf->hw.smq_link_type = SMQ_LINK_TYPE_LBK;
1968 		return hw->lbk_dwrr_mtu;
1969 	}
1970 
1971 	pfvf->hw.smq_link_type = SMQ_LINK_TYPE_RPM;
1972 	return hw->rpm_dwrr_mtu;
1973 }
1974 
1975 u16 otx2_get_max_mtu(struct otx2_nic *pfvf)
1976 {
1977 	struct nix_hw_info *rsp;
1978 	struct msg_req *req;
1979 	u16 max_mtu;
1980 	int rc;
1981 
1982 	mutex_lock(&pfvf->mbox.lock);
1983 
1984 	req = otx2_mbox_alloc_msg_nix_get_hw_info(&pfvf->mbox);
1985 	if (!req) {
1986 		rc =  -ENOMEM;
1987 		goto out;
1988 	}
1989 
1990 	rc = otx2_sync_mbox_msg(&pfvf->mbox);
1991 	if (!rc) {
1992 		rsp = (struct nix_hw_info *)
1993 		       otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
1994 		if (IS_ERR(rsp)) {
1995 			rc = PTR_ERR(rsp);
1996 			goto out;
1997 		}
1998 
1999 		/* HW counts VLAN insertion bytes (8 for double tag)
2000 		 * irrespective of whether SQE is requesting to insert VLAN
2001 		 * in the packet or not. Hence these 8 bytes have to be
2002 		 * discounted from max packet size otherwise HW will throw
2003 		 * SMQ errors
2004 		 */
2005 		max_mtu = rsp->max_mtu - 8 - OTX2_ETH_HLEN;
2006 
2007 		/* Also save DWRR MTU, needed for DWRR weight calculation */
2008 		pfvf->hw.dwrr_mtu = get_dwrr_mtu(pfvf, rsp);
2009 		if (!pfvf->hw.dwrr_mtu)
2010 			pfvf->hw.dwrr_mtu = 1;
2011 	}
2012 
2013 out:
2014 	mutex_unlock(&pfvf->mbox.lock);
2015 	if (rc) {
2016 		dev_warn(pfvf->dev,
2017 			 "Failed to get MTU from hardware setting default value(1500)\n");
2018 		max_mtu = 1500;
2019 	}
2020 	return max_mtu;
2021 }
2022 EXPORT_SYMBOL(otx2_get_max_mtu);
2023 
2024 int otx2_handle_ntuple_tc_features(struct net_device *netdev, netdev_features_t features)
2025 {
2026 	netdev_features_t changed = features ^ netdev->features;
2027 	struct otx2_nic *pfvf = netdev_priv(netdev);
2028 	bool ntuple = !!(features & NETIF_F_NTUPLE);
2029 	bool tc = !!(features & NETIF_F_HW_TC);
2030 
2031 	if ((changed & NETIF_F_NTUPLE) && !ntuple)
2032 		otx2_destroy_ntuple_flows(pfvf);
2033 
2034 	if ((changed & NETIF_F_NTUPLE) && ntuple) {
2035 		if (!pfvf->flow_cfg->max_flows) {
2036 			netdev_err(netdev,
2037 				   "Can't enable NTUPLE, MCAM entries not allocated\n");
2038 			return -EINVAL;
2039 		}
2040 	}
2041 
2042 	if ((changed & NETIF_F_HW_TC) && !tc &&
2043 	    otx2_tc_flower_rule_cnt(pfvf)) {
2044 		netdev_err(netdev, "Can't disable TC hardware offload while flows are active\n");
2045 		return -EBUSY;
2046 	}
2047 
2048 	if ((changed & NETIF_F_NTUPLE) && ntuple &&
2049 	    otx2_tc_flower_rule_cnt(pfvf) && !(changed & NETIF_F_HW_TC)) {
2050 		netdev_err(netdev,
2051 			   "Can't enable NTUPLE when TC flower offload is active, disable TC rules and retry\n");
2052 		return -EINVAL;
2053 	}
2054 
2055 	return 0;
2056 }
2057 EXPORT_SYMBOL(otx2_handle_ntuple_tc_features);
