xref: /linux/drivers/net/ethernet/microsoft/mana/mana_en.c (revision 0e50474fa514822e9d990874e554bf8043a201d7)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /* Copyright (c) 2021, Microsoft Corporation. */
3 
4 #include <uapi/linux/bpf.h>
5 
6 #include <linux/debugfs.h>
7 #include <linux/inetdevice.h>
8 #include <linux/etherdevice.h>
9 #include <linux/ethtool.h>
10 #include <linux/filter.h>
11 #include <linux/mm.h>
12 #include <linux/pci.h>
13 #include <linux/export.h>
14 
15 #include <net/checksum.h>
16 #include <net/ip6_checksum.h>
17 #include <net/netdev_lock.h>
18 #include <net/page_pool/helpers.h>
19 #include <net/xdp.h>
20 
21 #include <net/mana/mana.h>
22 #include <net/mana/mana_auxiliary.h>
23 #include <net/mana/hw_channel.h>
24 
25 static DEFINE_IDA(mana_adev_ida);
26 
27 static int mana_adev_idx_alloc(void)
28 {
29 	return ida_alloc(&mana_adev_ida, GFP_KERNEL);
30 }
31 
32 static void mana_adev_idx_free(int idx)
33 {
34 	ida_free(&mana_adev_ida, idx);
35 }
36 
37 static ssize_t mana_dbg_q_read(struct file *filp, char __user *buf, size_t count,
38 			       loff_t *pos)
39 {
40 	struct gdma_queue *gdma_q = filp->private_data;
41 
42 	return simple_read_from_buffer(buf, count, pos, gdma_q->queue_mem_ptr,
43 				       gdma_q->queue_size);
44 }
45 
46 static const struct file_operations mana_dbg_q_fops = {
47 	.owner  = THIS_MODULE,
48 	.open   = simple_open,
49 	.read   = mana_dbg_q_read,
50 };
51 
52 static bool mana_en_need_log(struct mana_port_context *apc, int err)
53 {
54 	if (apc && apc->ac && apc->ac->gdma_dev &&
55 	    apc->ac->gdma_dev->gdma_context)
56 		return mana_need_log(apc->ac->gdma_dev->gdma_context, err);
57 	else
58 		return true;
59 }
60 
61 static void mana_put_rx_page(struct mana_rxq *rxq, struct page *page,
62 			     bool from_pool)
63 {
64 	if (from_pool)
65 		page_pool_put_full_page(rxq->page_pool, page, false);
66 	else
67 		put_page(page);
68 }
69 
70 /* Microsoft Azure Network Adapter (MANA) functions */
71 
72 static int mana_open(struct net_device *ndev)
73 {
74 	struct mana_port_context *apc = netdev_priv(ndev);
75 	int err;
76 	err = mana_alloc_queues(ndev);
77 
78 	if (err) {
79 		netdev_err(ndev, "%s failed to allocate queues: %d\n", __func__, err);
80 		return err;
81 	}
82 
83 	apc->port_is_up = true;
84 
85 	/* Ensure port state updated before txq state */
86 	smp_wmb();
87 
88 	netif_tx_wake_all_queues(ndev);
89 	netdev_dbg(ndev, "%s successful\n", __func__);
90 	return 0;
91 }
92 
93 static int mana_close(struct net_device *ndev)
94 {
95 	struct mana_port_context *apc = netdev_priv(ndev);
96 
97 	if (!apc->port_is_up)
98 		return 0;
99 
100 	return mana_detach(ndev, true);
101 }
102 
103 static void mana_link_state_handle(struct work_struct *w)
104 {
105 	struct mana_context *ac;
106 	struct net_device *ndev;
107 	u32 link_event;
108 	bool link_up;
109 	int i;
110 
111 	ac = container_of(w, struct mana_context, link_change_work);
112 
113 	rtnl_lock();
114 
115 	link_event = READ_ONCE(ac->link_event);
116 
117 	if (link_event == HWC_DATA_HW_LINK_CONNECT)
118 		link_up = true;
119 	else if (link_event == HWC_DATA_HW_LINK_DISCONNECT)
120 		link_up = false;
121 	else
122 		goto out;
123 
124 	/* Process all ports */
125 	for (i = 0; i < ac->num_ports; i++) {
126 		ndev = ac->ports[i];
127 		if (!ndev)
128 			continue;
129 
130 		if (link_up) {
131 			netif_carrier_on(ndev);
132 
133 			__netdev_notify_peers(ndev);
134 		} else {
135 			netif_carrier_off(ndev);
136 		}
137 	}
138 
139 out:
140 	rtnl_unlock();
141 }
142 
143 static bool mana_can_tx(struct gdma_queue *wq)
144 {
145 	return mana_gd_wq_avail_space(wq) >= MAX_TX_WQE_SIZE;
146 }
147 
148 static unsigned int mana_checksum_info(struct sk_buff *skb)
149 {
150 	if (skb->protocol == htons(ETH_P_IP)) {
151 		struct iphdr *ip = ip_hdr(skb);
152 
153 		if (ip->protocol == IPPROTO_TCP)
154 			return IPPROTO_TCP;
155 
156 		if (ip->protocol == IPPROTO_UDP)
157 			return IPPROTO_UDP;
158 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
159 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
160 
161 		if (ip6->nexthdr == IPPROTO_TCP)
162 			return IPPROTO_TCP;
163 
164 		if (ip6->nexthdr == IPPROTO_UDP)
165 			return IPPROTO_UDP;
166 	}
167 
168 	/* No csum offloading */
169 	return 0;
170 }
171 
172 static void mana_add_sge(struct mana_tx_package *tp, struct mana_skb_head *ash,
173 			 int sg_i, dma_addr_t da, int sge_len, u32 gpa_mkey)
174 {
175 	ash->dma_handle[sg_i] = da;
176 	ash->size[sg_i] = sge_len;
177 
178 	tp->wqe_req.sgl[sg_i].address = da;
179 	tp->wqe_req.sgl[sg_i].mem_key = gpa_mkey;
180 	tp->wqe_req.sgl[sg_i].size = sge_len;
181 }
182 
183 static int mana_map_skb(struct sk_buff *skb, struct mana_port_context *apc,
184 			struct mana_tx_package *tp, int gso_hs)
185 {
186 	struct mana_skb_head *ash = (struct mana_skb_head *)skb->head;
187 	int hsg = 1; /* num of SGEs of linear part */
188 	struct gdma_dev *gd = apc->ac->gdma_dev;
189 	int skb_hlen = skb_headlen(skb);
190 	int sge0_len, sge1_len = 0;
191 	struct gdma_context *gc;
192 	struct device *dev;
193 	skb_frag_t *frag;
194 	dma_addr_t da;
195 	int sg_i;
196 	int i;
197 
198 	gc = gd->gdma_context;
199 	dev = gc->dev;
200 
201 	if (gso_hs && gso_hs < skb_hlen) {
202 		sge0_len = gso_hs;
203 		sge1_len = skb_hlen - gso_hs;
204 	} else {
205 		sge0_len = skb_hlen;
206 	}
207 
208 	da = dma_map_single(dev, skb->data, sge0_len, DMA_TO_DEVICE);
209 	if (dma_mapping_error(dev, da))
210 		return -ENOMEM;
211 
212 	mana_add_sge(tp, ash, 0, da, sge0_len, gd->gpa_mkey);
213 
214 	if (sge1_len) {
215 		sg_i = 1;
216 		da = dma_map_single(dev, skb->data + sge0_len, sge1_len,
217 				    DMA_TO_DEVICE);
218 		if (dma_mapping_error(dev, da))
219 			goto frag_err;
220 
221 		mana_add_sge(tp, ash, sg_i, da, sge1_len, gd->gpa_mkey);
222 		hsg = 2;
223 	}
224 
225 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
226 		sg_i = hsg + i;
227 
228 		frag = &skb_shinfo(skb)->frags[i];
229 		da = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
230 				      DMA_TO_DEVICE);
231 		if (dma_mapping_error(dev, da))
232 			goto frag_err;
233 
234 		mana_add_sge(tp, ash, sg_i, da, skb_frag_size(frag),
235 			     gd->gpa_mkey);
236 	}
237 
238 	return 0;
239 
240 frag_err:
241 	if (net_ratelimit())
242 		netdev_err(apc->ndev, "Failed to map skb of size %u to DMA\n",
243 			   skb->len);
244 	for (i = sg_i - 1; i >= hsg; i--)
245 		dma_unmap_page(dev, ash->dma_handle[i], ash->size[i],
246 			       DMA_TO_DEVICE);
247 
248 	for (i = hsg - 1; i >= 0; i--)
249 		dma_unmap_single(dev, ash->dma_handle[i], ash->size[i],
250 				 DMA_TO_DEVICE);
251 
252 	return -ENOMEM;
253 }
254 
255 /* Handle the case when GSO SKB linear length is too large.
256  * MANA NIC requires GSO packets to put only the packet header to SGE0.
257  * So, we need 2 SGEs for the skb linear part which contains more than the
258  * header.
259  * Return a positive value for the number of SGEs, or a negative value
260  * for an error.
261  */
262 static int mana_fix_skb_head(struct net_device *ndev, struct sk_buff *skb,
263 			     int gso_hs)
264 {
265 	int num_sge = 1 + skb_shinfo(skb)->nr_frags;
266 	int skb_hlen = skb_headlen(skb);
267 
268 	if (gso_hs < skb_hlen) {
269 		num_sge++;
270 	} else if (gso_hs > skb_hlen) {
271 		if (net_ratelimit())
272 			netdev_err(ndev,
273 				   "TX nonlinear head: hs:%d, skb_hlen:%d\n",
274 				   gso_hs, skb_hlen);
275 
276 		return -EINVAL;
277 	}
278 
279 	return num_sge;
280 }
281 
282 /* Get the GSO packet's header size */
283 static int mana_get_gso_hs(struct sk_buff *skb)
284 {
285 	int gso_hs;
286 
287 	if (skb->encapsulation) {
288 		gso_hs = skb_inner_tcp_all_headers(skb);
289 	} else {
290 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
291 			gso_hs = skb_transport_offset(skb) +
292 				 sizeof(struct udphdr);
293 		} else {
294 			gso_hs = skb_tcp_all_headers(skb);
295 		}
296 	}
297 
298 	return gso_hs;
299 }
300 
301 netdev_tx_t mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
302 {
303 	enum mana_tx_pkt_format pkt_fmt = MANA_SHORT_PKT_FMT;
304 	struct mana_port_context *apc = netdev_priv(ndev);
305 	int gso_hs = 0; /* zero for non-GSO pkts */
306 	u16 txq_idx = skb_get_queue_mapping(skb);
307 	struct gdma_dev *gd = apc->ac->gdma_dev;
308 	bool ipv4 = false, ipv6 = false;
309 	struct mana_tx_package pkg = {};
310 	struct netdev_queue *net_txq;
311 	struct mana_stats_tx *tx_stats;
312 	struct gdma_queue *gdma_sq;
313 	int err, len, num_gso_seg;
314 	unsigned int csum_type;
315 	struct mana_txq *txq;
316 	struct mana_cq *cq;
317 
318 	if (unlikely(!apc->port_is_up))
319 		goto tx_drop;
320 
321 	if (skb_cow_head(skb, MANA_HEADROOM))
322 		goto tx_drop_count;
323 
324 	if (unlikely(ipv6_hopopt_jumbo_remove(skb)))
325 		goto tx_drop_count;
326 
327 	txq = &apc->tx_qp[txq_idx].txq;
328 	gdma_sq = txq->gdma_sq;
329 	cq = &apc->tx_qp[txq_idx].tx_cq;
330 	tx_stats = &txq->stats;
331 
332 	pkg.tx_oob.s_oob.vcq_num = cq->gdma_id;
333 	pkg.tx_oob.s_oob.vsq_frame = txq->vsq_frame;
334 
335 	if (txq->vp_offset > MANA_SHORT_VPORT_OFFSET_MAX) {
336 		pkg.tx_oob.l_oob.long_vp_offset = txq->vp_offset;
337 		pkt_fmt = MANA_LONG_PKT_FMT;
338 	} else {
339 		pkg.tx_oob.s_oob.short_vp_offset = txq->vp_offset;
340 	}
341 
342 	if (skb_vlan_tag_present(skb)) {
343 		pkt_fmt = MANA_LONG_PKT_FMT;
344 		pkg.tx_oob.l_oob.inject_vlan_pri_tag = 1;
345 		pkg.tx_oob.l_oob.pcp = skb_vlan_tag_get_prio(skb);
346 		pkg.tx_oob.l_oob.dei = skb_vlan_tag_get_cfi(skb);
347 		pkg.tx_oob.l_oob.vlan_id = skb_vlan_tag_get_id(skb);
348 	}
349 
350 	pkg.tx_oob.s_oob.pkt_fmt = pkt_fmt;
351 
352 	if (pkt_fmt == MANA_SHORT_PKT_FMT) {
353 		pkg.wqe_req.inline_oob_size = sizeof(struct mana_tx_short_oob);
354 		u64_stats_update_begin(&tx_stats->syncp);
355 		tx_stats->short_pkt_fmt++;
356 		u64_stats_update_end(&tx_stats->syncp);
357 	} else {
358 		pkg.wqe_req.inline_oob_size = sizeof(struct mana_tx_oob);
359 		u64_stats_update_begin(&tx_stats->syncp);
360 		tx_stats->long_pkt_fmt++;
361 		u64_stats_update_end(&tx_stats->syncp);
362 	}
363 
364 	pkg.wqe_req.inline_oob_data = &pkg.tx_oob;
365 	pkg.wqe_req.flags = 0;
366 	pkg.wqe_req.client_data_unit = 0;
367 
368 	pkg.wqe_req.num_sge = 1 + skb_shinfo(skb)->nr_frags;
369 
370 	if (skb->protocol == htons(ETH_P_IP))
371 		ipv4 = true;
372 	else if (skb->protocol == htons(ETH_P_IPV6))
373 		ipv6 = true;
374 
375 	if (skb_is_gso(skb)) {
376 		int num_sge;
377 
378 		gso_hs = mana_get_gso_hs(skb);
379 
380 		num_sge = mana_fix_skb_head(ndev, skb, gso_hs);
381 		if (num_sge > 0)
382 			pkg.wqe_req.num_sge = num_sge;
383 		else
384 			goto tx_drop_count;
385 
386 		u64_stats_update_begin(&tx_stats->syncp);
387 		if (skb->encapsulation) {
388 			tx_stats->tso_inner_packets++;
389 			tx_stats->tso_inner_bytes += skb->len - gso_hs;
390 		} else {
391 			tx_stats->tso_packets++;
392 			tx_stats->tso_bytes += skb->len - gso_hs;
393 		}
394 		u64_stats_update_end(&tx_stats->syncp);
395 
396 		pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
397 		pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
398 
399 		pkg.tx_oob.s_oob.comp_iphdr_csum = 1;
400 		pkg.tx_oob.s_oob.comp_tcp_csum = 1;
401 		pkg.tx_oob.s_oob.trans_off = skb_transport_offset(skb);
402 
403 		pkg.wqe_req.client_data_unit = skb_shinfo(skb)->gso_size;
404 		pkg.wqe_req.flags = GDMA_WR_OOB_IN_SGL | GDMA_WR_PAD_BY_SGE0;
405 		if (ipv4) {
406 			ip_hdr(skb)->tot_len = 0;
407 			ip_hdr(skb)->check = 0;
408 			tcp_hdr(skb)->check =
409 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
410 						   ip_hdr(skb)->daddr, 0,
411 						   IPPROTO_TCP, 0);
412 		} else {
413 			ipv6_hdr(skb)->payload_len = 0;
414 			tcp_hdr(skb)->check =
415 				~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
416 						 &ipv6_hdr(skb)->daddr, 0,
417 						 IPPROTO_TCP, 0);
418 		}
419 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
420 		csum_type = mana_checksum_info(skb);
421 
422 		u64_stats_update_begin(&tx_stats->syncp);
423 		tx_stats->csum_partial++;
424 		u64_stats_update_end(&tx_stats->syncp);
425 
426 		if (csum_type == IPPROTO_TCP) {
427 			pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
428 			pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
429 
430 			pkg.tx_oob.s_oob.comp_tcp_csum = 1;
431 			pkg.tx_oob.s_oob.trans_off = skb_transport_offset(skb);
432 
433 		} else if (csum_type == IPPROTO_UDP) {
434 			pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
435 			pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
436 
437 			pkg.tx_oob.s_oob.comp_udp_csum = 1;
438 		} else {
439 			/* Can't do offload of this type of checksum */
440 			if (skb_checksum_help(skb))
441 				goto tx_drop_count;
442 		}
443 	}
444 
445 	WARN_ON_ONCE(pkg.wqe_req.num_sge > MAX_TX_WQE_SGL_ENTRIES);
446 
447 	if (pkg.wqe_req.num_sge <= ARRAY_SIZE(pkg.sgl_array)) {
448 		pkg.wqe_req.sgl = pkg.sgl_array;
449 	} else {
450 		pkg.sgl_ptr = kmalloc_array(pkg.wqe_req.num_sge,
451 					    sizeof(struct gdma_sge),
452 					    GFP_ATOMIC);
453 		if (!pkg.sgl_ptr)
454 			goto tx_drop_count;
455 
456 		pkg.wqe_req.sgl = pkg.sgl_ptr;
457 	}
458 
459 	if (mana_map_skb(skb, apc, &pkg, gso_hs)) {
460 		u64_stats_update_begin(&tx_stats->syncp);
461 		tx_stats->mana_map_err++;
462 		u64_stats_update_end(&tx_stats->syncp);
463 		goto free_sgl_ptr;
464 	}
465 
466 	skb_queue_tail(&txq->pending_skbs, skb);
467 
468 	len = skb->len;
469 	num_gso_seg = skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1;
470 	net_txq = netdev_get_tx_queue(ndev, txq_idx);
471 
472 	err = mana_gd_post_work_request(gdma_sq, &pkg.wqe_req,
473 					(struct gdma_posted_wqe_info *)skb->cb);
474 	if (!mana_can_tx(gdma_sq)) {
475 		netif_tx_stop_queue(net_txq);
476 		apc->eth_stats.stop_queue++;
477 	}
478 
479 	if (err) {
480 		(void)skb_dequeue_tail(&txq->pending_skbs);
481 		netdev_warn(ndev, "Failed to post TX OOB: %d\n", err);
482 		err = NETDEV_TX_BUSY;
483 		goto tx_busy;
484 	}
485 
486 	err = NETDEV_TX_OK;
487 	atomic_inc(&txq->pending_sends);
488 
489 	mana_gd_wq_ring_doorbell(gd->gdma_context, gdma_sq);
490 
491 	/* skb may be freed after mana_gd_post_work_request. Do not use it. */
492 	skb = NULL;
493 
494 	/* Populated the packet and bytes counters based on post GSO packet
495 	 * calculations
496 	 */
497 	tx_stats = &txq->stats;
