1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell OcteonTx2 RVU Ethernet driver
3  *
4  * Copyright (C) 2020 Marvell International Ltd.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 
11 #include <linux/interrupt.h>
12 #include <linux/pci.h>
13 #include <net/tso.h>
14 
15 #include "otx2_reg.h"
16 #include "otx2_common.h"
17 #include "otx2_struct.h"
18 
19 static void otx2_nix_rq_op_stats(struct queue_stats *stats,
20 				 struct otx2_nic *pfvf, int qidx)
21 {
22 	u64 incr = (u64)qidx << 32;
23 	u64 *ptr;
24 
25 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_OCTS);
26 	stats->bytes = otx2_atomic64_add(incr, ptr);
27 
28 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_RQ_OP_PKTS);
29 	stats->pkts = otx2_atomic64_add(incr, ptr);
30 }
31 
32 static void otx2_nix_sq_op_stats(struct queue_stats *stats,
33 				 struct otx2_nic *pfvf, int qidx)
34 {
35 	u64 incr = (u64)qidx << 32;
36 	u64 *ptr;
37 
38 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_OCTS);
39 	stats->bytes = otx2_atomic64_add(incr, ptr);
40 
41 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_PKTS);
42 	stats->pkts = otx2_atomic64_add(incr, ptr);
43 }
44 
45 void otx2_update_lmac_stats(struct otx2_nic *pfvf)
46 {
47 	struct msg_req *req;
48 
49 	if (!netif_running(pfvf->netdev))
50 		return;
51 
52 	otx2_mbox_lock(&pfvf->mbox);
53 	req = otx2_mbox_alloc_msg_cgx_stats(&pfvf->mbox);
54 	if (!req) {
55 		otx2_mbox_unlock(&pfvf->mbox);
56 		return;
57 	}
58 
59 	otx2_sync_mbox_msg(&pfvf->mbox);
60 	otx2_mbox_unlock(&pfvf->mbox);
61 }
62 
63 int otx2_update_rq_stats(struct otx2_nic *pfvf, int qidx)
64 {
65 	struct otx2_rcv_queue *rq = &pfvf->qset.rq[qidx];
66 
67 	if (!pfvf->qset.rq)
68 		return 0;
69 
70 	otx2_nix_rq_op_stats(&rq->stats, pfvf, qidx);
71 	return 1;
72 }
73 
74 int otx2_update_sq_stats(struct otx2_nic *pfvf, int qidx)
75 {
76 	struct otx2_snd_queue *sq = &pfvf->qset.sq[qidx];
77 
78 	if (!pfvf->qset.sq)
79 		return 0;
80 
81 	otx2_nix_sq_op_stats(&sq->stats, pfvf, qidx);
82 	return 1;
83 }
84 
85 void otx2_get_dev_stats(struct otx2_nic *pfvf)
86 {
87 	struct otx2_dev_stats *dev_stats = &pfvf->hw.dev_stats;
88 
89 #define OTX2_GET_RX_STATS(reg) \
90 	 otx2_read64(pfvf, NIX_LF_RX_STATX(reg))
91 #define OTX2_GET_TX_STATS(reg) \
92 	 otx2_read64(pfvf, NIX_LF_TX_STATX(reg))
93 
94 	dev_stats->rx_bytes = OTX2_GET_RX_STATS(RX_OCTS);
95 	dev_stats->rx_drops = OTX2_GET_RX_STATS(RX_DROP);
96 	dev_stats->rx_bcast_frames = OTX2_GET_RX_STATS(RX_BCAST);
97 	dev_stats->rx_mcast_frames = OTX2_GET_RX_STATS(RX_MCAST);
98 	dev_stats->rx_ucast_frames = OTX2_GET_RX_STATS(RX_UCAST);
99 	dev_stats->rx_frames = dev_stats->rx_bcast_frames +
100 			       dev_stats->rx_mcast_frames +
101 			       dev_stats->rx_ucast_frames;
102 
103 	dev_stats->tx_bytes = OTX2_GET_TX_STATS(TX_OCTS);
104 	dev_stats->tx_drops = OTX2_GET_TX_STATS(TX_DROP);
105 	dev_stats->tx_bcast_frames = OTX2_GET_TX_STATS(TX_BCAST);
106 	dev_stats->tx_mcast_frames = OTX2_GET_TX_STATS(TX_MCAST);
107 	dev_stats->tx_ucast_frames = OTX2_GET_TX_STATS(TX_UCAST);
108 	dev_stats->tx_frames = dev_stats->tx_bcast_frames +
109 			       dev_stats->tx_mcast_frames +
110 			       dev_stats->tx_ucast_frames;
111 }
112 
113 void otx2_get_stats64(struct net_device *netdev,
114 		      struct rtnl_link_stats64 *stats)
115 {
116 	struct otx2_nic *pfvf = netdev_priv(netdev);
117 	struct otx2_dev_stats *dev_stats;
118 
119 	otx2_get_dev_stats(pfvf);
120 
121 	dev_stats = &pfvf->hw.dev_stats;
122 	stats->rx_bytes = dev_stats->rx_bytes;
123 	stats->rx_packets = dev_stats->rx_frames;
124 	stats->rx_dropped = dev_stats->rx_drops;
125 	stats->multicast = dev_stats->rx_mcast_frames;
126 
127 	stats->tx_bytes = dev_stats->tx_bytes;
128 	stats->tx_packets = dev_stats->tx_frames;
129 	stats->tx_dropped = dev_stats->tx_drops;
130 }
131 
132 /* Sync MAC address with RVU AF */
133 static int otx2_hw_set_mac_addr(struct otx2_nic *pfvf, u8 *mac)
134 {
135 	struct nix_set_mac_addr *req;
136 	int err;
137 
138 	otx2_mbox_lock(&pfvf->mbox);
139 	req = otx2_mbox_alloc_msg_nix_set_mac_addr(&pfvf->mbox);
140 	if (!req) {
141 		otx2_mbox_unlock(&pfvf->mbox);
142 		return -ENOMEM;
143 	}
144 
145 	ether_addr_copy(req->mac_addr, mac);
146 
147 	err = otx2_sync_mbox_msg(&pfvf->mbox);
148 	otx2_mbox_unlock(&pfvf->mbox);
149 	return err;
150 }
151 
152 static int otx2_hw_get_mac_addr(struct otx2_nic *pfvf,
153 				struct net_device *netdev)
154 {
155 	struct nix_get_mac_addr_rsp *rsp;
156 	struct mbox_msghdr *msghdr;
157 	struct msg_req *req;
158 	int err;
159 
160 	otx2_mbox_lock(&pfvf->mbox);
161 	req = otx2_mbox_alloc_msg_nix_get_mac_addr(&pfvf->mbox);
162 	if (!req) {
163 		otx2_mbox_unlock(&pfvf->mbox);
164 		return -ENOMEM;
165 	}
166 
167 	err = otx2_sync_mbox_msg(&pfvf->mbox);
168 	if (err) {
169 		otx2_mbox_unlock(&pfvf->mbox);
170 		return err;
171 	}
172 
173 	msghdr = otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0, &req->hdr);
174 	if (IS_ERR(msghdr)) {
175 		otx2_mbox_unlock(&pfvf->mbox);
176 		return PTR_ERR(msghdr);
177 	}
178 	rsp = (struct nix_get_mac_addr_rsp *)msghdr;
179 	ether_addr_copy(netdev->dev_addr, rsp->mac_addr);
180 	otx2_mbox_unlock(&pfvf->mbox);
181 
182 	return 0;
183 }
184 
