xref: /linux/drivers/net/ethernet/microsoft/mana/mana_en.c (revision a34b0e4e21d6be3c3d620aa7f9dfbf0e9550c19e)
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 #include <linux/skbuff.h>
15 
16 #include <net/checksum.h>
17 #include <net/ip6_checksum.h>
18 #include <net/netdev_lock.h>
19 #include <net/page_pool/helpers.h>
20 #include <net/xdp.h>
21 
22 #include <net/mana/mana.h>
23 #include <net/mana/mana_auxiliary.h>
24 #include <net/mana/hw_channel.h>
25 
26 static DEFINE_IDA(mana_adev_ida);
27 
28 static int mana_adev_idx_alloc(void)
29 {
30 	return ida_alloc(&mana_adev_ida, GFP_KERNEL);
31 }
32 
33 static void mana_adev_idx_free(int idx)
34 {
35 	ida_free(&mana_adev_ida, idx);
36 }
37 
38 static ssize_t mana_dbg_q_read(struct file *filp, char __user *buf, size_t count,
39 			       loff_t *pos)
40 {
41 	struct gdma_queue *gdma_q = filp->private_data;
42 
43 	return simple_read_from_buffer(buf, count, pos, gdma_q->queue_mem_ptr,
44 				       gdma_q->queue_size);
45 }
46 
47 static const struct file_operations mana_dbg_q_fops = {
48 	.owner  = THIS_MODULE,
49 	.open   = simple_open,
50 	.read   = mana_dbg_q_read,
51 };
52 
53 static bool mana_en_need_log(struct mana_port_context *apc, int err)
54 {
55 	if (apc && apc->ac && apc->ac->gdma_dev &&
56 	    apc->ac->gdma_dev->gdma_context)
57 		return mana_need_log(apc->ac->gdma_dev->gdma_context, err);
58 	else
59 		return true;
60 }
61 
62 static void mana_put_rx_page(struct mana_rxq *rxq, struct page *page,
63 			     bool from_pool)
64 {
65 	if (from_pool)
66 		page_pool_put_full_page(rxq->page_pool, page, false);
67 	else
68 		put_page(page);
69 }
70 
71 /* Microsoft Azure Network Adapter (MANA) functions */
72 
73 static int mana_open(struct net_device *ndev)
74 {
75 	struct mana_port_context *apc = netdev_priv(ndev);
76 	int err;
77 	err = mana_alloc_queues(ndev);
78 
79 	if (err) {
80 		netdev_err(ndev, "%s failed to allocate queues: %d\n", __func__, err);
81 		return err;
82 	}
83 
84 	apc->port_is_up = true;
85 
86 	/* Ensure port state updated before txq state */
87 	smp_wmb();
88 
89 	netif_tx_wake_all_queues(ndev);
90 	netdev_dbg(ndev, "%s successful\n", __func__);
91 	return 0;
92 }
93 
94 static int mana_close(struct net_device *ndev)
95 {
96 	struct mana_port_context *apc = netdev_priv(ndev);
97 
98 	if (!apc->port_is_up)
99 		return 0;
100 
101 	return mana_detach(ndev, true);
102 }
103 
104 static void mana_link_state_handle(struct work_struct *w)
105 {
106 	struct mana_context *ac;
107 	struct net_device *ndev;
108 	u32 link_event;
109 	bool link_up;
110 	int i;
111 
112 	ac = container_of(w, struct mana_context, link_change_work);
113 
114 	rtnl_lock();
115 
116 	link_event = READ_ONCE(ac->link_event);
117 
118 	if (link_event == HWC_DATA_HW_LINK_CONNECT)
119 		link_up = true;
120 	else if (link_event == HWC_DATA_HW_LINK_DISCONNECT)
121 		link_up = false;
122 	else
123 		goto out;
124 
125 	/* Process all ports */
126 	for (i = 0; i < ac->num_ports; i++) {
127 		ndev = ac->ports[i];
128 		if (!ndev)
129 			continue;
130 
131 		if (link_up) {
132 			netif_carrier_on(ndev);
133 
134 			__netdev_notify_peers(ndev);
135 		} else {
136 			netif_carrier_off(ndev);
137 		}
138 	}
139 
140 out:
141 	rtnl_unlock();
142 }
143 
144 static bool mana_can_tx(struct gdma_queue *wq)
145 {
146 	return mana_gd_wq_avail_space(wq) >= MAX_TX_WQE_SIZE;
147 }
148 
149 static unsigned int mana_checksum_info(struct sk_buff *skb)
150 {
151 	if (skb->protocol == htons(ETH_P_IP)) {
152 		struct iphdr *ip = ip_hdr(skb);
153 
154 		if (ip->protocol == IPPROTO_TCP)
155 			return IPPROTO_TCP;
156 
157 		if (ip->protocol == IPPROTO_UDP)
158 			return IPPROTO_UDP;
159 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
160 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
161 
162 		if (ip6->nexthdr == IPPROTO_TCP)
163 			return IPPROTO_TCP;
164 
165 		if (ip6->nexthdr == IPPROTO_UDP)
166 			return IPPROTO_UDP;
167 	}
168 
169 	/* No csum offloading */
170 	return 0;
171 }
172 
173 static void mana_add_sge(struct mana_tx_package *tp, struct mana_skb_head *ash,
174 			 int sg_i, dma_addr_t da, int sge_len, u32 gpa_mkey)
175 {
176 	ash->dma_handle[sg_i] = da;
177 	ash->size[sg_i] = sge_len;
178 
179 	tp->wqe_req.sgl[sg_i].address = da;
180 	tp->wqe_req.sgl[sg_i].mem_key = gpa_mkey;
181 	tp->wqe_req.sgl[sg_i].size = sge_len;
182 }
183 
184 static int mana_map_skb(struct sk_buff *skb, struct mana_port_context *apc,
185 			struct mana_tx_package *tp, int gso_hs)
186 {
187 	struct mana_skb_head *ash = (struct mana_skb_head *)skb->head;
188 	int hsg = 1; /* num of SGEs of linear part */
189 	struct gdma_dev *gd = apc->ac->gdma_dev;
190 	int skb_hlen = skb_headlen(skb);
191 	int sge0_len, sge1_len = 0;
192 	struct gdma_context *gc;
193 	struct device *dev;
194 	skb_frag_t *frag;
195 	dma_addr_t da;
196 	int sg_i;
197 	int i;
198 
199 	gc = gd->gdma_context;
200 	dev = gc->dev;
201 
202 	if (gso_hs && gso_hs < skb_hlen) {
203 		sge0_len = gso_hs;
204 		sge1_len = skb_hlen - gso_hs;
205 	} else {
206 		sge0_len = skb_hlen;
207 	}
208 
209 	da = dma_map_single(dev, skb->data, sge0_len, DMA_TO_DEVICE);
210 	if (dma_mapping_error(dev, da))
211 		return -ENOMEM;
212 
213 	mana_add_sge(tp, ash, 0, da, sge0_len, gd->gpa_mkey);
214 
215 	if (sge1_len) {
216 		sg_i = 1;
217 		da = dma_map_single(dev, skb->data + sge0_len, sge1_len,
218 				    DMA_TO_DEVICE);
219 		if (dma_mapping_error(dev, da))
220 			goto frag_err;
221 
222 		mana_add_sge(tp, ash, sg_i, da, sge1_len, gd->gpa_mkey);
223 		hsg = 2;
224 	}
225 
226 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
227 		sg_i = hsg + i;
228 
229 		frag = &skb_shinfo(skb)->frags[i];
230 		da = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
231 				      DMA_TO_DEVICE);
232 		if (dma_mapping_error(dev, da))
233 			goto frag_err;
234 
235 		mana_add_sge(tp, ash, sg_i, da, skb_frag_size(frag),
236 			     gd->gpa_mkey);
237 	}
238 
239 	return 0;
240 
241 frag_err:
242 	if (net_ratelimit())
243 		netdev_err(apc->ndev, "Failed to map skb of size %u to DMA\n",
244 			   skb->len);
245 	for (i = sg_i - 1; i >= hsg; i--)
246 		dma_unmap_page(dev, ash->dma_handle[i], ash->size[i],
247 			       DMA_TO_DEVICE);
248 
249 	for (i = hsg - 1; i >= 0; i--)
250 		dma_unmap_single(dev, ash->dma_handle[i], ash->size[i],
251 				 DMA_TO_DEVICE);
252 
253 	return -ENOMEM;
254 }
255 
256 /* Handle the case when GSO SKB linear length is too large.
257  * MANA NIC requires GSO packets to put only the packet header to SGE0.
258  * So, we need 2 SGEs for the skb linear part which contains more than the
259  * header.
260  * Return a positive value for the number of SGEs, or a negative value
261  * for an error.
262  */
263 static int mana_fix_skb_head(struct net_device *ndev, struct sk_buff *skb,
264 			     int gso_hs)
265 {
266 	int num_sge = 1 + skb_shinfo(skb)->nr_frags;
267 	int skb_hlen = skb_headlen(skb);
268 
269 	if (gso_hs < skb_hlen) {
270 		num_sge++;
271 	} else if (gso_hs > skb_hlen) {
272 		if (net_ratelimit())
273 			netdev_err(ndev,
274 				   "TX nonlinear head: hs:%d, skb_hlen:%d\n",
275 				   gso_hs, skb_hlen);
276 
277 		return -EINVAL;
278 	}
279 
280 	return num_sge;
281 }
282 
283 /* Get the GSO packet's header size */
284 static int mana_get_gso_hs(struct sk_buff *skb)
285 {
286 	int gso_hs;
287 
288 	if (skb->encapsulation) {
289 		gso_hs = skb_inner_tcp_all_headers(skb);
290 	} else {
291 		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
292 			gso_hs = skb_transport_offset(skb) +
293 				 sizeof(struct udphdr);
294 		} else {
295 			gso_hs = skb_tcp_all_headers(skb);
296 		}
297 	}
298 
299 	return gso_hs;
300 }
301 
302 static void mana_per_port_queue_reset_work_handler(struct work_struct *work)
303 {
304 	struct mana_port_context *apc = container_of(work,
305 						     struct mana_port_context,
306 						     queue_reset_work);
307 	struct net_device *ndev = apc->ndev;
308 	int err;
309 
310 	rtnl_lock();
311 
312 	/* Pre-allocate buffers to prevent failure in mana_attach later */
313 	err = mana_pre_alloc_rxbufs(apc, ndev->mtu, apc->num_queues);
314 	if (err) {
315 		netdev_err(ndev, "Insufficient memory for reset post tx stall detection\n");
316 		goto out;
317 	}
318 
319 	err = mana_detach(ndev, false);
320 	if (err) {
321 		netdev_err(ndev, "mana_detach failed: %d\n", err);
322 		goto dealloc_pre_rxbufs;
323 	}
324 
325 	err = mana_attach(ndev);
326 	if (err)
327 		netdev_err(ndev, "mana_attach failed: %d\n", err);
328 
329 dealloc_pre_rxbufs:
330 	mana_pre_dealloc_rxbufs(apc);
331 out:
332 	rtnl_unlock();
333 }
334 
335 netdev_tx_t mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
336 {
337 	enum mana_tx_pkt_format pkt_fmt = MANA_SHORT_PKT_FMT;
338 	struct mana_port_context *apc = netdev_priv(ndev);
339 	int gso_hs = 0; /* zero for non-GSO pkts */
340 	u16 txq_idx = skb_get_queue_mapping(skb);
341 	struct gdma_dev *gd = apc->ac->gdma_dev;
342 	bool ipv4 = false, ipv6 = false;
343 	struct mana_tx_package pkg = {};
344 	struct netdev_queue *net_txq;
345 	struct mana_stats_tx *tx_stats;
346 	struct gdma_queue *gdma_sq;
347 	int err, len, num_gso_seg;
348 	unsigned int csum_type;
349 	struct mana_txq *txq;
350 	struct mana_cq *cq;
351 
352 	if (unlikely(!apc->port_is_up))
353 		goto tx_drop;
354 
355 	if (skb_cow_head(skb, MANA_HEADROOM))
356 		goto tx_drop_count;
357 
358 	if (unlikely(ipv6_hopopt_jumbo_remove(skb)))
359 		goto tx_drop_count;
360 
361 	txq = &apc->tx_qp[txq_idx].txq;
362 	gdma_sq = txq->gdma_sq;
363 	cq = &apc->tx_qp[txq_idx].tx_cq;
364 	tx_stats = &txq->stats;
365 
366 	BUILD_BUG_ON(MAX_TX_WQE_SGL_ENTRIES != MANA_MAX_TX_WQE_SGL_ENTRIES);
367 	if (MAX_SKB_FRAGS + 2 > MAX_TX_WQE_SGL_ENTRIES &&
368 	    skb_shinfo(skb)->nr_frags + 2 > MAX_TX_WQE_SGL_ENTRIES) {
369 		/* GSO skb with Hardware SGE limit exceeded is not expected here
370 		 * as they are handled in mana_features_check() callback
371 		 */
372 		if (skb_linearize(skb)) {
373 			netdev_warn_once(ndev, "Failed to linearize skb with nr_frags=%d and is_gso=%d\n",
374 					 skb_shinfo(skb)->nr_frags,
375 					 skb_is_gso(skb));
376 			goto tx_drop_count;
377 		}
378 		apc->eth_stats.tx_linear_pkt_cnt++;
379 	}
380 
381 	pkg.tx_oob.s_oob.vcq_num = cq->gdma_id;
382 	pkg.tx_oob.s_oob.vsq_frame = txq->vsq_frame;
383 
384 	if (txq->vp_offset > MANA_SHORT_VPORT_OFFSET_MAX) {
385 		pkg.tx_oob.l_oob.long_vp_offset = txq->vp_offset;
386 		pkt_fmt = MANA_LONG_PKT_FMT;
387 	} else {
388 		pkg.tx_oob.s_oob.short_vp_offset = txq->vp_offset;
389 	}
390 
391 	if (skb_vlan_tag_present(skb)) {
392 		pkt_fmt = MANA_LONG_PKT_FMT;
393 		pkg.tx_oob.l_oob.inject_vlan_pri_tag = 1;
394 		pkg.tx_oob.l_oob.pcp = skb_vlan_tag_get_prio(skb);
395 		pkg.tx_oob.l_oob.dei = skb_vlan_tag_get_cfi(skb);
396 		pkg.tx_oob.l_oob.vlan_id = skb_vlan_tag_get_id(skb);
397 	}
398 
399 	pkg.tx_oob.s_oob.pkt_fmt = pkt_fmt;
400 
401 	if (pkt_fmt == MANA_SHORT_PKT_FMT) {
402 		pkg.wqe_req.inline_oob_size = sizeof(struct mana_tx_short_oob);
403 		u64_stats_update_begin(&tx_stats->syncp);
404 		tx_stats->short_pkt_fmt++;
405 		u64_stats_update_end(&tx_stats->syncp);
406 	} else {
407 		pkg.wqe_req.inline_oob_size = sizeof(struct mana_tx_oob);
408 		u64_stats_update_begin(&tx_stats->syncp);
409 		tx_stats->long_pkt_fmt++;
410 		u64_stats_update_end(&tx_stats->syncp);
411 	}
412 
413 	pkg.wqe_req.inline_oob_data = &pkg.tx_oob;
414 	pkg.wqe_req.flags = 0;
415 	pkg.wqe_req.client_data_unit = 0;
416 
417 	pkg.wqe_req.num_sge = 1 + skb_shinfo(skb)->nr_frags;
418 
419 	if (skb->protocol == htons(ETH_P_IP))
420 		ipv4 = true;
421 	else if (skb->protocol == htons(ETH_P_IPV6))
422 		ipv6 = true;
423 
424 	if (skb_is_gso(skb)) {
425 		int num_sge;
426 
427 		gso_hs = mana_get_gso_hs(skb);
428 
429 		num_sge = mana_fix_skb_head(ndev, skb, gso_hs);
430 		if (num_sge > 0)
431 			pkg.wqe_req.num_sge = num_sge;
432 		else
433 			goto tx_drop_count;
434 
435 		u64_stats_update_begin(&tx_stats->syncp);
436 		if (skb->encapsulation) {
437 			tx_stats->tso_inner_packets++;
438 			tx_stats->tso_inner_bytes += skb->len - gso_hs;
439 		} else {
440 			tx_stats->tso_packets++;
441 			tx_stats->tso_bytes += skb->len - gso_hs;
442 		}
443 		u64_stats_update_end(&tx_stats->syncp);
444 
445 		pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
446 		pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
447 
448 		pkg.tx_oob.s_oob.comp_iphdr_csum = 1;
449 		pkg.tx_oob.s_oob.comp_tcp_csum = 1;
450 		pkg.tx_oob.s_oob.trans_off = skb_transport_offset(skb);
451 
452 		pkg.wqe_req.client_data_unit = skb_shinfo(skb)->gso_size;
453 		pkg.wqe_req.flags = GDMA_WR_OOB_IN_SGL | GDMA_WR_PAD_BY_SGE0;
454 		if (ipv4) {
455 			ip_hdr(skb)->tot_len = 0;