498 	u64_stats_update_begin(&tx_stats->syncp);
499 	tx_stats->packets += num_gso_seg;
500 	tx_stats->bytes += len + ((num_gso_seg - 1) * gso_hs);
501 	u64_stats_update_end(&tx_stats->syncp);
502 
503 tx_busy:
504 	if (netif_tx_queue_stopped(net_txq) && mana_can_tx(gdma_sq)) {
505 		netif_tx_wake_queue(net_txq);
506 		apc->eth_stats.wake_queue++;
507 	}
508 
509 	kfree(pkg.sgl_ptr);
510 	return err;
511 
512 free_sgl_ptr:
513 	kfree(pkg.sgl_ptr);
514 tx_drop_count:
515 	ndev->stats.tx_dropped++;
516 tx_drop:
517 	dev_kfree_skb_any(skb);
518 	return NETDEV_TX_OK;
519 }
520 
521 static void mana_get_stats64(struct net_device *ndev,
522 			     struct rtnl_link_stats64 *st)
523 {
524 	struct mana_port_context *apc = netdev_priv(ndev);
525 	unsigned int num_queues = apc->num_queues;
526 	struct mana_stats_rx *rx_stats;
527 	struct mana_stats_tx *tx_stats;
528 	unsigned int start;
529 	u64 packets, bytes;
530 	int q;
531 
532 	if (!apc->port_is_up)
533 		return;
534 
535 	netdev_stats_to_stats64(st, &ndev->stats);
536 
537 	for (q = 0; q < num_queues; q++) {
538 		rx_stats = &apc->rxqs[q]->stats;
539 
540 		do {
541 			start = u64_stats_fetch_begin(&rx_stats->syncp);
542 			packets = rx_stats->packets;
543 			bytes = rx_stats->bytes;
544 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
545 
546 		st->rx_packets += packets;
547 		st->rx_bytes += bytes;
548 	}
549 
550 	for (q = 0; q < num_queues; q++) {
551 		tx_stats = &apc->tx_qp[q].txq.stats;
552 
553 		do {
554 			start = u64_stats_fetch_begin(&tx_stats->syncp);
555 			packets = tx_stats->packets;
556 			bytes = tx_stats->bytes;
557 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
558 
559 		st->tx_packets += packets;
560 		st->tx_bytes += bytes;
561 	}
562 }
563 
564 static int mana_get_tx_queue(struct net_device *ndev, struct sk_buff *skb,
565 			     int old_q)
566 {
567 	struct mana_port_context *apc = netdev_priv(ndev);
568 	u32 hash = skb_get_hash(skb);
569 	struct sock *sk = skb->sk;
570 	int txq;
571 
572 	txq = apc->indir_table[hash & (apc->indir_table_sz - 1)];
573 
574 	if (txq != old_q && sk && sk_fullsock(sk) &&
575 	    rcu_access_pointer(sk->sk_dst_cache))
576 		sk_tx_queue_set(sk, txq);
577 
578 	return txq;
579 }
580 
581 static u16 mana_select_queue(struct net_device *ndev, struct sk_buff *skb,
582 			     struct net_device *sb_dev)
583 {
584 	int txq;
585 
586 	if (ndev->real_num_tx_queues == 1)
587 		return 0;
588 
589 	txq = sk_tx_queue_get(skb->sk);
590 
591 	if (txq < 0 || skb->ooo_okay || txq >= ndev->real_num_tx_queues) {
592 		if (skb_rx_queue_recorded(skb))
593 			txq = skb_get_rx_queue(skb);
594 		else
595 			txq = mana_get_tx_queue(ndev, skb, txq);
596 	}
597 
598 	return txq;
599 }
600 
601 /* Release pre-allocated RX buffers */
602 void mana_pre_dealloc_rxbufs(struct mana_port_context *mpc)
603 {
604 	struct device *dev;
605 	int i;
606 
607 	dev = mpc->ac->gdma_dev->gdma_context->dev;
608 
609 	if (!mpc->rxbufs_pre)
610 		goto out1;
611 
612 	if (!mpc->das_pre)
613 		goto out2;
614 
615 	while (mpc->rxbpre_total) {
616 		i = --mpc->rxbpre_total;
617 		dma_unmap_single(dev, mpc->das_pre[i], mpc->rxbpre_datasize,
618 				 DMA_FROM_DEVICE);
619 		put_page(virt_to_head_page(mpc->rxbufs_pre[i]));
620 	}
621 
622 	kfree(mpc->das_pre);
623 	mpc->das_pre = NULL;
624 
625 out2:
626 	kfree(mpc->rxbufs_pre);
627 	mpc->rxbufs_pre = NULL;
628 
629 out1:
630 	mpc->rxbpre_datasize = 0;
631 	mpc->rxbpre_alloc_size = 0;
632 	mpc->rxbpre_headroom = 0;
633 }
634 
635 /* Get a buffer from the pre-allocated RX buffers */
636 static void *mana_get_rxbuf_pre(struct mana_rxq *rxq, dma_addr_t *da)
637 {
638 	struct net_device *ndev = rxq->ndev;
639 	struct mana_port_context *mpc;
640 	void *va;
641 
642 	mpc = netdev_priv(ndev);
643 
644 	if (!mpc->rxbufs_pre || !mpc->das_pre || !mpc->rxbpre_total) {
645 		netdev_err(ndev, "No RX pre-allocated bufs\n");
646 		return NULL;
647 	}
648 
649 	/* Check sizes to catch unexpected coding error */
650 	if (mpc->rxbpre_datasize != rxq->datasize) {
651 		netdev_err(ndev, "rxbpre_datasize mismatch: %u: %u\n",
652 			   mpc->rxbpre_datasize, rxq->datasize);
653 		return NULL;
654 	}
655 
656 	if (mpc->rxbpre_alloc_size != rxq->alloc_size) {
657 		netdev_err(ndev, "rxbpre_alloc_size mismatch: %u: %u\n",
658 			   mpc->rxbpre_alloc_size, rxq->alloc_size);
659 		return NULL;
660 	}
661 
662 	if (mpc->rxbpre_headroom != rxq->headroom) {
663 		netdev_err(ndev, "rxbpre_headroom mismatch: %u: %u\n",
664 			   mpc->rxbpre_headroom, rxq->headroom);
665 		return NULL;
666 	}
667 
668 	mpc->rxbpre_total--;
669 
670 	*da = mpc->das_pre[mpc->rxbpre_total];
671 	va = mpc->rxbufs_pre[mpc->rxbpre_total];
672 	mpc->rxbufs_pre[mpc->rxbpre_total] = NULL;
673 
674 	/* Deallocate the array after all buffers are gone */
675 	if (!mpc->rxbpre_total)
676 		mana_pre_dealloc_rxbufs(mpc);
677 
678 	return va;
679 }
680 
681 /* Get RX buffer's data size, alloc size, XDP headroom based on MTU */
682 static void mana_get_rxbuf_cfg(struct mana_port_context *apc,
683 			       int mtu, u32 *datasize, u32 *alloc_size,
684 			       u32 *headroom, u32 *frag_count)
685 {
686 	u32 len, buf_size;
687 
688 	/* Calculate datasize first (consistent across all cases) */
689 	*datasize = mtu + ETH_HLEN;
690 
691 	/* For xdp and jumbo frames make sure only one packet fits per page */
692 	if (mtu + MANA_RXBUF_PAD > PAGE_SIZE / 2 || mana_xdp_get(apc)) {
693 		if (mana_xdp_get(apc)) {
694 			*headroom = XDP_PACKET_HEADROOM;
695 			*alloc_size = PAGE_SIZE;
696 		} else {
697 			*headroom = 0; /* no support for XDP */
698 			*alloc_size = SKB_DATA_ALIGN(mtu + MANA_RXBUF_PAD +
699 						     *headroom);
700 		}
701 
702 		*frag_count = 1;
703 		return;
704 	}
705 
706 	/* Standard MTU case - optimize for multiple packets per page */
707 	*headroom = 0;
708 
709 	/* Calculate base buffer size needed */
710 	len = SKB_DATA_ALIGN(mtu + MANA_RXBUF_PAD + *headroom);
711 	buf_size = ALIGN(len, MANA_RX_FRAG_ALIGNMENT);
712 
713 	/* Calculate how many packets can fit in a page */
714 	*frag_count = PAGE_SIZE / buf_size;
715 	*alloc_size = buf_size;
716 }
717 
718 int mana_pre_alloc_rxbufs(struct mana_port_context *mpc, int new_mtu, int num_queues)
719 {
720 	struct device *dev;
721 	struct page *page;
722 	dma_addr_t da;
723 	int num_rxb;
724 	void *va;
725 	int i;
726 
727 	mana_get_rxbuf_cfg(mpc, new_mtu, &mpc->rxbpre_datasize,
728 			   &mpc->rxbpre_alloc_size, &mpc->rxbpre_headroom,
729 			   &mpc->rxbpre_frag_count);
730 
731 	dev = mpc->ac->gdma_dev->gdma_context->dev;
732 
733 	num_rxb = num_queues * mpc->rx_queue_size;
734 
735 	WARN(mpc->rxbufs_pre, "mana rxbufs_pre exists\n");
736 	mpc->rxbufs_pre = kmalloc_array(num_rxb, sizeof(void *), GFP_KERNEL);
737 	if (!mpc->rxbufs_pre)
738 		goto error;
739 
740 	mpc->das_pre = kmalloc_array(num_rxb, sizeof(dma_addr_t), GFP_KERNEL);
741 	if (!mpc->das_pre)
742 		goto error;
743 
744 	mpc->rxbpre_total = 0;
745 
746 	for (i = 0; i < num_rxb; i++) {
747 		page = dev_alloc_pages(get_order(mpc->rxbpre_alloc_size));
748 		if (!page)
749 			goto error;
750 
751 		va = page_to_virt(page);
752 
753 		da = dma_map_single(dev, va + mpc->rxbpre_headroom,
754 				    mpc->rxbpre_datasize, DMA_FROM_DEVICE);
755 		if (dma_mapping_error(dev, da)) {
756 			put_page(page);
757 			goto error;
758 		}
759 
760 		mpc->rxbufs_pre[i] = va;
761 		mpc->das_pre[i] = da;
762 		mpc->rxbpre_total = i + 1;
763 	}
764 
765 	return 0;
766 
767 error:
768 	netdev_err(mpc->ndev, "Failed to pre-allocate RX buffers for %d queues\n", num_queues);
769 	mana_pre_dealloc_rxbufs(mpc);
770 	return -ENOMEM;
771 }
772 
773 static int mana_change_mtu(struct net_device *ndev, int new_mtu)
774 {
775 	struct mana_port_context *mpc = netdev_priv(ndev);
776 	unsigned int old_mtu = ndev->mtu;
777 	int err;
778 
779 	/* Pre-allocate buffers to prevent failure in mana_attach later */
780 	err = mana_pre_alloc_rxbufs(mpc, new_mtu, mpc->num_queues);
781 	if (err) {
782 		netdev_err(ndev, "Insufficient memory for new MTU\n");
783 		return err;
784 	}
785 
786 	err = mana_detach(ndev, false);
787 	if (err) {
788 		netdev_err(ndev, "mana_detach failed: %d\n", err);
789 		goto out;
790 	}
791 
792 	WRITE_ONCE(ndev->mtu, new_mtu);
793 
794 	err = mana_attach(ndev);
795 	if (err) {
796 		netdev_err(ndev, "mana_attach failed: %d\n", err);
797 		WRITE_ONCE(ndev->mtu, old_mtu);
798 	}
799 
800 out:
801 	mana_pre_dealloc_rxbufs(mpc);
802 	return err;
803 }
804 
805 static int mana_shaper_set(struct net_shaper_binding *binding,
806 			   const struct net_shaper *shaper,
807 			   struct netlink_ext_ack *extack)
808 {
809 	struct mana_port_context *apc = netdev_priv(binding->netdev);
810 	u32 old_speed, rate;
811 	int err;
812 
813 	if (shaper->handle.scope != NET_SHAPER_SCOPE_NETDEV) {
814 		NL_SET_ERR_MSG_MOD(extack, "net shaper scope should be netdev");
815 		return -EINVAL;
816 	}
817 
818 	if (apc->handle.id && shaper->handle.id != apc->handle.id) {
819 		NL_SET_ERR_MSG_MOD(extack, "Cannot create multiple shapers");
820 		return -EOPNOTSUPP;
821 	}
822 
823 	if (!shaper->bw_max || (shaper->bw_max % 100000000)) {
824 		NL_SET_ERR_MSG_MOD(extack, "Please use multiples of 100Mbps for bandwidth");
825 		return -EINVAL;
826 	}
827 
828 	rate = div_u64(shaper->bw_max, 1000); /* Convert bps to Kbps */
829 	rate = div_u64(rate, 1000);	      /* Convert Kbps to Mbps */
830 
831 	/* Get current speed */
832 	err = mana_query_link_cfg(apc);
833 	old_speed = (err) ? SPEED_UNKNOWN : apc->speed;
834 
835 	if (!err) {
836 		err = mana_set_bw_clamp(apc, rate, TRI_STATE_TRUE);
837 		apc->speed = (err) ? old_speed : rate;
838 		apc->handle = (err) ? apc->handle : shaper->handle;
839 	}
840 
841 	return err;
842 }
843 
844 static int mana_shaper_del(struct net_shaper_binding *binding,
845 			   const struct net_shaper_handle *handle,
846 			   struct netlink_ext_ack *extack)
847 {
848 	struct mana_port_context *apc = netdev_priv(binding->netdev);
849 	int err;
850 
851 	err = mana_set_bw_clamp(apc, 0, TRI_STATE_FALSE);
852 
853 	if (!err) {
854 		/* Reset mana port context parameters */
855 		apc->handle.id = 0;
856 		apc->handle.scope = NET_SHAPER_SCOPE_UNSPEC;
857 		apc->speed = 0;
858 	}
859 
860 	return err;
861 }
862 
863 static void mana_shaper_cap(struct net_shaper_binding *binding,
864 			    enum net_shaper_scope scope,
865 			    unsigned long *flags)
866 {
867 	*flags = BIT(NET_SHAPER_A_CAPS_SUPPORT_BW_MAX) |
868 		 BIT(NET_SHAPER_A_CAPS_SUPPORT_METRIC_BPS);
869 }
870 
871 static const struct net_shaper_ops mana_shaper_ops = {
872 	.set = mana_shaper_set,
873 	.delete = mana_shaper_del,
874 	.capabilities = mana_shaper_cap,
875 };
876 
877 static const struct net_device_ops mana_devops = {
878 	.ndo_open		= mana_open,
879 	.ndo_stop		= mana_close,
880 	.ndo_select_queue	= mana_select_queue,
881 	.ndo_start_xmit		= mana_start_xmit,
882 	.ndo_validate_addr	= eth_validate_addr,
883 	.ndo_get_stats64	= mana_get_stats64,
884 	.ndo_bpf		= mana_bpf,
885 	.ndo_xdp_xmit		= mana_xdp_xmit,
886 	.ndo_change_mtu		= mana_change_mtu,
887 	.net_shaper_ops         = &mana_shaper_ops,
888 };
889 
890 static void mana_cleanup_port_context(struct mana_port_context *apc)
891 {
892 	/*
893 	 * make sure subsequent cleanup attempts don't end up removing already
894 	 * cleaned dentry pointer
895 	 */
896 	debugfs_remove(apc->mana_port_debugfs);
897 	apc->mana_port_debugfs = NULL;
898 	kfree(apc->rxqs);
899 	apc->rxqs = NULL;
900 }
901 
902 static void mana_cleanup_indir_table(struct mana_port_context *apc)
903 {
904 	apc->indir_table_sz = 0;
905 	kfree(apc->indir_table);
906 	kfree(apc->rxobj_table);
907 }
908 
909 static int mana_init_port_context(struct mana_port_context *apc)
910 {
911 	apc->rxqs = kcalloc(apc->num_queues, sizeof(struct mana_rxq *),
912 			    GFP_KERNEL);
913 
914 	return !apc->rxqs ? -ENOMEM : 0;
915 }
916 
917 static int mana_send_request(struct mana_context *ac, void *in_buf,
918 			     u32 in_len, void *out_buf, u32 out_len)
919 {
920 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
921 	struct gdma_resp_hdr *resp = out_buf;
922 	struct gdma_req_hdr *req = in_buf;
923 	struct device *dev = gc->dev;
924 	static atomic_t activity_id;
925 	int err;
926 
927 	req->dev_id = gc->mana.dev_id;
928 	req->activity_id = atomic_inc_return(&activity_id);
929 
930 	err = mana_gd_send_request(gc, in_len, in_buf, out_len,
931 				   out_buf);
932 	if (err || resp->status) {
933 		if (err == -EOPNOTSUPP)
934 			return err;
935 
936 		if (req->req.msg_type != MANA_QUERY_PHY_STAT &&
937 		    mana_need_log(gc, err))
938 			dev_err(dev, "Failed to send mana message: %d, 0x%x\n",
939 				err, resp->status);
940 		return err ? err : -EPROTO;
941 	}
942 
943 	if (req->dev_id.as_uint32 != resp->dev_id.as_uint32 ||
944 	    req->activity_id != resp->activity_id) {
945 		dev_err(dev, "Unexpected mana message response: %x,%x,%x,%x\n",
946 			req->dev_id.as_uint32, resp->dev_id.as_uint32,
947 			req->activity_id, resp->activity_id);
948 		return -EPROTO;
949 	}
950 
951 	return 0;
952 }
953 
954 static int mana_verify_resp_hdr(const struct gdma_resp_hdr *resp_hdr,
955 				const enum mana_command_code expected_code,
956 				const u32 min_size)
957 {
958 	if (resp_hdr->response.msg_type != expected_code)
959 		return -EPROTO;
960 
961 	if (resp_hdr->response.msg_version < GDMA_MESSAGE_V1)
962 		return -EPROTO;
963 
964 	if (resp_hdr->response.msg_size < min_size)
965 		return -EPROTO;
966 
967 	return 0;
968 }
969 
970 static int mana_pf_register_hw_vport(struct mana_port_context *apc)
971 {
972 	struct mana_register_hw_vport_resp resp = {};
973 	struct mana_register_hw_vport_req req = {};
974 	int err;
975 
976 	mana_gd_init_req_hdr(&req.hdr, MANA_REGISTER_HW_PORT,
977 			     sizeof(req), sizeof(resp));
978 	req.attached_gfid = 1;
979 	req.is_pf_default_vport = 1;
980 	req.allow_all_ether_types = 1;
981 
982 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
983 				sizeof(resp));
984 	if (err) {
985 		netdev_err(apc->ndev, "Failed to register hw vPort: %d\n", err);
986 		return err;
987 	}
988 
989 	err = mana_verify_resp_hdr(&resp.hdr, MANA_REGISTER_HW_PORT,
990 				   sizeof(resp));
991 	if (err || resp.hdr.status) {
992 		netdev_err(apc->ndev, "Failed to register hw vPort: %d, 0x%x\n",
993 			   err, resp.hdr.status);