185 int otx2_set_mac_address(struct net_device *netdev, void *p)
186 {
187 	struct otx2_nic *pfvf = netdev_priv(netdev);
188 	struct sockaddr *addr = p;
189 
190 	if (!is_valid_ether_addr(addr->sa_data))
191 		return -EADDRNOTAVAIL;
192 
193 	if (!otx2_hw_set_mac_addr(pfvf, addr->sa_data))
194 		memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
195 	else
196 		return -EPERM;
197 
198 	return 0;
199 }
200 
201 int otx2_hw_set_mtu(struct otx2_nic *pfvf, int mtu)
202 {
203 	struct nix_frs_cfg *req;
204 	int err;
205 
206 	otx2_mbox_lock(&pfvf->mbox);
207 	req = otx2_mbox_alloc_msg_nix_set_hw_frs(&pfvf->mbox);
208 	if (!req) {
209 		otx2_mbox_unlock(&pfvf->mbox);
210 		return -ENOMEM;
211 	}
212 
213 	/* SMQ config limits maximum pkt size that can be transmitted */
214 	req->update_smq = true;
215 	pfvf->max_frs = mtu +  OTX2_ETH_HLEN;
216 	req->maxlen = pfvf->max_frs;
217 
218 	err = otx2_sync_mbox_msg(&pfvf->mbox);
219 	otx2_mbox_unlock(&pfvf->mbox);
220 	return err;
221 }
222 
223 int otx2_config_pause_frm(struct otx2_nic *pfvf)
224 {
225 	struct cgx_pause_frm_cfg *req;
226 	int err;
227 
228 	otx2_mbox_lock(&pfvf->mbox);
229 	req = otx2_mbox_alloc_msg_cgx_cfg_pause_frm(&pfvf->mbox);
230 	if (!req)
231 		return -ENOMEM;
232 
233 	req->rx_pause = !!(pfvf->flags & OTX2_FLAG_RX_PAUSE_ENABLED);
234 	req->tx_pause = !!(pfvf->flags & OTX2_FLAG_TX_PAUSE_ENABLED);
235 	req->set = 1;
236 
237 	err = otx2_sync_mbox_msg(&pfvf->mbox);
238 	otx2_mbox_unlock(&pfvf->mbox);
239 	return err;
240 }
241 
242 int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)
243 {
244 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
245 	struct nix_rss_flowkey_cfg *req;
246 	int err;
247 
248 	otx2_mbox_lock(&pfvf->mbox);
249 	req = otx2_mbox_alloc_msg_nix_rss_flowkey_cfg(&pfvf->mbox);
250 	if (!req) {
251 		otx2_mbox_unlock(&pfvf->mbox);
252 		return -ENOMEM;
253 	}
254 	req->mcam_index = -1; /* Default or reserved index */
255 	req->flowkey_cfg = rss->flowkey_cfg;
256 	req->group = DEFAULT_RSS_CONTEXT_GROUP;
257 
258 	err = otx2_sync_mbox_msg(&pfvf->mbox);
259 	otx2_mbox_unlock(&pfvf->mbox);
260 	return err;
261 }
262 
263 int otx2_set_rss_table(struct otx2_nic *pfvf)
264 {
265 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
266 	struct mbox *mbox = &pfvf->mbox;
267 	struct nix_aq_enq_req *aq;
268 	int idx, err;
269 
270 	otx2_mbox_lock(mbox);
271 	/* Get memory to put this msg */
272 	for (idx = 0; idx < rss->rss_size; idx++) {
273 		aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
274 		if (!aq) {
275 			/* The shared memory buffer can be full.
276 			 * Flush it and retry
277 			 */
278 			err = otx2_sync_mbox_msg(mbox);
279 			if (err) {
280 				otx2_mbox_unlock(mbox);
281 				return err;
282 			}
283 			aq = otx2_mbox_alloc_msg_nix_aq_enq(mbox);
284 			if (!aq) {
285 				otx2_mbox_unlock(mbox);
286 				return -ENOMEM;
287 			}
288 		}
289 
290 		aq->rss.rq = rss->ind_tbl[idx];
291 
292 		/* Fill AQ info */
293 		aq->qidx = idx;
294 		aq->ctype = NIX_AQ_CTYPE_RSS;
295 		aq->op = NIX_AQ_INSTOP_INIT;
296 	}
297 	err = otx2_sync_mbox_msg(mbox);
298 	otx2_mbox_unlock(mbox);
299 	return err;
300 }
301 
302 void otx2_set_rss_key(struct otx2_nic *pfvf)
303 {
304 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
305 	u64 *key = (u64 *)&rss->key[4];
306 	int idx;
307 
308 	/* 352bit or 44byte key needs to be configured as below
309 	 * NIX_LF_RX_SECRETX0 = key<351:288>
310 	 * NIX_LF_RX_SECRETX1 = key<287:224>
311 	 * NIX_LF_RX_SECRETX2 = key<223:160>
312 	 * NIX_LF_RX_SECRETX3 = key<159:96>
313 	 * NIX_LF_RX_SECRETX4 = key<95:32>
314 	 * NIX_LF_RX_SECRETX5<63:32> = key<31:0>
315 	 */
316 	otx2_write64(pfvf, NIX_LF_RX_SECRETX(5),
317 		     (u64)(*((u32 *)&rss->key)) << 32);
318 	idx = sizeof(rss->key) / sizeof(u64);
319 	while (idx > 0) {
320 		idx--;
321 		otx2_write64(pfvf, NIX_LF_RX_SECRETX(idx), *key++);
322 	}
323 }
324 
325 int otx2_rss_init(struct otx2_nic *pfvf)
326 {
327 	struct otx2_rss_info *rss = &pfvf->hw.rss_info;
328 	int idx, ret = 0;
329 
330 	rss->rss_size = sizeof(rss->ind_tbl);
331 
332 	/* Init RSS key if it is not setup already */
333 	if (!rss->enable)
334 		netdev_rss_key_fill(rss->key, sizeof(rss->key));
335 	otx2_set_rss_key(pfvf);
336 
337 	if (!netif_is_rxfh_configured(pfvf->netdev)) {
338 		/* Default indirection table */
339 		for (idx = 0; idx < rss->rss_size; idx++)
340 			rss->ind_tbl[idx] =
341 				ethtool_rxfh_indir_default(idx,
342 							   pfvf->hw.rx_queues);
343 	}
344 	ret = otx2_set_rss_table(pfvf);
345 	if (ret)
346 		return ret;
347 
348 	/* Flowkey or hash config to be used for generating flow tag */
349 	rss->flowkey_cfg = rss->enable ? rss->flowkey_cfg :
350 			   NIX_FLOW_KEY_TYPE_IPV4 | NIX_FLOW_KEY_TYPE_IPV6 |
351 			   NIX_FLOW_KEY_TYPE_TCP | NIX_FLOW_KEY_TYPE_UDP |
352 			   NIX_FLOW_KEY_TYPE_SCTP;
353 
354 	ret = otx2_set_flowkey_cfg(pfvf);
355 	if (ret)
356 		return ret;
357 
358 	rss->enable = true;
359 	return 0;
360 }
361 
362 void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
363 {
364 	/* Configure CQE interrupt coalescing parameters
365 	 *
366 	 * HW triggers an irq when ECOUNT > cq_ecount_wait, hence
367 	 * set 1 less than cq_ecount_wait. And cq_time_wait is in
368 	 * usecs, convert that to 100ns count.