456 			ip_hdr(skb)->check = 0;
457 			tcp_hdr(skb)->check =
458 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
459 						   ip_hdr(skb)->daddr, 0,
460 						   IPPROTO_TCP, 0);
461 		} else {
462 			ipv6_hdr(skb)->payload_len = 0;
463 			tcp_hdr(skb)->check =
464 				~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
465 						 &ipv6_hdr(skb)->daddr, 0,
466 						 IPPROTO_TCP, 0);
467 		}
468 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
469 		csum_type = mana_checksum_info(skb);
470 
471 		u64_stats_update_begin(&tx_stats->syncp);
472 		tx_stats->csum_partial++;
473 		u64_stats_update_end(&tx_stats->syncp);
474 
475 		if (csum_type == IPPROTO_TCP) {
476 			pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
477 			pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
478 
479 			pkg.tx_oob.s_oob.comp_tcp_csum = 1;
480 			pkg.tx_oob.s_oob.trans_off = skb_transport_offset(skb);
481 
482 		} else if (csum_type == IPPROTO_UDP) {
483 			pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
484 			pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
485 
486 			pkg.tx_oob.s_oob.comp_udp_csum = 1;
487 		} else {
488 			/* Can't do offload of this type of checksum */
489 			if (skb_checksum_help(skb))
490 				goto tx_drop_count;
491 		}
492 	}
493 
494 	if (pkg.wqe_req.num_sge <= ARRAY_SIZE(pkg.sgl_array)) {
495 		pkg.wqe_req.sgl = pkg.sgl_array;
496 	} else {
497 		pkg.sgl_ptr = kmalloc_array(pkg.wqe_req.num_sge,
498 					    sizeof(struct gdma_sge),
499 					    GFP_ATOMIC);
500 		if (!pkg.sgl_ptr)
501 			goto tx_drop_count;
502 
503 		pkg.wqe_req.sgl = pkg.sgl_ptr;
504 	}
505 
506 	if (mana_map_skb(skb, apc, &pkg, gso_hs)) {
507 		u64_stats_update_begin(&tx_stats->syncp);
508 		tx_stats->mana_map_err++;
509 		u64_stats_update_end(&tx_stats->syncp);
510 		goto free_sgl_ptr;
511 	}
512 
513 	skb_queue_tail(&txq->pending_skbs, skb);
514 
515 	len = skb->len;
516 	num_gso_seg = skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1;
517 	net_txq = netdev_get_tx_queue(ndev, txq_idx);
518 
519 	err = mana_gd_post_work_request(gdma_sq, &pkg.wqe_req,
520 					(struct gdma_posted_wqe_info *)skb->cb);
521 	if (!mana_can_tx(gdma_sq)) {
522 		netif_tx_stop_queue(net_txq);
523 		apc->eth_stats.stop_queue++;
524 	}
525 
526 	if (err) {
527 		(void)skb_dequeue_tail(&txq->pending_skbs);
528 		mana_unmap_skb(skb, apc);
529 		netdev_warn(ndev, "Failed to post TX OOB: %d\n", err);
530 		goto free_sgl_ptr;
531 	}
532 
533 	err = NETDEV_TX_OK;
534 	atomic_inc(&txq->pending_sends);
535 
536 	mana_gd_wq_ring_doorbell(gd->gdma_context, gdma_sq);
537 
538 	/* skb may be freed after mana_gd_post_work_request. Do not use it. */
539 	skb = NULL;
540 
541 	/* Populated the packet and bytes counters based on post GSO packet
542 	 * calculations
543 	 */
544 	tx_stats = &txq->stats;
545 	u64_stats_update_begin(&tx_stats->syncp);
546 	tx_stats->packets += num_gso_seg;
547 	tx_stats->bytes += len + ((num_gso_seg - 1) * gso_hs);
548 	u64_stats_update_end(&tx_stats->syncp);
549 
550 	if (netif_tx_queue_stopped(net_txq) && mana_can_tx(gdma_sq)) {
551 		netif_tx_wake_queue(net_txq);
552 		apc->eth_stats.wake_queue++;
553 	}
554 
555 	kfree(pkg.sgl_ptr);
556 	return err;
557 
558 free_sgl_ptr:
559 	kfree(pkg.sgl_ptr);
560 tx_drop_count:
561 	ndev->stats.tx_dropped++;
562 tx_drop:
563 	dev_kfree_skb_any(skb);
564 	return NETDEV_TX_OK;
565 }
566 
567 #if (MAX_SKB_FRAGS + 2 > MANA_MAX_TX_WQE_SGL_ENTRIES)
568 static netdev_features_t mana_features_check(struct sk_buff *skb,
569 					     struct net_device *ndev,
570 					     netdev_features_t features)
571 {
572 	if (skb_shinfo(skb)->nr_frags + 2 > MAX_TX_WQE_SGL_ENTRIES) {
573 		/* Exceeds HW SGE limit.
574 		 * GSO case:
575 		 *   Disable GSO so the stack will software-segment the skb
576 		 *   into smaller skbs that fit the SGE budget.
577 		 * Non-GSO case:
578 		 *   The xmit path will attempt skb_linearize() as a fallback.
579 		 */
580 		features &= ~NETIF_F_GSO_MASK;
581 	}
582 	return features;
583 }
584 #endif
585 
586 static void mana_get_stats64(struct net_device *ndev,
587 			     struct rtnl_link_stats64 *st)
588 {
589 	struct mana_port_context *apc = netdev_priv(ndev);
590 	unsigned int num_queues = apc->num_queues;
591 	struct mana_stats_rx *rx_stats;
592 	struct mana_stats_tx *tx_stats;
593 	unsigned int start;
594 	u64 packets, bytes;
595 	int q;
596 
597 	if (!apc->port_is_up)
598 		return;
599 
600 	netdev_stats_to_stats64(st, &ndev->stats);
601 
602 	if (apc->ac->hwc_timeout_occurred)
603 		netdev_warn_once(ndev, "HWC timeout occurred\n");
604 
605 	st->rx_missed_errors = apc->ac->hc_stats.hc_rx_discards_no_wqe;
606 
607 	for (q = 0; q < num_queues; q++) {
608 		rx_stats = &apc->rxqs[q]->stats;
609 
610 		do {
611 			start = u64_stats_fetch_begin(&rx_stats->syncp);
612 			packets = rx_stats->packets;
613 			bytes = rx_stats->bytes;
614 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
615 
616 		st->rx_packets += packets;
617 		st->rx_bytes += bytes;
618 	}
619 
620 	for (q = 0; q < num_queues; q++) {
621 		tx_stats = &apc->tx_qp[q].txq.stats;
622 
623 		do {
624 			start = u64_stats_fetch_begin(&tx_stats->syncp);
625 			packets = tx_stats->packets;
626 			bytes = tx_stats->bytes;
627 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
628 
629 		st->tx_packets += packets;
630 		st->tx_bytes += bytes;
631 	}
632 }
633 
634 static int mana_get_tx_queue(struct net_device *ndev, struct sk_buff *skb,
635 			     int old_q)
636 {
637 	struct mana_port_context *apc = netdev_priv(ndev);
638 	u32 hash = skb_get_hash(skb);
639 	struct sock *sk = skb->sk;
640 	int txq;
641 
642 	txq = apc->indir_table[hash & (apc->indir_table_sz - 1)];
643 
644 	if (txq != old_q && sk && sk_fullsock(sk) &&
645 	    rcu_access_pointer(sk->sk_dst_cache))
646 		sk_tx_queue_set(sk, txq);
647 
648 	return txq;
649 }
650 
651 static u16 mana_select_queue(struct net_device *ndev, struct sk_buff *skb,
652 			     struct net_device *sb_dev)
653 {
654 	int txq;
655 
656 	if (ndev->real_num_tx_queues == 1)
657 		return 0;
658 
659 	txq = sk_tx_queue_get(skb->sk);
660 
661 	if (txq < 0 || skb->ooo_okay || txq >= ndev->real_num_tx_queues) {
662 		if (skb_rx_queue_recorded(skb))
663 			txq = skb_get_rx_queue(skb);
664 		else
665 			txq = mana_get_tx_queue(ndev, skb, txq);
666 	}
667 
668 	return txq;
669 }
670 
671 /* Release pre-allocated RX buffers */
672 void mana_pre_dealloc_rxbufs(struct mana_port_context *mpc)
673 {
674 	struct device *dev;
675 	int i;
676 
677 	dev = mpc->ac->gdma_dev->gdma_context->dev;
678 
679 	if (!mpc->rxbufs_pre)
680 		goto out1;
681 
682 	if (!mpc->das_pre)
683 		goto out2;
684 
685 	while (mpc->rxbpre_total) {
686 		i = --mpc->rxbpre_total;
687 		dma_unmap_single(dev, mpc->das_pre[i], mpc->rxbpre_datasize,
688 				 DMA_FROM_DEVICE);
689 		put_page(virt_to_head_page(mpc->rxbufs_pre[i]));
690 	}
691 
692 	kfree(mpc->das_pre);
693 	mpc->das_pre = NULL;
694 
695 out2:
696 	kfree(mpc->rxbufs_pre);
697 	mpc->rxbufs_pre = NULL;
698 
699 out1:
700 	mpc->rxbpre_datasize = 0;
701 	mpc->rxbpre_alloc_size = 0;
702 	mpc->rxbpre_headroom = 0;
703 }
704 
705 /* Get a buffer from the pre-allocated RX buffers */
706 static void *mana_get_rxbuf_pre(struct mana_rxq *rxq, dma_addr_t *da)
707 {
708 	struct net_device *ndev = rxq->ndev;
709 	struct mana_port_context *mpc;
710 	void *va;
711 
712 	mpc = netdev_priv(ndev);
713 
714 	if (!mpc->rxbufs_pre || !mpc->das_pre || !mpc->rxbpre_total) {
715 		netdev_err(ndev, "No RX pre-allocated bufs\n");
716 		return NULL;
717 	}
718 
719 	/* Check sizes to catch unexpected coding error */
720 	if (mpc->rxbpre_datasize != rxq->datasize) {
721 		netdev_err(ndev, "rxbpre_datasize mismatch: %u: %u\n",
722 			   mpc->rxbpre_datasize, rxq->datasize);
723 		return NULL;
724 	}
725 
726 	if (mpc->rxbpre_alloc_size != rxq->alloc_size) {
727 		netdev_err(ndev, "rxbpre_alloc_size mismatch: %u: %u\n",
728 			   mpc->rxbpre_alloc_size, rxq->alloc_size);
729 		return NULL;
730 	}
731 
732 	if (mpc->rxbpre_headroom != rxq->headroom) {
733 		netdev_err(ndev, "rxbpre_headroom mismatch: %u: %u\n",
734 			   mpc->rxbpre_headroom, rxq->headroom);
735 		return NULL;
736 	}
737 
738 	mpc->rxbpre_total--;
739 
740 	*da = mpc->das_pre[mpc->rxbpre_total];
741 	va = mpc->rxbufs_pre[mpc->rxbpre_total];
742 	mpc->rxbufs_pre[mpc->rxbpre_total] = NULL;
743 
744 	/* Deallocate the array after all buffers are gone */
745 	if (!mpc->rxbpre_total)
746 		mana_pre_dealloc_rxbufs(mpc);
747 
748 	return va;
749 }
750 
751 /* Get RX buffer's data size, alloc size, XDP headroom based on MTU */
752 static void mana_get_rxbuf_cfg(struct mana_port_context *apc,
753 			       int mtu, u32 *datasize, u32 *alloc_size,
754 			       u32 *headroom, u32 *frag_count)
755 {
756 	u32 len, buf_size;
757 
758 	/* Calculate datasize first (consistent across all cases) */
759 	*datasize = mtu + ETH_HLEN;
760 
761 	/* For xdp and jumbo frames make sure only one packet fits per page */
762 	if (mtu + MANA_RXBUF_PAD > PAGE_SIZE / 2 || mana_xdp_get(apc)) {
763 		if (mana_xdp_get(apc)) {
764 			*headroom = XDP_PACKET_HEADROOM;
765 			*alloc_size = PAGE_SIZE;
766 		} else {
767 			*headroom = 0; /* no support for XDP */
768 			*alloc_size = SKB_DATA_ALIGN(mtu + MANA_RXBUF_PAD +
769 						     *headroom);
770 		}
771 
772 		*frag_count = 1;
773 		return;
774 	}
775 
776 	/* Standard MTU case - optimize for multiple packets per page */
777 	*headroom = 0;
778 
779 	/* Calculate base buffer size needed */
780 	len = SKB_DATA_ALIGN(mtu + MANA_RXBUF_PAD + *headroom);
781 	buf_size = ALIGN(len, MANA_RX_FRAG_ALIGNMENT);
782 
783 	/* Calculate how many packets can fit in a page */
784 	*frag_count = PAGE_SIZE / buf_size;
785 	*alloc_size = buf_size;
786 }
787 
788 int mana_pre_alloc_rxbufs(struct mana_port_context *mpc, int new_mtu, int num_queues)
789 {
790 	struct device *dev;
791 	struct page *page;
792 	dma_addr_t da;
793 	int num_rxb;
794 	void *va;
795 	int i;
796 
797 	mana_get_rxbuf_cfg(mpc, new_mtu, &mpc->rxbpre_datasize,
798 			   &mpc->rxbpre_alloc_size, &mpc->rxbpre_headroom,
799 			   &mpc->rxbpre_frag_count);
800 
801 	dev = mpc->ac->gdma_dev->gdma_context->dev;
802 
803 	num_rxb = num_queues * mpc->rx_queue_size;
804 
805 	WARN(mpc->rxbufs_pre, "mana rxbufs_pre exists\n");
806 	mpc->rxbufs_pre = kmalloc_array(num_rxb, sizeof(void *), GFP_KERNEL);
807 	if (!mpc->rxbufs_pre)
808 		goto error;
809 
810 	mpc->das_pre = kmalloc_array(num_rxb, sizeof(dma_addr_t), GFP_KERNEL);
811 	if (!mpc->das_pre)
812 		goto error;
813 
814 	mpc->rxbpre_total = 0;
815 
816 	for (i = 0; i < num_rxb; i++) {
817 		page = dev_alloc_pages(get_order(mpc->rxbpre_alloc_size));
818 		if (!page)
819 			goto error;
820 
821 		va = page_to_virt(page);
822 
823 		da = dma_map_single(dev, va + mpc->rxbpre_headroom,
824 				    mpc->rxbpre_datasize, DMA_FROM_DEVICE);
825 		if (dma_mapping_error(dev, da)) {
826 			put_page(page);
827 			goto error;
828 		}
829 
830 		mpc->rxbufs_pre[i] = va;
831 		mpc->das_pre[i] = da;
832 		mpc->rxbpre_total = i + 1;
833 	}
834 
835 	return 0;
836 
837 error:
838 	netdev_err(mpc->ndev, "Failed to pre-allocate RX buffers for %d queues\n", num_queues);
839 	mana_pre_dealloc_rxbufs(mpc);
840 	return -ENOMEM;
841 }
842 
843 static int mana_change_mtu(struct net_device *ndev, int new_mtu)
844 {
845 	struct mana_port_context *mpc = netdev_priv(ndev);
846 	unsigned int old_mtu = ndev->mtu;
847 	int err;
848 
849 	/* Pre-allocate buffers to prevent failure in mana_attach later */
850 	err = mana_pre_alloc_rxbufs(mpc, new_mtu, mpc->num_queues);
851 	if (err) {
852 		netdev_err(ndev, "Insufficient memory for new MTU\n");
853 		return err;
854 	}
855 
856 	err = mana_detach(ndev, false);
857 	if (err) {
858 		netdev_err(ndev, "mana_detach failed: %d\n", err);
859 		goto out;
860 	}
861 
862 	WRITE_ONCE(ndev->mtu, new_mtu);
863 
864 	err = mana_attach(ndev);
865 	if (err) {
866 		netdev_err(ndev, "mana_attach failed: %d\n", err);
867 		WRITE_ONCE(ndev->mtu, old_mtu);
868 	}
869 
870 out:
871 	mana_pre_dealloc_rxbufs(mpc);
872 	return err;
873 }
874 
875 static void mana_tx_timeout(struct net_device *netdev, unsigned int txqueue)
876 {
877 	struct mana_port_context *apc = netdev_priv(netdev);
878 	struct mana_context *ac = apc->ac;
879 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
880 
881 	/* Already in service, hence tx queue reset is not required.*/
882 	if (gc->in_service)
883 		return;
884 
885 	/* Note: If there are pending queue reset work for this port(apc),
886 	 * subsequent request queued up from here are ignored. This is because
887 	 * we are using the same work instance per port(apc).