994 		return err ? err : -EPROTO;
995 	}
996 
997 	apc->port_handle = resp.hw_vport_handle;
998 	return 0;
999 }
1000 
1001 static void mana_pf_deregister_hw_vport(struct mana_port_context *apc)
1002 {
1003 	struct mana_deregister_hw_vport_resp resp = {};
1004 	struct mana_deregister_hw_vport_req req = {};
1005 	int err;
1006 
1007 	mana_gd_init_req_hdr(&req.hdr, MANA_DEREGISTER_HW_PORT,
1008 			     sizeof(req), sizeof(resp));
1009 	req.hw_vport_handle = apc->port_handle;
1010 
1011 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1012 				sizeof(resp));
1013 	if (err) {
1014 		if (mana_en_need_log(apc, err))
1015 			netdev_err(apc->ndev, "Failed to unregister hw vPort: %d\n",
1016 				   err);
1017 
1018 		return;
1019 	}
1020 
1021 	err = mana_verify_resp_hdr(&resp.hdr, MANA_DEREGISTER_HW_PORT,
1022 				   sizeof(resp));
1023 	if (err || resp.hdr.status)
1024 		netdev_err(apc->ndev,
1025 			   "Failed to deregister hw vPort: %d, 0x%x\n",
1026 			   err, resp.hdr.status);
1027 }
1028 
1029 static int mana_pf_register_filter(struct mana_port_context *apc)
1030 {
1031 	struct mana_register_filter_resp resp = {};
1032 	struct mana_register_filter_req req = {};
1033 	int err;
1034 
1035 	mana_gd_init_req_hdr(&req.hdr, MANA_REGISTER_FILTER,
1036 			     sizeof(req), sizeof(resp));
1037 	req.vport = apc->port_handle;
1038 	memcpy(req.mac_addr, apc->mac_addr, ETH_ALEN);
1039 
1040 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1041 				sizeof(resp));
1042 	if (err) {
1043 		netdev_err(apc->ndev, "Failed to register filter: %d\n", err);
1044 		return err;
1045 	}
1046 
1047 	err = mana_verify_resp_hdr(&resp.hdr, MANA_REGISTER_FILTER,
1048 				   sizeof(resp));
1049 	if (err || resp.hdr.status) {
1050 		netdev_err(apc->ndev, "Failed to register filter: %d, 0x%x\n",
1051 			   err, resp.hdr.status);
1052 		return err ? err : -EPROTO;
1053 	}
1054 
1055 	apc->pf_filter_handle = resp.filter_handle;
1056 	return 0;
1057 }
1058 
1059 static void mana_pf_deregister_filter(struct mana_port_context *apc)
1060 {
1061 	struct mana_deregister_filter_resp resp = {};
1062 	struct mana_deregister_filter_req req = {};
1063 	int err;
1064 
1065 	mana_gd_init_req_hdr(&req.hdr, MANA_DEREGISTER_FILTER,
1066 			     sizeof(req), sizeof(resp));
1067 	req.filter_handle = apc->pf_filter_handle;
1068 
1069 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1070 				sizeof(resp));
1071 	if (err) {
1072 		if (mana_en_need_log(apc, err))
1073 			netdev_err(apc->ndev, "Failed to unregister filter: %d\n",
1074 				   err);
1075 
1076 		return;
1077 	}
1078 
1079 	err = mana_verify_resp_hdr(&resp.hdr, MANA_DEREGISTER_FILTER,
1080 				   sizeof(resp));
1081 	if (err || resp.hdr.status)
1082 		netdev_err(apc->ndev,
1083 			   "Failed to deregister filter: %d, 0x%x\n",
1084 			   err, resp.hdr.status);
1085 }
1086 
1087 static int mana_query_device_cfg(struct mana_context *ac, u32 proto_major_ver,
1088 				 u32 proto_minor_ver, u32 proto_micro_ver,
1089 				 u16 *max_num_vports, u8 *bm_hostmode)
1090 {
1091 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
1092 	struct mana_query_device_cfg_resp resp = {};
1093 	struct mana_query_device_cfg_req req = {};
1094 	struct device *dev = gc->dev;
1095 	int err = 0;
1096 
1097 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_DEV_CONFIG,
1098 			     sizeof(req), sizeof(resp));
1099 
1100 	req.hdr.resp.msg_version = GDMA_MESSAGE_V3;
1101 
1102 	req.proto_major_ver = proto_major_ver;
1103 	req.proto_minor_ver = proto_minor_ver;
1104 	req.proto_micro_ver = proto_micro_ver;
1105 
1106 	err = mana_send_request(ac, &req, sizeof(req), &resp, sizeof(resp));
1107 	if (err) {
1108 		dev_err(dev, "Failed to query config: %d", err);
1109 		return err;
1110 	}
1111 
1112 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_DEV_CONFIG,
1113 				   sizeof(resp));
1114 	if (err || resp.hdr.status) {
1115 		dev_err(dev, "Invalid query result: %d, 0x%x\n", err,
1116 			resp.hdr.status);
1117 		if (!err)
1118 			err = -EPROTO;
1119 		return err;
1120 	}
1121 
1122 	*max_num_vports = resp.max_num_vports;
1123 
1124 	if (resp.hdr.response.msg_version >= GDMA_MESSAGE_V2)
1125 		gc->adapter_mtu = resp.adapter_mtu;
1126 	else
1127 		gc->adapter_mtu = ETH_FRAME_LEN;
1128 
1129 	if (resp.hdr.response.msg_version >= GDMA_MESSAGE_V3)
1130 		*bm_hostmode = resp.bm_hostmode;
1131 	else
1132 		*bm_hostmode = 0;
1133 
1134 	debugfs_create_u16("adapter-MTU", 0400, gc->mana_pci_debugfs, &gc->adapter_mtu);
1135 
1136 	return 0;
1137 }
1138 
1139 static int mana_query_vport_cfg(struct mana_port_context *apc, u32 vport_index,
1140 				u32 *max_sq, u32 *max_rq, u32 *num_indir_entry)
1141 {
1142 	struct mana_query_vport_cfg_resp resp = {};
1143 	struct mana_query_vport_cfg_req req = {};
1144 	int err;
1145 
1146 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_VPORT_CONFIG,
1147 			     sizeof(req), sizeof(resp));
1148 
1149 	req.vport_index = vport_index;
1150 
1151 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1152 				sizeof(resp));
1153 	if (err)
1154 		return err;
1155 
1156 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_VPORT_CONFIG,
1157 				   sizeof(resp));
1158 	if (err)
1159 		return err;
1160 
1161 	if (resp.hdr.status)
1162 		return -EPROTO;
1163 
1164 	*max_sq = resp.max_num_sq;
1165 	*max_rq = resp.max_num_rq;
1166 	if (resp.num_indirection_ent > 0 &&
1167 	    resp.num_indirection_ent <= MANA_INDIRECT_TABLE_MAX_SIZE &&
1168 	    is_power_of_2(resp.num_indirection_ent)) {
1169 		*num_indir_entry = resp.num_indirection_ent;
1170 	} else {
1171 		netdev_warn(apc->ndev,
1172 			    "Setting indirection table size to default %d for vPort %d\n",
1173 			    MANA_INDIRECT_TABLE_DEF_SIZE, apc->port_idx);
1174 		*num_indir_entry = MANA_INDIRECT_TABLE_DEF_SIZE;
1175 	}
1176 
1177 	apc->port_handle = resp.vport;
1178 	ether_addr_copy(apc->mac_addr, resp.mac_addr);
1179 
1180 	return 0;
1181 }
1182 
1183 void mana_uncfg_vport(struct mana_port_context *apc)
1184 {
1185 	mutex_lock(&apc->vport_mutex);
1186 	apc->vport_use_count--;
1187 	WARN_ON(apc->vport_use_count < 0);
1188 	mutex_unlock(&apc->vport_mutex);
1189 }
1190 EXPORT_SYMBOL_NS(mana_uncfg_vport, "NET_MANA");
1191 
1192 int mana_cfg_vport(struct mana_port_context *apc, u32 protection_dom_id,
1193 		   u32 doorbell_pg_id)
1194 {
1195 	struct mana_config_vport_resp resp = {};
1196 	struct mana_config_vport_req req = {};
1197 	int err;
1198 
1199 	/* This function is used to program the Ethernet port in the hardware
1200 	 * table. It can be called from the Ethernet driver or the RDMA driver.
1201 	 *
1202 	 * For Ethernet usage, the hardware supports only one active user on a
1203 	 * physical port. The driver checks on the port usage before programming
1204 	 * the hardware when creating the RAW QP (RDMA driver) or exposing the
1205 	 * device to kernel NET layer (Ethernet driver).
1206 	 *
1207 	 * Because the RDMA driver doesn't know in advance which QP type the
1208 	 * user will create, it exposes the device with all its ports. The user
1209 	 * may not be able to create RAW QP on a port if this port is already
1210 	 * in used by the Ethernet driver from the kernel.
1211 	 *
1212 	 * This physical port limitation only applies to the RAW QP. For RC QP,
1213 	 * the hardware doesn't have this limitation. The user can create RC
1214 	 * QPs on a physical port up to the hardware limits independent of the
1215 	 * Ethernet usage on the same port.
1216 	 */
1217 	mutex_lock(&apc->vport_mutex);
1218 	if (apc->vport_use_count > 0) {
1219 		mutex_unlock(&apc->vport_mutex);
1220 		return -EBUSY;
1221 	}
1222 	apc->vport_use_count++;
1223 	mutex_unlock(&apc->vport_mutex);
1224 
1225 	mana_gd_init_req_hdr(&req.hdr, MANA_CONFIG_VPORT_TX,
1226 			     sizeof(req), sizeof(resp));
1227 	req.vport = apc->port_handle;
1228 	req.pdid = protection_dom_id;
1229 	req.doorbell_pageid = doorbell_pg_id;
1230 
1231 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1232 				sizeof(resp));
1233 	if (err) {
1234 		netdev_err(apc->ndev, "Failed to configure vPort: %d\n", err);
1235 		goto out;
1236 	}
1237 
1238 	err = mana_verify_resp_hdr(&resp.hdr, MANA_CONFIG_VPORT_TX,
1239 				   sizeof(resp));
1240 	if (err || resp.hdr.status) {
1241 		netdev_err(apc->ndev, "Failed to configure vPort: %d, 0x%x\n",
1242 			   err, resp.hdr.status);
1243 		if (!err)
1244 			err = -EPROTO;
1245 
1246 		goto out;
1247 	}
1248 
1249 	apc->tx_shortform_allowed = resp.short_form_allowed;
1250 	apc->tx_vp_offset = resp.tx_vport_offset;
1251 
1252 	netdev_info(apc->ndev, "Configured vPort %llu PD %u DB %u\n",
1253 		    apc->port_handle, protection_dom_id, doorbell_pg_id);
1254 out:
1255 	if (err)
1256 		mana_uncfg_vport(apc);
1257 
1258 	return err;
1259 }
1260 EXPORT_SYMBOL_NS(mana_cfg_vport, "NET_MANA");
1261 
1262 static int mana_cfg_vport_steering(struct mana_port_context *apc,
1263 				   enum TRI_STATE rx,
1264 				   bool update_default_rxobj, bool update_key,
1265 				   bool update_tab)
1266 {
1267 	struct mana_cfg_rx_steer_req_v2 *req;
1268 	struct mana_cfg_rx_steer_resp resp = {};
1269 	struct net_device *ndev = apc->ndev;
1270 	u32 req_buf_size;
1271 	int err;
1272 
1273 	req_buf_size = struct_size(req, indir_tab, apc->indir_table_sz);
1274 	req = kzalloc(req_buf_size, GFP_KERNEL);
1275 	if (!req)
1276 		return -ENOMEM;
1277 
1278 	mana_gd_init_req_hdr(&req->hdr, MANA_CONFIG_VPORT_RX, req_buf_size,
1279 			     sizeof(resp));
1280 
1281 	req->hdr.req.msg_version = GDMA_MESSAGE_V2;
1282 
1283 	req->vport = apc->port_handle;
1284 	req->num_indir_entries = apc->indir_table_sz;
1285 	req->indir_tab_offset = offsetof(struct mana_cfg_rx_steer_req_v2,
1286 					 indir_tab);
1287 	req->rx_enable = rx;
1288 	req->rss_enable = apc->rss_state;
1289 	req->update_default_rxobj = update_default_rxobj;
1290 	req->update_hashkey = update_key;
1291 	req->update_indir_tab = update_tab;
1292 	req->default_rxobj = apc->default_rxobj;
1293 	req->cqe_coalescing_enable = 0;
1294 
1295 	if (update_key)
1296 		memcpy(&req->hashkey, apc->hashkey, MANA_HASH_KEY_SIZE);
1297 
1298 	if (update_tab)
1299 		memcpy(req->indir_tab, apc->rxobj_table,
1300 		       flex_array_size(req, indir_tab, req->num_indir_entries));
1301 
1302 	err = mana_send_request(apc->ac, req, req_buf_size, &resp,
1303 				sizeof(resp));
1304 	if (err) {
1305 		if (mana_en_need_log(apc, err))
1306 			netdev_err(ndev, "Failed to configure vPort RX: %d\n", err);
1307 
1308 		goto out;
1309 	}
1310 
1311 	err = mana_verify_resp_hdr(&resp.hdr, MANA_CONFIG_VPORT_RX,
1312 				   sizeof(resp));
1313 	if (err) {
1314 		netdev_err(ndev, "vPort RX configuration failed: %d\n", err);
1315 		goto out;
1316 	}
1317 
1318 	if (resp.hdr.status) {
1319 		netdev_err(ndev, "vPort RX configuration failed: 0x%x\n",
1320 			   resp.hdr.status);
1321 		err = -EPROTO;
1322 	}
1323 
1324 	netdev_info(ndev, "Configured steering vPort %llu entries %u\n",
1325 		    apc->port_handle, apc->indir_table_sz);
1326 out:
1327 	kfree(req);
1328 	return err;
1329 }
1330 
1331 int mana_query_link_cfg(struct mana_port_context *apc)
1332 {
1333 	struct net_device *ndev = apc->ndev;
1334 	struct mana_query_link_config_resp resp = {};
1335 	struct mana_query_link_config_req req = {};
1336 	int err;
1337 
1338 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_LINK_CONFIG,
1339 			     sizeof(req), sizeof(resp));
1340 
1341 	req.vport = apc->port_handle;
1342 	req.hdr.resp.msg_version = GDMA_MESSAGE_V2;
1343 
1344 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1345 				sizeof(resp));
1346 
1347 	if (err) {
1348 		if (err == -EOPNOTSUPP) {
1349 			netdev_info_once(ndev, "MANA_QUERY_LINK_CONFIG not supported\n");
1350 			return err;
1351 		}
1352 		netdev_err(ndev, "Failed to query link config: %d\n", err);
1353 		return err;
1354 	}
1355 
1356 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_LINK_CONFIG,
1357 				   sizeof(resp));
1358 
1359 	if (err || resp.hdr.status) {
1360 		netdev_err(ndev, "Failed to query link config: %d, 0x%x\n", err,
1361 			   resp.hdr.status);
1362 		if (!err)
1363 			err = -EOPNOTSUPP;
1364 		return err;
1365 	}
1366 
1367 	if (resp.qos_unconfigured) {
1368 		err = -EINVAL;
1369 		return err;
1370 	}
1371 	apc->speed = resp.link_speed_mbps;
1372 	apc->max_speed = resp.qos_speed_mbps;
1373 	return 0;
1374 }
1375 
1376 int mana_set_bw_clamp(struct mana_port_context *apc, u32 speed,
1377 		      int enable_clamping)
1378 {
1379 	struct mana_set_bw_clamp_resp resp = {};
1380 	struct mana_set_bw_clamp_req req = {};
1381 	struct net_device *ndev = apc->ndev;
1382 	int err;
1383 
1384 	mana_gd_init_req_hdr(&req.hdr, MANA_SET_BW_CLAMP,
1385 			     sizeof(req), sizeof(resp));
1386 	req.vport = apc->port_handle;
1387 	req.link_speed_mbps = speed;
1388 	req.enable_clamping = enable_clamping;
1389 
1390 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1391 				sizeof(resp));
1392 
1393 	if (err) {
1394 		if (err == -EOPNOTSUPP) {
1395 			netdev_info_once(ndev, "MANA_SET_BW_CLAMP not supported\n");
1396 			return err;
1397 		}
1398 		netdev_err(ndev, "Failed to set bandwidth clamp for speed %u, err = %d",
1399 			   speed, err);
1400 		return err;
1401 	}
1402 
1403 	err = mana_verify_resp_hdr(&resp.hdr, MANA_SET_BW_CLAMP,
1404 				   sizeof(resp));
1405 
1406 	if (err || resp.hdr.status) {
1407 		netdev_err(ndev, "Failed to set bandwidth clamp: %d, 0x%x\n", err,
1408 			   resp.hdr.status);
1409 		if (!err)
1410 			err = -EOPNOTSUPP;
1411 		return err;
1412 	}
1413 
1414 	if (resp.qos_unconfigured)
1415 		netdev_info(ndev, "QoS is unconfigured\n");
1416 
1417 	return 0;
1418 }
1419 
1420 int mana_create_wq_obj(struct mana_port_context *apc,
1421 		       mana_handle_t vport,
1422 		       u32 wq_type, struct mana_obj_spec *wq_spec,
1423 		       struct mana_obj_spec *cq_spec,
1424 		       mana_handle_t *wq_obj)
1425 {
1426 	struct mana_create_wqobj_resp resp = {};
1427 	struct mana_create_wqobj_req req = {};
1428 	struct net_device *ndev = apc->ndev;
1429 	int err;
1430 
1431 	mana_gd_init_req_hdr(&req.hdr, MANA_CREATE_WQ_OBJ,
1432 			     sizeof(req), sizeof(resp));
1433 	req.vport = vport;
1434 	req.wq_type = wq_type;
1435 	req.wq_gdma_region = wq_spec->gdma_region;
1436 	req.cq_gdma_region = cq_spec->gdma_region;
1437 	req.wq_size = wq_spec->queue_size;