369 	 */
370 	otx2_write64(pfvf, NIX_LF_CINTX_WAIT(qidx),
371 		     ((u64)(pfvf->hw.cq_time_wait * 10) << 48) |
372 		     ((u64)pfvf->hw.cq_qcount_wait << 32) |
373 		     (pfvf->hw.cq_ecount_wait - 1));
374 }
375 
376 dma_addr_t otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
377 			   gfp_t gfp)
378 {
379 	dma_addr_t iova;
380 
381 	/* Check if request can be accommodated in previous allocated page */
382 	if (pool->page && ((pool->page_offset + pool->rbsize) <=
383 	    (PAGE_SIZE << pool->rbpage_order))) {
384 		pool->pageref++;
385 		goto ret;
386 	}
387 
388 	otx2_get_page(pool);
389 
390 	/* Allocate a new page */
391 	pool->page = alloc_pages(gfp | __GFP_COMP | __GFP_NOWARN,
392 				 pool->rbpage_order);
393 	if (unlikely(!pool->page))
394 		return -ENOMEM;
395 
396 	pool->page_offset = 0;
397 ret:
398 	iova = (u64)otx2_dma_map_page(pfvf, pool->page, pool->page_offset,
399 				      pool->rbsize, DMA_FROM_DEVICE);
400 	if (!iova) {
401 		if (!pool->page_offset)
402 			__free_pages(pool->page, pool->rbpage_order);
403 		pool->page = NULL;
404 		return -ENOMEM;
405 	}
406 	pool->page_offset += pool->rbsize;
407 	return iova;
408 }
409 
410 void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
411 {
412 	struct otx2_nic *pfvf = netdev_priv(netdev);
413 
414 	schedule_work(&pfvf->reset_task);
415 }
416 
417 void otx2_get_mac_from_af(struct net_device *netdev)
418 {
419 	struct otx2_nic *pfvf = netdev_priv(netdev);
420 	int err;
421 
422 	err = otx2_hw_get_mac_addr(pfvf, netdev);
423 	if (err)
424 		dev_warn(pfvf->dev, "Failed to read mac from hardware\n");
425 
426 	/* If AF doesn't provide a valid MAC, generate a random one */
427 	if (!is_valid_ether_addr(netdev->dev_addr))
428 		eth_hw_addr_random(netdev);
429 }
430 
431 static int otx2_get_link(struct otx2_nic *pfvf)
432 {
433 	int link = 0;
434 	u16 map;
435 
436 	/* cgx lmac link */
437 	if (pfvf->hw.tx_chan_base >= CGX_CHAN_BASE) {
438 		map = pfvf->hw.tx_chan_base & 0x7FF;
439 		link = 4 * ((map >> 8) & 0xF) + ((map >> 4) & 0xF);
440 	}
441 	/* LBK channel */
442 	if (pfvf->hw.tx_chan_base < SDP_CHAN_BASE)
443 		link = 12;
444 
445 	return link;
446 }
447 
448 int otx2_txschq_config(struct otx2_nic *pfvf, int lvl)
449 {
450 	struct otx2_hw *hw = &pfvf->hw;
451 	struct nix_txschq_config *req;
452 	u64 schq, parent;
453 
454 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&pfvf->mbox);
455 	if (!req)
456 		return -ENOMEM;
457 
458 	req->lvl = lvl;
459 	req->num_regs = 1;
460 
461 	schq = hw->txschq_list[lvl][0];
462 	/* Set topology e.t.c configuration */
463 	if (lvl == NIX_TXSCH_LVL_SMQ) {
464 		req->reg[0] = NIX_AF_SMQX_CFG(schq);
465 		req->regval[0] = ((pfvf->netdev->mtu  + OTX2_ETH_HLEN) << 8) |
466 				   OTX2_MIN_MTU;
467 
468 		req->regval[0] |= (0x20ULL << 51) | (0x80ULL << 39) |
469 				  (0x2ULL << 36);
470 		req->num_regs++;
471 		/* MDQ config */
472 		parent =  hw->txschq_list[NIX_TXSCH_LVL_TL4][0];
473 		req->reg[1] = NIX_AF_MDQX_PARENT(schq);
474 		req->regval[1] = parent << 16;
475 		req->num_regs++;
476 		/* Set DWRR quantum */
477 		req->reg[2] = NIX_AF_MDQX_SCHEDULE(schq);
478 		req->regval[2] =  DFLT_RR_QTM;
479 	} else if (lvl == NIX_TXSCH_LVL_TL4) {
480 		parent =  hw->txschq_list[NIX_TXSCH_LVL_TL3][0];
481 		req->reg[0] = NIX_AF_TL4X_PARENT(schq);
482 		req->regval[0] = parent << 16;
483 		req->num_regs++;
484 		req->reg[1] = NIX_AF_TL4X_SCHEDULE(schq);
485 		req->regval[1] = DFLT_RR_QTM;
486 	} else if (lvl == NIX_TXSCH_LVL_TL3) {
487 		parent = hw->txschq_list[NIX_TXSCH_LVL_TL2][0];
488 		req->reg[0] = NIX_AF_TL3X_PARENT(schq);
489 		req->regval[0] = parent << 16;
490 		req->num_regs++;
491 		req->reg[1] = NIX_AF_TL3X_SCHEDULE(schq);
492 		req->regval[1] = DFLT_RR_QTM;
493 	} else if (lvl == NIX_TXSCH_LVL_TL2) {
494 		parent =  hw->txschq_list[NIX_TXSCH_LVL_TL1][0];
495 		req->reg[0] = NIX_AF_TL2X_PARENT(schq);
496 		req->regval[0] = parent << 16;
497 
498 		req->num_regs++;
499 		req->reg[1] = NIX_AF_TL2X_SCHEDULE(schq);
500 		req->regval[1] = TXSCH_TL1_DFLT_RR_PRIO << 24 | DFLT_RR_QTM;
501 
502 		req->num_regs++;
503 		req->reg[2] = NIX_AF_TL3_TL2X_LINKX_CFG(schq,
504 							otx2_get_link(pfvf));
505 		/* Enable this queue and backpressure */
506 		req->regval[2] = BIT_ULL(13) | BIT_ULL(12);
507 
508 	} else if (lvl == NIX_TXSCH_LVL_TL1) {
509 		/* Default config for TL1.
510 		 * For VF this is always ignored.
511 		 */
512 
513 		/* Set DWRR quantum */
514 		req->reg[0] = NIX_AF_TL1X_SCHEDULE(schq);
515 		req->regval[0] = TXSCH_TL1_DFLT_RR_QTM;
516 
517 		req->num_regs++;
518 		req->reg[1] = NIX_AF_TL1X_TOPOLOGY(schq);
519 		req->regval[1] = (TXSCH_TL1_DFLT_RR_PRIO << 1);
520 
521 		req->num_regs++;
522 		req->reg[2] = NIX_AF_TL1X_CIR(schq);
523 		req->regval[2] = 0;
524 	}
525 
526 	return otx2_sync_mbox_msg(&pfvf->mbox);
527 }
528 
529 int otx2_txsch_alloc(struct otx2_nic *pfvf)
530 {
531 	struct nix_txsch_alloc_req *req;
532 	int lvl;
533 
534 	/* Get memory to put this msg */
535 	req = otx2_mbox_alloc_msg_nix_txsch_alloc(&pfvf->mbox);
536 	if (!req)
537 		return -ENOMEM;
538 
539 	/* Request one schq per level */
540 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
541 		req->schq[lvl] = 1;
542 
543 	return otx2_sync_mbox_msg(&pfvf->mbox);
544 }
545 
546 int otx2_txschq_stop(struct otx2_nic *pfvf)
547 {
548 	struct nix_txsch_free_req *free_req;
549 	int lvl, schq, err;
550 
551 	otx2_mbox_lock(&pfvf->mbox);
552 	/* Free the transmit schedulers */
553 	free_req = otx2_mbox_alloc_msg_nix_txsch_free(&pfvf->mbox);
554 	if (!free_req) {
555 		otx2_mbox_unlock(&pfvf->mbox);
556 		return -ENOMEM;
557 	}
558 
559 	free_req->flags = TXSCHQ_FREE_ALL;
560 	err = otx2_sync_mbox_msg(&pfvf->mbox);
561 	otx2_mbox_unlock(&pfvf->mbox);
562 
563 	/* Clear the txschq list */
564 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++) {
565 		for (schq = 0; schq < MAX_TXSCHQ_PER_FUNC; schq++)
566 			pfvf->hw.txschq_list[lvl][schq] = 0;
567 	}
568 	return err;
569 }
570 
571 void otx2_sqb_flush(struct otx2_nic *pfvf)
572 {
573 	int qidx, sqe_tail, sqe_head;
574 	u64 incr, *ptr, val;
575 
576 	ptr = (u64 *)otx2_get_regaddr(pfvf, NIX_LF_SQ_OP_STATUS);
577 	for (qidx = 0; qidx < pfvf->hw.tx_queues; qidx++) {
578 		incr = (u64)qidx << 32;
579 		while (1) {
580 			val = otx2_atomic64_add(incr, ptr);
581 			sqe_head = (val >> 20) & 0x3F;
582 			sqe_tail = (val >> 28) & 0x3F;
583 			if (sqe_head == sqe_tail)
584 				break;
585 			usleep_range(1, 3);
586 		}
587 	}
588 }
589 
590 /* RED and drop levels of CQ on packet reception.