888 	 */
889 	queue_work(ac->per_port_queue_reset_wq, &apc->queue_reset_work);
890 }
891 
892 static int mana_shaper_set(struct net_shaper_binding *binding,
893 			   const struct net_shaper *shaper,
894 			   struct netlink_ext_ack *extack)
895 {
896 	struct mana_port_context *apc = netdev_priv(binding->netdev);
897 	u32 old_speed, rate;
898 	int err;
899 
900 	if (shaper->handle.scope != NET_SHAPER_SCOPE_NETDEV) {
901 		NL_SET_ERR_MSG_MOD(extack, "net shaper scope should be netdev");
902 		return -EINVAL;
903 	}
904 
905 	if (apc->handle.id && shaper->handle.id != apc->handle.id) {
906 		NL_SET_ERR_MSG_MOD(extack, "Cannot create multiple shapers");
907 		return -EOPNOTSUPP;
908 	}
909 
910 	if (!shaper->bw_max || (shaper->bw_max % 100000000)) {
911 		NL_SET_ERR_MSG_MOD(extack, "Please use multiples of 100Mbps for bandwidth");
912 		return -EINVAL;
913 	}
914 
915 	rate = div_u64(shaper->bw_max, 1000); /* Convert bps to Kbps */
916 	rate = div_u64(rate, 1000);	      /* Convert Kbps to Mbps */
917 
918 	/* Get current speed */
919 	err = mana_query_link_cfg(apc);
920 	old_speed = (err) ? SPEED_UNKNOWN : apc->speed;
921 
922 	if (!err) {
923 		err = mana_set_bw_clamp(apc, rate, TRI_STATE_TRUE);
924 		apc->speed = (err) ? old_speed : rate;
925 		apc->handle = (err) ? apc->handle : shaper->handle;
926 	}
927 
928 	return err;
929 }
930 
931 static int mana_shaper_del(struct net_shaper_binding *binding,
932 			   const struct net_shaper_handle *handle,
933 			   struct netlink_ext_ack *extack)
934 {
935 	struct mana_port_context *apc = netdev_priv(binding->netdev);
936 	int err;
937 
938 	err = mana_set_bw_clamp(apc, 0, TRI_STATE_FALSE);
939 
940 	if (!err) {
941 		/* Reset mana port context parameters */
942 		apc->handle.id = 0;
943 		apc->handle.scope = NET_SHAPER_SCOPE_UNSPEC;
944 		apc->speed = apc->max_speed;
945 	}
946 
947 	return err;
948 }
949 
950 static void mana_shaper_cap(struct net_shaper_binding *binding,
951 			    enum net_shaper_scope scope,
952 			    unsigned long *flags)
953 {
954 	*flags = BIT(NET_SHAPER_A_CAPS_SUPPORT_BW_MAX) |
955 		 BIT(NET_SHAPER_A_CAPS_SUPPORT_METRIC_BPS);
956 }
957 
958 static const struct net_shaper_ops mana_shaper_ops = {
959 	.set = mana_shaper_set,
960 	.delete = mana_shaper_del,
961 	.capabilities = mana_shaper_cap,
962 };
963 
964 static const struct net_device_ops mana_devops = {
965 	.ndo_open		= mana_open,
966 	.ndo_stop		= mana_close,
967 	.ndo_select_queue	= mana_select_queue,
968 #if (MAX_SKB_FRAGS + 2 > MANA_MAX_TX_WQE_SGL_ENTRIES)
969 	.ndo_features_check	= mana_features_check,
970 #endif
971 	.ndo_start_xmit		= mana_start_xmit,
972 	.ndo_validate_addr	= eth_validate_addr,
973 	.ndo_get_stats64	= mana_get_stats64,
974 	.ndo_bpf		= mana_bpf,
975 	.ndo_xdp_xmit		= mana_xdp_xmit,
976 	.ndo_change_mtu		= mana_change_mtu,
977 	.ndo_tx_timeout		= mana_tx_timeout,
978 	.net_shaper_ops         = &mana_shaper_ops,
979 };
980 
981 static void mana_cleanup_port_context(struct mana_port_context *apc)
982 {
983 	/*
984 	 * make sure subsequent cleanup attempts don't end up removing already
985 	 * cleaned dentry pointer
986 	 */
987 	debugfs_remove(apc->mana_port_debugfs);
988 	apc->mana_port_debugfs = NULL;
989 	kfree(apc->rxqs);
990 	apc->rxqs = NULL;
991 }
992 
993 static void mana_cleanup_indir_table(struct mana_port_context *apc)
994 {
995 	apc->indir_table_sz = 0;
996 	kfree(apc->indir_table);
997 	kfree(apc->rxobj_table);
998 }
999 
1000 static int mana_init_port_context(struct mana_port_context *apc)
1001 {
1002 	apc->rxqs = kcalloc(apc->num_queues, sizeof(struct mana_rxq *),
1003 			    GFP_KERNEL);
1004 
1005 	return !apc->rxqs ? -ENOMEM : 0;
1006 }
1007 
1008 static int mana_send_request(struct mana_context *ac, void *in_buf,
1009 			     u32 in_len, void *out_buf, u32 out_len)
1010 {
1011 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
1012 	struct gdma_resp_hdr *resp = out_buf;
1013 	struct gdma_req_hdr *req = in_buf;
1014 	struct device *dev = gc->dev;
1015 	static atomic_t activity_id;
1016 	int err;
1017 
1018 	req->dev_id = gc->mana.dev_id;
1019 	req->activity_id = atomic_inc_return(&activity_id);
1020 
1021 	err = mana_gd_send_request(gc, in_len, in_buf, out_len,
1022 				   out_buf);
1023 	if (err || resp->status) {
1024 		if (err == -EOPNOTSUPP)
1025 			return err;
1026 
1027 		if (req->req.msg_type != MANA_QUERY_PHY_STAT &&
1028 		    mana_need_log(gc, err))
1029 			dev_err(dev, "Failed to send mana message: %d, 0x%x\n",
1030 				err, resp->status);
1031 		return err ? err : -EPROTO;
1032 	}
1033 
1034 	if (req->dev_id.as_uint32 != resp->dev_id.as_uint32 ||
1035 	    req->activity_id != resp->activity_id) {
1036 		dev_err(dev, "Unexpected mana message response: %x,%x,%x,%x\n",
1037 			req->dev_id.as_uint32, resp->dev_id.as_uint32,
1038 			req->activity_id, resp->activity_id);
1039 		return -EPROTO;
1040 	}
1041 
1042 	return 0;
1043 }
1044 
1045 static int mana_verify_resp_hdr(const struct gdma_resp_hdr *resp_hdr,
1046 				const enum mana_command_code expected_code,
1047 				const u32 min_size)
1048 {
1049 	if (resp_hdr->response.msg_type != expected_code)
1050 		return -EPROTO;
1051 
1052 	if (resp_hdr->response.msg_version < GDMA_MESSAGE_V1)
1053 		return -EPROTO;
1054 
1055 	if (resp_hdr->response.msg_size < min_size)
1056 		return -EPROTO;
1057 
1058 	return 0;
1059 }
1060 
1061 static int mana_pf_register_hw_vport(struct mana_port_context *apc)
1062 {
1063 	struct mana_register_hw_vport_resp resp = {};
1064 	struct mana_register_hw_vport_req req = {};
1065 	int err;
1066 
1067 	mana_gd_init_req_hdr(&req.hdr, MANA_REGISTER_HW_PORT,
1068 			     sizeof(req), sizeof(resp));
1069 	req.attached_gfid = 1;
1070 	req.is_pf_default_vport = 1;
1071 	req.allow_all_ether_types = 1;
1072 
1073 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1074 				sizeof(resp));
1075 	if (err) {
1076 		netdev_err(apc->ndev, "Failed to register hw vPort: %d\n", err);
1077 		return err;
1078 	}
1079 
1080 	err = mana_verify_resp_hdr(&resp.hdr, MANA_REGISTER_HW_PORT,
1081 				   sizeof(resp));
1082 	if (err || resp.hdr.status) {
1083 		netdev_err(apc->ndev, "Failed to register hw vPort: %d, 0x%x\n",
1084 			   err, resp.hdr.status);
1085 		return err ? err : -EPROTO;
1086 	}
1087 
1088 	apc->port_handle = resp.hw_vport_handle;
1089 	return 0;
1090 }
1091 
1092 static void mana_pf_deregister_hw_vport(struct mana_port_context *apc)
1093 {
1094 	struct mana_deregister_hw_vport_resp resp = {};
1095 	struct mana_deregister_hw_vport_req req = {};
1096 	int err;
1097 
1098 	mana_gd_init_req_hdr(&req.hdr, MANA_DEREGISTER_HW_PORT,
1099 			     sizeof(req), sizeof(resp));
1100 	req.hw_vport_handle = apc->port_handle;
1101 
1102 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1103 				sizeof(resp));
1104 	if (err) {
1105 		if (mana_en_need_log(apc, err))
1106 			netdev_err(apc->ndev, "Failed to unregister hw vPort: %d\n",
1107 				   err);
1108 
1109 		return;
1110 	}
1111 
1112 	err = mana_verify_resp_hdr(&resp.hdr, MANA_DEREGISTER_HW_PORT,
1113 				   sizeof(resp));
1114 	if (err || resp.hdr.status)
1115 		netdev_err(apc->ndev,
1116 			   "Failed to deregister hw vPort: %d, 0x%x\n",
1117 			   err, resp.hdr.status);
1118 }
1119 
1120 static int mana_pf_register_filter(struct mana_port_context *apc)
1121 {
1122 	struct mana_register_filter_resp resp = {};
1123 	struct mana_register_filter_req req = {};
1124 	int err;
1125 
1126 	mana_gd_init_req_hdr(&req.hdr, MANA_REGISTER_FILTER,
1127 			     sizeof(req), sizeof(resp));
1128 	req.vport = apc->port_handle;
1129 	memcpy(req.mac_addr, apc->mac_addr, ETH_ALEN);
1130 
1131 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1132 				sizeof(resp));
1133 	if (err) {
1134 		netdev_err(apc->ndev, "Failed to register filter: %d\n", err);
1135 		return err;
1136 	}
1137 
1138 	err = mana_verify_resp_hdr(&resp.hdr, MANA_REGISTER_FILTER,
1139 				   sizeof(resp));
1140 	if (err || resp.hdr.status) {
1141 		netdev_err(apc->ndev, "Failed to register filter: %d, 0x%x\n",
1142 			   err, resp.hdr.status);
1143 		return err ? err : -EPROTO;
1144 	}
1145 
1146 	apc->pf_filter_handle = resp.filter_handle;
1147 	return 0;
1148 }
1149 
1150 static void mana_pf_deregister_filter(struct mana_port_context *apc)
1151 {
1152 	struct mana_deregister_filter_resp resp = {};
1153 	struct mana_deregister_filter_req req = {};
1154 	int err;
1155 
1156 	mana_gd_init_req_hdr(&req.hdr, MANA_DEREGISTER_FILTER,
1157 			     sizeof(req), sizeof(resp));
1158 	req.filter_handle = apc->pf_filter_handle;
1159 
1160 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1161 				sizeof(resp));
1162 	if (err) {
1163 		if (mana_en_need_log(apc, err))
1164 			netdev_err(apc->ndev, "Failed to unregister filter: %d\n",
1165 				   err);
1166 
1167 		return;
1168 	}
1169 
1170 	err = mana_verify_resp_hdr(&resp.hdr, MANA_DEREGISTER_FILTER,
1171 				   sizeof(resp));
1172 	if (err || resp.hdr.status)
1173 		netdev_err(apc->ndev,
1174 			   "Failed to deregister filter: %d, 0x%x\n",
1175 			   err, resp.hdr.status);
1176 }
1177 
1178 static int mana_query_device_cfg(struct mana_context *ac, u32 proto_major_ver,
1179 				 u32 proto_minor_ver, u32 proto_micro_ver,
1180 				 u16 *max_num_vports, u8 *bm_hostmode)
1181 {
1182 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
1183 	struct mana_query_device_cfg_resp resp = {};
1184 	struct mana_query_device_cfg_req req = {};
1185 	struct device *dev = gc->dev;
1186 	int err = 0;
1187 
1188 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_DEV_CONFIG,
1189 			     sizeof(req), sizeof(resp));
1190 
1191 	req.hdr.resp.msg_version = GDMA_MESSAGE_V3;
1192 
1193 	req.proto_major_ver = proto_major_ver;
1194 	req.proto_minor_ver = proto_minor_ver;
1195 	req.proto_micro_ver = proto_micro_ver;
1196 
1197 	err = mana_send_request(ac, &req, sizeof(req), &resp, sizeof(resp));
1198 	if (err) {
1199 		dev_err(dev, "Failed to query config: %d", err);
1200 		return err;
1201 	}
1202 
1203 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_DEV_CONFIG,
1204 				   sizeof(resp));
1205 	if (err || resp.hdr.status) {
1206 		dev_err(dev, "Invalid query result: %d, 0x%x\n", err,
1207 			resp.hdr.status);
1208 		if (!err)
1209 			err = -EPROTO;
1210 		return err;
1211 	}
1212 
1213 	*max_num_vports = resp.max_num_vports;
1214 
1215 	if (resp.hdr.response.msg_version >= GDMA_MESSAGE_V2)
1216 		gc->adapter_mtu = resp.adapter_mtu;
1217 	else
1218 		gc->adapter_mtu = ETH_FRAME_LEN;
1219 
1220 	if (resp.hdr.response.msg_version >= GDMA_MESSAGE_V3)
1221 		*bm_hostmode = resp.bm_hostmode;
1222 	else
1223 		*bm_hostmode = 0;
1224 
1225 	debugfs_create_u16("adapter-MTU", 0400, gc->mana_pci_debugfs, &gc->adapter_mtu);
1226 
1227 	return 0;
1228 }
1229 
1230 static int mana_query_vport_cfg(struct mana_port_context *apc, u32 vport_index,
1231 				u32 *max_sq, u32 *max_rq, u32 *num_indir_entry)
1232 {
1233 	struct mana_query_vport_cfg_resp resp = {};
1234 	struct mana_query_vport_cfg_req req = {};
1235 	int err;
1236 
1237 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_VPORT_CONFIG,
1238 			     sizeof(req), sizeof(resp));
1239 
1240 	req.vport_index = vport_index;
1241 
1242 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1243 				sizeof(resp));
1244 	if (err)
1245 		return err;
1246 
1247 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_VPORT_CONFIG,
1248 				   sizeof(resp));
1249 	if (err)
1250 		return err;
1251 
1252 	if (resp.hdr.status)
1253 		return -EPROTO;
1254 
1255 	*max_sq = resp.max_num_sq;
1256 	*max_rq = resp.max_num_rq;
1257 	if (resp.num_indirection_ent > 0 &&
1258 	    resp.num_indirection_ent <= MANA_INDIRECT_TABLE_MAX_SIZE &&
1259 	    is_power_of_2(resp.num_indirection_ent)) {
1260 		*num_indir_entry = resp.num_indirection_ent;
1261 	} else {
1262 		netdev_warn(apc->ndev,
1263 			    "Setting indirection table size to default %d for vPort %d\n",
1264 			    MANA_INDIRECT_TABLE_DEF_SIZE, apc->port_idx);
1265 		*num_indir_entry = MANA_INDIRECT_TABLE_DEF_SIZE;
1266 	}
1267 
1268 	apc->port_handle = resp.vport;
1269 	ether_addr_copy(apc->mac_addr, resp.mac_addr);
1270 
1271 	return 0;
1272 }
1273 
1274 void mana_uncfg_vport(struct mana_port_context *apc)
1275 {
1276 	mutex_lock(&apc->vport_mutex);
1277 	apc->vport_use_count--;
1278 	WARN_ON(apc->vport_use_count < 0);
1279 	mutex_unlock(&apc->vport_mutex);
1280 }
1281 EXPORT_SYMBOL_NS(mana_uncfg_vport, "NET_MANA");
1282 
1283 int mana_cfg_vport(struct mana_port_context *apc, u32 protection_dom_id,
1284 		   u32 doorbell_pg_id)
1285 {
1286 	struct mana_config_vport_resp resp = {};
1287 	struct mana_config_vport_req req = {};
1288 	int err;
1289 
1290 	/* This function is used to program the Ethernet port in the hardware
1291 	 * table. It can be called from the Ethernet driver or the RDMA driver.