1438 	req.cq_size = cq_spec->queue_size;
1439 	req.cq_moderation_ctx_id = cq_spec->modr_ctx_id;
1440 	req.cq_parent_qid = cq_spec->attached_eq;
1441 
1442 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1443 				sizeof(resp));
1444 	if (err) {
1445 		netdev_err(ndev, "Failed to create WQ object: %d\n", err);
1446 		goto out;
1447 	}
1448 
1449 	err = mana_verify_resp_hdr(&resp.hdr, MANA_CREATE_WQ_OBJ,
1450 				   sizeof(resp));
1451 	if (err || resp.hdr.status) {
1452 		netdev_err(ndev, "Failed to create WQ object: %d, 0x%x\n", err,
1453 			   resp.hdr.status);
1454 		if (!err)
1455 			err = -EPROTO;
1456 		goto out;
1457 	}
1458 
1459 	if (resp.wq_obj == INVALID_MANA_HANDLE) {
1460 		netdev_err(ndev, "Got an invalid WQ object handle\n");
1461 		err = -EPROTO;
1462 		goto out;
1463 	}
1464 
1465 	*wq_obj = resp.wq_obj;
1466 	wq_spec->queue_index = resp.wq_id;
1467 	cq_spec->queue_index = resp.cq_id;
1468 
1469 	return 0;
1470 out:
1471 	return err;
1472 }
1473 EXPORT_SYMBOL_NS(mana_create_wq_obj, "NET_MANA");
1474 
1475 void mana_destroy_wq_obj(struct mana_port_context *apc, u32 wq_type,
1476 			 mana_handle_t wq_obj)
1477 {
1478 	struct mana_destroy_wqobj_resp resp = {};
1479 	struct mana_destroy_wqobj_req req = {};
1480 	struct net_device *ndev = apc->ndev;
1481 	int err;
1482 
1483 	mana_gd_init_req_hdr(&req.hdr, MANA_DESTROY_WQ_OBJ,
1484 			     sizeof(req), sizeof(resp));
1485 	req.wq_type = wq_type;
1486 	req.wq_obj_handle = wq_obj;
1487 
1488 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1489 				sizeof(resp));
1490 	if (err) {
1491 		if (mana_en_need_log(apc, err))
1492 			netdev_err(ndev, "Failed to destroy WQ object: %d\n", err);
1493 
1494 		return;
1495 	}
1496 
1497 	err = mana_verify_resp_hdr(&resp.hdr, MANA_DESTROY_WQ_OBJ,
1498 				   sizeof(resp));
1499 	if (err || resp.hdr.status)
1500 		netdev_err(ndev, "Failed to destroy WQ object: %d, 0x%x\n", err,
1501 			   resp.hdr.status);
1502 }
1503 EXPORT_SYMBOL_NS(mana_destroy_wq_obj, "NET_MANA");
1504 
1505 static void mana_destroy_eq(struct mana_context *ac)
1506 {
1507 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
1508 	struct gdma_queue *eq;
1509 	int i;
1510 
1511 	if (!ac->eqs)
1512 		return;
1513 
1514 	debugfs_remove_recursive(ac->mana_eqs_debugfs);
1515 	ac->mana_eqs_debugfs = NULL;
1516 
1517 	for (i = 0; i < gc->max_num_queues; i++) {
1518 		eq = ac->eqs[i].eq;
1519 		if (!eq)
1520 			continue;
1521 
1522 		mana_gd_destroy_queue(gc, eq);
1523 	}
1524 
1525 	kfree(ac->eqs);
1526 	ac->eqs = NULL;
1527 }
1528 
1529 static void mana_create_eq_debugfs(struct mana_context *ac, int i)
1530 {
1531 	struct mana_eq eq = ac->eqs[i];
1532 	char eqnum[32];
1533 
1534 	sprintf(eqnum, "eq%d", i);
1535 	eq.mana_eq_debugfs = debugfs_create_dir(eqnum, ac->mana_eqs_debugfs);
1536 	debugfs_create_u32("head", 0400, eq.mana_eq_debugfs, &eq.eq->head);
1537 	debugfs_create_u32("tail", 0400, eq.mana_eq_debugfs, &eq.eq->tail);
1538 	debugfs_create_file("eq_dump", 0400, eq.mana_eq_debugfs, eq.eq, &mana_dbg_q_fops);
1539 }
1540 
1541 static int mana_create_eq(struct mana_context *ac)
1542 {
1543 	struct gdma_dev *gd = ac->gdma_dev;
1544 	struct gdma_context *gc = gd->gdma_context;
1545 	struct gdma_queue_spec spec = {};
1546 	int err;
1547 	int i;
1548 
1549 	ac->eqs = kcalloc(gc->max_num_queues, sizeof(struct mana_eq),
1550 			  GFP_KERNEL);
1551 	if (!ac->eqs)
1552 		return -ENOMEM;
1553 
1554 	spec.type = GDMA_EQ;
1555 	spec.monitor_avl_buf = false;
1556 	spec.queue_size = EQ_SIZE;
1557 	spec.eq.callback = NULL;
1558 	spec.eq.context = ac->eqs;
1559 	spec.eq.log2_throttle_limit = LOG2_EQ_THROTTLE;
1560 
1561 	ac->mana_eqs_debugfs = debugfs_create_dir("EQs", gc->mana_pci_debugfs);
1562 
1563 	for (i = 0; i < gc->max_num_queues; i++) {
1564 		spec.eq.msix_index = (i + 1) % gc->num_msix_usable;
1565 		err = mana_gd_create_mana_eq(gd, &spec, &ac->eqs[i].eq);
1566 		if (err) {
1567 			dev_err(gc->dev, "Failed to create EQ %d : %d\n", i, err);
1568 			goto out;
1569 		}
1570 		mana_create_eq_debugfs(ac, i);
1571 	}
1572 
1573 	return 0;
1574 out:
1575 	mana_destroy_eq(ac);
1576 	return err;
1577 }
1578 
1579 static int mana_fence_rq(struct mana_port_context *apc, struct mana_rxq *rxq)
1580 {
1581 	struct mana_fence_rq_resp resp = {};
1582 	struct mana_fence_rq_req req = {};
1583 	int err;
1584 
1585 	init_completion(&rxq->fence_event);
1586 
1587 	mana_gd_init_req_hdr(&req.hdr, MANA_FENCE_RQ,
1588 			     sizeof(req), sizeof(resp));
1589 	req.wq_obj_handle =  rxq->rxobj;
1590 
1591 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1592 				sizeof(resp));
1593 	if (err) {
1594 		netdev_err(apc->ndev, "Failed to fence RQ %u: %d\n",
1595 			   rxq->rxq_idx, err);
1596 		return err;
1597 	}
1598 
1599 	err = mana_verify_resp_hdr(&resp.hdr, MANA_FENCE_RQ, sizeof(resp));
1600 	if (err || resp.hdr.status) {
1601 		netdev_err(apc->ndev, "Failed to fence RQ %u: %d, 0x%x\n",
1602 			   rxq->rxq_idx, err, resp.hdr.status);
1603 		if (!err)
1604 			err = -EPROTO;
1605 
1606 		return err;
1607 	}
1608 
1609 	if (wait_for_completion_timeout(&rxq->fence_event, 10 * HZ) == 0) {
1610 		netdev_err(apc->ndev, "Failed to fence RQ %u: timed out\n",
1611 			   rxq->rxq_idx);
1612 		return -ETIMEDOUT;
1613 	}
1614 
1615 	return 0;
1616 }
1617 
1618 static void mana_fence_rqs(struct mana_port_context *apc)
1619 {
1620 	unsigned int rxq_idx;
1621 	struct mana_rxq *rxq;
1622 	int err;
1623 
1624 	for (rxq_idx = 0; rxq_idx < apc->num_queues; rxq_idx++) {
1625 		rxq = apc->rxqs[rxq_idx];
1626 		err = mana_fence_rq(apc, rxq);
1627 
1628 		/* In case of any error, use sleep instead. */
1629 		if (err)
1630 			msleep(100);
1631 	}
1632 }
1633 
1634 static int mana_move_wq_tail(struct gdma_queue *wq, u32 num_units)
1635 {
1636 	u32 used_space_old;
1637 	u32 used_space_new;
1638 
1639 	used_space_old = wq->head - wq->tail;
1640 	used_space_new = wq->head - (wq->tail + num_units);
1641 
1642 	if (WARN_ON_ONCE(used_space_new > used_space_old))
1643 		return -ERANGE;
1644 
1645 	wq->tail += num_units;
1646 	return 0;
1647 }
1648 
1649 static void mana_unmap_skb(struct sk_buff *skb, struct mana_port_context *apc)
1650 {
1651 	struct mana_skb_head *ash = (struct mana_skb_head *)skb->head;
1652 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
1653 	struct device *dev = gc->dev;
1654 	int hsg, i;
1655 
1656 	/* Number of SGEs of linear part */
1657 	hsg = (skb_is_gso(skb) && skb_headlen(skb) > ash->size[0]) ? 2 : 1;
1658 
1659 	for (i = 0; i < hsg; i++)
1660 		dma_unmap_single(dev, ash->dma_handle[i], ash->size[i],
1661 				 DMA_TO_DEVICE);
1662 
1663 	for (i = hsg; i < skb_shinfo(skb)->nr_frags + hsg; i++)
1664 		dma_unmap_page(dev, ash->dma_handle[i], ash->size[i],
1665 			       DMA_TO_DEVICE);
1666 }
1667 
1668 static void mana_poll_tx_cq(struct mana_cq *cq)
1669 {
1670 	struct gdma_comp *completions = cq->gdma_comp_buf;
1671 	struct gdma_posted_wqe_info *wqe_info;
1672 	unsigned int pkt_transmitted = 0;
1673 	unsigned int wqe_unit_cnt = 0;
1674 	struct mana_txq *txq = cq->txq;
1675 	struct mana_port_context *apc;
1676 	struct netdev_queue *net_txq;
1677 	struct gdma_queue *gdma_wq;
1678 	unsigned int avail_space;
1679 	struct net_device *ndev;
1680 	struct sk_buff *skb;
1681 	bool txq_stopped;
1682 	int comp_read;
1683 	int i;
1684 
1685 	ndev = txq->ndev;
1686 	apc = netdev_priv(ndev);
1687 
1688 	comp_read = mana_gd_poll_cq(cq->gdma_cq, completions,
1689 				    CQE_POLLING_BUFFER);
1690 
1691 	if (comp_read < 1)
1692 		return;
1693 
1694 	for (i = 0; i < comp_read; i++) {
1695 		struct mana_tx_comp_oob *cqe_oob;
1696 
1697 		if (WARN_ON_ONCE(!completions[i].is_sq))
1698 			return;
1699 
1700 		cqe_oob = (struct mana_tx_comp_oob *)completions[i].cqe_data;
1701 		if (WARN_ON_ONCE(cqe_oob->cqe_hdr.client_type !=
1702 				 MANA_CQE_COMPLETION))
1703 			return;
1704 
1705 		switch (cqe_oob->cqe_hdr.cqe_type) {
1706 		case CQE_TX_OKAY:
1707 			break;
1708 
1709 		case CQE_TX_SA_DROP:
1710 		case CQE_TX_MTU_DROP:
1711 		case CQE_TX_INVALID_OOB:
1712 		case CQE_TX_INVALID_ETH_TYPE:
1713 		case CQE_TX_HDR_PROCESSING_ERROR:
1714 		case CQE_TX_VF_DISABLED:
1715 		case CQE_TX_VPORT_IDX_OUT_OF_RANGE:
1716 		case CQE_TX_VPORT_DISABLED:
1717 		case CQE_TX_VLAN_TAGGING_VIOLATION:
1718 			if (net_ratelimit())
1719 				netdev_err(ndev, "TX: CQE error %d\n",
1720 					   cqe_oob->cqe_hdr.cqe_type);
1721 
1722 			apc->eth_stats.tx_cqe_err++;
1723 			break;
1724 
1725 		default:
1726 			/* If the CQE type is unknown, log an error,
1727 			 * and still free the SKB, update tail, etc.
1728 			 */
1729 			if (net_ratelimit())
1730 				netdev_err(ndev, "TX: unknown CQE type %d\n",
1731 					   cqe_oob->cqe_hdr.cqe_type);
1732 
1733 			apc->eth_stats.tx_cqe_unknown_type++;
1734 			break;
1735 		}
1736 
1737 		if (WARN_ON_ONCE(txq->gdma_txq_id != completions[i].wq_num))
1738 			return;
1739 
1740 		skb = skb_dequeue(&txq->pending_skbs);
1741 		if (WARN_ON_ONCE(!skb))
1742 			return;
1743 
1744 		wqe_info = (struct gdma_posted_wqe_info *)skb->cb;
1745 		wqe_unit_cnt += wqe_info->wqe_size_in_bu;
1746 
1747 		mana_unmap_skb(skb, apc);
1748 
1749 		napi_consume_skb(skb, cq->budget);
1750 
1751 		pkt_transmitted++;
1752 	}
1753 
1754 	if (WARN_ON_ONCE(wqe_unit_cnt == 0))
1755 		return;
1756 
1757 	mana_move_wq_tail(txq->gdma_sq, wqe_unit_cnt);
1758 
1759 	gdma_wq = txq->gdma_sq;
1760 	avail_space = mana_gd_wq_avail_space(gdma_wq);
1761 
1762 	/* Ensure tail updated before checking q stop */
1763 	smp_mb();
1764 
1765 	net_txq = txq->net_txq;
1766 	txq_stopped = netif_tx_queue_stopped(net_txq);
1767 
1768 	/* Ensure checking txq_stopped before apc->port_is_up. */
1769 	smp_rmb();
1770 
1771 	if (txq_stopped && apc->port_is_up && avail_space >= MAX_TX_WQE_SIZE) {
1772 		netif_tx_wake_queue(net_txq);
1773 		apc->eth_stats.wake_queue++;
1774 	}
1775 
1776 	if (atomic_sub_return(pkt_transmitted, &txq->pending_sends) < 0)
1777 		WARN_ON_ONCE(1);
1778 
1779 	cq->work_done = pkt_transmitted;
1780 }
1781 
1782 static void mana_post_pkt_rxq(struct mana_rxq *rxq)
1783 {
1784 	struct mana_recv_buf_oob *recv_buf_oob;
1785 	u32 curr_index;
1786 	int err;
1787 
1788 	curr_index = rxq->buf_index++;
1789 	if (rxq->buf_index == rxq->num_rx_buf)
1790 		rxq->buf_index = 0;
1791 
1792 	recv_buf_oob = &rxq->rx_oobs[curr_index];
1793 
1794 	err = mana_gd_post_work_request(rxq->gdma_rq, &recv_buf_oob->wqe_req,
1795 					&recv_buf_oob->wqe_inf);
1796 	if (WARN_ON_ONCE(err))
1797 		return;
1798 
1799 	WARN_ON_ONCE(recv_buf_oob->wqe_inf.wqe_size_in_bu != 1);
1800 }
1801 
1802 static struct sk_buff *mana_build_skb(struct mana_rxq *rxq, void *buf_va,
1803 				      uint pkt_len, struct xdp_buff *xdp)
1804 {
1805 	struct sk_buff *skb = napi_build_skb(buf_va, rxq->alloc_size);
1806 
1807 	if (!skb)
1808 		return NULL;
1809 
1810 	if (xdp->data_hard_start) {
1811 		u32 metasize = xdp->data - xdp->data_meta;
1812 
1813 		skb_reserve(skb, xdp->data - xdp->data_hard_start);
1814 		skb_put(skb, xdp->data_end - xdp->data);
1815 		if (metasize)
1816 			skb_metadata_set(skb, metasize);
1817 		return skb;
1818 	}
1819 
1820 	skb_reserve(skb, rxq->headroom);
1821 	skb_put(skb, pkt_len);
1822 
1823 	return skb;
1824 }
1825 
1826 static void mana_rx_skb(void *buf_va, bool from_pool,
1827 			struct mana_rxcomp_oob *cqe, struct mana_rxq *rxq)
1828 {
1829 	struct mana_stats_rx *rx_stats = &rxq->stats;
1830 	struct net_device *ndev = rxq->ndev;
1831 	uint pkt_len = cqe->ppi[0].pkt_len;
1832 	u16 rxq_idx = rxq->rxq_idx;
1833 	struct napi_struct *napi;
1834 	struct xdp_buff xdp = {};
1835 	struct sk_buff *skb;
1836 	u32 hash_value;
1837 	u32 act;
1838 
1839 	rxq->rx_cq.work_done++;
1840 	napi = &rxq->rx_cq.napi;
1841 
1842 	if (!buf_va) {
1843 		++ndev->stats.rx_dropped;
1844 		return;
1845 	}
1846 
1847 	act = mana_run_xdp(ndev, rxq, &xdp, buf_va, pkt_len);
1848 
1849 	if (act == XDP_REDIRECT && !rxq->xdp_rc)
1850 		return;
1851 
1852 	if (act != XDP_PASS && act != XDP_TX)
1853 		goto drop_xdp;
1854 
1855 	skb = mana_build_skb(rxq, buf_va, pkt_len, &xdp);
1856 
1857 	if (!skb)
1858 		goto drop;
1859 
1860 	if (from_pool)
1861 		skb_mark_for_recycle(skb);
1862 
1863 	skb->dev = napi->dev;
1864 
1865 	skb->protocol = eth_type_trans(skb, ndev);
1866 	skb_checksum_none_assert(skb);
1867 	skb_record_rx_queue(skb, rxq_idx);
1868 
1869 	if ((ndev->features & NETIF_F_RXCSUM) && cqe->rx_iphdr_csum_succeed) {
1870 		if (cqe->rx_tcp_csum_succeed || cqe->rx_udp_csum_succeed)
1871 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1872 	}
1873 
1874 	if (cqe->rx_hashtype != 0 && (ndev->features & NETIF_F_RXHASH)) {
1875 		hash_value = cqe->ppi[0].pkt_hash;
1876 
1877 		if (cqe->rx_hashtype & MANA_HASH_L4)
1878 			skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L4);
1879 		else
1880 			skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L3);
1881 	}
1882 
1883 	if (cqe->rx_vlantag_present) {
1884 		u16 vlan_tci = cqe->rx_vlan_id;
1885 
1886 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1887 	}
1888 
1889 	u64_stats_update_begin(&rx_stats->syncp);
1890 	rx_stats->packets++;
1891 	rx_stats->bytes += pkt_len;
1892 
1893 	if (act == XDP_TX)
1894 		rx_stats->xdp_tx++;
1895 	u64_stats_update_end(&rx_stats->syncp);
1896 
1897 	if (act == XDP_TX) {
1898 		skb_set_queue_mapping(skb, rxq_idx);
1899 		mana_xdp_tx(skb, ndev);
1900 		return;
1901 	}
1902 
1903 	napi_gro_receive(napi, skb);
1904 
1905 	return;
1906 
1907 drop_xdp:
1908 	u64_stats_update_begin(&rx_stats->syncp);
1909 	rx_stats->xdp_drop++;
1910 	u64_stats_update_end(&rx_stats->syncp);
1911 
1912 drop:
1913 	if (from_pool) {
1914 		if (rxq->frag_count == 1)
1915 			page_pool_recycle_direct(rxq->page_pool,
1916 						 virt_to_head_page(buf_va));
1917 		else
1918 			page_pool_free_va(rxq->page_pool, buf_va, true);
1919 	} else {
1920 		WARN_ON_ONCE(rxq->xdp_save_va);
1921 		/* Save for reuse */
1922 		rxq->xdp_save_va = buf_va;
1923 	}
1924 
1925 	++ndev->stats.rx_dropped;
1926 
1927 	return;
1928 }
1929 
1930 static void *mana_get_rxfrag(struct mana_rxq *rxq, struct device *dev,
1931 			     dma_addr_t *da, bool *from_pool)
1932 {
1933 	struct page *page;
1934 	u32 offset;
1935 	void *va;
1936 	*from_pool = false;
1937 
1938 	/* Don't use fragments for jumbo frames or XDP where it's 1 fragment
1939 	 * per page.