591  * For CQ level is measure of emptiness ( 0x0 = full, 255 = empty).
592  */
593 #define RQ_PASS_LVL_CQ(skid, qsize)	((((skid) + 16) * 256) / (qsize))
594 #define RQ_DROP_LVL_CQ(skid, qsize)	(((skid) * 256) / (qsize))
595 
596 /* RED and drop levels of AURA for packet reception.
597  * For AURA level is measure of fullness (0x0 = empty, 255 = full).
598  * Eg: For RQ length 1K, for pass/drop level 204/230.
599  * RED accepts pkts if free pointers > 102 & <= 205.
600  * Drops pkts if free pointers < 102.
601  */
602 #define RQ_BP_LVL_AURA   (255 - ((85 * 256) / 100)) /* BP when 85% is full */
603 #define RQ_PASS_LVL_AURA (255 - ((95 * 256) / 100)) /* RED when 95% is full */
604 #define RQ_DROP_LVL_AURA (255 - ((99 * 256) / 100)) /* Drop when 99% is full */
605 
606 /* Send skid of 2000 packets required for CQ size of 4K CQEs. */
607 #define SEND_CQ_SKID	2000
608 
609 static int otx2_rq_init(struct otx2_nic *pfvf, u16 qidx, u16 lpb_aura)
610 {
611 	struct otx2_qset *qset = &pfvf->qset;
612 	struct nix_aq_enq_req *aq;
613 
614 	/* Get memory to put this msg */
615 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
616 	if (!aq)
617 		return -ENOMEM;
618 
619 	aq->rq.cq = qidx;
620 	aq->rq.ena = 1;
621 	aq->rq.pb_caching = 1;
622 	aq->rq.lpb_aura = lpb_aura; /* Use large packet buffer aura */
623 	aq->rq.lpb_sizem1 = (DMA_BUFFER_LEN(pfvf->rbsize) / 8) - 1;
624 	aq->rq.xqe_imm_size = 0; /* Copying of packet to CQE not needed */
625 	aq->rq.flow_tagw = 32; /* Copy full 32bit flow_tag to CQE header */
626 	aq->rq.qint_idx = 0;
627 	aq->rq.lpb_drop_ena = 1; /* Enable RED dropping for AURA */
628 	aq->rq.xqe_drop_ena = 1; /* Enable RED dropping for CQ/SSO */
629 	aq->rq.xqe_pass = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
630 	aq->rq.xqe_drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
631 	aq->rq.lpb_aura_pass = RQ_PASS_LVL_AURA;
632 	aq->rq.lpb_aura_drop = RQ_DROP_LVL_AURA;
633 
634 	/* Fill AQ info */
635 	aq->qidx = qidx;
636 	aq->ctype = NIX_AQ_CTYPE_RQ;
637 	aq->op = NIX_AQ_INSTOP_INIT;
638 
639 	return otx2_sync_mbox_msg(&pfvf->mbox);
640 }
641 
642 static int otx2_sq_init(struct otx2_nic *pfvf, u16 qidx, u16 sqb_aura)
643 {
644 	struct otx2_qset *qset = &pfvf->qset;
645 	struct otx2_snd_queue *sq;
646 	struct nix_aq_enq_req *aq;
647 	struct otx2_pool *pool;
648 	int err;
649 
650 	pool = &pfvf->qset.pool[sqb_aura];
651 	sq = &qset->sq[qidx];
652 	sq->sqe_size = NIX_SQESZ_W16 ? 64 : 128;
653 	sq->sqe_cnt = qset->sqe_cnt;
654 
655 	err = qmem_alloc(pfvf->dev, &sq->sqe, 1, sq->sqe_size);
656 	if (err)
657 		return err;
658 
659 	err = qmem_alloc(pfvf->dev, &sq->tso_hdrs, qset->sqe_cnt,
660 			 TSO_HEADER_SIZE);
661 	if (err)
662 		return err;
663 
664 	sq->sqe_base = sq->sqe->base;
665 	sq->sg = kcalloc(qset->sqe_cnt, sizeof(struct sg_list), GFP_KERNEL);
666 	if (!sq->sg)
667 		return -ENOMEM;
668 
669 	sq->head = 0;
670 	sq->sqe_per_sqb = (pfvf->hw.sqb_size / sq->sqe_size) - 1;
671 	sq->num_sqbs = (qset->sqe_cnt + sq->sqe_per_sqb) / sq->sqe_per_sqb;
672 	/* Set SQE threshold to 10% of total SQEs */
673 	sq->sqe_thresh = ((sq->num_sqbs * sq->sqe_per_sqb) * 10) / 100;
674 	sq->aura_id = sqb_aura;
675 	sq->aura_fc_addr = pool->fc_addr->base;
676 	sq->lmt_addr = (__force u64 *)(pfvf->reg_base + LMT_LF_LMTLINEX(qidx));
677 	sq->io_addr = (__force u64)otx2_get_regaddr(pfvf, NIX_LF_OP_SENDX(0));
678 
679 	sq->stats.bytes = 0;
680 	sq->stats.pkts = 0;
681 
682 	/* Get memory to put this msg */
683 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
684 	if (!aq)
685 		return -ENOMEM;
686 
687 	aq->sq.cq = pfvf->hw.rx_queues + qidx;
688 	aq->sq.max_sqe_size = NIX_MAXSQESZ_W16; /* 128 byte */
689 	aq->sq.cq_ena = 1;
690 	aq->sq.ena = 1;
691 	/* Only one SMQ is allocated, map all SQ's to that SMQ  */
692 	aq->sq.smq = pfvf->hw.txschq_list[NIX_TXSCH_LVL_SMQ][0];
693 	aq->sq.smq_rr_quantum = DFLT_RR_QTM;
694 	aq->sq.default_chan = pfvf->hw.tx_chan_base;
695 	aq->sq.sqe_stype = NIX_STYPE_STF; /* Cache SQB */
696 	aq->sq.sqb_aura = sqb_aura;
697 	aq->sq.sq_int_ena = NIX_SQINT_BITS;
698 	aq->sq.qint_idx = 0;
699 	/* Due pipelining impact minimum 2000 unused SQ CQE's
700 	 * need to maintain to avoid CQ overflow.