2058 
2059 int otx2_set_hw_capabilities(struct otx2_nic *pfvf)
2060 {
2061 	struct mbox *mbox = &pfvf->mbox;
2062 	struct otx2_hw *hw = &pfvf->hw;
2063 	struct get_hw_cap_rsp *rsp;
2064 	struct msg_req *req;
2065 	int ret = -ENOMEM;
2066 
2067 	mutex_lock(&mbox->lock);
2068 
2069 	req = otx2_mbox_alloc_msg_get_hw_cap(mbox);
2070 	if (!req)
2071 		goto fail;
2072 
2073 	ret = otx2_sync_mbox_msg(mbox);
2074 	if (ret)
2075 		goto fail;
2076 
2077 	rsp = (struct get_hw_cap_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox,
2078 							 0, &req->hdr);
2079 	if (IS_ERR(rsp)) {
2080 		ret = -EINVAL;
2081 		goto fail;
2082 	}
2083 
2084 	if (rsp->hw_caps & HW_CAP_MACSEC)
2085 		__set_bit(CN10K_HW_MACSEC, &hw->cap_flag);
2086 
2087 	mutex_unlock(&mbox->lock);
2088 
2089 	return 0;
2090 fail:
2091 	dev_err(pfvf->dev, "Cannot get MACSEC capability from AF\n");
2092 	mutex_unlock(&mbox->lock);
2093 	return ret;
2094 }
2095 
2096 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
2097 int __weak								\
2098 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf,		\
2099 				struct _req_type *req,			\
2100 				struct _rsp_type *rsp)			\
2101 {									\
2102 	/* Nothing to do here */					\
2103 	return 0;							\
2104 }									\
2105 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
2106 MBOX_UP_CGX_MESSAGES
2107 MBOX_UP_MCS_MESSAGES
2108 #undef M
2109 
2110 dma_addr_t otx2_dma_map_skb_frag(struct otx2_nic *pfvf,
2111 				 struct sk_buff *skb, int seg, int *len)
2112 {
2113 	enum dma_data_direction dir = DMA_TO_DEVICE;
2114 	const skb_frag_t *frag;
2115 	struct page *page;
2116 	int offset;
2117 
2118 	/* Crypto hardware need write permission for ipsec crypto offload */
2119 	if (unlikely(xfrm_offload(skb))) {
2120 		dir = DMA_BIDIRECTIONAL;
2121 		skb = skb_unshare(skb, GFP_ATOMIC);
2122 	}
2123 
2124 	/* First segment is always skb->data */
2125 	if (!seg) {
2126 		page = virt_to_page(skb->data);
2127 		offset = offset_in_page(skb->data);
2128 		*len = skb_headlen(skb);
2129 	} else {
2130 		frag = &skb_shinfo(skb)->frags[seg - 1];
2131 		page = skb_frag_page(frag);
2132 		offset = skb_frag_off(frag);
2133 		*len = skb_frag_size(frag);
2134 	}
2135 	return otx2_dma_map_page(pfvf, page, offset, *len, dir);
2136 }
2137 
2138 void otx2_dma_unmap_skb_frags(struct otx2_nic *pfvf, struct sg_list *sg)
2139 {
2140 	enum dma_data_direction dir = DMA_TO_DEVICE;
2141 	struct sk_buff *skb = NULL;
2142 	int seg;
2143 
2144 	skb = (struct sk_buff *)sg->skb;
2145 	if (unlikely(xfrm_offload(skb)))
2146 		dir = DMA_BIDIRECTIONAL;
2147 
2148 	for (seg = 0; seg < sg->num_segs; seg++) {
2149 		otx2_dma_unmap_page(pfvf, sg->dma_addr[seg],
2150 				    sg->size[seg], dir);
2151 	}
2152 	sg->num_segs = 0;
2153 }
2154