1292 	 *
1293 	 * For Ethernet usage, the hardware supports only one active user on a
1294 	 * physical port. The driver checks on the port usage before programming
1295 	 * the hardware when creating the RAW QP (RDMA driver) or exposing the
1296 	 * device to kernel NET layer (Ethernet driver).
1297 	 *
1298 	 * Because the RDMA driver doesn't know in advance which QP type the
1299 	 * user will create, it exposes the device with all its ports. The user
1300 	 * may not be able to create RAW QP on a port if this port is already
1301 	 * in used by the Ethernet driver from the kernel.
1302 	 *
1303 	 * This physical port limitation only applies to the RAW QP. For RC QP,
1304 	 * the hardware doesn't have this limitation. The user can create RC
1305 	 * QPs on a physical port up to the hardware limits independent of the
1306 	 * Ethernet usage on the same port.
1307 	 */
1308 	mutex_lock(&apc->vport_mutex);
1309 	if (apc->vport_use_count > 0) {
1310 		mutex_unlock(&apc->vport_mutex);
1311 		return -EBUSY;
1312 	}
1313 	apc->vport_use_count++;
1314 	mutex_unlock(&apc->vport_mutex);
1315 
1316 	mana_gd_init_req_hdr(&req.hdr, MANA_CONFIG_VPORT_TX,
1317 			     sizeof(req), sizeof(resp));
1318 	req.vport = apc->port_handle;
1319 	req.pdid = protection_dom_id;
1320 	req.doorbell_pageid = doorbell_pg_id;
1321 
1322 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1323 				sizeof(resp));
1324 	if (err) {
1325 		netdev_err(apc->ndev, "Failed to configure vPort: %d\n", err);
1326 		goto out;
1327 	}
1328 
1329 	err = mana_verify_resp_hdr(&resp.hdr, MANA_CONFIG_VPORT_TX,
1330 				   sizeof(resp));
1331 	if (err || resp.hdr.status) {
1332 		netdev_err(apc->ndev, "Failed to configure vPort: %d, 0x%x\n",
1333 			   err, resp.hdr.status);
1334 		if (!err)
1335 			err = -EPROTO;
1336 
1337 		goto out;
1338 	}
1339 
1340 	apc->tx_shortform_allowed = resp.short_form_allowed;
1341 	apc->tx_vp_offset = resp.tx_vport_offset;
1342 
1343 	netdev_info(apc->ndev, "Configured vPort %llu PD %u DB %u\n",
1344 		    apc->port_handle, protection_dom_id, doorbell_pg_id);
1345 out:
1346 	if (err)
1347 		mana_uncfg_vport(apc);
1348 
1349 	return err;
1350 }
1351 EXPORT_SYMBOL_NS(mana_cfg_vport, "NET_MANA");
1352 
1353 static int mana_cfg_vport_steering(struct mana_port_context *apc,
1354 				   enum TRI_STATE rx,
1355 				   bool update_default_rxobj, bool update_key,
1356 				   bool update_tab)
1357 {
1358 	struct mana_cfg_rx_steer_req_v2 *req;
1359 	struct mana_cfg_rx_steer_resp resp = {};
1360 	struct net_device *ndev = apc->ndev;
1361 	u32 req_buf_size;
1362 	int err;
1363 
1364 	req_buf_size = struct_size(req, indir_tab, apc->indir_table_sz);
1365 	req = kzalloc(req_buf_size, GFP_KERNEL);
1366 	if (!req)
1367 		return -ENOMEM;
1368 
1369 	mana_gd_init_req_hdr(&req->hdr, MANA_CONFIG_VPORT_RX, req_buf_size,
1370 			     sizeof(resp));
1371 
1372 	req->hdr.req.msg_version = GDMA_MESSAGE_V2;
1373 
1374 	req->vport = apc->port_handle;
1375 	req->num_indir_entries = apc->indir_table_sz;
1376 	req->indir_tab_offset = offsetof(struct mana_cfg_rx_steer_req_v2,
1377 					 indir_tab);
1378 	req->rx_enable = rx;
1379 	req->rss_enable = apc->rss_state;
1380 	req->update_default_rxobj = update_default_rxobj;
1381 	req->update_hashkey = update_key;
1382 	req->update_indir_tab = update_tab;
1383 	req->default_rxobj = apc->default_rxobj;
1384 	req->cqe_coalescing_enable = 0;
1385 
1386 	if (update_key)
1387 		memcpy(&req->hashkey, apc->hashkey, MANA_HASH_KEY_SIZE);
1388 
1389 	if (update_tab)
1390 		memcpy(req->indir_tab, apc->rxobj_table,
1391 		       flex_array_size(req, indir_tab, req->num_indir_entries));
1392 
1393 	err = mana_send_request(apc->ac, req, req_buf_size, &resp,
1394 				sizeof(resp));
1395 	if (err) {
1396 		if (mana_en_need_log(apc, err))
1397 			netdev_err(ndev, "Failed to configure vPort RX: %d\n", err);
1398 
1399 		goto out;
1400 	}
1401 
1402 	err = mana_verify_resp_hdr(&resp.hdr, MANA_CONFIG_VPORT_RX,
1403 				   sizeof(resp));
1404 	if (err) {
1405 		netdev_err(ndev, "vPort RX configuration failed: %d\n", err);
1406 		goto out;
1407 	}
1408 
1409 	if (resp.hdr.status) {
1410 		netdev_err(ndev, "vPort RX configuration failed: 0x%x\n",
1411 			   resp.hdr.status);
1412 		err = -EPROTO;
1413 	}
1414 
1415 	netdev_info(ndev, "Configured steering vPort %llu entries %u\n",
1416 		    apc->port_handle, apc->indir_table_sz);
1417 out:
1418 	kfree(req);
1419 	return err;
1420 }
1421 
1422 int mana_query_link_cfg(struct mana_port_context *apc)
1423 {
1424 	struct net_device *ndev = apc->ndev;
1425 	struct mana_query_link_config_resp resp = {};
1426 	struct mana_query_link_config_req req = {};
1427 	int err;
1428 
1429 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_LINK_CONFIG,
1430 			     sizeof(req), sizeof(resp));
1431 
1432 	req.vport = apc->port_handle;
1433 	req.hdr.resp.msg_version = GDMA_MESSAGE_V2;
1434 
1435 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1436 				sizeof(resp));
1437 
1438 	if (err) {
1439 		if (err == -EOPNOTSUPP) {
1440 			netdev_info_once(ndev, "MANA_QUERY_LINK_CONFIG not supported\n");
1441 			return err;
1442 		}
1443 		netdev_err(ndev, "Failed to query link config: %d\n", err);
1444 		return err;
1445 	}
1446 
1447 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_LINK_CONFIG,
1448 				   sizeof(resp));
1449 
1450 	if (err || resp.hdr.status) {
1451 		netdev_err(ndev, "Failed to query link config: %d, 0x%x\n", err,
1452 			   resp.hdr.status);
1453 		if (!err)
1454 			err = -EOPNOTSUPP;
1455 		return err;
1456 	}
1457 
1458 	if (resp.qos_unconfigured) {
1459 		err = -EINVAL;
1460 		return err;
1461 	}
1462 	apc->speed = resp.link_speed_mbps;
1463 	apc->max_speed = resp.qos_speed_mbps;
1464 	return 0;
1465 }
1466 
1467 int mana_set_bw_clamp(struct mana_port_context *apc, u32 speed,
1468 		      int enable_clamping)
1469 {
1470 	struct mana_set_bw_clamp_resp resp = {};
1471 	struct mana_set_bw_clamp_req req = {};
1472 	struct net_device *ndev = apc->ndev;
1473 	int err;
1474 
1475 	mana_gd_init_req_hdr(&req.hdr, MANA_SET_BW_CLAMP,
1476 			     sizeof(req), sizeof(resp));
1477 	req.vport = apc->port_handle;
1478 	req.link_speed_mbps = speed;
1479 	req.enable_clamping = enable_clamping;
1480 
1481 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1482 				sizeof(resp));
1483 
1484 	if (err) {
1485 		if (err == -EOPNOTSUPP) {
1486 			netdev_info_once(ndev, "MANA_SET_BW_CLAMP not supported\n");
1487 			return err;
1488 		}
1489 		netdev_err(ndev, "Failed to set bandwidth clamp for speed %u, err = %d",
1490 			   speed, err);
1491 		return err;
1492 	}
1493 
1494 	err = mana_verify_resp_hdr(&resp.hdr, MANA_SET_BW_CLAMP,
1495 				   sizeof(resp));
1496 
1497 	if (err || resp.hdr.status) {
1498 		netdev_err(ndev, "Failed to set bandwidth clamp: %d, 0x%x\n", err,
1499 			   resp.hdr.status);
1500 		if (!err)
1501 			err = -EOPNOTSUPP;
1502 		return err;
1503 	}
1504 
1505 	if (resp.qos_unconfigured)
1506 		netdev_info(ndev, "QoS is unconfigured\n");
1507 
1508 	return 0;
1509 }
1510 
1511 int mana_create_wq_obj(struct mana_port_context *apc,
1512 		       mana_handle_t vport,
1513 		       u32 wq_type, struct mana_obj_spec *wq_spec,
1514 		       struct mana_obj_spec *cq_spec,
1515 		       mana_handle_t *wq_obj)
1516 {
1517 	struct mana_create_wqobj_resp resp = {};
1518 	struct mana_create_wqobj_req req = {};
1519 	struct net_device *ndev = apc->ndev;
1520 	int err;
1521 
1522 	mana_gd_init_req_hdr(&req.hdr, MANA_CREATE_WQ_OBJ,
1523 			     sizeof(req), sizeof(resp));
1524 	req.vport = vport;
1525 	req.wq_type = wq_type;
1526 	req.wq_gdma_region = wq_spec->gdma_region;
1527 	req.cq_gdma_region = cq_spec->gdma_region;
1528 	req.wq_size = wq_spec->queue_size;
1529 	req.cq_size = cq_spec->queue_size;
1530 	req.cq_moderation_ctx_id = cq_spec->modr_ctx_id;
1531 	req.cq_parent_qid = cq_spec->attached_eq;
1532 
1533 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1534 				sizeof(resp));
1535 	if (err) {
1536 		netdev_err(ndev, "Failed to create WQ object: %d\n", err);
1537 		goto out;
1538 	}
1539 
1540 	err = mana_verify_resp_hdr(&resp.hdr, MANA_CREATE_WQ_OBJ,
1541 				   sizeof(resp));
1542 	if (err || resp.hdr.status) {
1543 		netdev_err(ndev, "Failed to create WQ object: %d, 0x%x\n", err,
1544 			   resp.hdr.status);
1545 		if (!err)
1546 			err = -EPROTO;
1547 		goto out;
1548 	}
1549 
1550 	if (resp.wq_obj == INVALID_MANA_HANDLE) {
1551 		netdev_err(ndev, "Got an invalid WQ object handle\n");
1552 		err = -EPROTO;
1553 		goto out;
1554 	}
1555 
1556 	*wq_obj = resp.wq_obj;
1557 	wq_spec->queue_index = resp.wq_id;
1558 	cq_spec->queue_index = resp.cq_id;
1559 
1560 	return 0;
1561 out:
1562 	return err;
1563 }
1564 EXPORT_SYMBOL_NS(mana_create_wq_obj, "NET_MANA");
1565 
1566 void mana_destroy_wq_obj(struct mana_port_context *apc, u32 wq_type,
1567 			 mana_handle_t wq_obj)
1568 {
1569 	struct mana_destroy_wqobj_resp resp = {};
1570 	struct mana_destroy_wqobj_req req = {};
1571 	struct net_device *ndev = apc->ndev;
1572 	int err;
1573 
1574 	mana_gd_init_req_hdr(&req.hdr, MANA_DESTROY_WQ_OBJ,
1575 			     sizeof(req), sizeof(resp));
1576 	req.wq_type = wq_type;
1577 	req.wq_obj_handle = wq_obj;
1578 
1579 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1580 				sizeof(resp));
1581 	if (err) {
1582 		if (mana_en_need_log(apc, err))
1583 			netdev_err(ndev, "Failed to destroy WQ object: %d\n", err);
1584 
1585 		return;
1586 	}
1587 
1588 	err = mana_verify_resp_hdr(&resp.hdr, MANA_DESTROY_WQ_OBJ,
1589 				   sizeof(resp));
1590 	if (err || resp.hdr.status)
1591 		netdev_err(ndev, "Failed to destroy WQ object: %d, 0x%x\n", err,
1592 			   resp.hdr.status);
1593 }
1594 EXPORT_SYMBOL_NS(mana_destroy_wq_obj, "NET_MANA");
1595 
1596 static void mana_destroy_eq(struct mana_context *ac)
1597 {
1598 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
1599 	struct gdma_queue *eq;
1600 	int i;
1601 
1602 	if (!ac->eqs)
1603 		return;
1604 
1605 	debugfs_remove_recursive(ac->mana_eqs_debugfs);
1606 	ac->mana_eqs_debugfs = NULL;
1607 
1608 	for (i = 0; i < gc->max_num_queues; i++) {
1609 		eq = ac->eqs[i].eq;
1610 		if (!eq)
1611 			continue;
1612 
1613 		mana_gd_destroy_queue(gc, eq);
1614 	}
1615 
1616 	kfree(ac->eqs);
1617 	ac->eqs = NULL;
1618 }
1619 
1620 static void mana_create_eq_debugfs(struct mana_context *ac, int i)
1621 {
1622 	struct mana_eq eq = ac->eqs[i];
1623 	char eqnum[32];
1624 
1625 	sprintf(eqnum, "eq%d", i);
1626 	eq.mana_eq_debugfs = debugfs_create_dir(eqnum, ac->mana_eqs_debugfs);
1627 	debugfs_create_u32("head", 0400, eq.mana_eq_debugfs, &eq.eq->head);
1628 	debugfs_create_u32("tail", 0400, eq.mana_eq_debugfs, &eq.eq->tail);
1629 	debugfs_create_file("eq_dump", 0400, eq.mana_eq_debugfs, eq.eq, &mana_dbg_q_fops);
1630 }
1631 
1632 static int mana_create_eq(struct mana_context *ac)
1633 {
1634 	struct gdma_dev *gd = ac->gdma_dev;
1635 	struct gdma_context *gc = gd->gdma_context;
1636 	struct gdma_queue_spec spec = {};
1637 	int err;
1638 	int i;
1639 