701 	 */
702 	aq->sq.cq_limit = ((SEND_CQ_SKID * 256) / (sq->sqe_cnt));
703 
704 	/* Fill AQ info */
705 	aq->qidx = qidx;
706 	aq->ctype = NIX_AQ_CTYPE_SQ;
707 	aq->op = NIX_AQ_INSTOP_INIT;
708 
709 	return otx2_sync_mbox_msg(&pfvf->mbox);
710 }
711 
712 static int otx2_cq_init(struct otx2_nic *pfvf, u16 qidx)
713 {
714 	struct otx2_qset *qset = &pfvf->qset;
715 	struct nix_aq_enq_req *aq;
716 	struct otx2_cq_queue *cq;
717 	int err, pool_id;
718 
719 	cq = &qset->cq[qidx];
720 	cq->cq_idx = qidx;
721 	if (qidx < pfvf->hw.rx_queues) {
722 		cq->cq_type = CQ_RX;
723 		cq->cint_idx = qidx;
724 		cq->cqe_cnt = qset->rqe_cnt;
725 	} else {
726 		cq->cq_type = CQ_TX;
727 		cq->cint_idx = qidx - pfvf->hw.rx_queues;
728 		cq->cqe_cnt = qset->sqe_cnt;
729 	}
730 	cq->cqe_size = pfvf->qset.xqe_size;
731 
732 	/* Allocate memory for CQEs */
733 	err = qmem_alloc(pfvf->dev, &cq->cqe, cq->cqe_cnt, cq->cqe_size);
734 	if (err)
735 		return err;
736 
737 	/* Save CQE CPU base for faster reference */
738 	cq->cqe_base = cq->cqe->base;
739 	/* In case where all RQs auras point to single pool,
740 	 * all CQs receive buffer pool also point to same pool.
741 	 */
742 	pool_id = ((cq->cq_type == CQ_RX) &&
743 		   (pfvf->hw.rqpool_cnt != pfvf->hw.rx_queues)) ? 0 : qidx;
744 	cq->rbpool = &qset->pool[pool_id];
745 	cq->refill_task_sched = false;
746 
747 	/* Get memory to put this msg */
748 	aq = otx2_mbox_alloc_msg_nix_aq_enq(&pfvf->mbox);
749 	if (!aq)
750 		return -ENOMEM;
751 
752 	aq->cq.ena = 1;
753 	aq->cq.qsize = Q_SIZE(cq->cqe_cnt, 4);
754 	aq->cq.caching = 1;
755 	aq->cq.base = cq->cqe->iova;
756 	aq->cq.cint_idx = cq->cint_idx;
757 	aq->cq.cq_err_int_ena = NIX_CQERRINT_BITS;
758 	aq->cq.qint_idx = 0;
759 	aq->cq.avg_level = 255;
760 
761 	if (qidx < pfvf->hw.rx_queues) {
762 		aq->cq.drop = RQ_DROP_LVL_CQ(pfvf->hw.rq_skid, cq->cqe_cnt);
763 		aq->cq.drop_ena = 1;
764 
765 		/* Enable receive CQ backpressure */
766 		aq->cq.bp_ena = 1;
767 		aq->cq.bpid = pfvf->bpid[0];
768 
769 		/* Set backpressure level is same as cq pass level */
770 		aq->cq.bp = RQ_PASS_LVL_CQ(pfvf->hw.rq_skid, qset->rqe_cnt);
771 	}
772 
773 	/* Fill AQ info */
774 	aq->qidx = qidx;
775 	aq->ctype = NIX_AQ_CTYPE_CQ;
776 	aq->op = NIX_AQ_INSTOP_INIT;
777 
778 	return otx2_sync_mbox_msg(&pfvf->mbox);
779 }
780 
781 static void otx2_pool_refill_task(struct work_struct *work)
782 {
783 	struct otx2_cq_queue *cq;
784 	struct otx2_pool *rbpool;
785 	struct refill_work *wrk;
786 	int qidx, free_ptrs = 0;
787 	struct otx2_nic *pfvf;
788 	s64 bufptr;
789 
790 	wrk = container_of(work, struct refill_work, pool_refill_work.work);
791 	pfvf = wrk->pf;
792 	qidx = wrk - pfvf->refill_wrk;
793 	cq = &pfvf->qset.cq[qidx];
794 	rbpool = cq->rbpool;
795 	free_ptrs = cq->pool_ptrs;
796 
797 	while (cq->pool_ptrs) {
798 		bufptr = otx2_alloc_rbuf(pfvf, rbpool, GFP_KERNEL);
799 		if (bufptr <= 0) {
800 			/* Schedule a WQ if we fails to free atleast half of the
801 			 * pointers else enable napi for this RQ.
802 			 */
803 			if (!((free_ptrs - cq->pool_ptrs) > free_ptrs / 2)) {
804 				struct delayed_work *dwork;
805 
806 				dwork = &wrk->pool_refill_work;
807 				schedule_delayed_work(dwork,
808 						      msecs_to_jiffies(100));
809 			} else {
810 				cq->refill_task_sched = false;
811 			}
812 			return;
813 		}
814 		otx2_aura_freeptr(pfvf, qidx, bufptr + OTX2_HEAD_ROOM);
815 		cq->pool_ptrs--;
816 	}
817 	cq->refill_task_sched = false;
818 }
819 
820 int otx2_config_nix_queues(struct otx2_nic *pfvf)
821 {
822 	int qidx, err;
823 
824 	/* Initialize RX queues */
825 	for (qidx = 0; qidx < pfvf->hw.rx_queues; qidx++) {
826 		u16 lpb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, qidx);
827 
828 		err = otx2_rq_init(pfvf, qidx, lpb_aura);
829 		if (err)
830 			return err;
831 	}
832 
833 	/* Initialize TX queues */
834 	for (qidx = 0; qidx < pfvf->hw.tx_queues; qidx++) {
835 		u16 sqb_aura = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
836 
837 		err = otx2_sq_init(pfvf, qidx, sqb_aura);
838 		if (err)
839 			return err;
840 	}
841 
842 	/* Initialize completion queues */
843 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
844 		err = otx2_cq_init(pfvf, qidx);
845 		if (err)
846 			return err;
847 	}
848 
849 	/* Initialize work queue for receive buffer refill */
850 	pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
851 					sizeof(struct refill_work), GFP_KERNEL);
852 	if (!pfvf->refill_wrk)
853 		return -ENOMEM;
854 
855 	for (qidx = 0; qidx < pfvf->qset.cq_cnt; qidx++) {
856 		pfvf->refill_wrk[qidx].pf = pfvf;
857 		INIT_DELAYED_WORK(&pfvf->refill_wrk[qidx].pool_refill_work,
858 				  otx2_pool_refill_task);
859 	}
860 	return 0;
861 }
862 
863 int otx2_config_nix(struct otx2_nic *pfvf)
864 {
865 	struct nix_lf_alloc_req  *nixlf;
866 	struct nix_lf_alloc_rsp *rsp;
867 	int err;
868 
869 	pfvf->qset.xqe_size = NIX_XQESZ_W16 ? 128 : 512;
870 
871 	/* Get memory to put this msg */
872 	nixlf = otx2_mbox_alloc_msg_nix_lf_alloc(&pfvf->mbox);
873 	if (!nixlf)
874 		return -ENOMEM;
875 
876 	/* Set RQ/SQ/CQ counts */
877 	nixlf->rq_cnt = pfvf->hw.rx_queues;
878 	nixlf->sq_cnt = pfvf->hw.tx_queues;
879 	nixlf->cq_cnt = pfvf->qset.cq_cnt;
880 	nixlf->rss_sz = MAX_RSS_INDIR_TBL_SIZE;
881 	nixlf->rss_grps = 1; /* Single RSS indir table supported, for now */
882 	nixlf->xqe_sz = NIX_XQESZ_W16;
883 	/* We don't know absolute NPA LF idx attached.
884 	 * AF will replace 'RVU_DEFAULT_PF_FUNC' with
885 	 * NPA LF attached to this RVU PF/VF.
886 	 */
887 	nixlf->npa_func = RVU_DEFAULT_PF_FUNC;
888 	/* Disable alignment pad, enable L2 length check,
889 	 * enable L4 TCP/UDP checksum verification.