1940 	 */
1941 	if (rxq->frag_count == 1) {
1942 		/* Reuse XDP dropped page if available */
1943 		if (rxq->xdp_save_va) {
1944 			va = rxq->xdp_save_va;
1945 			page = virt_to_head_page(va);
1946 			rxq->xdp_save_va = NULL;
1947 		} else {
1948 			page = page_pool_dev_alloc_pages(rxq->page_pool);
1949 			if (!page)
1950 				return NULL;
1951 
1952 			*from_pool = true;
1953 			va = page_to_virt(page);
1954 		}
1955 
1956 		*da = dma_map_single(dev, va + rxq->headroom, rxq->datasize,
1957 				     DMA_FROM_DEVICE);
1958 		if (dma_mapping_error(dev, *da)) {
1959 			mana_put_rx_page(rxq, page, *from_pool);
1960 			return NULL;
1961 		}
1962 
1963 		return va;
1964 	}
1965 
1966 	page =  page_pool_dev_alloc_frag(rxq->page_pool, &offset,
1967 					 rxq->alloc_size);
1968 	if (!page)
1969 		return NULL;
1970 
1971 	va  = page_to_virt(page) + offset;
1972 	*da = page_pool_get_dma_addr(page) + offset + rxq->headroom;
1973 	*from_pool = true;
1974 
1975 	return va;
1976 }
1977 
1978 /* Allocate frag for rx buffer, and save the old buf */
1979 static void mana_refill_rx_oob(struct device *dev, struct mana_rxq *rxq,
1980 			       struct mana_recv_buf_oob *rxoob, void **old_buf,
1981 			       bool *old_fp)
1982 {
1983 	bool from_pool;
1984 	dma_addr_t da;
1985 	void *va;
1986 
1987 	va = mana_get_rxfrag(rxq, dev, &da, &from_pool);
1988 	if (!va)
1989 		return;
1990 	if (!rxoob->from_pool || rxq->frag_count == 1)
1991 		dma_unmap_single(dev, rxoob->sgl[0].address, rxq->datasize,
1992 				 DMA_FROM_DEVICE);
1993 	*old_buf = rxoob->buf_va;
1994 	*old_fp = rxoob->from_pool;
1995 
1996 	rxoob->buf_va = va;
1997 	rxoob->sgl[0].address = da;
1998 	rxoob->from_pool = from_pool;
1999 }
2000 
2001 static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
2002 				struct gdma_comp *cqe)
2003 {
2004 	struct mana_rxcomp_oob *oob = (struct mana_rxcomp_oob *)cqe->cqe_data;
2005 	struct gdma_context *gc = rxq->gdma_rq->gdma_dev->gdma_context;
2006 	struct net_device *ndev = rxq->ndev;
2007 	struct mana_recv_buf_oob *rxbuf_oob;
2008 	struct mana_port_context *apc;
2009 	struct device *dev = gc->dev;
2010 	void *old_buf = NULL;
2011 	u32 curr, pktlen;
2012 	bool old_fp;
2013 
2014 	apc = netdev_priv(ndev);
2015 
2016 	switch (oob->cqe_hdr.cqe_type) {
2017 	case CQE_RX_OKAY:
2018 		break;
2019 
2020 	case CQE_RX_TRUNCATED:
2021 		++ndev->stats.rx_dropped;
2022 		rxbuf_oob = &rxq->rx_oobs[rxq->buf_index];
2023 		netdev_warn_once(ndev, "Dropped a truncated packet\n");
2024 		goto drop;
2025 
2026 	case CQE_RX_COALESCED_4:
2027 		netdev_err(ndev, "RX coalescing is unsupported\n");
2028 		apc->eth_stats.rx_coalesced_err++;
2029 		return;
2030 
2031 	case CQE_RX_OBJECT_FENCE:
2032 		complete(&rxq->fence_event);
2033 		return;
2034 
2035 	default:
2036 		netdev_err(ndev, "Unknown RX CQE type = %d\n",
2037 			   oob->cqe_hdr.cqe_type);
2038 		apc->eth_stats.rx_cqe_unknown_type++;
2039 		return;
2040 	}
2041 
2042 	pktlen = oob->ppi[0].pkt_len;
2043 
2044 	if (pktlen == 0) {
2045 		/* data packets should never have packetlength of zero */
2046 		netdev_err(ndev, "RX pkt len=0, rq=%u, cq=%u, rxobj=0x%llx\n",
2047 			   rxq->gdma_id, cq->gdma_id, rxq->rxobj);
2048 		return;
2049 	}
2050 
2051 	curr = rxq->buf_index;
2052 	rxbuf_oob = &rxq->rx_oobs[curr];
2053 	WARN_ON_ONCE(rxbuf_oob->wqe_inf.wqe_size_in_bu != 1);
2054 
2055 	mana_refill_rx_oob(dev, rxq, rxbuf_oob, &old_buf, &old_fp);
2056 
2057 	/* Unsuccessful refill will have old_buf == NULL.
2058 	 * In this case, mana_rx_skb() will drop the packet.
2059 	 */
2060 	mana_rx_skb(old_buf, old_fp, oob, rxq);
2061 
2062 drop:
2063 	mana_move_wq_tail(rxq->gdma_rq, rxbuf_oob->wqe_inf.wqe_size_in_bu);
2064 
2065 	mana_post_pkt_rxq(rxq);
2066 }
2067 
2068 static void mana_poll_rx_cq(struct mana_cq *cq)
2069 {
2070 	struct gdma_comp *comp = cq->gdma_comp_buf;
2071 	struct mana_rxq *rxq = cq->rxq;
2072 	int comp_read, i;
2073 
2074 	comp_read = mana_gd_poll_cq(cq->gdma_cq, comp, CQE_POLLING_BUFFER);
2075 	WARN_ON_ONCE(comp_read > CQE_POLLING_BUFFER);
2076 
2077 	rxq->xdp_flush = false;
2078 
2079 	for (i = 0; i < comp_read; i++) {
2080 		if (WARN_ON_ONCE(comp[i].is_sq))
2081 			return;
2082 
2083 		/* verify recv cqe references the right rxq */
2084 		if (WARN_ON_ONCE(comp[i].wq_num != cq->rxq->gdma_id))
2085 			return;
2086 
2087 		mana_process_rx_cqe(rxq, cq, &comp[i]);
2088 	}
2089 
2090 	if (comp_read > 0) {
2091 		struct gdma_context *gc = rxq->gdma_rq->gdma_dev->gdma_context;
2092 
2093 		mana_gd_wq_ring_doorbell(gc, rxq->gdma_rq);
2094 	}
2095 
2096 	if (rxq->xdp_flush)
2097 		xdp_do_flush();
2098 }
2099 
2100 static int mana_cq_handler(void *context, struct gdma_queue *gdma_queue)
2101 {
2102 	struct mana_cq *cq = context;
2103 	int w;
2104 
2105 	WARN_ON_ONCE(cq->gdma_cq != gdma_queue);
2106 
2107 	if (cq->type == MANA_CQ_TYPE_RX)
2108 		mana_poll_rx_cq(cq);
2109 	else
2110 		mana_poll_tx_cq(cq);
2111 
2112 	w = cq->work_done;
2113 	cq->work_done_since_doorbell += w;
2114 
2115 	if (w < cq->budget) {
2116 		mana_gd_ring_cq(gdma_queue, SET_ARM_BIT);
2117 		cq->work_done_since_doorbell = 0;
2118 		napi_complete_done(&cq->napi, w);
2119 	} else if (cq->work_done_since_doorbell >
2120 		   cq->gdma_cq->queue_size / COMP_ENTRY_SIZE * 4) {
2121 		/* MANA hardware requires at least one doorbell ring every 8
2122 		 * wraparounds of CQ even if there is no need to arm the CQ.
2123 		 * This driver rings the doorbell as soon as we have exceeded
2124 		 * 4 wraparounds.
2125 		 */
2126 		mana_gd_ring_cq(gdma_queue, 0);
2127 		cq->work_done_since_doorbell = 0;
2128 	}
2129 
2130 	return w;
2131 }
2132 
2133 static int mana_poll(struct napi_struct *napi, int budget)
2134 {
2135 	struct mana_cq *cq = container_of(napi, struct mana_cq, napi);
2136 	int w;
2137 
2138 	cq->work_done = 0;
2139 	cq->budget = budget;
2140 
2141 	w = mana_cq_handler(cq, cq->gdma_cq);
2142 
2143 	return min(w, budget);
2144 }
2145 
2146 static void mana_schedule_napi(void *context, struct gdma_queue *gdma_queue)
2147 {
2148 	struct mana_cq *cq = context;
2149 
2150 	napi_schedule_irqoff(&cq->napi);
2151 }
2152 
2153 static void mana_deinit_cq(struct mana_port_context *apc, struct mana_cq *cq)
2154 {
2155 	struct gdma_dev *gd = apc->ac->gdma_dev;
2156 
2157 	if (!cq->gdma_cq)
2158 		return;
2159 
2160 	mana_gd_destroy_queue(gd->gdma_context, cq->gdma_cq);
2161 }
2162 
2163 static void mana_deinit_txq(struct mana_port_context *apc, struct mana_txq *txq)
2164 {
2165 	struct gdma_dev *gd = apc->ac->gdma_dev;
2166 
2167 	if (!txq->gdma_sq)
2168 		return;
2169 
2170 	mana_gd_destroy_queue(gd->gdma_context, txq->gdma_sq);
2171 }
2172 
2173 static void mana_destroy_txq(struct mana_port_context *apc)
2174 {
2175 	struct napi_struct *napi;
2176 	int i;
2177 
2178 	if (!apc->tx_qp)
2179 		return;
2180 
2181 	for (i = 0; i < apc->num_queues; i++) {
2182 		debugfs_remove_recursive(apc->tx_qp[i].mana_tx_debugfs);
2183 		apc->tx_qp[i].mana_tx_debugfs = NULL;
2184 
2185 		napi = &apc->tx_qp[i].tx_cq.napi;
2186 		if (apc->tx_qp[i].txq.napi_initialized) {
2187 			napi_synchronize(napi);
2188 			napi_disable_locked(napi);
2189 			netif_napi_del_locked(napi);
2190 			apc->tx_qp[i].txq.napi_initialized = false;
2191 		}
2192 		mana_destroy_wq_obj(apc, GDMA_SQ, apc->tx_qp[i].tx_object);
2193 
2194 		mana_deinit_cq(apc, &apc->tx_qp[i].tx_cq);
2195 
2196 		mana_deinit_txq(apc, &apc->tx_qp[i].txq);
2197 	}
2198 
2199 	kfree(apc->tx_qp);
2200 	apc->tx_qp = NULL;
2201 }
2202 
2203 static void mana_create_txq_debugfs(struct mana_port_context *apc, int idx)
2204 {
2205 	struct mana_tx_qp *tx_qp = &apc->tx_qp[idx];
2206 	char qnum[32];
2207 
2208 	sprintf(qnum, "TX-%d", idx);
2209 	tx_qp->mana_tx_debugfs = debugfs_create_dir(qnum, apc->mana_port_debugfs);
2210 	debugfs_create_u32("sq_head", 0400, tx_qp->mana_tx_debugfs,
2211 			   &tx_qp->txq.gdma_sq->head);
2212 	debugfs_create_u32("sq_tail", 0400, tx_qp->mana_tx_debugfs,
2213 			   &tx_qp->txq.gdma_sq->tail);
2214 	debugfs_create_u32("sq_pend_skb_qlen", 0400, tx_qp->mana_tx_debugfs,
2215 			   &tx_qp->txq.pending_skbs.qlen);
2216 	debugfs_create_u32("cq_head", 0400, tx_qp->mana_tx_debugfs,
2217 			   &tx_qp->tx_cq.gdma_cq->head);
2218 	debugfs_create_u32("cq_tail", 0400, tx_qp->mana_tx_debugfs,
2219 			   &tx_qp->tx_cq.gdma_cq->tail);
2220 	debugfs_create_u32("cq_budget", 0400, tx_qp->mana_tx_debugfs,
2221 			   &tx_qp->tx_cq.budget);
2222 	debugfs_create_file("txq_dump", 0400, tx_qp->mana_tx_debugfs,
2223 			    tx_qp->txq.gdma_sq, &mana_dbg_q_fops);
2224 	debugfs_create_file("cq_dump", 0400, tx_qp->mana_tx_debugfs,
2225 			    tx_qp->tx_cq.gdma_cq, &mana_dbg_q_fops);
2226 }
2227 
2228 static int mana_create_txq(struct mana_port_context *apc,
2229 			   struct net_device *net)
2230 {
2231 	struct mana_context *ac = apc->ac;
2232 	struct gdma_dev *gd = ac->gdma_dev;
2233 	struct mana_obj_spec wq_spec;
2234 	struct mana_obj_spec cq_spec;
2235 	struct gdma_queue_spec spec;
2236 	struct gdma_context *gc;
2237 	struct mana_txq *txq;
2238 	struct mana_cq *cq;
2239 	u32 txq_size;
2240 	u32 cq_size;
2241 	int err;
2242 	int i;
2243 
2244 	apc->tx_qp = kcalloc(apc->num_queues, sizeof(struct mana_tx_qp),
2245 			     GFP_KERNEL);
2246 	if (!apc->tx_qp)
2247 		return -ENOMEM;
2248 
2249 	/*  The minimum size of the WQE is 32 bytes, hence
2250 	 *  apc->tx_queue_size represents the maximum number of WQEs
2251 	 *  the SQ can store. This value is then used to size other queues
2252 	 *  to prevent overflow.