1640 	ac->eqs = kcalloc(gc->max_num_queues, sizeof(struct mana_eq),
1641 			  GFP_KERNEL);
1642 	if (!ac->eqs)
1643 		return -ENOMEM;
1644 
1645 	spec.type = GDMA_EQ;
1646 	spec.monitor_avl_buf = false;
1647 	spec.queue_size = EQ_SIZE;
1648 	spec.eq.callback = NULL;
1649 	spec.eq.context = ac->eqs;
1650 	spec.eq.log2_throttle_limit = LOG2_EQ_THROTTLE;
1651 
1652 	ac->mana_eqs_debugfs = debugfs_create_dir("EQs", gc->mana_pci_debugfs);
1653 
1654 	for (i = 0; i < gc->max_num_queues; i++) {
1655 		spec.eq.msix_index = (i + 1) % gc->num_msix_usable;
1656 		err = mana_gd_create_mana_eq(gd, &spec, &ac->eqs[i].eq);
1657 		if (err) {
1658 			dev_err(gc->dev, "Failed to create EQ %d : %d\n", i, err);
1659 			goto out;
1660 		}
1661 		mana_create_eq_debugfs(ac, i);
1662 	}
1663 
1664 	return 0;
1665 out:
1666 	mana_destroy_eq(ac);
1667 	return err;
1668 }
1669 
1670 static int mana_fence_rq(struct mana_port_context *apc, struct mana_rxq *rxq)
1671 {
1672 	struct mana_fence_rq_resp resp = {};
1673 	struct mana_fence_rq_req req = {};
1674 	int err;
1675 
1676 	init_completion(&rxq->fence_event);
1677 
1678 	mana_gd_init_req_hdr(&req.hdr, MANA_FENCE_RQ,
1679 			     sizeof(req), sizeof(resp));
1680 	req.wq_obj_handle =  rxq->rxobj;
1681 
1682 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1683 				sizeof(resp));
1684 	if (err) {
1685 		netdev_err(apc->ndev, "Failed to fence RQ %u: %d\n",
1686 			   rxq->rxq_idx, err);
1687 		return err;
1688 	}
1689 
1690 	err = mana_verify_resp_hdr(&resp.hdr, MANA_FENCE_RQ, sizeof(resp));
1691 	if (err || resp.hdr.status) {
1692 		netdev_err(apc->ndev, "Failed to fence RQ %u: %d, 0x%x\n",
1693 			   rxq->rxq_idx, err, resp.hdr.status);
1694 		if (!err)
1695 			err = -EPROTO;
1696 
1697 		return err;
1698 	}
1699 
1700 	if (wait_for_completion_timeout(&rxq->fence_event, 10 * HZ) == 0) {
1701 		netdev_err(apc->ndev, "Failed to fence RQ %u: timed out\n",
1702 			   rxq->rxq_idx);
1703 		return -ETIMEDOUT;
1704 	}
1705 
1706 	return 0;
1707 }
1708 
1709 static void mana_fence_rqs(struct mana_port_context *apc)
1710 {
1711 	unsigned int rxq_idx;
1712 	struct mana_rxq *rxq;
1713 	int err;
1714 
1715 	for (rxq_idx = 0; rxq_idx < apc->num_queues; rxq_idx++) {
1716 		rxq = apc->rxqs[rxq_idx];
1717 		err = mana_fence_rq(apc, rxq);
1718 
1719 		/* In case of any error, use sleep instead. */
1720 		if (err)
1721 			msleep(100);
1722 	}
1723 }
1724 
1725 static int mana_move_wq_tail(struct gdma_queue *wq, u32 num_units)
1726 {
1727 	u32 used_space_old;
1728 	u32 used_space_new;
1729 
1730 	used_space_old = wq->head - wq->tail;
1731 	used_space_new = wq->head - (wq->tail + num_units);
1732 
1733 	if (WARN_ON_ONCE(used_space_new > used_space_old))
1734 		return -ERANGE;
1735 
1736 	wq->tail += num_units;
1737 	return 0;
1738 }
1739 
1740 void mana_unmap_skb(struct sk_buff *skb, struct mana_port_context *apc)
1741 {
1742 	struct mana_skb_head *ash = (struct mana_skb_head *)skb->head;
1743 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
1744 	struct device *dev = gc->dev;
1745 	int hsg, i;
1746 
1747 	/* Number of SGEs of linear part */
1748 	hsg = (skb_is_gso(skb) && skb_headlen(skb) > ash->size[0]) ? 2 : 1;
1749 
1750 	for (i = 0; i < hsg; i++)
1751 		dma_unmap_single(dev, ash->dma_handle[i], ash->size[i],
1752 				 DMA_TO_DEVICE);
1753 
1754 	for (i = hsg; i < skb_shinfo(skb)->nr_frags + hsg; i++)
1755 		dma_unmap_page(dev, ash->dma_handle[i], ash->size[i],
1756 			       DMA_TO_DEVICE);
1757 }
1758 
1759 static void mana_poll_tx_cq(struct mana_cq *cq)
1760 {
1761 	struct gdma_comp *completions = cq->gdma_comp_buf;
1762 	struct gdma_posted_wqe_info *wqe_info;
1763 	unsigned int pkt_transmitted = 0;
1764 	unsigned int wqe_unit_cnt = 0;
1765 	struct mana_txq *txq = cq->txq;
1766 	struct mana_port_context *apc;
1767 	struct netdev_queue *net_txq;
1768 	struct gdma_queue *gdma_wq;
1769 	unsigned int avail_space;
1770 	struct net_device *ndev;
1771 	struct sk_buff *skb;
1772 	bool txq_stopped;
1773 	int comp_read;
1774 	int i;
1775 
1776 	ndev = txq->ndev;
1777 	apc = netdev_priv(ndev);
1778 
1779 	comp_read = mana_gd_poll_cq(cq->gdma_cq, completions,
1780 				    CQE_POLLING_BUFFER);
1781 
1782 	if (comp_read < 1)
1783 		return;
1784 
1785 	for (i = 0; i < comp_read; i++) {
1786 		struct mana_tx_comp_oob *cqe_oob;
1787 
1788 		if (WARN_ON_ONCE(!completions[i].is_sq))
1789 			return;
1790 
1791 		cqe_oob = (struct mana_tx_comp_oob *)completions[i].cqe_data;
1792 		if (WARN_ON_ONCE(cqe_oob->cqe_hdr.client_type !=
1793 				 MANA_CQE_COMPLETION))
1794 			return;
1795 
1796 		switch (cqe_oob->cqe_hdr.cqe_type) {
1797 		case CQE_TX_OKAY:
1798 			break;
1799 
1800 		case CQE_TX_SA_DROP:
1801 		case CQE_TX_MTU_DROP:
1802 		case CQE_TX_INVALID_OOB:
1803 		case CQE_TX_INVALID_ETH_TYPE:
1804 		case CQE_TX_HDR_PROCESSING_ERROR:
1805 		case CQE_TX_VF_DISABLED:
1806 		case CQE_TX_VPORT_IDX_OUT_OF_RANGE:
1807 		case CQE_TX_VPORT_DISABLED:
1808 		case CQE_TX_VLAN_TAGGING_VIOLATION:
1809 			if (net_ratelimit())
1810 				netdev_err(ndev, "TX: CQE error %d\n",
1811 					   cqe_oob->cqe_hdr.cqe_type);
1812 
1813 			apc->eth_stats.tx_cqe_err++;
1814 			break;
1815 
1816 		default:
1817 			/* If the CQE type is unknown, log an error,
1818 			 * and still free the SKB, update tail, etc.
1819 			 */
1820 			if (net_ratelimit())
1821 				netdev_err(ndev, "TX: unknown CQE type %d\n",
1822 					   cqe_oob->cqe_hdr.cqe_type);
1823 
1824 			apc->eth_stats.tx_cqe_unknown_type++;
1825 			break;
1826 		}
1827 
1828 		if (WARN_ON_ONCE(txq->gdma_txq_id != completions[i].wq_num))
1829 			return;
1830 
1831 		skb = skb_dequeue(&txq->pending_skbs);
1832 		if (WARN_ON_ONCE(!skb))
1833 			return;
1834 
1835 		wqe_info = (struct gdma_posted_wqe_info *)skb->cb;
1836 		wqe_unit_cnt += wqe_info->wqe_size_in_bu;
1837 
1838 		mana_unmap_skb(skb, apc);
1839 
1840 		napi_consume_skb(skb, cq->budget);
1841 
1842 		pkt_transmitted++;
1843 	}
1844 
1845 	if (WARN_ON_ONCE(wqe_unit_cnt == 0))
1846 		return;
1847 
1848 	mana_move_wq_tail(txq->gdma_sq, wqe_unit_cnt);
1849 
1850 	gdma_wq = txq->gdma_sq;
1851 	avail_space = mana_gd_wq_avail_space(gdma_wq);
1852 
1853 	/* Ensure tail updated before checking q stop */
1854 	smp_mb();
1855 
1856 	net_txq = txq->net_txq;
1857 	txq_stopped = netif_tx_queue_stopped(net_txq);
1858 
1859 	/* Ensure checking txq_stopped before apc->port_is_up. */
1860 	smp_rmb();
1861 
1862 	if (txq_stopped && apc->port_is_up && avail_space >= MAX_TX_WQE_SIZE) {
1863 		netif_tx_wake_queue(net_txq);
1864 		apc->eth_stats.wake_queue++;
1865 	}
1866 
1867 	if (atomic_sub_return(pkt_transmitted, &txq->pending_sends) < 0)
1868 		WARN_ON_ONCE(1);
1869 
1870 	cq->work_done = pkt_transmitted;
1871 }
1872 
1873 static void mana_post_pkt_rxq(struct mana_rxq *rxq)
1874 {
1875 	struct mana_recv_buf_oob *recv_buf_oob;
1876 	u32 curr_index;
1877 	int err;
1878 
1879 	curr_index = rxq->buf_index++;
1880 	if (rxq->buf_index == rxq->num_rx_buf)
1881 		rxq->buf_index = 0;
1882 
1883 	recv_buf_oob = &rxq->rx_oobs[curr_index];
1884 
1885 	err = mana_gd_post_work_request(rxq->gdma_rq, &recv_buf_oob->wqe_req,
1886 					&recv_buf_oob->wqe_inf);
1887 	if (WARN_ON_ONCE(err))
1888 		return;
1889 
1890 	WARN_ON_ONCE(recv_buf_oob->wqe_inf.wqe_size_in_bu != 1);
1891 }
1892 
1893 static struct sk_buff *mana_build_skb(struct mana_rxq *rxq, void *buf_va,
1894 				      uint pkt_len, struct xdp_buff *xdp)
1895 {
1896 	struct sk_buff *skb = napi_build_skb(buf_va, rxq->alloc_size);
1897 
1898 	if (!skb)
1899 		return NULL;
1900 
1901 	if (xdp->data_hard_start) {
1902 		u32 metasize = xdp->data - xdp->data_meta;
1903 
1904 		skb_reserve(skb, xdp->data - xdp->data_hard_start);
1905 		skb_put(skb, xdp->data_end - xdp->data);
1906 		if (metasize)
1907 			skb_metadata_set(skb, metasize);
1908 		return skb;
1909 	}
1910 
1911 	skb_reserve(skb, rxq->headroom);
1912 	skb_put(skb, pkt_len);
1913 
1914 	return skb;
1915 }
1916 
1917 static void mana_rx_skb(void *buf_va, bool from_pool,
1918 			struct mana_rxcomp_oob *cqe, struct mana_rxq *rxq)
1919 {
1920 	struct mana_stats_rx *rx_stats = &rxq->stats;
1921 	struct net_device *ndev = rxq->ndev;
1922 	uint pkt_len = cqe->ppi[0].pkt_len;
1923 	u16 rxq_idx = rxq->rxq_idx;
1924 	struct napi_struct *napi;
1925 	struct xdp_buff xdp = {};
1926 	struct sk_buff *skb;
1927 	u32 hash_value;
1928 	u32 act;
1929 
1930 	rxq->rx_cq.work_done++;
1931 	napi = &rxq->rx_cq.napi;
1932 
1933 	if (!buf_va) {
1934 		++ndev->stats.rx_dropped;
1935 		return;
1936 	}
1937 
1938 	act = mana_run_xdp(ndev, rxq, &xdp, buf_va, pkt_len);
1939 
1940 	if (act == XDP_REDIRECT && !rxq->xdp_rc)
1941 		return;
1942 
1943 	if (act != XDP_PASS && act != XDP_TX)
1944 		goto drop_xdp;
1945 
1946 	skb = mana_build_skb(rxq, buf_va, pkt_len, &xdp);
1947 
1948 	if (!skb)
1949 		goto drop;
1950 
1951 	if (from_pool)
1952 		skb_mark_for_recycle(skb);
1953 
1954 	skb->dev = napi->dev;
1955 
1956 	skb->protocol = eth_type_trans(skb, ndev);
1957 	skb_checksum_none_assert(skb);
1958 	skb_record_rx_queue(skb, rxq_idx);
1959 
1960 	if ((ndev->features & NETIF_F_RXCSUM) && cqe->rx_iphdr_csum_succeed) {
1961 		if (cqe->rx_tcp_csum_succeed || cqe->rx_udp_csum_succeed)
1962 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1963 	}
1964 
1965 	if (cqe->rx_hashtype != 0 && (ndev->features & NETIF_F_RXHASH)) {
1966 		hash_value = cqe->ppi[0].pkt_hash;
1967 
1968 		if (cqe->rx_hashtype & MANA_HASH_L4)
1969 			skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L4);
1970 		else
1971 			skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L3);
1972 	}
1973 
1974 	if (cqe->rx_vlantag_present) {
1975 		u16 vlan_tci = cqe->rx_vlan_id;
1976 
1977 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1978 	}
1979 
1980 	u64_stats_update_begin(&rx_stats->syncp);
1981 	rx_stats->packets++;
1982 	rx_stats->bytes += pkt_len;
1983 
1984 	if (act == XDP_TX)
1985 		rx_stats->xdp_tx++;
1986 	u64_stats_update_end(&rx_stats->syncp);
1987 
1988 	if (act == XDP_TX) {
1989 		skb_set_queue_mapping(skb, rxq_idx);
1990 		mana_xdp_tx(skb, ndev);
1991 		return;
1992 	}
1993 
1994 	napi_gro_receive(napi, skb);
1995 
1996 	return;
1997 
1998 drop_xdp:
1999 	u64_stats_update_begin(&rx_stats->syncp);
2000 	rx_stats->xdp_drop++;
2001 	u64_stats_update_end(&rx_stats->syncp);
2002 
2003 drop:
2004 	if (from_pool) {
2005 		if (rxq->frag_count == 1)
2006 			page_pool_recycle_direct(rxq->page_pool,
2007 						 virt_to_head_page(buf_va));
2008 		else
2009 			page_pool_free_va(rxq->page_pool, buf_va, true);
2010 	} else {
2011 		WARN_ON_ONCE(rxq->xdp_save_va);
2012 		/* Save for reuse */
2013 		rxq->xdp_save_va = buf_va;
2014 	}
2015 
2016 	++ndev->stats.rx_dropped;
2017 
2018 	return;
2019 }
2020 
2021 static void *mana_get_rxfrag(struct mana_rxq *rxq, struct device *dev,
2022 			     dma_addr_t *da, bool *from_pool)
2023 {
2024 	struct page *page;
2025 	u32 offset;
2026 	void *va;
2027 	*from_pool = false;
2028 
2029 	/* Don't use fragments for jumbo frames or XDP where it's 1 fragment
2030 	 * per page.