890 	 */
891 	nixlf->rx_cfg = BIT_ULL(33) | BIT_ULL(35) | BIT_ULL(37);
892 
893 	err = otx2_sync_mbox_msg(&pfvf->mbox);
894 	if (err)
895 		return err;
896 
897 	rsp = (struct nix_lf_alloc_rsp *)otx2_mbox_get_rsp(&pfvf->mbox.mbox, 0,
898 							   &nixlf->hdr);
899 	if (IS_ERR(rsp))
900 		return PTR_ERR(rsp);
901 
902 	if (rsp->qints < 1)
903 		return -ENXIO;
904 
905 	return rsp->hdr.rc;
906 }
907 
908 void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
909 {
910 	struct otx2_qset *qset = &pfvf->qset;
911 	struct otx2_hw *hw = &pfvf->hw;
912 	struct otx2_snd_queue *sq;
913 	int sqb, qidx;
914 	u64 iova, pa;
915 
916 	for (qidx = 0; qidx < hw->tx_queues; qidx++) {
917 		sq = &qset->sq[qidx];
918 		if (!sq->sqb_ptrs)
919 			continue;
920 		for (sqb = 0; sqb < sq->sqb_count; sqb++) {
921 			if (!sq->sqb_ptrs[sqb])
922 				continue;
923 			iova = sq->sqb_ptrs[sqb];
924 			pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
925 			dma_unmap_page_attrs(pfvf->dev, iova, hw->sqb_size,
926 					     DMA_FROM_DEVICE,
927 					     DMA_ATTR_SKIP_CPU_SYNC);
928 			put_page(virt_to_page(phys_to_virt(pa)));
929 		}
930 		sq->sqb_count = 0;
931 	}
932 }
933 
934 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
935 {
936 	int pool_id, pool_start = 0, pool_end = 0, size = 0;
937 	u64 iova, pa;
938 
939 	if (type == AURA_NIX_SQ) {
940 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
941 		pool_end =  pool_start + pfvf->hw.sqpool_cnt;
942 		size = pfvf->hw.sqb_size;
943 	}
944 	if (type == AURA_NIX_RQ) {
945 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
946 		pool_end = pfvf->hw.rqpool_cnt;
947 		size = pfvf->rbsize;
948 	}
949 
950 	/* Free SQB and RQB pointers from the aura pool */
951 	for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
952 		iova = otx2_aura_allocptr(pfvf, pool_id);
953 		while (iova) {
954 			if (type == AURA_NIX_RQ)
955 				iova -= OTX2_HEAD_ROOM;
956 
957 			pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
958 			dma_unmap_page_attrs(pfvf->dev, iova, size,
959 					     DMA_FROM_DEVICE,
960 					     DMA_ATTR_SKIP_CPU_SYNC);
961 			put_page(virt_to_page(phys_to_virt(pa)));
962 			iova = otx2_aura_allocptr(pfvf, pool_id);
963 		}
964 	}
965 }
966 
967 void otx2_aura_pool_free(struct otx2_nic *pfvf)
968 {
969 	struct otx2_pool *pool;
970 	int pool_id;
971 
972 	if (!pfvf->qset.pool)
973 		return;
974 
975 	for (pool_id = 0; pool_id < pfvf->hw.pool_cnt; pool_id++) {
976 		pool = &pfvf->qset.pool[pool_id];
977 		qmem_free(pfvf->dev, pool->stack);
978 		qmem_free(pfvf->dev, pool->fc_addr);
979 	}
980 	devm_kfree(pfvf->dev, pfvf->qset.pool);
981 }
982 
983 static int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
984 			  int pool_id, int numptrs)
985 {
986 	struct npa_aq_enq_req *aq;
987 	struct otx2_pool *pool;
988 	int err;
989 
990 	pool = &pfvf->qset.pool[pool_id];
991 
992 	/* Allocate memory for HW to update Aura count.
993 	 * Alloc one cache line, so that it fits all FC_STYPE modes.
994 	 */
995 	if (!pool->fc_addr) {
996 		err = qmem_alloc(pfvf->dev, &pool->fc_addr, 1, OTX2_ALIGN);
997 		if (err)
998 			return err;
999 	}
1000 
1001 	/* Initialize this aura's context via AF */
1002 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1003 	if (!aq) {
1004 		/* Shared mbox memory buffer is full, flush it and retry */
1005 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1006 		if (err)
1007 			return err;
1008 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1009 		if (!aq)
1010 			return -ENOMEM;
1011 	}
1012 
1013 	aq->aura_id = aura_id;
1014 	/* Will be filled by AF with correct pool context address */
1015 	aq->aura.pool_addr = pool_id;
1016 	aq->aura.pool_caching = 1;
1017 	aq->aura.shift = ilog2(numptrs) - 8;
1018 	aq->aura.count = numptrs;
1019 	aq->aura.limit = numptrs;
1020 	aq->aura.avg_level = 255;
1021 	aq->aura.ena = 1;
1022 	aq->aura.fc_ena = 1;
1023 	aq->aura.fc_addr = pool->fc_addr->iova;
1024 	aq->aura.fc_hyst_bits = 0; /* Store count on all updates */
1025 
1026 	/* Enable backpressure for RQ aura */
1027 	if (aura_id < pfvf->hw.rqpool_cnt) {
1028 		aq->aura.bp_ena = 0;
1029 		aq->aura.nix0_bpid = pfvf->bpid[0];
1030 		/* Set backpressure level for RQ's Aura */
1031 		aq->aura.bp = RQ_BP_LVL_AURA;
1032 	}
1033 
1034 	/* Fill AQ info */
1035 	aq->ctype = NPA_AQ_CTYPE_AURA;
1036 	aq->op = NPA_AQ_INSTOP_INIT;
1037 
1038 	return 0;
1039 }
1040 
1041 static int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
1042 			  int stack_pages, int numptrs, int buf_size)
1043 {
1044 	struct npa_aq_enq_req *aq;
1045 	struct otx2_pool *pool;
1046 	int err;
1047 
1048 	pool = &pfvf->qset.pool[pool_id];
1049 	/* Alloc memory for stack which is used to store buffer pointers */
1050 	err = qmem_alloc(pfvf->dev, &pool->stack,
1051 			 stack_pages, pfvf->hw.stack_pg_bytes);
1052 	if (err)
1053 		return err;
1054 
1055 	pool->rbsize = buf_size;
1056 	pool->rbpage_order = get_order(buf_size);
1057 
1058 	/* Initialize this pool's context via AF */
1059 	aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1060 	if (!aq) {
1061 		/* Shared mbox memory buffer is full, flush it and retry */
1062 		err = otx2_sync_mbox_msg(&pfvf->mbox);
1063 		if (err) {
1064 			qmem_free(pfvf->dev, pool->stack);
1065 			return err;
1066 		}
1067 		aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
1068 		if (!aq) {
1069 			qmem_free(pfvf->dev, pool->stack);
1070 			return -ENOMEM;
1071 		}
1072 	}
1073 
1074 	aq->aura_id = pool_id;
1075 	aq->pool.stack_base = pool->stack->iova;
1076 	aq->pool.stack_caching = 1;
1077 	aq->pool.ena = 1;
1078 	aq->pool.buf_size = buf_size / 128;
1079 	aq->pool.stack_max_pages = stack_pages;
1080 	aq->pool.shift = ilog2(numptrs) - 8;
1081 	aq->pool.ptr_start = 0;
1082 	aq->pool.ptr_end = ~0ULL;
1083 
1084 	/* Fill AQ info */
1085 	aq->ctype = NPA_AQ_CTYPE_POOL;
1086 	aq->op = NPA_AQ_INSTOP_INIT;
1087 
1088 	return 0;
1089 }
1090 
1091 int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
1092 {
1093 	int qidx, pool_id, stack_pages, num_sqbs;
1094 	struct otx2_qset *qset = &pfvf->qset;
1095 	struct otx2_hw *hw = &pfvf->hw;
1096 	struct otx2_snd_queue *sq;
1097 	struct otx2_pool *pool;
1098 	int err, ptr;
1099 	s64 bufptr;
1100 
1101 	/* Calculate number of SQBs needed.