2253 	 *  Also note that the txq_size is always going to be MANA_PAGE_ALIGNED,
2254 	 *  as min val of apc->tx_queue_size is 128 and that would make
2255 	 *  txq_size 128*32 = 4096 and the other higher values of apc->tx_queue_size
2256 	 *  are always power of two
2257 	 */
2258 	txq_size = apc->tx_queue_size * 32;
2259 
2260 	cq_size = apc->tx_queue_size * COMP_ENTRY_SIZE;
2261 
2262 	gc = gd->gdma_context;
2263 
2264 	for (i = 0; i < apc->num_queues; i++) {
2265 		apc->tx_qp[i].tx_object = INVALID_MANA_HANDLE;
2266 
2267 		/* Create SQ */
2268 		txq = &apc->tx_qp[i].txq;
2269 
2270 		u64_stats_init(&txq->stats.syncp);
2271 		txq->ndev = net;
2272 		txq->net_txq = netdev_get_tx_queue(net, i);
2273 		txq->vp_offset = apc->tx_vp_offset;
2274 		txq->napi_initialized = false;
2275 		skb_queue_head_init(&txq->pending_skbs);
2276 
2277 		memset(&spec, 0, sizeof(spec));
2278 		spec.type = GDMA_SQ;
2279 		spec.monitor_avl_buf = true;
2280 		spec.queue_size = txq_size;
2281 		err = mana_gd_create_mana_wq_cq(gd, &spec, &txq->gdma_sq);
2282 		if (err)
2283 			goto out;
2284 
2285 		/* Create SQ's CQ */
2286 		cq = &apc->tx_qp[i].tx_cq;
2287 		cq->type = MANA_CQ_TYPE_TX;
2288 
2289 		cq->txq = txq;
2290 
2291 		memset(&spec, 0, sizeof(spec));
2292 		spec.type = GDMA_CQ;
2293 		spec.monitor_avl_buf = false;
2294 		spec.queue_size = cq_size;
2295 		spec.cq.callback = mana_schedule_napi;
2296 		spec.cq.parent_eq = ac->eqs[i].eq;
2297 		spec.cq.context = cq;
2298 		err = mana_gd_create_mana_wq_cq(gd, &spec, &cq->gdma_cq);
2299 		if (err)
2300 			goto out;
2301 
2302 		memset(&wq_spec, 0, sizeof(wq_spec));
2303 		memset(&cq_spec, 0, sizeof(cq_spec));
2304 
2305 		wq_spec.gdma_region = txq->gdma_sq->mem_info.dma_region_handle;
2306 		wq_spec.queue_size = txq->gdma_sq->queue_size;
2307 
2308 		cq_spec.gdma_region = cq->gdma_cq->mem_info.dma_region_handle;
2309 		cq_spec.queue_size = cq->gdma_cq->queue_size;
2310 		cq_spec.modr_ctx_id = 0;
2311 		cq_spec.attached_eq = cq->gdma_cq->cq.parent->id;
2312 
2313 		err = mana_create_wq_obj(apc, apc->port_handle, GDMA_SQ,
2314 					 &wq_spec, &cq_spec,
2315 					 &apc->tx_qp[i].tx_object);
2316 
2317 		if (err)
2318 			goto out;
2319 
2320 		txq->gdma_sq->id = wq_spec.queue_index;
2321 		cq->gdma_cq->id = cq_spec.queue_index;
2322 
2323 		txq->gdma_sq->mem_info.dma_region_handle =
2324 			GDMA_INVALID_DMA_REGION;
2325 		cq->gdma_cq->mem_info.dma_region_handle =
2326 			GDMA_INVALID_DMA_REGION;
2327 
2328 		txq->gdma_txq_id = txq->gdma_sq->id;
2329 
2330 		cq->gdma_id = cq->gdma_cq->id;
2331 
2332 		if (WARN_ON(cq->gdma_id >= gc->max_num_cqs)) {
2333 			err = -EINVAL;
2334 			goto out;
2335 		}
2336 
2337 		gc->cq_table[cq->gdma_id] = cq->gdma_cq;
2338 
2339 		mana_create_txq_debugfs(apc, i);
2340 
2341 		set_bit(NAPI_STATE_NO_BUSY_POLL, &cq->napi.state);
2342 		netif_napi_add_locked(net, &cq->napi, mana_poll);
2343 		napi_enable_locked(&cq->napi);
2344 		txq->napi_initialized = true;
2345 
2346 		mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
2347 	}
2348 
2349 	return 0;
2350 out:
2351 	netdev_err(net, "Failed to create %d TX queues, %d\n",
2352 		   apc->num_queues, err);
2353 	mana_destroy_txq(apc);
2354 	return err;
2355 }
2356 
2357 static void mana_destroy_rxq(struct mana_port_context *apc,
2358 			     struct mana_rxq *rxq, bool napi_initialized)
2359 
2360 {
2361 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
2362 	struct mana_recv_buf_oob *rx_oob;
2363 	struct device *dev = gc->dev;
2364 	struct napi_struct *napi;
2365 	struct page *page;
2366 	int i;
2367 
2368 	if (!rxq)
2369 		return;
2370 
2371 	debugfs_remove_recursive(rxq->mana_rx_debugfs);
2372 	rxq->mana_rx_debugfs = NULL;
2373 
2374 	napi = &rxq->rx_cq.napi;
2375 
2376 	if (napi_initialized) {
2377 		napi_synchronize(napi);
2378 
2379 		napi_disable_locked(napi);
2380 		netif_napi_del_locked(napi);
2381 	}
2382 	xdp_rxq_info_unreg(&rxq->xdp_rxq);
2383 
2384 	mana_destroy_wq_obj(apc, GDMA_RQ, rxq->rxobj);
2385 
2386 	mana_deinit_cq(apc, &rxq->rx_cq);
2387 
2388 	if (rxq->xdp_save_va)
2389 		put_page(virt_to_head_page(rxq->xdp_save_va));
2390 
2391 	for (i = 0; i < rxq->num_rx_buf; i++) {
2392 		rx_oob = &rxq->rx_oobs[i];
2393 
2394 		if (!rx_oob->buf_va)
2395 			continue;
2396 
2397 		page = virt_to_head_page(rx_oob->buf_va);
2398 
2399 		if (rxq->frag_count == 1 || !rx_oob->from_pool) {
2400 			dma_unmap_single(dev, rx_oob->sgl[0].address,
2401 					 rx_oob->sgl[0].size, DMA_FROM_DEVICE);
2402 			mana_put_rx_page(rxq, page, rx_oob->from_pool);
2403 		} else {
2404 			page_pool_free_va(rxq->page_pool, rx_oob->buf_va, true);
2405 		}
2406 
2407 		rx_oob->buf_va = NULL;
2408 	}
2409 
2410 	page_pool_destroy(rxq->page_pool);
2411 
2412 	if (rxq->gdma_rq)
2413 		mana_gd_destroy_queue(gc, rxq->gdma_rq);
2414 
2415 	kfree(rxq);
2416 }
2417 
2418 static int mana_fill_rx_oob(struct mana_recv_buf_oob *rx_oob, u32 mem_key,
2419 			    struct mana_rxq *rxq, struct device *dev)
2420 {
2421 	struct mana_port_context *mpc = netdev_priv(rxq->ndev);
2422 	bool from_pool = false;
2423 	dma_addr_t da;
2424 	void *va;
2425 
2426 	if (mpc->rxbufs_pre)
2427 		va = mana_get_rxbuf_pre(rxq, &da);
2428 	else
2429 		va = mana_get_rxfrag(rxq, dev, &da, &from_pool);
2430 
2431 	if (!va)
2432 		return -ENOMEM;
2433 
2434 	rx_oob->buf_va = va;
2435 	rx_oob->from_pool = from_pool;
2436 
2437 	rx_oob->sgl[0].address = da;
2438 	rx_oob->sgl[0].size = rxq->datasize;
2439 	rx_oob->sgl[0].mem_key = mem_key;
2440 
2441 	return 0;
2442 }
2443 
2444 #define MANA_WQE_HEADER_SIZE 16
2445 #define MANA_WQE_SGE_SIZE 16
2446 
2447 static int mana_alloc_rx_wqe(struct mana_port_context *apc,
2448 			     struct mana_rxq *rxq, u32 *rxq_size, u32 *cq_size)
2449 {
2450 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
2451 	struct mana_recv_buf_oob *rx_oob;
2452 	struct device *dev = gc->dev;
2453 	u32 buf_idx;
2454 	int ret;
2455 
2456 	WARN_ON(rxq->datasize == 0);
2457 
2458 	*rxq_size = 0;
2459 	*cq_size = 0;
2460 
2461 	for (buf_idx = 0; buf_idx < rxq->num_rx_buf; buf_idx++) {
2462 		rx_oob = &rxq->rx_oobs[buf_idx];
2463 		memset(rx_oob, 0, sizeof(*rx_oob));
2464 
2465 		rx_oob->num_sge = 1;
2466 
2467 		ret = mana_fill_rx_oob(rx_oob, apc->ac->gdma_dev->gpa_mkey, rxq,
2468 				       dev);
2469 		if (ret)
2470 			return ret;
2471 
2472 		rx_oob->wqe_req.sgl = rx_oob->sgl;
2473 		rx_oob->wqe_req.num_sge = rx_oob->num_sge;
2474 		rx_oob->wqe_req.inline_oob_size = 0;
2475 		rx_oob->wqe_req.inline_oob_data = NULL;
2476 		rx_oob->wqe_req.flags = 0;
2477 		rx_oob->wqe_req.client_data_unit = 0;
2478 
2479 		*rxq_size += ALIGN(MANA_WQE_HEADER_SIZE +
2480 				   MANA_WQE_SGE_SIZE * rx_oob->num_sge, 32);
2481 		*cq_size += COMP_ENTRY_SIZE;
2482 	}
2483 
2484 	return 0;
2485 }
2486 
2487 static int mana_push_wqe(struct mana_rxq *rxq)
2488 {
2489 	struct mana_recv_buf_oob *rx_oob;
2490 	u32 buf_idx;
2491 	int err;
2492 
2493 	for (buf_idx = 0; buf_idx < rxq->num_rx_buf; buf_idx++) {
2494 		rx_oob = &rxq->rx_oobs[buf_idx];
2495 
2496 		err = mana_gd_post_and_ring(rxq->gdma_rq, &rx_oob->wqe_req,
2497 					    &rx_oob->wqe_inf);
2498 		if (err)
2499 			return -ENOSPC;
2500 	}
2501 
2502 	return 0;
2503 }
2504 
2505 static int mana_create_page_pool(struct mana_rxq *rxq, struct gdma_context *gc)
2506 {
2507 	struct mana_port_context *mpc = netdev_priv(rxq->ndev);
2508 	struct page_pool_params pprm = {};
2509 	int ret;
2510 
2511 	pprm.pool_size = mpc->rx_queue_size / rxq->frag_count + 1;
2512 	pprm.nid = gc->numa_node;
2513 	pprm.napi = &rxq->rx_cq.napi;
2514 	pprm.netdev = rxq->ndev;
2515 	pprm.order = get_order(rxq->alloc_size);
2516 	pprm.queue_idx = rxq->rxq_idx;
2517 	pprm.dev = gc->dev;
2518 
2519 	/* Let the page pool do the dma map when page sharing with multiple
2520 	 * fragments enabled for rx buffers.
2521 	 */
2522 	if (rxq->frag_count > 1) {
2523 		pprm.flags =  PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
2524 		pprm.max_len = PAGE_SIZE;
2525 		pprm.dma_dir = DMA_FROM_DEVICE;
2526 	}
2527 
2528 	rxq->page_pool = page_pool_create(&pprm);
2529 
2530 	if (IS_ERR(rxq->page_pool)) {
2531 		ret = PTR_ERR(rxq->page_pool);
2532 		rxq->page_pool = NULL;
2533 		return ret;
2534 	}
2535 
2536 	return 0;
2537 }
2538 
2539 static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
2540 					u32 rxq_idx, struct mana_eq *eq,
2541 					struct net_device *ndev)
2542 {
2543 	struct gdma_dev *gd = apc->ac->gdma_dev;
2544 	struct mana_obj_spec wq_spec;
2545 	struct mana_obj_spec cq_spec;
2546 	struct gdma_queue_spec spec;
2547 	struct mana_cq *cq = NULL;
2548 	struct gdma_context *gc;
2549 	u32 cq_size, rq_size;
2550 	struct mana_rxq *rxq;
2551 	int err;
2552 
2553 	gc = gd->gdma_context;
2554 
2555 	rxq = kzalloc(struct_size(rxq, rx_oobs, apc->rx_queue_size),
2556 		      GFP_KERNEL);
2557 	if (!rxq)
2558 		return NULL;
2559 
2560 	rxq->ndev = ndev;
2561 	rxq->num_rx_buf = apc->rx_queue_size;
2562 	rxq->rxq_idx = rxq_idx;
2563 	rxq->rxobj = INVALID_MANA_HANDLE;
2564 
2565 	mana_get_rxbuf_cfg(apc, ndev->mtu, &rxq->datasize, &rxq->alloc_size,
2566 			   &rxq->headroom, &rxq->frag_count);
2567 	/* Create page pool for RX queue */
2568 	err = mana_create_page_pool(rxq, gc);
2569 	if (err) {
2570 		netdev_err(ndev, "Create page pool err:%d\n", err);
2571 		goto out;
2572 	}
2573 
2574 	err = mana_alloc_rx_wqe(apc, rxq, &rq_size, &cq_size);
2575 	if (err)
2576 		goto out;
2577 
2578 	rq_size = MANA_PAGE_ALIGN(rq_size);
2579 	cq_size = MANA_PAGE_ALIGN(cq_size);
2580 
2581 	/* Create RQ */
2582 	memset(&spec, 0, sizeof(spec));
2583 	spec.type = GDMA_RQ;
2584 	spec.monitor_avl_buf = true;
2585 	spec.queue_size = rq_size;
2586 	err = mana_gd_create_mana_wq_cq(gd, &spec, &rxq->gdma_rq);
2587 	if (err)
2588 		goto out;
2589 
2590 	/* Create RQ's CQ */
2591 	cq = &rxq->rx_cq;
2592 	cq->type = MANA_CQ_TYPE_RX;
2593 	cq->rxq = rxq;
2594 
2595 	memset(&spec, 0, sizeof(spec));
2596 	spec.type = GDMA_CQ;
2597 	spec.monitor_avl_buf = false;
2598 	spec.queue_size = cq_size;
2599 	spec.cq.callback = mana_schedule_napi;
2600 	spec.cq.parent_eq = eq->eq;
2601 	spec.cq.context = cq;
2602 	err = mana_gd_create_mana_wq_cq(gd, &spec, &cq->gdma_cq);
2603 	if (err)
2604 		goto out;
2605 
2606 	memset(&wq_spec, 0, sizeof(wq_spec));
2607 	memset(&cq_spec, 0, sizeof(cq_spec));
2608 	wq_spec.gdma_region = rxq->gdma_rq->mem_info.dma_region_handle;
2609 	wq_spec.queue_size = rxq->gdma_rq->queue_size;
2610 
2611 	cq_spec.gdma_region = cq->gdma_cq->mem_info.dma_region_handle;
2612 	cq_spec.queue_size = cq->gdma_cq->queue_size;
2613 	cq_spec.modr_ctx_id = 0;
2614 	cq_spec.attached_eq = cq->gdma_cq->cq.parent->id;
2615 
2616 	err = mana_create_wq_obj(apc, apc->port_handle, GDMA_RQ,
2617 				 &wq_spec, &cq_spec, &rxq->rxobj);
2618 	if (err)
2619 		goto out;
2620 
2621 	rxq->gdma_rq->id = wq_spec.queue_index;
2622 	cq->gdma_cq->id = cq_spec.queue_index;
2623 
2624 	rxq->gdma_rq->mem_info.dma_region_handle = GDMA_INVALID_DMA_REGION;
2625 	cq->gdma_cq->mem_info.dma_region_handle = GDMA_INVALID_DMA_REGION;
2626 
2627 	rxq->gdma_id = rxq->gdma_rq->id;
2628 	cq->gdma_id = cq->gdma_cq->id;
2629 
2630 	err = mana_push_wqe(rxq);
2631 	if (err)
2632 		goto out;
2633 
2634 	if (WARN_ON(cq->gdma_id >= gc->max_num_cqs)) {
2635 		err = -EINVAL;
2636 		goto out;
2637 	}
2638 
2639 	gc->cq_table[cq->gdma_id] = cq->gdma_cq;
2640 
2641 	netif_napi_add_weight_locked(ndev, &cq->napi, mana_poll, 1);
2642 
2643 	WARN_ON(xdp_rxq_info_reg(&rxq->xdp_rxq, ndev, rxq_idx,
2644 				 cq->napi.napi_id));
2645 	WARN_ON(xdp_rxq_info_reg_mem_model(&rxq->xdp_rxq, MEM_TYPE_PAGE_POOL,
2646 					   rxq->page_pool));
2647 
2648 	napi_enable_locked(&cq->napi);
2649 
2650 	mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
2651 out:
2652 	if (!err)
2653 		return rxq;
2654 
2655 	netdev_err(ndev, "Failed to create RXQ: err = %d\n", err);
2656 
2657 	mana_destroy_rxq(apc, rxq, false);
2658 
2659 	if (cq)
2660 		mana_deinit_cq(apc, cq);
2661 
2662 	return NULL;
2663 }
2664 
2665 static void mana_create_rxq_debugfs(struct mana_port_context *apc, int idx)
2666 {
2667 	struct mana_rxq *rxq;
2668 	char qnum[32];
2669 
2670 	rxq = apc->rxqs[idx];
2671 
2672 	sprintf(qnum, "RX-%d", idx);
2673 	rxq->mana_rx_debugfs = debugfs_create_dir(qnum, apc->mana_port_debugfs);
2674 	debugfs_create_u32("rq_head", 0400, rxq->mana_rx_debugfs, &rxq->gdma_rq->head);
2675 	debugfs_create_u32("rq_tail", 0400, rxq->mana_rx_debugfs, &rxq->gdma_rq->tail);
2676 	debugfs_create_u32("rq_nbuf", 0400, rxq->mana_rx_debugfs, &rxq->num_rx_buf);
2677 	debugfs_create_u32("cq_head", 0400, rxq->mana_rx_debugfs,
2678 			   &rxq->rx_cq.gdma_cq->head);
2679 	debugfs_create_u32("cq_tail", 0400, rxq->mana_rx_debugfs,
2680 			   &rxq->rx_cq.gdma_cq->tail);
2681 	debugfs_create_u32("cq_budget", 0400, rxq->mana_rx_debugfs, &rxq->rx_cq.budget);
2682 	debugfs_create_file("rxq_dump", 0400, rxq->mana_rx_debugfs, rxq->gdma_rq, &mana_dbg_q_fops);
2683 	debugfs_create_file("cq_dump", 0400, rxq->mana_rx_debugfs, rxq->rx_cq.gdma_cq,
2684 			    &mana_dbg_q_fops);
2685 }
2686 
2687 static int mana_add_rx_queues(struct mana_port_context *apc,
2688 			      struct net_device *ndev)
2689 {
2690 	struct mana_context *ac = apc->ac;
2691 	struct mana_rxq *rxq;
2692 	int err = 0;
2693 	int i;
2694 
2695 	for (i = 0; i < apc->num_queues; i++) {
2696 		rxq = mana_create_rxq(apc, i, &ac->eqs[i], ndev);
2697 		if (!rxq) {
2698 			err = -ENOMEM;
2699 			netdev_err(ndev, "Failed to create rxq %d : %d\n", i, err);
2700 			goto out;
2701 		}
2702 
2703 		u64_stats_init(&rxq->stats.syncp);
2704 
2705 		apc->rxqs[i] = rxq;
2706 
2707 		mana_create_rxq_debugfs(apc, i);
2708 	}
2709 
2710 	apc->default_rxobj = apc->rxqs[0]->rxobj;
2711 out:
2712 	return err;
2713 }
2714 
2715 static void mana_destroy_vport(struct mana_port_context *apc)
2716 {
2717 	struct gdma_dev *gd = apc->ac->gdma_dev;
2718 	struct mana_rxq *rxq;
2719 	u32 rxq_idx;
2720 
2721 	for (rxq_idx = 0; rxq_idx < apc->num_queues; rxq_idx++) {
2722 		rxq = apc->rxqs[rxq_idx];
2723 		if (!rxq)
2724 			continue;
2725 
2726 		mana_destroy_rxq(apc, rxq, true);
2727 		apc->rxqs[rxq_idx] = NULL;
2728 	}
2729 
2730 	mana_destroy_txq(apc);
2731 	mana_uncfg_vport(apc);
2732 
2733 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode)
2734 		mana_pf_deregister_hw_vport(apc);
2735 }
2736 
2737 static int mana_create_vport(struct mana_port_context *apc,
2738 			     struct net_device *net)
2739 {