2031 	 */
2032 	if (rxq->frag_count == 1) {
2033 		/* Reuse XDP dropped page if available */
2034 		if (rxq->xdp_save_va) {
2035 			va = rxq->xdp_save_va;
2036 			page = virt_to_head_page(va);
2037 			rxq->xdp_save_va = NULL;
2038 		} else {
2039 			page = page_pool_dev_alloc_pages(rxq->page_pool);
2040 			if (!page)
2041 				return NULL;
2042 
2043 			*from_pool = true;
2044 			va = page_to_virt(page);
2045 		}
2046 
2047 		*da = dma_map_single(dev, va + rxq->headroom, rxq->datasize,
2048 				     DMA_FROM_DEVICE);
2049 		if (dma_mapping_error(dev, *da)) {
2050 			mana_put_rx_page(rxq, page, *from_pool);
2051 			return NULL;
2052 		}
2053 
2054 		return va;
2055 	}
2056 
2057 	page =  page_pool_dev_alloc_frag(rxq->page_pool, &offset,
2058 					 rxq->alloc_size);
2059 	if (!page)
2060 		return NULL;
2061 
2062 	va  = page_to_virt(page) + offset;
2063 	*da = page_pool_get_dma_addr(page) + offset + rxq->headroom;
2064 	*from_pool = true;
2065 
2066 	return va;
2067 }
2068 
2069 /* Allocate frag for rx buffer, and save the old buf */
2070 static void mana_refill_rx_oob(struct device *dev, struct mana_rxq *rxq,
2071 			       struct mana_recv_buf_oob *rxoob, void **old_buf,
2072 			       bool *old_fp)
2073 {
2074 	bool from_pool;
2075 	dma_addr_t da;
2076 	void *va;
2077 
2078 	va = mana_get_rxfrag(rxq, dev, &da, &from_pool);
2079 	if (!va)
2080 		return;
2081 	if (!rxoob->from_pool || rxq->frag_count == 1)
2082 		dma_unmap_single(dev, rxoob->sgl[0].address, rxq->datasize,
2083 				 DMA_FROM_DEVICE);
2084 	*old_buf = rxoob->buf_va;
2085 	*old_fp = rxoob->from_pool;
2086 
2087 	rxoob->buf_va = va;
2088 	rxoob->sgl[0].address = da;
2089 	rxoob->from_pool = from_pool;
2090 }
2091 
2092 static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
2093 				struct gdma_comp *cqe)
2094 {
2095 	struct mana_rxcomp_oob *oob = (struct mana_rxcomp_oob *)cqe->cqe_data;
2096 	struct gdma_context *gc = rxq->gdma_rq->gdma_dev->gdma_context;
2097 	struct net_device *ndev = rxq->ndev;
2098 	struct mana_recv_buf_oob *rxbuf_oob;
2099 	struct mana_port_context *apc;
2100 	struct device *dev = gc->dev;
2101 	void *old_buf = NULL;
2102 	u32 curr, pktlen;
2103 	bool old_fp;
2104 
2105 	apc = netdev_priv(ndev);
2106 
2107 	switch (oob->cqe_hdr.cqe_type) {
2108 	case CQE_RX_OKAY:
2109 		break;
2110 
2111 	case CQE_RX_TRUNCATED:
2112 		++ndev->stats.rx_dropped;
2113 		rxbuf_oob = &rxq->rx_oobs[rxq->buf_index];
2114 		netdev_warn_once(ndev, "Dropped a truncated packet\n");
2115 		goto drop;
2116 
2117 	case CQE_RX_COALESCED_4:
2118 		netdev_err(ndev, "RX coalescing is unsupported\n");
2119 		apc->eth_stats.rx_coalesced_err++;
2120 		return;
2121 
2122 	case CQE_RX_OBJECT_FENCE:
2123 		complete(&rxq->fence_event);
2124 		return;
2125 
2126 	default:
2127 		netdev_err(ndev, "Unknown RX CQE type = %d\n",
2128 			   oob->cqe_hdr.cqe_type);
2129 		apc->eth_stats.rx_cqe_unknown_type++;
2130 		return;
2131 	}
2132 
2133 	pktlen = oob->ppi[0].pkt_len;
2134 
2135 	if (pktlen == 0) {
2136 		/* data packets should never have packetlength of zero */
2137 		netdev_err(ndev, "RX pkt len=0, rq=%u, cq=%u, rxobj=0x%llx\n",
2138 			   rxq->gdma_id, cq->gdma_id, rxq->rxobj);
2139 		return;
2140 	}
2141 
2142 	curr = rxq->buf_index;
2143 	rxbuf_oob = &rxq->rx_oobs[curr];
2144 	WARN_ON_ONCE(rxbuf_oob->wqe_inf.wqe_size_in_bu != 1);
2145 
2146 	mana_refill_rx_oob(dev, rxq, rxbuf_oob, &old_buf, &old_fp);
2147 
2148 	/* Unsuccessful refill will have old_buf == NULL.
2149 	 * In this case, mana_rx_skb() will drop the packet.
2150 	 */
2151 	mana_rx_skb(old_buf, old_fp, oob, rxq);
2152 
2153 drop:
2154 	mana_move_wq_tail(rxq->gdma_rq, rxbuf_oob->wqe_inf.wqe_size_in_bu);
2155 
2156 	mana_post_pkt_rxq(rxq);
2157 }
2158 
2159 static void mana_poll_rx_cq(struct mana_cq *cq)
2160 {
2161 	struct gdma_comp *comp = cq->gdma_comp_buf;
2162 	struct mana_rxq *rxq = cq->rxq;
2163 	int comp_read, i;
2164 
2165 	comp_read = mana_gd_poll_cq(cq->gdma_cq, comp, CQE_POLLING_BUFFER);
2166 	WARN_ON_ONCE(comp_read > CQE_POLLING_BUFFER);
2167 
2168 	rxq->xdp_flush = false;
2169 
2170 	for (i = 0; i < comp_read; i++) {
2171 		if (WARN_ON_ONCE(comp[i].is_sq))
2172 			return;
2173 
2174 		/* verify recv cqe references the right rxq */
2175 		if (WARN_ON_ONCE(comp[i].wq_num != cq->rxq->gdma_id))
2176 			return;
2177 
2178 		mana_process_rx_cqe(rxq, cq, &comp[i]);
2179 	}
2180 
2181 	if (comp_read > 0) {
2182 		struct gdma_context *gc = rxq->gdma_rq->gdma_dev->gdma_context;
2183 
2184 		mana_gd_wq_ring_doorbell(gc, rxq->gdma_rq);
2185 	}
2186 
2187 	if (rxq->xdp_flush)
2188 		xdp_do_flush();
2189 }
2190 
2191 static int mana_cq_handler(void *context, struct gdma_queue *gdma_queue)
2192 {
2193 	struct mana_cq *cq = context;
2194 	int w;
2195 
2196 	WARN_ON_ONCE(cq->gdma_cq != gdma_queue);
2197 
2198 	if (cq->type == MANA_CQ_TYPE_RX)
2199 		mana_poll_rx_cq(cq);
2200 	else
2201 		mana_poll_tx_cq(cq);
2202 
2203 	w = cq->work_done;
2204 	cq->work_done_since_doorbell += w;
2205 
2206 	if (w < cq->budget) {
2207 		mana_gd_ring_cq(gdma_queue, SET_ARM_BIT);
2208 		cq->work_done_since_doorbell = 0;
2209 		napi_complete_done(&cq->napi, w);
2210 	} else if (cq->work_done_since_doorbell >
2211 		   cq->gdma_cq->queue_size / COMP_ENTRY_SIZE * 4) {
2212 		/* MANA hardware requires at least one doorbell ring every 8
2213 		 * wraparounds of CQ even if there is no need to arm the CQ.
2214 		 * This driver rings the doorbell as soon as we have exceeded
2215 		 * 4 wraparounds.
2216 		 */
2217 		mana_gd_ring_cq(gdma_queue, 0);
2218 		cq->work_done_since_doorbell = 0;
2219 	}
2220 
2221 	return w;
2222 }
2223 
2224 static int mana_poll(struct napi_struct *napi, int budget)
2225 {
2226 	struct mana_cq *cq = container_of(napi, struct mana_cq, napi);
2227 	int w;
2228 
2229 	cq->work_done = 0;
2230 	cq->budget = budget;
2231 
2232 	w = mana_cq_handler(cq, cq->gdma_cq);
2233 
2234 	return min(w, budget);
2235 }
2236 
2237 static void mana_schedule_napi(void *context, struct gdma_queue *gdma_queue)
2238 {
2239 	struct mana_cq *cq = context;
2240 
2241 	napi_schedule_irqoff(&cq->napi);
2242 }
2243 
2244 static void mana_deinit_cq(struct mana_port_context *apc, struct mana_cq *cq)
2245 {
2246 	struct gdma_dev *gd = apc->ac->gdma_dev;
2247 
2248 	if (!cq->gdma_cq)
2249 		return;
2250 
2251 	mana_gd_destroy_queue(gd->gdma_context, cq->gdma_cq);
2252 }
2253 
2254 static void mana_deinit_txq(struct mana_port_context *apc, struct mana_txq *txq)
2255 {
2256 	struct gdma_dev *gd = apc->ac->gdma_dev;
2257 
2258 	if (!txq->gdma_sq)
2259 		return;
2260 
2261 	mana_gd_destroy_queue(gd->gdma_context, txq->gdma_sq);
2262 }
2263 
2264 static void mana_destroy_txq(struct mana_port_context *apc)
2265 {
2266 	struct napi_struct *napi;
2267 	int i;
2268 
2269 	if (!apc->tx_qp)
2270 		return;
2271 
2272 	for (i = 0; i < apc->num_queues; i++) {
2273 		debugfs_remove_recursive(apc->tx_qp[i].mana_tx_debugfs);
2274 		apc->tx_qp[i].mana_tx_debugfs = NULL;
2275 
2276 		napi = &apc->tx_qp[i].tx_cq.napi;
2277 		if (apc->tx_qp[i].txq.napi_initialized) {
2278 			napi_synchronize(napi);
2279 			napi_disable_locked(napi);
2280 			netif_napi_del_locked(napi);
2281 			apc->tx_qp[i].txq.napi_initialized = false;
2282 		}
2283 		mana_destroy_wq_obj(apc, GDMA_SQ, apc->tx_qp[i].tx_object);
2284 
2285 		mana_deinit_cq(apc, &apc->tx_qp[i].tx_cq);
2286 
2287 		mana_deinit_txq(apc, &apc->tx_qp[i].txq);
2288 	}
2289 
2290 	kfree(apc->tx_qp);
2291 	apc->tx_qp = NULL;
2292 }
2293 
2294 static void mana_create_txq_debugfs(struct mana_port_context *apc, int idx)
2295 {
2296 	struct mana_tx_qp *tx_qp = &apc->tx_qp[idx];
2297 	char qnum[32];
2298 
2299 	sprintf(qnum, "TX-%d", idx);
2300 	tx_qp->mana_tx_debugfs = debugfs_create_dir(qnum, apc->mana_port_debugfs);
2301 	debugfs_create_u32("sq_head", 0400, tx_qp->mana_tx_debugfs,
2302 			   &tx_qp->txq.gdma_sq->head);
2303 	debugfs_create_u32("sq_tail", 0400, tx_qp->mana_tx_debugfs,
2304 			   &tx_qp->txq.gdma_sq->tail);
2305 	debugfs_create_u32("sq_pend_skb_qlen", 0400, tx_qp->mana_tx_debugfs,
2306 			   &tx_qp->txq.pending_skbs.qlen);
2307 	debugfs_create_u32("cq_head", 0400, tx_qp->mana_tx_debugfs,
2308 			   &tx_qp->tx_cq.gdma_cq->head);
2309 	debugfs_create_u32("cq_tail", 0400, tx_qp->mana_tx_debugfs,
2310 			   &tx_qp->tx_cq.gdma_cq->tail);
2311 	debugfs_create_u32("cq_budget", 0400, tx_qp->mana_tx_debugfs,
2312 			   &tx_qp->tx_cq.budget);
2313 	debugfs_create_file("txq_dump", 0400, tx_qp->mana_tx_debugfs,
2314 			    tx_qp->txq.gdma_sq, &mana_dbg_q_fops);
2315 	debugfs_create_file("cq_dump", 0400, tx_qp->mana_tx_debugfs,
2316 			    tx_qp->tx_cq.gdma_cq, &mana_dbg_q_fops);
2317 }
2318 
2319 static int mana_create_txq(struct mana_port_context *apc,
2320 			   struct net_device *net)
2321 {
2322 	struct mana_context *ac = apc->ac;
2323 	struct gdma_dev *gd = ac->gdma_dev;
2324 	struct mana_obj_spec wq_spec;
2325 	struct mana_obj_spec cq_spec;
2326 	struct gdma_queue_spec spec;
2327 	struct gdma_context *gc;
2328 	struct mana_txq *txq;
2329 	struct mana_cq *cq;
2330 	u32 txq_size;
2331 	u32 cq_size;
2332 	int err;
2333 	int i;
2334 
2335 	apc->tx_qp = kcalloc(apc->num_queues, sizeof(struct mana_tx_qp),
2336 			     GFP_KERNEL);
2337 	if (!apc->tx_qp)
2338 		return -ENOMEM;
2339 
2340 	/*  The minimum size of the WQE is 32 bytes, hence
2341 	 *  apc->tx_queue_size represents the maximum number of WQEs
2342 	 *  the SQ can store. This value is then used to size other queues
2343 	 *  to prevent overflow.