1102 	 *
1103 	 * For a 128byte SQE, and 4K size SQB, 31 SQEs will fit in one SQB.
1104 	 * Last SQE is used for pointing to next SQB.
1105 	 */
1106 	num_sqbs = (hw->sqb_size / 128) - 1;
1107 	num_sqbs = (qset->sqe_cnt + num_sqbs) / num_sqbs;
1108 
1109 	/* Get no of stack pages needed */
1110 	stack_pages =
1111 		(num_sqbs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1112 
1113 	for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1114 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1115 		/* Initialize aura context */
1116 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_sqbs);
1117 		if (err)
1118 			goto fail;
1119 
1120 		/* Initialize pool context */
1121 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1122 				     num_sqbs, hw->sqb_size);
1123 		if (err)
1124 			goto fail;
1125 	}
1126 
1127 	/* Flush accumulated messages */
1128 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1129 	if (err)
1130 		goto fail;
1131 
1132 	/* Allocate pointers and free them to aura/pool */
1133 	for (qidx = 0; qidx < hw->tx_queues; qidx++) {
1134 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_SQ, qidx);
1135 		pool = &pfvf->qset.pool[pool_id];
1136 
1137 		sq = &qset->sq[qidx];
1138 		sq->sqb_count = 0;
1139 		sq->sqb_ptrs = kcalloc(num_sqbs, sizeof(u64 *), GFP_KERNEL);
1140 		if (!sq->sqb_ptrs)
1141 			return -ENOMEM;
1142 
1143 		for (ptr = 0; ptr < num_sqbs; ptr++) {
1144 			bufptr = otx2_alloc_rbuf(pfvf, pool, GFP_KERNEL);
1145 			if (bufptr <= 0)
1146 				return bufptr;
1147 			otx2_aura_freeptr(pfvf, pool_id, bufptr);
1148 			sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
1149 		}
1150 		otx2_get_page(pool);
1151 	}
1152 
1153 	return 0;
1154 fail:
1155 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1156 	otx2_aura_pool_free(pfvf);
1157 	return err;
1158 }
1159 
1160 int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
1161 {
1162 	struct otx2_hw *hw = &pfvf->hw;
1163 	int stack_pages, pool_id, rq;
1164 	struct otx2_pool *pool;
1165 	int err, ptr, num_ptrs;
1166 	s64 bufptr;
1167 
1168 	num_ptrs = pfvf->qset.rqe_cnt;
1169 
1170 	stack_pages =
1171 		(num_ptrs + hw->stack_pg_ptrs - 1) / hw->stack_pg_ptrs;
1172 
1173 	for (rq = 0; rq < hw->rx_queues; rq++) {
1174 		pool_id = otx2_get_pool_idx(pfvf, AURA_NIX_RQ, rq);
1175 		/* Initialize aura context */
1176 		err = otx2_aura_init(pfvf, pool_id, pool_id, num_ptrs);
1177 		if (err)
1178 			goto fail;
1179 	}
1180 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1181 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
1182 				     num_ptrs, pfvf->rbsize);
1183 		if (err)
1184 			goto fail;
1185 	}
1186 
1187 	/* Flush accumulated messages */
1188 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1189 	if (err)
1190 		goto fail;
1191 
1192 	/* Allocate pointers and free them to aura/pool */
1193 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
1194 		pool = &pfvf->qset.pool[pool_id];
1195 		for (ptr = 0; ptr < num_ptrs; ptr++) {
1196 			bufptr = otx2_alloc_rbuf(pfvf, pool, GFP_KERNEL);
1197 			if (bufptr <= 0)
1198 				return bufptr;
1199 			otx2_aura_freeptr(pfvf, pool_id,
1200 					  bufptr + OTX2_HEAD_ROOM);
1201 		}
1202 		otx2_get_page(pool);
1203 	}
1204 
1205 	return 0;
1206 fail:
1207 	otx2_mbox_reset(&pfvf->mbox.mbox, 0);
1208 	otx2_aura_pool_free(pfvf);
1209 	return err;
1210 }
1211 
1212 int otx2_config_npa(struct otx2_nic *pfvf)
1213 {
1214 	struct otx2_qset *qset = &pfvf->qset;
1215 	struct npa_lf_alloc_req  *npalf;
1216 	struct otx2_hw *hw = &pfvf->hw;
1217 	int aura_cnt;
1218 
1219 	/* Pool - Stack of free buffer pointers
1220 	 * Aura - Alloc/frees pointers from/to pool for NIX DMA.
1221 	 */
1222 
1223 	if (!hw->pool_cnt)
1224 		return -EINVAL;
1225 
1226 	qset->pool = devm_kzalloc(pfvf->dev, sizeof(struct otx2_pool) *
1227 				  hw->pool_cnt, GFP_KERNEL);
1228 	if (!qset->pool)
1229 		return -ENOMEM;
1230 
1231 	/* Get memory to put this msg */
1232 	npalf = otx2_mbox_alloc_msg_npa_lf_alloc(&pfvf->mbox);
1233 	if (!npalf)
1234 		return -ENOMEM;
1235 
1236 	/* Set aura and pool counts */
1237 	npalf->nr_pools = hw->pool_cnt;
1238 	aura_cnt = ilog2(roundup_pow_of_two(hw->pool_cnt));
1239 	npalf->aura_sz = (aura_cnt >= ilog2(128)) ? (aura_cnt - 6) : 1;
1240 
1241 	return otx2_sync_mbox_msg(&pfvf->mbox);
1242 }
1243 
1244 int otx2_detach_resources(struct mbox *mbox)
1245 {
1246 	struct rsrc_detach *detach;
1247 
1248 	otx2_mbox_lock(mbox);
1249 	detach = otx2_mbox_alloc_msg_detach_resources(mbox);
1250 	if (!detach) {
1251 		otx2_mbox_unlock(mbox);
1252 		return -ENOMEM;
1253 	}
1254 
1255 	/* detach all */
1256 	detach->partial = false;
1257 
1258 	/* Send detach request to AF */
1259 	otx2_mbox_msg_send(&mbox->mbox, 0);
1260 	otx2_mbox_unlock(mbox);
1261 	return 0;
1262 }
1263 
1264 int otx2_attach_npa_nix(struct otx2_nic *pfvf)
1265 {
1266 	struct rsrc_attach *attach;
1267 	struct msg_req *msix;
1268 	int err;
1269 
1270 	otx2_mbox_lock(&pfvf->mbox);
1271 	/* Get memory to put this msg */
1272 	attach = otx2_mbox_alloc_msg_attach_resources(&pfvf->mbox);
1273 	if (!attach) {
1274 		otx2_mbox_unlock(&pfvf->mbox);
1275 		return -ENOMEM;
1276 	}
1277 
1278 	attach->npalf = true;
1279 	attach->nixlf = true;
1280 
1281 	/* Send attach request to AF */
1282 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1283 	if (err) {
1284 		otx2_mbox_unlock(&pfvf->mbox);
1285 		return err;
1286 	}
1287 
1288 	pfvf->nix_blkaddr = BLKADDR_NIX0;
1289 
1290 	/* If the platform has two NIX blocks then LF may be
1291 	 * allocated from NIX1.