2740 	struct gdma_dev *gd = apc->ac->gdma_dev;
2741 	int err;
2742 
2743 	apc->default_rxobj = INVALID_MANA_HANDLE;
2744 
2745 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode) {
2746 		err = mana_pf_register_hw_vport(apc);
2747 		if (err)
2748 			return err;
2749 	}
2750 
2751 	err = mana_cfg_vport(apc, gd->pdid, gd->doorbell);
2752 	if (err)
2753 		return err;
2754 
2755 	return mana_create_txq(apc, net);
2756 }
2757 
2758 static int mana_rss_table_alloc(struct mana_port_context *apc)
2759 {
2760 	if (!apc->indir_table_sz) {
2761 		netdev_err(apc->ndev,
2762 			   "Indirection table size not set for vPort %d\n",
2763 			   apc->port_idx);
2764 		return -EINVAL;
2765 	}
2766 
2767 	apc->indir_table = kcalloc(apc->indir_table_sz, sizeof(u32), GFP_KERNEL);
2768 	if (!apc->indir_table)
2769 		return -ENOMEM;
2770 
2771 	apc->rxobj_table = kcalloc(apc->indir_table_sz, sizeof(mana_handle_t), GFP_KERNEL);
2772 	if (!apc->rxobj_table) {
2773 		kfree(apc->indir_table);
2774 		return -ENOMEM;
2775 	}
2776 
2777 	return 0;
2778 }
2779 
2780 static void mana_rss_table_init(struct mana_port_context *apc)
2781 {
2782 	int i;
2783 
2784 	for (i = 0; i < apc->indir_table_sz; i++)
2785 		apc->indir_table[i] =
2786 			ethtool_rxfh_indir_default(i, apc->num_queues);
2787 }
2788 
2789 int mana_config_rss(struct mana_port_context *apc, enum TRI_STATE rx,
2790 		    bool update_hash, bool update_tab)
2791 {
2792 	u32 queue_idx;
2793 	int err;
2794 	int i;
2795 
2796 	if (update_tab) {
2797 		for (i = 0; i < apc->indir_table_sz; i++) {
2798 			queue_idx = apc->indir_table[i];
2799 			apc->rxobj_table[i] = apc->rxqs[queue_idx]->rxobj;
2800 		}
2801 	}
2802 
2803 	err = mana_cfg_vport_steering(apc, rx, true, update_hash, update_tab);
2804 	if (err)
2805 		return err;
2806 
2807 	mana_fence_rqs(apc);
2808 
2809 	return 0;
2810 }
2811 
2812 void mana_query_gf_stats(struct mana_port_context *apc)
2813 {
2814 	struct mana_query_gf_stat_resp resp = {};
2815 	struct mana_query_gf_stat_req req = {};
2816 	struct net_device *ndev = apc->ndev;
2817 	int err;
2818 
2819 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_GF_STAT,
2820 			     sizeof(req), sizeof(resp));
2821 	req.hdr.resp.msg_version = GDMA_MESSAGE_V2;
2822 	req.req_stats = STATISTICS_FLAGS_RX_DISCARDS_NO_WQE |
2823 			STATISTICS_FLAGS_RX_ERRORS_VPORT_DISABLED |
2824 			STATISTICS_FLAGS_HC_RX_BYTES |
2825 			STATISTICS_FLAGS_HC_RX_UCAST_PACKETS |
2826 			STATISTICS_FLAGS_HC_RX_UCAST_BYTES |
2827 			STATISTICS_FLAGS_HC_RX_MCAST_PACKETS |
2828 			STATISTICS_FLAGS_HC_RX_MCAST_BYTES |
2829 			STATISTICS_FLAGS_HC_RX_BCAST_PACKETS |
2830 			STATISTICS_FLAGS_HC_RX_BCAST_BYTES |
2831 			STATISTICS_FLAGS_TX_ERRORS_GF_DISABLED |
2832 			STATISTICS_FLAGS_TX_ERRORS_VPORT_DISABLED |
2833 			STATISTICS_FLAGS_TX_ERRORS_INVAL_VPORT_OFFSET_PACKETS |
2834 			STATISTICS_FLAGS_TX_ERRORS_VLAN_ENFORCEMENT |
2835 			STATISTICS_FLAGS_TX_ERRORS_ETH_TYPE_ENFORCEMENT |
2836 			STATISTICS_FLAGS_TX_ERRORS_SA_ENFORCEMENT |
2837 			STATISTICS_FLAGS_TX_ERRORS_SQPDID_ENFORCEMENT |
2838 			STATISTICS_FLAGS_TX_ERRORS_CQPDID_ENFORCEMENT |
2839 			STATISTICS_FLAGS_TX_ERRORS_MTU_VIOLATION |
2840 			STATISTICS_FLAGS_TX_ERRORS_INVALID_OOB |
2841 			STATISTICS_FLAGS_HC_TX_BYTES |
2842 			STATISTICS_FLAGS_HC_TX_UCAST_PACKETS |
2843 			STATISTICS_FLAGS_HC_TX_UCAST_BYTES |
2844 			STATISTICS_FLAGS_HC_TX_MCAST_PACKETS |
2845 			STATISTICS_FLAGS_HC_TX_MCAST_BYTES |
2846 			STATISTICS_FLAGS_HC_TX_BCAST_PACKETS |
2847 			STATISTICS_FLAGS_HC_TX_BCAST_BYTES |
2848 			STATISTICS_FLAGS_TX_ERRORS_GDMA_ERROR;
2849 
2850 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
2851 				sizeof(resp));
2852 	if (err) {
2853 		netdev_err(ndev, "Failed to query GF stats: %d\n", err);
2854 		return;
2855 	}
2856 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_GF_STAT,
2857 				   sizeof(resp));
2858 	if (err || resp.hdr.status) {
2859 		netdev_err(ndev, "Failed to query GF stats: %d, 0x%x\n", err,
2860 			   resp.hdr.status);
2861 		return;
2862 	}
2863 
2864 	apc->eth_stats.hc_rx_discards_no_wqe = resp.rx_discards_nowqe;
2865 	apc->eth_stats.hc_rx_err_vport_disabled = resp.rx_err_vport_disabled;
2866 	apc->eth_stats.hc_rx_bytes = resp.hc_rx_bytes;
2867 	apc->eth_stats.hc_rx_ucast_pkts = resp.hc_rx_ucast_pkts;
2868 	apc->eth_stats.hc_rx_ucast_bytes = resp.hc_rx_ucast_bytes;
2869 	apc->eth_stats.hc_rx_bcast_pkts = resp.hc_rx_bcast_pkts;
2870 	apc->eth_stats.hc_rx_bcast_bytes = resp.hc_rx_bcast_bytes;
2871 	apc->eth_stats.hc_rx_mcast_pkts = resp.hc_rx_mcast_pkts;
2872 	apc->eth_stats.hc_rx_mcast_bytes = resp.hc_rx_mcast_bytes;
2873 	apc->eth_stats.hc_tx_err_gf_disabled = resp.tx_err_gf_disabled;
2874 	apc->eth_stats.hc_tx_err_vport_disabled = resp.tx_err_vport_disabled;
2875 	apc->eth_stats.hc_tx_err_inval_vportoffset_pkt =
2876 					     resp.tx_err_inval_vport_offset_pkt;
2877 	apc->eth_stats.hc_tx_err_vlan_enforcement =
2878 					     resp.tx_err_vlan_enforcement;
2879 	apc->eth_stats.hc_tx_err_eth_type_enforcement =
2880 					     resp.tx_err_ethtype_enforcement;
2881 	apc->eth_stats.hc_tx_err_sa_enforcement = resp.tx_err_SA_enforcement;
2882 	apc->eth_stats.hc_tx_err_sqpdid_enforcement =
2883 					     resp.tx_err_SQPDID_enforcement;
2884 	apc->eth_stats.hc_tx_err_cqpdid_enforcement =
2885 					     resp.tx_err_CQPDID_enforcement;
2886 	apc->eth_stats.hc_tx_err_mtu_violation = resp.tx_err_mtu_violation;
2887 	apc->eth_stats.hc_tx_err_inval_oob = resp.tx_err_inval_oob;
2888 	apc->eth_stats.hc_tx_bytes = resp.hc_tx_bytes;
2889 	apc->eth_stats.hc_tx_ucast_pkts = resp.hc_tx_ucast_pkts;
2890 	apc->eth_stats.hc_tx_ucast_bytes = resp.hc_tx_ucast_bytes;
2891 	apc->eth_stats.hc_tx_bcast_pkts = resp.hc_tx_bcast_pkts;
2892 	apc->eth_stats.hc_tx_bcast_bytes = resp.hc_tx_bcast_bytes;
2893 	apc->eth_stats.hc_tx_mcast_pkts = resp.hc_tx_mcast_pkts;
2894 	apc->eth_stats.hc_tx_mcast_bytes = resp.hc_tx_mcast_bytes;
2895 	apc->eth_stats.hc_tx_err_gdma = resp.tx_err_gdma;
2896 }
2897 
2898 void mana_query_phy_stats(struct mana_port_context *apc)
2899 {
2900 	struct mana_query_phy_stat_resp resp = {};
2901 	struct mana_query_phy_stat_req req = {};
2902 	struct net_device *ndev = apc->ndev;
2903 	int err;
2904 
2905 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_PHY_STAT,
2906 			     sizeof(req), sizeof(resp));
2907 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
2908 				sizeof(resp));
2909 	if (err)
2910 		return;
2911 
2912 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_PHY_STAT,
2913 				   sizeof(resp));
2914 	if (err || resp.hdr.status) {
2915 		netdev_err(ndev,
2916 			   "Failed to query PHY stats: %d, resp:0x%x\n",
2917 				err, resp.hdr.status);
2918 		return;
2919 	}
2920 
2921 	/* Aggregate drop counters */
2922 	apc->phy_stats.rx_pkt_drop_phy = resp.rx_pkt_drop_phy;
2923 	apc->phy_stats.tx_pkt_drop_phy = resp.tx_pkt_drop_phy;
2924 
2925 	/* Per TC traffic Counters */
2926 	apc->phy_stats.rx_pkt_tc0_phy = resp.rx_pkt_tc0_phy;
2927 	apc->phy_stats.tx_pkt_tc0_phy = resp.tx_pkt_tc0_phy;
2928 	apc->phy_stats.rx_pkt_tc1_phy = resp.rx_pkt_tc1_phy;
2929 	apc->phy_stats.tx_pkt_tc1_phy = resp.tx_pkt_tc1_phy;
2930 	apc->phy_stats.rx_pkt_tc2_phy = resp.rx_pkt_tc2_phy;
2931 	apc->phy_stats.tx_pkt_tc2_phy = resp.tx_pkt_tc2_phy;
2932 	apc->phy_stats.rx_pkt_tc3_phy = resp.rx_pkt_tc3_phy;
2933 	apc->phy_stats.tx_pkt_tc3_phy = resp.tx_pkt_tc3_phy;
2934 	apc->phy_stats.rx_pkt_tc4_phy = resp.rx_pkt_tc4_phy;
2935 	apc->phy_stats.tx_pkt_tc4_phy = resp.tx_pkt_tc4_phy;
2936 	apc->phy_stats.rx_pkt_tc5_phy = resp.rx_pkt_tc5_phy;
2937 	apc->phy_stats.tx_pkt_tc5_phy = resp.tx_pkt_tc5_phy;
2938 	apc->phy_stats.rx_pkt_tc6_phy = resp.rx_pkt_tc6_phy;
2939 	apc->phy_stats.tx_pkt_tc6_phy = resp.tx_pkt_tc6_phy;
2940 	apc->phy_stats.rx_pkt_tc7_phy = resp.rx_pkt_tc7_phy;
2941 	apc->phy_stats.tx_pkt_tc7_phy = resp.tx_pkt_tc7_phy;
2942 
2943 	/* Per TC byte Counters */
2944 	apc->phy_stats.rx_byte_tc0_phy = resp.rx_byte_tc0_phy;
2945 	apc->phy_stats.tx_byte_tc0_phy = resp.tx_byte_tc0_phy;
2946 	apc->phy_stats.rx_byte_tc1_phy = resp.rx_byte_tc1_phy;
2947 	apc->phy_stats.tx_byte_tc1_phy = resp.tx_byte_tc1_phy;
2948 	apc->phy_stats.rx_byte_tc2_phy = resp.rx_byte_tc2_phy;
2949 	apc->phy_stats.tx_byte_tc2_phy = resp.tx_byte_tc2_phy;
2950 	apc->phy_stats.rx_byte_tc3_phy = resp.rx_byte_tc3_phy;
2951 	apc->phy_stats.tx_byte_tc3_phy = resp.tx_byte_tc3_phy;
2952 	apc->phy_stats.rx_byte_tc4_phy = resp.rx_byte_tc4_phy;
2953 	apc->phy_stats.tx_byte_tc4_phy = resp.tx_byte_tc4_phy;
2954 	apc->phy_stats.rx_byte_tc5_phy = resp.rx_byte_tc5_phy;
2955 	apc->phy_stats.tx_byte_tc5_phy = resp.tx_byte_tc5_phy;
2956 	apc->phy_stats.rx_byte_tc6_phy = resp.rx_byte_tc6_phy;
2957 	apc->phy_stats.tx_byte_tc6_phy = resp.tx_byte_tc6_phy;
2958 	apc->phy_stats.rx_byte_tc7_phy = resp.rx_byte_tc7_phy;
2959 	apc->phy_stats.tx_byte_tc7_phy = resp.tx_byte_tc7_phy;
2960 
2961 	/* Per TC pause Counters */
2962 	apc->phy_stats.rx_pause_tc0_phy = resp.rx_pause_tc0_phy;
2963 	apc->phy_stats.tx_pause_tc0_phy = resp.tx_pause_tc0_phy;
2964 	apc->phy_stats.rx_pause_tc1_phy = resp.rx_pause_tc1_phy;
2965 	apc->phy_stats.tx_pause_tc1_phy = resp.tx_pause_tc1_phy;
2966 	apc->phy_stats.rx_pause_tc2_phy = resp.rx_pause_tc2_phy;
2967 	apc->phy_stats.tx_pause_tc2_phy = resp.tx_pause_tc2_phy;
2968 	apc->phy_stats.rx_pause_tc3_phy = resp.rx_pause_tc3_phy;
2969 	apc->phy_stats.tx_pause_tc3_phy = resp.tx_pause_tc3_phy;
2970 	apc->phy_stats.rx_pause_tc4_phy = resp.rx_pause_tc4_phy;
2971 	apc->phy_stats.tx_pause_tc4_phy = resp.tx_pause_tc4_phy;
2972 	apc->phy_stats.rx_pause_tc5_phy = resp.rx_pause_tc5_phy;
2973 	apc->phy_stats.tx_pause_tc5_phy = resp.tx_pause_tc5_phy;
2974 	apc->phy_stats.rx_pause_tc6_phy = resp.rx_pause_tc6_phy;
2975 	apc->phy_stats.tx_pause_tc6_phy = resp.tx_pause_tc6_phy;
2976 	apc->phy_stats.rx_pause_tc7_phy = resp.rx_pause_tc7_phy;
2977 	apc->phy_stats.tx_pause_tc7_phy = resp.tx_pause_tc7_phy;
2978 }
2979 
2980 static int mana_init_port(struct net_device *ndev)
2981 {
2982 	struct mana_port_context *apc = netdev_priv(ndev);
2983 	struct gdma_dev *gd = apc->ac->gdma_dev;
2984 	u32 max_txq, max_rxq, max_queues;
2985 	int port_idx = apc->port_idx;
2986 	struct gdma_context *gc;
2987 	char vport[32];
2988 	int err;
2989 
2990 	err = mana_init_port_context(apc);
2991 	if (err)
2992 		return err;
2993 
2994 	gc = gd->gdma_context;
2995 
2996 	err = mana_query_vport_cfg(apc, port_idx, &max_txq, &max_rxq,
2997 				   &apc->indir_table_sz);
2998 	if (err) {
2999 		netdev_err(ndev, "Failed to query info for vPort %d\n",
3000 			   port_idx);
3001 		goto reset_apc;
3002 	}
3003 
3004 	max_queues = min_t(u32, max_txq, max_rxq);
3005 	if (apc->max_queues > max_queues)
3006 		apc->max_queues = max_queues;
3007 
3008 	if (apc->num_queues > apc->max_queues)
3009 		apc->num_queues = apc->max_queues;
3010 
3011 	eth_hw_addr_set(ndev, apc->mac_addr);
3012 	sprintf(vport, "vport%d", port_idx);
3013 	apc->mana_port_debugfs = debugfs_create_dir(vport, gc->mana_pci_debugfs);
3014 	return 0;
3015 
3016 reset_apc:
3017 	mana_cleanup_port_context(apc);
3018 	return err;
3019 }
3020 
3021 int mana_alloc_queues(struct net_device *ndev)
3022 {
3023 	struct mana_port_context *apc = netdev_priv(ndev);
3024 	struct gdma_dev *gd = apc->ac->gdma_dev;
3025 	int err;
3026 
3027 	err = mana_create_vport(apc, ndev);
3028 	if (err) {
3029 		netdev_err(ndev, "Failed to create vPort %u : %d\n", apc->port_idx, err);
3030 		return err;
3031 	}
3032 
3033 	err = netif_set_real_num_tx_queues(ndev, apc->num_queues);
3034 	if (err) {
3035 		netdev_err(ndev,
3036 			   "netif_set_real_num_tx_queues () failed for ndev with num_queues %u : %d\n",
3037 			   apc->num_queues, err);
3038 		goto destroy_vport;
3039 	}
3040 
3041 	err = mana_add_rx_queues(apc, ndev);
3042 	if (err)
3043 		goto destroy_vport;
3044 
3045 	apc->rss_state = apc->num_queues > 1 ? TRI_STATE_TRUE : TRI_STATE_FALSE;
3046 
3047 	err = netif_set_real_num_rx_queues(ndev, apc->num_queues);
3048 	if (err) {
3049 		netdev_err(ndev,
3050 			   "netif_set_real_num_rx_queues () failed for ndev with num_queues %u : %d\n",
3051 			   apc->num_queues, err);
3052 		goto destroy_vport;
3053 	}
3054 
3055 	mana_rss_table_init(apc);
3056 
3057 	err = mana_config_rss(apc, TRI_STATE_TRUE, true, true);
3058 	if (err) {
3059 		netdev_err(ndev, "Failed to configure RSS table: %d\n", err);
3060 		goto destroy_vport;
3061 	}
3062 
3063 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode) {
3064 		err = mana_pf_register_filter(apc);
3065 		if (err)
3066 			goto destroy_vport;
3067 	}
3068 
3069 	mana_chn_setxdp(apc, mana_xdp_get(apc));
3070 
3071 	return 0;
3072 
3073 destroy_vport:
3074 	mana_destroy_vport(apc);
3075 	return err;
3076 }
3077 
3078 int mana_attach(struct net_device *ndev)
3079 {
3080 	struct mana_port_context *apc = netdev_priv(ndev);
3081 	int err;
3082 
3083 	ASSERT_RTNL();
3084 
3085 	err = mana_init_port(ndev);
3086 	if (err)
3087 		return err;
3088 
3089 	if (apc->port_st_save) {
3090 		err = mana_alloc_queues(ndev);
3091 		if (err) {
3092 			mana_cleanup_port_context(apc);
3093 			return err;
3094 		}
3095 	}
3096 
3097 	apc->port_is_up = apc->port_st_save;
3098 
3099 	/* Ensure port state updated before txq state */
3100 	smp_wmb();
3101 
3102 	netif_device_attach(ndev);
3103 
3104 	return 0;
3105 }
3106 
3107 static int mana_dealloc_queues(struct net_device *ndev)
3108 {
3109 	struct mana_port_context *apc = netdev_priv(ndev);
3110 	unsigned long timeout = jiffies + 120 * HZ;
3111 	struct gdma_dev *gd = apc->ac->gdma_dev;
3112 	struct mana_txq *txq;
3113 	struct sk_buff *skb;
3114 	int i, err;
3115 	u32 tsleep;
3116 
3117 	if (apc->port_is_up)
3118 		return -EINVAL;
3119 
3120 	mana_chn_setxdp(apc, NULL);
3121 
3122 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode)
3123 		mana_pf_deregister_filter(apc);
3124 
3125 	/* No packet can be transmitted now since apc->port_is_up is false.
3126 	 * There is still a tiny chance that mana_poll_tx_cq() can re-enable
3127 	 * a txq because it may not timely see apc->port_is_up being cleared
3128 	 * to false, but it doesn't matter since mana_start_xmit() drops any
3129 	 * new packets due to apc->port_is_up being false.
3130 	 *
3131 	 * Drain all the in-flight TX packets.
3132 	 * A timeout of 120 seconds for all the queues is used.
3133 	 * This will break the while loop when h/w is not responding.
3134 	 * This value of 120 has been decided here considering max
3135 	 * number of queues.