2344 	 *  Also note that the txq_size is always going to be MANA_PAGE_ALIGNED,
2345 	 *  as min val of apc->tx_queue_size is 128 and that would make
2346 	 *  txq_size 128*32 = 4096 and the other higher values of apc->tx_queue_size
2347 	 *  are always power of two
2348 	 */
2349 	txq_size = apc->tx_queue_size * 32;
2350 
2351 	cq_size = apc->tx_queue_size * COMP_ENTRY_SIZE;
2352 
2353 	gc = gd->gdma_context;
2354 
2355 	for (i = 0; i < apc->num_queues; i++) {
2356 		apc->tx_qp[i].tx_object = INVALID_MANA_HANDLE;
2357 
2358 		/* Create SQ */
2359 		txq = &apc->tx_qp[i].txq;
2360 
2361 		u64_stats_init(&txq->stats.syncp);
2362 		txq->ndev = net;
2363 		txq->net_txq = netdev_get_tx_queue(net, i);
2364 		txq->vp_offset = apc->tx_vp_offset;
2365 		txq->napi_initialized = false;
2366 		skb_queue_head_init(&txq->pending_skbs);
2367 
2368 		memset(&spec, 0, sizeof(spec));
2369 		spec.type = GDMA_SQ;
2370 		spec.monitor_avl_buf = true;
2371 		spec.queue_size = txq_size;
2372 		err = mana_gd_create_mana_wq_cq(gd, &spec, &txq->gdma_sq);
2373 		if (err)
2374 			goto out;
2375 
2376 		/* Create SQ's CQ */
2377 		cq = &apc->tx_qp[i].tx_cq;
2378 		cq->type = MANA_CQ_TYPE_TX;
2379 
2380 		cq->txq = txq;
2381 
2382 		memset(&spec, 0, sizeof(spec));
2383 		spec.type = GDMA_CQ;
2384 		spec.monitor_avl_buf = false;
2385 		spec.queue_size = cq_size;
2386 		spec.cq.callback = mana_schedule_napi;
2387 		spec.cq.parent_eq = ac->eqs[i].eq;
2388 		spec.cq.context = cq;
2389 		err = mana_gd_create_mana_wq_cq(gd, &spec, &cq->gdma_cq);
2390 		if (err)
2391 			goto out;
2392 
2393 		memset(&wq_spec, 0, sizeof(wq_spec));
2394 		memset(&cq_spec, 0, sizeof(cq_spec));
2395 
2396 		wq_spec.gdma_region = txq->gdma_sq->mem_info.dma_region_handle;
2397 		wq_spec.queue_size = txq->gdma_sq->queue_size;
2398 
2399 		cq_spec.gdma_region = cq->gdma_cq->mem_info.dma_region_handle;
2400 		cq_spec.queue_size = cq->gdma_cq->queue_size;
2401 		cq_spec.modr_ctx_id = 0;
2402 		cq_spec.attached_eq = cq->gdma_cq->cq.parent->id;
2403 
2404 		err = mana_create_wq_obj(apc, apc->port_handle, GDMA_SQ,
2405 					 &wq_spec, &cq_spec,
2406 					 &apc->tx_qp[i].tx_object);
2407 
2408 		if (err)
2409 			goto out;
2410 
2411 		txq->gdma_sq->id = wq_spec.queue_index;
2412 		cq->gdma_cq->id = cq_spec.queue_index;
2413 
2414 		txq->gdma_sq->mem_info.dma_region_handle =
2415 			GDMA_INVALID_DMA_REGION;
2416 		cq->gdma_cq->mem_info.dma_region_handle =
2417 			GDMA_INVALID_DMA_REGION;
2418 
2419 		txq->gdma_txq_id = txq->gdma_sq->id;
2420 
2421 		cq->gdma_id = cq->gdma_cq->id;
2422 
2423 		if (WARN_ON(cq->gdma_id >= gc->max_num_cqs)) {
2424 			err = -EINVAL;
2425 			goto out;
2426 		}
2427 
2428 		gc->cq_table[cq->gdma_id] = cq->gdma_cq;
2429 
2430 		mana_create_txq_debugfs(apc, i);
2431 
2432 		set_bit(NAPI_STATE_NO_BUSY_POLL, &cq->napi.state);
2433 		netif_napi_add_locked(net, &cq->napi, mana_poll);
2434 		napi_enable_locked(&cq->napi);
2435 		txq->napi_initialized = true;
2436 
2437 		mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
2438 	}
2439 
2440 	return 0;
2441 out:
2442 	netdev_err(net, "Failed to create %d TX queues, %d\n",
2443 		   apc->num_queues, err);
2444 	mana_destroy_txq(apc);
2445 	return err;
2446 }
2447 
2448 static void mana_destroy_rxq(struct mana_port_context *apc,
2449 			     struct mana_rxq *rxq, bool napi_initialized)
2450 
2451 {
2452 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
2453 	struct mana_recv_buf_oob *rx_oob;
2454 	struct device *dev = gc->dev;
2455 	struct napi_struct *napi;
2456 	struct page *page;
2457 	int i;
2458 
2459 	if (!rxq)
2460 		return;
2461 
2462 	debugfs_remove_recursive(rxq->mana_rx_debugfs);
2463 	rxq->mana_rx_debugfs = NULL;
2464 
2465 	napi = &rxq->rx_cq.napi;
2466 
2467 	if (napi_initialized) {
2468 		napi_synchronize(napi);
2469 
2470 		napi_disable_locked(napi);
2471 		netif_napi_del_locked(napi);
2472 	}
2473 	xdp_rxq_info_unreg(&rxq->xdp_rxq);
2474 
2475 	mana_destroy_wq_obj(apc, GDMA_RQ, rxq->rxobj);
2476 
2477 	mana_deinit_cq(apc, &rxq->rx_cq);
2478 
2479 	if (rxq->xdp_save_va)
2480 		put_page(virt_to_head_page(rxq->xdp_save_va));
2481 
2482 	for (i = 0; i < rxq->num_rx_buf; i++) {
2483 		rx_oob = &rxq->rx_oobs[i];
2484 
2485 		if (!rx_oob->buf_va)
2486 			continue;
2487 
2488 		page = virt_to_head_page(rx_oob->buf_va);
2489 
2490 		if (rxq->frag_count == 1 || !rx_oob->from_pool) {
2491 			dma_unmap_single(dev, rx_oob->sgl[0].address,
2492 					 rx_oob->sgl[0].size, DMA_FROM_DEVICE);
2493 			mana_put_rx_page(rxq, page, rx_oob->from_pool);
2494 		} else {
2495 			page_pool_free_va(rxq->page_pool, rx_oob->buf_va, true);
2496 		}
2497 
2498 		rx_oob->buf_va = NULL;
2499 	}
2500 
2501 	page_pool_destroy(rxq->page_pool);
2502 
2503 	if (rxq->gdma_rq)
2504 		mana_gd_destroy_queue(gc, rxq->gdma_rq);
2505 
2506 	kfree(rxq);
2507 }
2508 
2509 static int mana_fill_rx_oob(struct mana_recv_buf_oob *rx_oob, u32 mem_key,
2510 			    struct mana_rxq *rxq, struct device *dev)
2511 {
2512 	struct mana_port_context *mpc = netdev_priv(rxq->ndev);
2513 	bool from_pool = false;
2514 	dma_addr_t da;
2515 	void *va;
2516 
2517 	if (mpc->rxbufs_pre)
2518 		va = mana_get_rxbuf_pre(rxq, &da);
2519 	else
2520 		va = mana_get_rxfrag(rxq, dev, &da, &from_pool);
2521 
2522 	if (!va)
2523 		return -ENOMEM;
2524 
2525 	rx_oob->buf_va = va;
2526 	rx_oob->from_pool = from_pool;
2527 
2528 	rx_oob->sgl[0].address = da;
2529 	rx_oob->sgl[0].size = rxq->datasize;
2530 	rx_oob->sgl[0].mem_key = mem_key;
2531 
2532 	return 0;
2533 }
2534 
2535 #define MANA_WQE_HEADER_SIZE 16
2536 #define MANA_WQE_SGE_SIZE 16
2537 
2538 static int mana_alloc_rx_wqe(struct mana_port_context *apc,
2539 			     struct mana_rxq *rxq, u32 *rxq_size, u32 *cq_size)
2540 {
2541 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
2542 	struct mana_recv_buf_oob *rx_oob;
2543 	struct device *dev = gc->dev;
2544 	u32 buf_idx;
2545 	int ret;
2546 
2547 	WARN_ON(rxq->datasize == 0);
2548 
2549 	*rxq_size = 0;
2550 	*cq_size = 0;
2551 
2552 	for (buf_idx = 0; buf_idx < rxq->num_rx_buf; buf_idx++) {
2553 		rx_oob = &rxq->rx_oobs[buf_idx];
2554 		memset(rx_oob, 0, sizeof(*rx_oob));
2555 
2556 		rx_oob->num_sge = 1;
2557 
2558 		ret = mana_fill_rx_oob(rx_oob, apc->ac->gdma_dev->gpa_mkey, rxq,
2559 				       dev);
2560 		if (ret)
2561 			return ret;
2562 
2563 		rx_oob->wqe_req.sgl = rx_oob->sgl;
2564 		rx_oob->wqe_req.num_sge = rx_oob->num_sge;
2565 		rx_oob->wqe_req.inline_oob_size = 0;
2566 		rx_oob->wqe_req.inline_oob_data = NULL;
2567 		rx_oob->wqe_req.flags = 0;
2568 		rx_oob->wqe_req.client_data_unit = 0;
2569 
2570 		*rxq_size += ALIGN(MANA_WQE_HEADER_SIZE +
2571 				   MANA_WQE_SGE_SIZE * rx_oob->num_sge, 32);
2572 		*cq_size += COMP_ENTRY_SIZE;
2573 	}
2574 
2575 	return 0;
2576 }
2577 
2578 static int mana_push_wqe(struct mana_rxq *rxq)
2579 {
2580 	struct mana_recv_buf_oob *rx_oob;
2581 	u32 buf_idx;
2582 	int err;
2583 
2584 	for (buf_idx = 0; buf_idx < rxq->num_rx_buf; buf_idx++) {
2585 		rx_oob = &rxq->rx_oobs[buf_idx];
2586 
2587 		err = mana_gd_post_and_ring(rxq->gdma_rq, &rx_oob->wqe_req,
2588 					    &rx_oob->wqe_inf);
2589 		if (err)
2590 			return -ENOSPC;
2591 	}
2592 
2593 	return 0;
2594 }
2595 
2596 static int mana_create_page_pool(struct mana_rxq *rxq, struct gdma_context *gc)
2597 {
2598 	struct mana_port_context *mpc = netdev_priv(rxq->ndev);
2599 	struct page_pool_params pprm = {};
2600 	int ret;
2601 
2602 	pprm.pool_size = mpc->rx_queue_size / rxq->frag_count + 1;
2603 	pprm.nid = gc->numa_node;
2604 	pprm.napi = &rxq->rx_cq.napi;
2605 	pprm.netdev = rxq->ndev;
2606 	pprm.order = get_order(rxq->alloc_size);
2607 	pprm.queue_idx = rxq->rxq_idx;
2608 	pprm.dev = gc->dev;
2609 
2610 	/* Let the page pool do the dma map when page sharing with multiple
2611 	 * fragments enabled for rx buffers.
2612 	 */
2613 	if (rxq->frag_count > 1) {
2614 		pprm.flags =  PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
2615 		pprm.max_len = PAGE_SIZE;
2616 		pprm.dma_dir = DMA_FROM_DEVICE;
2617 	}
2618 
2619 	rxq->page_pool = page_pool_create(&pprm);
2620 
2621 	if (IS_ERR(rxq->page_pool)) {
2622 		ret = PTR_ERR(rxq->page_pool);
2623 		rxq->page_pool = NULL;
2624 		return ret;
2625 	}
2626 
2627 	return 0;
2628 }
2629 
2630 static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
2631 					u32 rxq_idx, struct mana_eq *eq,
2632 					struct net_device *ndev)
2633 {
2634 	struct gdma_dev *gd = apc->ac->gdma_dev;
2635 	struct mana_obj_spec wq_spec;
2636 	struct mana_obj_spec cq_spec;
2637 	struct gdma_queue_spec spec;
2638 	struct mana_cq *cq = NULL;
2639 	struct gdma_context *gc;
2640 	u32 cq_size, rq_size;
2641 	struct mana_rxq *rxq;
2642 	int err;
2643 
2644 	gc = gd->gdma_context;
2645 
2646 	rxq = kzalloc(struct_size(rxq, rx_oobs, apc->rx_queue_size),
2647 		      GFP_KERNEL);
2648 	if (!rxq)
2649 		return NULL;
2650 
2651 	rxq->ndev = ndev;
2652 	rxq->num_rx_buf = apc->rx_queue_size;
2653 	rxq->rxq_idx = rxq_idx;
2654 	rxq->rxobj = INVALID_MANA_HANDLE;
2655 
2656 	mana_get_rxbuf_cfg(apc, ndev->mtu, &rxq->datasize, &rxq->alloc_size,
2657 			   &rxq->headroom, &rxq->frag_count);
2658 	/* Create page pool for RX queue */
2659 	err = mana_create_page_pool(rxq, gc);
2660 	if (err) {
2661 		netdev_err(ndev, "Create page pool err:%d\n", err);
2662 		goto out;
2663 	}
2664 
2665 	err = mana_alloc_rx_wqe(apc, rxq, &rq_size, &cq_size);
2666 	if (err)
2667 		goto out;
2668 
2669 	rq_size = MANA_PAGE_ALIGN(rq_size);
2670 	cq_size = MANA_PAGE_ALIGN(cq_size);
2671 
2672 	/* Create RQ */
2673 	memset(&spec, 0, sizeof(spec));
2674 	spec.type = GDMA_RQ;
2675 	spec.monitor_avl_buf = true;
2676 	spec.queue_size = rq_size;
2677 	err = mana_gd_create_mana_wq_cq(gd, &spec, &rxq->gdma_rq);
2678 	if (err)
2679 		goto out;
2680 
2681 	/* Create RQ's CQ */
2682 	cq = &rxq->rx_cq;
2683 	cq->type = MANA_CQ_TYPE_RX;
2684 	cq->rxq = rxq;
2685 
2686 	memset(&spec, 0, sizeof(spec));
2687 	spec.type = GDMA_CQ;
2688 	spec.monitor_avl_buf = false;
2689 	spec.queue_size = cq_size;
2690 	spec.cq.callback = mana_schedule_napi;
2691 	spec.cq.parent_eq = eq->eq;
2692 	spec.cq.context = cq;
2693 	err = mana_gd_create_mana_wq_cq(gd, &spec, &cq->gdma_cq);
2694 	if (err)
2695 		goto out;
2696 
2697 	memset(&wq_spec, 0, sizeof(wq_spec));
2698 	memset(&cq_spec, 0, sizeof(cq_spec));
2699 	wq_spec.gdma_region = rxq->gdma_rq->mem_info.dma_region_handle;
2700 	wq_spec.queue_size = rxq->gdma_rq->queue_size;
2701 
2702 	cq_spec.gdma_region = cq->gdma_cq->mem_info.dma_region_handle;
2703 	cq_spec.queue_size = cq->gdma_cq->queue_size;
2704 	cq_spec.modr_ctx_id = 0;
2705 	cq_spec.attached_eq = cq->gdma_cq->cq.parent->id;
2706 
2707 	err = mana_create_wq_obj(apc, apc->port_handle, GDMA_RQ,
2708 				 &wq_spec, &cq_spec, &rxq->rxobj);
2709 	if (err)
2710 		goto out;
2711 
2712 	rxq->gdma_rq->id = wq_spec.queue_index;
2713 	cq->gdma_cq->id = cq_spec.queue_index;
2714 
2715 	rxq->gdma_rq->mem_info.dma_region_handle = GDMA_INVALID_DMA_REGION;