1292 	 */
1293 	if (otx2_read64(pfvf, RVU_PF_BLOCK_ADDRX_DISC(BLKADDR_NIX1)) & 0x1FFULL)
1294 		pfvf->nix_blkaddr = BLKADDR_NIX1;
1295 
1296 	/* Get NPA and NIX MSIX vector offsets */
1297 	msix = otx2_mbox_alloc_msg_msix_offset(&pfvf->mbox);
1298 	if (!msix) {
1299 		otx2_mbox_unlock(&pfvf->mbox);
1300 		return -ENOMEM;
1301 	}
1302 
1303 	err = otx2_sync_mbox_msg(&pfvf->mbox);
1304 	if (err) {
1305 		otx2_mbox_unlock(&pfvf->mbox);
1306 		return err;
1307 	}
1308 	otx2_mbox_unlock(&pfvf->mbox);
1309 
1310 	if (pfvf->hw.npa_msixoff == MSIX_VECTOR_INVALID ||
1311 	    pfvf->hw.nix_msixoff == MSIX_VECTOR_INVALID) {
1312 		dev_err(pfvf->dev,
1313 			"RVUPF: Invalid MSIX vector offset for NPA/NIX\n");
1314 		return -EINVAL;
1315 	}
1316 
1317 	return 0;
1318 }
1319 
1320 void otx2_ctx_disable(struct mbox *mbox, int type, bool npa)
1321 {
1322 	struct hwctx_disable_req *req;
1323 
1324 	otx2_mbox_lock(mbox);
1325 	/* Request AQ to disable this context */
1326 	if (npa)
1327 		req = otx2_mbox_alloc_msg_npa_hwctx_disable(mbox);
1328 	else
1329 		req = otx2_mbox_alloc_msg_nix_hwctx_disable(mbox);
1330 
1331 	if (!req) {
1332 		otx2_mbox_unlock(mbox);
1333 		return;
1334 	}
1335 
1336 	req->ctype = type;
1337 
1338 	if (otx2_sync_mbox_msg(mbox))
1339 		dev_err(mbox->pfvf->dev, "%s failed to disable context\n",
1340 			__func__);
1341 
1342 	otx2_mbox_unlock(mbox);
1343 }
1344 
1345 int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
1346 {
1347 	struct nix_bp_cfg_req *req;
1348 
1349 	if (enable)
1350 		req = otx2_mbox_alloc_msg_nix_bp_enable(&pfvf->mbox);
1351 	else
1352 		req = otx2_mbox_alloc_msg_nix_bp_disable(&pfvf->mbox);
1353 
1354 	if (!req)
1355 		return -ENOMEM;
1356 
1357 	req->chan_base = 0;
1358 	req->chan_cnt = 1;
1359 	req->bpid_per_chan = 0;
1360 
1361 	return otx2_sync_mbox_msg(&pfvf->mbox);
1362 }
1363 
1364 /* Mbox message handlers */
1365 void mbox_handler_cgx_stats(struct otx2_nic *pfvf,
1366 			    struct cgx_stats_rsp *rsp)
1367 {
1368 	int id;
1369 
1370 	for (id = 0; id < CGX_RX_STATS_COUNT; id++)
1371 		pfvf->hw.cgx_rx_stats[id] = rsp->rx_stats[id];
1372 	for (id = 0; id < CGX_TX_STATS_COUNT; id++)
1373 		pfvf->hw.cgx_tx_stats[id] = rsp->tx_stats[id];
1374 }
1375 
1376 void mbox_handler_nix_txsch_alloc(struct otx2_nic *pf,
1377 				  struct nix_txsch_alloc_rsp *rsp)
1378 {
1379 	int lvl, schq;
1380 
1381 	/* Setup transmit scheduler list */
1382 	for (lvl = 0; lvl < NIX_TXSCH_LVL_CNT; lvl++)
1383 		for (schq = 0; schq < rsp->schq[lvl]; schq++)
1384 			pf->hw.txschq_list[lvl][schq] =
1385 				rsp->schq_list[lvl][schq];
1386 }
1387 
1388 void mbox_handler_npa_lf_alloc(struct otx2_nic *pfvf,
1389 			       struct npa_lf_alloc_rsp *rsp)
1390 {
1391 	pfvf->hw.stack_pg_ptrs = rsp->stack_pg_ptrs;
1392 	pfvf->hw.stack_pg_bytes = rsp->stack_pg_bytes;
1393 }
1394 
1395 void mbox_handler_nix_lf_alloc(struct otx2_nic *pfvf,
1396 			       struct nix_lf_alloc_rsp *rsp)
1397 {
1398 	pfvf->hw.sqb_size = rsp->sqb_size;
1399 	pfvf->hw.rx_chan_base = rsp->rx_chan_base;
1400 	pfvf->hw.tx_chan_base = rsp->tx_chan_base;
1401 	pfvf->hw.lso_tsov4_idx = rsp->lso_tsov4_idx;
1402 	pfvf->hw.lso_tsov6_idx = rsp->lso_tsov6_idx;
1403 }
1404 
1405 void mbox_handler_msix_offset(struct otx2_nic *pfvf,
1406 			      struct msix_offset_rsp *rsp)
1407 {
1408 	pfvf->hw.npa_msixoff = rsp->npa_msixoff;
1409 	pfvf->hw.nix_msixoff = rsp->nix_msixoff;
1410 }
1411 
1412 void mbox_handler_nix_bp_enable(struct otx2_nic *pfvf,
1413 				struct nix_bp_cfg_rsp *rsp)
1414 {
1415 	int chan, chan_id;
1416 
1417 	for (chan = 0; chan < rsp->chan_cnt; chan++) {
1418 		chan_id = ((rsp->chan_bpid[chan] >> 10) & 0x7F);
1419 		pfvf->bpid[chan_id] = rsp->chan_bpid[chan] & 0x3FF;
1420 	}
1421 }
1422 
1423 void otx2_free_cints(struct otx2_nic *pfvf, int n)
1424 {
1425 	struct otx2_qset *qset = &pfvf->qset;
1426 	struct otx2_hw *hw = &pfvf->hw;
1427 	int irq, qidx;
1428 
1429 	for (qidx = 0, irq = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1430 	     qidx < n;
1431 	     qidx++, irq++) {
1432 		int vector = pci_irq_vector(pfvf->pdev, irq);
1433 
1434 		irq_set_affinity_hint(vector, NULL);
1435 		free_cpumask_var(hw->affinity_mask[irq]);
1436 		free_irq(vector, &qset->napi[qidx]);
1437 	}
1438 }
1439 
1440 void otx2_set_cints_affinity(struct otx2_nic *pfvf)
1441 {
1442 	struct otx2_hw *hw = &pfvf->hw;
1443 	int vec, cpu, irq, cint;
1444 
1445 	vec = hw->nix_msixoff + NIX_LF_CINT_VEC_START;
1446 	cpu = cpumask_first(cpu_online_mask);
1447 
1448 	/* CQ interrupts */
1449 	for (cint = 0; cint < pfvf->hw.cint_cnt; cint++, vec++) {
1450 		if (!alloc_cpumask_var(&hw->affinity_mask[vec], GFP_KERNEL))
1451 			return;
1452 
1453 		cpumask_set_cpu(cpu, hw->affinity_mask[vec]);
1454 
1455 		irq = pci_irq_vector(pfvf->pdev, vec);
1456 		irq_set_affinity_hint(irq, hw->affinity_mask[vec]);
1457 
1458 		cpu = cpumask_next(cpu, cpu_online_mask);
1459 		if (unlikely(cpu >= nr_cpu_ids))
1460 			cpu = 0;
1461 	}
1462 }
1463 
1464 #define M(_name, _id, _fn_name, _req_type, _rsp_type)			\
1465 int __weak								\
1466 otx2_mbox_up_handler_ ## _fn_name(struct otx2_nic *pfvf,		\
1467 				struct _req_type *req,			\
1468 				struct _rsp_type *rsp)			\
1469 {									\
1470 	/* Nothing to do here */					\
1471 	return 0;							\
1472 }									\
1473 EXPORT_SYMBOL(otx2_mbox_up_handler_ ## _fn_name);
1474 MBOX_UP_CGX_MESSAGES
1475 #undef M
1476