3136 	 */
3137 
3138 	for (i = 0; i < apc->num_queues; i++) {
3139 		txq = &apc->tx_qp[i].txq;
3140 		tsleep = 1000;
3141 		while (atomic_read(&txq->pending_sends) > 0 &&
3142 		       time_before(jiffies, timeout)) {
3143 			usleep_range(tsleep, tsleep + 1000);
3144 			tsleep <<= 1;
3145 		}
3146 		if (atomic_read(&txq->pending_sends)) {
3147 			err = pcie_flr(to_pci_dev(gd->gdma_context->dev));
3148 			if (err) {
3149 				netdev_err(ndev, "flr failed %d with %d pkts pending in txq %u\n",
3150 					   err, atomic_read(&txq->pending_sends),
3151 					   txq->gdma_txq_id);
3152 			}
3153 			break;
3154 		}
3155 	}
3156 
3157 	for (i = 0; i < apc->num_queues; i++) {
3158 		txq = &apc->tx_qp[i].txq;
3159 		while ((skb = skb_dequeue(&txq->pending_skbs))) {
3160 			mana_unmap_skb(skb, apc);
3161 			dev_kfree_skb_any(skb);
3162 		}
3163 		atomic_set(&txq->pending_sends, 0);
3164 	}
3165 	/* We're 100% sure the queues can no longer be woken up, because
3166 	 * we're sure now mana_poll_tx_cq() can't be running.
3167 	 */
3168 
3169 	apc->rss_state = TRI_STATE_FALSE;
3170 	err = mana_config_rss(apc, TRI_STATE_FALSE, false, false);
3171 	if (err && mana_en_need_log(apc, err))
3172 		netdev_err(ndev, "Failed to disable vPort: %d\n", err);
3173 
3174 	/* Even in err case, still need to cleanup the vPort */
3175 	mana_destroy_vport(apc);
3176 
3177 	return 0;
3178 }
3179 
3180 int mana_detach(struct net_device *ndev, bool from_close)
3181 {
3182 	struct mana_port_context *apc = netdev_priv(ndev);
3183 	int err;
3184 
3185 	ASSERT_RTNL();
3186 
3187 	apc->port_st_save = apc->port_is_up;
3188 	apc->port_is_up = false;
3189 
3190 	/* Ensure port state updated before txq state */
3191 	smp_wmb();
3192 
3193 	netif_tx_disable(ndev);
3194 
3195 	if (apc->port_st_save) {
3196 		err = mana_dealloc_queues(ndev);
3197 		if (err) {
3198 			netdev_err(ndev, "%s failed to deallocate queues: %d\n", __func__, err);
3199 			return err;
3200 		}
3201 	}
3202 
3203 	if (!from_close) {
3204 		netif_device_detach(ndev);
3205 		mana_cleanup_port_context(apc);
3206 	}
3207 
3208 	return 0;
3209 }
3210 
3211 static int mana_probe_port(struct mana_context *ac, int port_idx,
3212 			   struct net_device **ndev_storage)
3213 {
3214 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
3215 	struct mana_port_context *apc;
3216 	struct net_device *ndev;
3217 	int err;
3218 
3219 	ndev = alloc_etherdev_mq(sizeof(struct mana_port_context),
3220 				 gc->max_num_queues);
3221 	if (!ndev)
3222 		return -ENOMEM;
3223 
3224 	*ndev_storage = ndev;
3225 
3226 	apc = netdev_priv(ndev);
3227 	apc->ac = ac;
3228 	apc->ndev = ndev;
3229 	apc->max_queues = gc->max_num_queues;
3230 	apc->num_queues = gc->max_num_queues;
3231 	apc->tx_queue_size = DEF_TX_BUFFERS_PER_QUEUE;
3232 	apc->rx_queue_size = DEF_RX_BUFFERS_PER_QUEUE;
3233 	apc->port_handle = INVALID_MANA_HANDLE;
3234 	apc->pf_filter_handle = INVALID_MANA_HANDLE;
3235 	apc->port_idx = port_idx;
3236 
3237 	mutex_init(&apc->vport_mutex);
3238 	apc->vport_use_count = 0;
3239 
3240 	ndev->netdev_ops = &mana_devops;
3241 	ndev->ethtool_ops = &mana_ethtool_ops;
3242 	ndev->mtu = ETH_DATA_LEN;
3243 	ndev->max_mtu = gc->adapter_mtu - ETH_HLEN;
3244 	ndev->min_mtu = ETH_MIN_MTU;
3245 	ndev->needed_headroom = MANA_HEADROOM;
3246 	ndev->dev_port = port_idx;
3247 	SET_NETDEV_DEV(ndev, gc->dev);
3248 
3249 	netif_set_tso_max_size(ndev, GSO_MAX_SIZE);
3250 
3251 	netif_carrier_off(ndev);
3252 
3253 	netdev_rss_key_fill(apc->hashkey, MANA_HASH_KEY_SIZE);
3254 
3255 	err = mana_init_port(ndev);
3256 	if (err)
3257 		goto free_net;
3258 
3259 	err = mana_rss_table_alloc(apc);
3260 	if (err)
3261 		goto reset_apc;
3262 
3263 	netdev_lockdep_set_classes(ndev);
3264 
3265 	ndev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3266 	ndev->hw_features |= NETIF_F_RXCSUM;
3267 	ndev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3268 	ndev->hw_features |= NETIF_F_RXHASH;
3269 	ndev->features = ndev->hw_features | NETIF_F_HW_VLAN_CTAG_TX |
3270 			 NETIF_F_HW_VLAN_CTAG_RX;
3271 	ndev->vlan_features = ndev->features;
3272 	xdp_set_features_flag(ndev, NETDEV_XDP_ACT_BASIC |
3273 			      NETDEV_XDP_ACT_REDIRECT |
3274 			      NETDEV_XDP_ACT_NDO_XMIT);
3275 
3276 	err = register_netdev(ndev);
3277 	if (err) {
3278 		netdev_err(ndev, "Unable to register netdev.\n");
3279 		goto free_indir;
3280 	}
3281 
3282 	netif_carrier_on(ndev);
3283 
3284 	debugfs_create_u32("current_speed", 0400, apc->mana_port_debugfs, &apc->speed);
3285 
3286 	return 0;
3287 
3288 free_indir:
3289 	mana_cleanup_indir_table(apc);
3290 reset_apc:
3291 	mana_cleanup_port_context(apc);
3292 free_net:
3293 	*ndev_storage = NULL;
3294 	netdev_err(ndev, "Failed to probe vPort %d: %d\n", port_idx, err);
3295 	free_netdev(ndev);
3296 	return err;
3297 }
3298 
3299 static void adev_release(struct device *dev)
3300 {
3301 	struct mana_adev *madev = container_of(dev, struct mana_adev, adev.dev);
3302 
3303 	kfree(madev);
3304 }
3305 
3306 static void remove_adev(struct gdma_dev *gd)
3307 {
3308 	struct auxiliary_device *adev = gd->adev;
3309 	int id = adev->id;
3310 
3311 	auxiliary_device_delete(adev);
3312 	auxiliary_device_uninit(adev);
3313 
3314 	mana_adev_idx_free(id);
3315 	gd->adev = NULL;
3316 }
3317 
3318 static int add_adev(struct gdma_dev *gd, const char *name)
3319 {
3320 	struct auxiliary_device *adev;
3321 	struct mana_adev *madev;
3322 	int ret;
3323 
3324 	madev = kzalloc(sizeof(*madev), GFP_KERNEL);
3325 	if (!madev)
3326 		return -ENOMEM;
3327 
3328 	adev = &madev->adev;
3329 	ret = mana_adev_idx_alloc();
3330 	if (ret < 0)
3331 		goto idx_fail;
3332 	adev->id = ret;
3333 
3334 	adev->name = name;
3335 	adev->dev.parent = gd->gdma_context->dev;
3336 	adev->dev.release = adev_release;
3337 	madev->mdev = gd;
3338 
3339 	ret = auxiliary_device_init(adev);
3340 	if (ret)
3341 		goto init_fail;
3342 
3343 	/* madev is owned by the auxiliary device */
3344 	madev = NULL;
3345 	ret = auxiliary_device_add(adev);
3346 	if (ret)
3347 		goto add_fail;
3348 
3349 	gd->adev = adev;
3350 	dev_dbg(gd->gdma_context->dev,
3351 		"Auxiliary device added successfully\n");
3352 	return 0;
3353 
3354 add_fail:
3355 	auxiliary_device_uninit(adev);
3356 
3357 init_fail:
3358 	mana_adev_idx_free(adev->id);
3359 
3360 idx_fail:
3361 	kfree(madev);
3362 
3363 	return ret;
3364 }
3365 
3366 static void mana_rdma_service_handle(struct work_struct *work)
3367 {
3368 	struct mana_service_work *serv_work =
3369 		container_of(work, struct mana_service_work, work);
3370 	struct gdma_dev *gd = serv_work->gdma_dev;
3371 	struct device *dev = gd->gdma_context->dev;
3372 	int ret;
3373 
3374 	if (READ_ONCE(gd->rdma_teardown))
3375 		goto out;
3376 
3377 	switch (serv_work->event) {
3378 	case GDMA_SERVICE_TYPE_RDMA_SUSPEND:
3379 		if (!gd->adev || gd->is_suspended)
3380 			break;
3381 
3382 		remove_adev(gd);
3383 		gd->is_suspended = true;
3384 		break;
3385 
3386 	case GDMA_SERVICE_TYPE_RDMA_RESUME:
3387 		if (!gd->is_suspended)
3388 			break;
3389 
3390 		ret = add_adev(gd, "rdma");
3391 		if (ret)
3392 			dev_err(dev, "Failed to add adev on resume: %d\n", ret);
3393 		else
3394 			gd->is_suspended = false;
3395 		break;
3396 
3397 	default:
3398 		dev_warn(dev, "unknown adev service event %u\n",
3399 			 serv_work->event);
3400 		break;
3401 	}
3402 
3403 out:
3404 	kfree(serv_work);
3405 }
3406 
3407 int mana_rdma_service_event(struct gdma_context *gc, enum gdma_service_type event)
3408 {
3409 	struct gdma_dev *gd = &gc->mana_ib;
3410 	struct mana_service_work *serv_work;
3411 
3412 	if (gd->dev_id.type != GDMA_DEVICE_MANA_IB) {
3413 		/* RDMA device is not detected on pci */
3414 		return 0;
3415 	}
3416 
3417 	serv_work = kzalloc(sizeof(*serv_work), GFP_ATOMIC);
3418 	if (!serv_work)
3419 		return -ENOMEM;
3420 
3421 	serv_work->event = event;
3422 	serv_work->gdma_dev = gd;
3423 
3424 	INIT_WORK(&serv_work->work, mana_rdma_service_handle);
3425 	queue_work(gc->service_wq, &serv_work->work);
3426 
3427 	return 0;
3428 }
3429 
3430 int mana_probe(struct gdma_dev *gd, bool resuming)
3431 {
3432 	struct gdma_context *gc = gd->gdma_context;
3433 	struct mana_context *ac = gd->driver_data;
3434 	struct device *dev = gc->dev;
3435 	u8 bm_hostmode = 0;
3436 	u16 num_ports = 0;
3437 	int err;
3438 	int i;
3439 
3440 	dev_info(dev,
3441 		 "Microsoft Azure Network Adapter protocol version: %d.%d.%d\n",
3442 		 MANA_MAJOR_VERSION, MANA_MINOR_VERSION, MANA_MICRO_VERSION);
3443 
3444 	err = mana_gd_register_device(gd);
3445 	if (err)
3446 		return err;
3447 
3448 	if (!resuming) {
3449 		ac = kzalloc(sizeof(*ac), GFP_KERNEL);
3450 		if (!ac)
3451 			return -ENOMEM;
3452 
3453 		ac->gdma_dev = gd;
3454 		gd->driver_data = ac;
3455 	}
3456 
3457 	err = mana_create_eq(ac);
3458 	if (err) {
3459 		dev_err(dev, "Failed to create EQs: %d\n", err);
3460 		goto out;
3461 	}
3462 
3463 	err = mana_query_device_cfg(ac, MANA_MAJOR_VERSION, MANA_MINOR_VERSION,
3464 				    MANA_MICRO_VERSION, &num_ports, &bm_hostmode);
3465 	if (err)
3466 		goto out;
3467 
3468 	ac->bm_hostmode = bm_hostmode;
3469 
3470 	if (!resuming) {
3471 		ac->num_ports = num_ports;
3472 
3473 		INIT_WORK(&ac->link_change_work, mana_link_state_handle);
3474 	} else {
3475 		if (ac->num_ports != num_ports) {
3476 			dev_err(dev, "The number of vPorts changed: %d->%d\n",
3477 				ac->num_ports, num_ports);
3478 			err = -EPROTO;
3479 			goto out;
3480 		}
3481 
3482 		enable_work(&ac->link_change_work);
3483 	}
3484 
3485 	if (ac->num_ports == 0)
3486 		dev_err(dev, "Failed to detect any vPort\n");
3487 
3488 	if (ac->num_ports > MAX_PORTS_IN_MANA_DEV)
3489 		ac->num_ports = MAX_PORTS_IN_MANA_DEV;
3490 
3491 	if (!resuming) {
3492 		for (i = 0; i < ac->num_ports; i++) {
3493 			err = mana_probe_port(ac, i, &ac->ports[i]);
3494 			/* we log the port for which the probe failed and stop
3495 			 * probes for subsequent ports.
3496 			 * Note that we keep running ports, for which the probes
3497 			 * were successful, unless add_adev fails too
3498 			 */
3499 			if (err) {
3500 				dev_err(dev, "Probe Failed for port %d\n", i);
3501 				break;
3502 			}
3503 		}
3504 	} else {
3505 		for (i = 0; i < ac->num_ports; i++) {
3506 			rtnl_lock();
3507 			err = mana_attach(ac->ports[i]);
3508 			rtnl_unlock();
3509 			/* we log the port for which the attach failed and stop
3510 			 * attach for subsequent ports
3511 			 * Note that we keep running ports, for which the attach
3512 			 * were successful, unless add_adev fails too
3513 			 */
3514 			if (err) {
3515 				dev_err(dev, "Attach Failed for port %d\n", i);
3516 				break;
3517 			}
3518 		}
3519 	}
3520 
3521 	err = add_adev(gd, "eth");
3522 out:
3523 	if (err) {
3524 		mana_remove(gd, false);
3525 	} else {
3526 		dev_dbg(dev, "gd=%p, id=%u, num_ports=%d, type=%u, instance=%u\n",
3527 			gd, gd->dev_id.as_uint32, ac->num_ports,
3528 			gd->dev_id.type, gd->dev_id.instance);
3529 		dev_dbg(dev, "%s succeeded\n", __func__);
3530 	}
3531 
3532 	return err;
3533 }
3534 
3535 void mana_remove(struct gdma_dev *gd, bool suspending)
3536 {
3537 	struct gdma_context *gc = gd->gdma_context;
3538 	struct mana_context *ac = gd->driver_data;
3539 	struct mana_port_context *apc;
3540 	struct device *dev = gc->dev;
3541 	struct net_device *ndev;
3542 	int err;
3543 	int i;
3544 
3545 	disable_work_sync(&ac->link_change_work);
3546 
3547 	/* adev currently doesn't support suspending, always remove it */
3548 	if (gd->adev)
3549 		remove_adev(gd);
3550 
3551 	for (i = 0; i < ac->num_ports; i++) {
3552 		ndev = ac->ports[i];
3553 		apc = netdev_priv(ndev);
3554 		if (!ndev) {
3555 			if (i == 0)
3556 				dev_err(dev, "No net device to remove\n");
3557 			goto out;
3558 		}
3559 
3560 		/* All cleanup actions should stay after rtnl_lock(), otherwise
3561 		 * other functions may access partially cleaned up data.
3562 		 */
3563 		rtnl_lock();
3564 
3565 		err = mana_detach(ndev, false);
3566 		if (err)
3567 			netdev_err(ndev, "Failed to detach vPort %d: %d\n",
3568 				   i, err);
3569 
3570 		if (suspending) {
3571 			/* No need to unregister the ndev. */
3572 			rtnl_unlock();
3573 			continue;
3574 		}
3575 
3576 		unregister_netdevice(ndev);
3577 		mana_cleanup_indir_table(apc);
3578 
3579 		rtnl_unlock();
3580 
3581 		free_netdev(ndev);
3582 	}
3583 
3584 	mana_destroy_eq(ac);
3585 out:
3586 	mana_gd_deregister_device(gd);
3587 
3588 	if (suspending)
3589 		return;
3590 
3591 	gd->driver_data = NULL;
3592 	gd->gdma_context = NULL;
3593 	kfree(ac);
3594 	dev_dbg(dev, "%s succeeded\n", __func__);
3595 }
3596 
3597 int mana_rdma_probe(struct gdma_dev *gd)
3598 {
3599 	int err = 0;
3600 
3601 	if (gd->dev_id.type != GDMA_DEVICE_MANA_IB) {
3602 		/* RDMA device is not detected on pci */
3603 		return err;
3604 	}
3605 
3606 	err = mana_gd_register_device(gd);
3607 	if (err)
3608 		return err;
3609 
3610 	err = add_adev(gd, "rdma");
3611 	if (err)
3612 		mana_gd_deregister_device(gd);
3613 
3614 	return err;
3615 }
3616 
3617 void mana_rdma_remove(struct gdma_dev *gd)
3618 {
3619 	struct gdma_context *gc = gd->gdma_context;
3620 
3621 	if (gd->dev_id.type != GDMA_DEVICE_MANA_IB) {
3622 		/* RDMA device is not detected on pci */
3623 		return;
3624 	}
3625 
3626 	WRITE_ONCE(gd->rdma_teardown, true);
3627 	flush_workqueue(gc->service_wq);
3628 
3629 	if (gd->adev)
3630 		remove_adev(gd);
3631 
3632 	mana_gd_deregister_device(gd);
3633 }
3634 
3635 struct net_device *mana_get_primary_netdev(struct mana_context *ac,
3636 					   u32 port_index,
3637 					   netdevice_tracker *tracker)
3638 {
3639 	struct net_device *ndev;
3640 
3641 	if (port_index >= ac->num_ports)
3642 		return NULL;
3643 
3644 	rcu_read_lock();
3645 
3646 	/* If mana is used in netvsc, the upper netdevice should be returned. */
3647 	ndev = netdev_master_upper_dev_get_rcu(ac->ports[port_index]);
3648 
3649 	/* If there is no upper device, use the parent Ethernet device */
3650 	if (!ndev)
3651 		ndev = ac->ports[port_index];
3652 
3653 	netdev_hold(ndev, tracker, GFP_ATOMIC);
3654 	rcu_read_unlock();
3655 
3656 	return ndev;
3657 }
3658 EXPORT_SYMBOL_NS(mana_get_primary_netdev, "NET_MANA");
3659