2716 	cq->gdma_cq->mem_info.dma_region_handle = GDMA_INVALID_DMA_REGION;
2717 
2718 	rxq->gdma_id = rxq->gdma_rq->id;
2719 	cq->gdma_id = cq->gdma_cq->id;
2720 
2721 	err = mana_push_wqe(rxq);
2722 	if (err)
2723 		goto out;
2724 
2725 	if (WARN_ON(cq->gdma_id >= gc->max_num_cqs)) {
2726 		err = -EINVAL;
2727 		goto out;
2728 	}
2729 
2730 	gc->cq_table[cq->gdma_id] = cq->gdma_cq;
2731 
2732 	netif_napi_add_weight_locked(ndev, &cq->napi, mana_poll, 1);
2733 
2734 	WARN_ON(xdp_rxq_info_reg(&rxq->xdp_rxq, ndev, rxq_idx,
2735 				 cq->napi.napi_id));
2736 	WARN_ON(xdp_rxq_info_reg_mem_model(&rxq->xdp_rxq, MEM_TYPE_PAGE_POOL,
2737 					   rxq->page_pool));
2738 
2739 	napi_enable_locked(&cq->napi);
2740 
2741 	mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
2742 out:
2743 	if (!err)
2744 		return rxq;
2745 
2746 	netdev_err(ndev, "Failed to create RXQ: err = %d\n", err);
2747 
2748 	mana_destroy_rxq(apc, rxq, false);
2749 
2750 	if (cq)
2751 		mana_deinit_cq(apc, cq);
2752 
2753 	return NULL;
2754 }
2755 
2756 static void mana_create_rxq_debugfs(struct mana_port_context *apc, int idx)
2757 {
2758 	struct mana_rxq *rxq;
2759 	char qnum[32];
2760 
2761 	rxq = apc->rxqs[idx];
2762 
2763 	sprintf(qnum, "RX-%d", idx);
2764 	rxq->mana_rx_debugfs = debugfs_create_dir(qnum, apc->mana_port_debugfs);
2765 	debugfs_create_u32("rq_head", 0400, rxq->mana_rx_debugfs, &rxq->gdma_rq->head);
2766 	debugfs_create_u32("rq_tail", 0400, rxq->mana_rx_debugfs, &rxq->gdma_rq->tail);
2767 	debugfs_create_u32("rq_nbuf", 0400, rxq->mana_rx_debugfs, &rxq->num_rx_buf);
2768 	debugfs_create_u32("cq_head", 0400, rxq->mana_rx_debugfs,
2769 			   &rxq->rx_cq.gdma_cq->head);
2770 	debugfs_create_u32("cq_tail", 0400, rxq->mana_rx_debugfs,
2771 			   &rxq->rx_cq.gdma_cq->tail);
2772 	debugfs_create_u32("cq_budget", 0400, rxq->mana_rx_debugfs, &rxq->rx_cq.budget);
2773 	debugfs_create_file("rxq_dump", 0400, rxq->mana_rx_debugfs, rxq->gdma_rq, &mana_dbg_q_fops);
2774 	debugfs_create_file("cq_dump", 0400, rxq->mana_rx_debugfs, rxq->rx_cq.gdma_cq,
2775 			    &mana_dbg_q_fops);
2776 }
2777 
2778 static int mana_add_rx_queues(struct mana_port_context *apc,
2779 			      struct net_device *ndev)
2780 {
2781 	struct mana_context *ac = apc->ac;
2782 	struct mana_rxq *rxq;
2783 	int err = 0;
2784 	int i;
2785 
2786 	for (i = 0; i < apc->num_queues; i++) {
2787 		rxq = mana_create_rxq(apc, i, &ac->eqs[i], ndev);
2788 		if (!rxq) {
2789 			err = -ENOMEM;
2790 			netdev_err(ndev, "Failed to create rxq %d : %d\n", i, err);
2791 			goto out;
2792 		}
2793 
2794 		u64_stats_init(&rxq->stats.syncp);
2795 
2796 		apc->rxqs[i] = rxq;
2797 
2798 		mana_create_rxq_debugfs(apc, i);
2799 	}
2800 
2801 	apc->default_rxobj = apc->rxqs[0]->rxobj;
2802 out:
2803 	return err;
2804 }
2805 
2806 static void mana_destroy_vport(struct mana_port_context *apc)
2807 {
2808 	struct gdma_dev *gd = apc->ac->gdma_dev;
2809 	struct mana_rxq *rxq;
2810 	u32 rxq_idx;
2811 
2812 	for (rxq_idx = 0; rxq_idx < apc->num_queues; rxq_idx++) {
2813 		rxq = apc->rxqs[rxq_idx];
2814 		if (!rxq)
2815 			continue;
2816 
2817 		mana_destroy_rxq(apc, rxq, true);
2818 		apc->rxqs[rxq_idx] = NULL;
2819 	}
2820 
2821 	mana_destroy_txq(apc);
2822 	mana_uncfg_vport(apc);
2823 
2824 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode)
2825 		mana_pf_deregister_hw_vport(apc);
2826 }
2827 
2828 static int mana_create_vport(struct mana_port_context *apc,
2829 			     struct net_device *net)
2830 {
2831 	struct gdma_dev *gd = apc->ac->gdma_dev;
2832 	int err;
2833 
2834 	apc->default_rxobj = INVALID_MANA_HANDLE;
2835 
2836 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode) {
2837 		err = mana_pf_register_hw_vport(apc);
2838 		if (err)
2839 			return err;
2840 	}
2841 
2842 	err = mana_cfg_vport(apc, gd->pdid, gd->doorbell);
2843 	if (err)
2844 		return err;
2845 
2846 	return mana_create_txq(apc, net);
2847 }
2848 
2849 static int mana_rss_table_alloc(struct mana_port_context *apc)
2850 {
2851 	if (!apc->indir_table_sz) {
2852 		netdev_err(apc->ndev,
2853 			   "Indirection table size not set for vPort %d\n",
2854 			   apc->port_idx);
2855 		return -EINVAL;
2856 	}
2857 
2858 	apc->indir_table = kcalloc(apc->indir_table_sz, sizeof(u32), GFP_KERNEL);
2859 	if (!apc->indir_table)
2860 		return -ENOMEM;
2861 
2862 	apc->rxobj_table = kcalloc(apc->indir_table_sz, sizeof(mana_handle_t), GFP_KERNEL);
2863 	if (!apc->rxobj_table) {
2864 		kfree(apc->indir_table);
2865 		return -ENOMEM;
2866 	}
2867 
2868 	return 0;
2869 }
2870 
2871 static void mana_rss_table_init(struct mana_port_context *apc)
2872 {
2873 	int i;
2874 
2875 	for (i = 0; i < apc->indir_table_sz; i++)
2876 		apc->indir_table[i] =
2877 			ethtool_rxfh_indir_default(i, apc->num_queues);
2878 }
2879 
2880 int mana_config_rss(struct mana_port_context *apc, enum TRI_STATE rx,
2881 		    bool update_hash, bool update_tab)
2882 {
2883 	u32 queue_idx;
2884 	int err;
2885 	int i;
2886 
2887 	if (update_tab) {
2888 		for (i = 0; i < apc->indir_table_sz; i++) {
2889 			queue_idx = apc->indir_table[i];
2890 			apc->rxobj_table[i] = apc->rxqs[queue_idx]->rxobj;
2891 		}
2892 	}
2893 
2894 	err = mana_cfg_vport_steering(apc, rx, true, update_hash, update_tab);
2895 	if (err)
2896 		return err;
2897 
2898 	mana_fence_rqs(apc);
2899 
2900 	return 0;
2901 }
2902 
2903 int mana_query_gf_stats(struct mana_context *ac)
2904 {
2905 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
2906 	struct mana_query_gf_stat_resp resp = {};
2907 	struct mana_query_gf_stat_req req = {};
2908 	struct device *dev = gc->dev;
2909 	int err;
2910 
2911 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_GF_STAT,
2912 			     sizeof(req), sizeof(resp));
2913 	req.hdr.resp.msg_version = GDMA_MESSAGE_V2;
2914 	req.req_stats = STATISTICS_FLAGS_RX_DISCARDS_NO_WQE |
2915 			STATISTICS_FLAGS_RX_ERRORS_VPORT_DISABLED |
2916 			STATISTICS_FLAGS_HC_RX_BYTES |
2917 			STATISTICS_FLAGS_HC_RX_UCAST_PACKETS |
2918 			STATISTICS_FLAGS_HC_RX_UCAST_BYTES |
2919 			STATISTICS_FLAGS_HC_RX_MCAST_PACKETS |
2920 			STATISTICS_FLAGS_HC_RX_MCAST_BYTES |
2921 			STATISTICS_FLAGS_HC_RX_BCAST_PACKETS |
2922 			STATISTICS_FLAGS_HC_RX_BCAST_BYTES |
2923 			STATISTICS_FLAGS_TX_ERRORS_GF_DISABLED |
2924 			STATISTICS_FLAGS_TX_ERRORS_VPORT_DISABLED |
2925 			STATISTICS_FLAGS_TX_ERRORS_INVAL_VPORT_OFFSET_PACKETS |
2926 			STATISTICS_FLAGS_TX_ERRORS_VLAN_ENFORCEMENT |
2927 			STATISTICS_FLAGS_TX_ERRORS_ETH_TYPE_ENFORCEMENT |
2928 			STATISTICS_FLAGS_TX_ERRORS_SA_ENFORCEMENT |
2929 			STATISTICS_FLAGS_TX_ERRORS_SQPDID_ENFORCEMENT |
2930 			STATISTICS_FLAGS_TX_ERRORS_CQPDID_ENFORCEMENT |
2931 			STATISTICS_FLAGS_TX_ERRORS_MTU_VIOLATION |
2932 			STATISTICS_FLAGS_TX_ERRORS_INVALID_OOB |
2933 			STATISTICS_FLAGS_HC_TX_BYTES |
2934 			STATISTICS_FLAGS_HC_TX_UCAST_PACKETS |
2935 			STATISTICS_FLAGS_HC_TX_UCAST_BYTES |
2936 			STATISTICS_FLAGS_HC_TX_MCAST_PACKETS |
2937 			STATISTICS_FLAGS_HC_TX_MCAST_BYTES |
2938 			STATISTICS_FLAGS_HC_TX_BCAST_PACKETS |
2939 			STATISTICS_FLAGS_HC_TX_BCAST_BYTES |
2940 			STATISTICS_FLAGS_TX_ERRORS_GDMA_ERROR;
2941 
2942 	err = mana_send_request(ac, &req, sizeof(req), &resp,
2943 				sizeof(resp));
2944 	if (err) {
2945 		dev_err(dev, "Failed to query GF stats: %d\n", err);
2946 		return err;
2947 	}
2948 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_GF_STAT,
2949 				   sizeof(resp));
2950 	if (err || resp.hdr.status) {
2951 		dev_err(dev, "Failed to query GF stats: %d, 0x%x\n", err,
2952 			resp.hdr.status);
2953 		return err;
2954 	}
2955 
2956 	ac->hc_stats.hc_rx_discards_no_wqe = resp.rx_discards_nowqe;
2957 	ac->hc_stats.hc_rx_err_vport_disabled = resp.rx_err_vport_disabled;
2958 	ac->hc_stats.hc_rx_bytes = resp.hc_rx_bytes;
2959 	ac->hc_stats.hc_rx_ucast_pkts = resp.hc_rx_ucast_pkts;
2960 	ac->hc_stats.hc_rx_ucast_bytes = resp.hc_rx_ucast_bytes;
2961 	ac->hc_stats.hc_rx_bcast_pkts = resp.hc_rx_bcast_pkts;
2962 	ac->hc_stats.hc_rx_bcast_bytes = resp.hc_rx_bcast_bytes;
2963 	ac->hc_stats.hc_rx_mcast_pkts = resp.hc_rx_mcast_pkts;
2964 	ac->hc_stats.hc_rx_mcast_bytes = resp.hc_rx_mcast_bytes;
2965 	ac->hc_stats.hc_tx_err_gf_disabled = resp.tx_err_gf_disabled;
2966 	ac->hc_stats.hc_tx_err_vport_disabled = resp.tx_err_vport_disabled;
2967 	ac->hc_stats.hc_tx_err_inval_vportoffset_pkt =
2968 					     resp.tx_err_inval_vport_offset_pkt;
2969 	ac->hc_stats.hc_tx_err_vlan_enforcement =
2970 					     resp.tx_err_vlan_enforcement;
2971 	ac->hc_stats.hc_tx_err_eth_type_enforcement =
2972 					     resp.tx_err_ethtype_enforcement;
2973 	ac->hc_stats.hc_tx_err_sa_enforcement = resp.tx_err_SA_enforcement;
2974 	ac->hc_stats.hc_tx_err_sqpdid_enforcement =
2975 					     resp.tx_err_SQPDID_enforcement;
2976 	ac->hc_stats.hc_tx_err_cqpdid_enforcement =
2977 					     resp.tx_err_CQPDID_enforcement;
2978 	ac->hc_stats.hc_tx_err_mtu_violation = resp.tx_err_mtu_violation;
2979 	ac->hc_stats.hc_tx_err_inval_oob = resp.tx_err_inval_oob;
2980 	ac->hc_stats.hc_tx_bytes = resp.hc_tx_bytes;
2981 	ac->hc_stats.hc_tx_ucast_pkts = resp.hc_tx_ucast_pkts;
2982 	ac->hc_stats.hc_tx_ucast_bytes = resp.hc_tx_ucast_bytes;
2983 	ac->hc_stats.hc_tx_bcast_pkts = resp.hc_tx_bcast_pkts;
2984 	ac->hc_stats.hc_tx_bcast_bytes = resp.hc_tx_bcast_bytes;
2985 	ac->hc_stats.hc_tx_mcast_pkts = resp.hc_tx_mcast_pkts;
2986 	ac->hc_stats.hc_tx_mcast_bytes = resp.hc_tx_mcast_bytes;
2987 	ac->hc_stats.hc_tx_err_gdma = resp.tx_err_gdma;
2988 
2989 	return 0;
2990 }
2991 
2992 void mana_query_phy_stats(struct mana_port_context *apc)
2993 {
2994 	struct mana_query_phy_stat_resp resp = {};
2995 	struct mana_query_phy_stat_req req = {};
2996 	struct net_device *ndev = apc->ndev;
2997 	int err;
2998 
2999 	mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_PHY_STAT,
3000 			     sizeof(req), sizeof(resp));
3001 	err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
3002 				sizeof(resp));
3003 	if (err)
3004 		return;
3005 
3006 	err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_PHY_STAT,
3007 				   sizeof(resp));
3008 	if (err || resp.hdr.status) {
3009 		netdev_err(ndev,
3010 			   "Failed to query PHY stats: %d, resp:0x%x\n",
3011 				err, resp.hdr.status);
3012 		return;
3013 	}
3014 
3015 	/* Aggregate drop counters */
3016 	apc->phy_stats.rx_pkt_drop_phy = resp.rx_pkt_drop_phy;
3017 	apc->phy_stats.tx_pkt_drop_phy = resp.tx_pkt_drop_phy;
3018 
3019 	/* Per TC traffic Counters */
3020 	apc->phy_stats.rx_pkt_tc0_phy = resp.rx_pkt_tc0_phy;
3021 	apc->phy_stats.tx_pkt_tc0_phy = resp.tx_pkt_tc0_phy;
3022 	apc->phy_stats.rx_pkt_tc1_phy = resp.rx_pkt_tc1_phy;
3023 	apc->phy_stats.tx_pkt_tc1_phy = resp.tx_pkt_tc1_phy;
3024 	apc->phy_stats.rx_pkt_tc2_phy = resp.rx_pkt_tc2_phy;
3025 	apc->phy_stats.tx_pkt_tc2_phy = resp.tx_pkt_tc2_phy;
3026 	apc->phy_stats.rx_pkt_tc3_phy = resp.rx_pkt_tc3_phy;
3027 	apc->phy_stats.tx_pkt_tc3_phy = resp.tx_pkt_tc3_phy;
3028 	apc->phy_stats.rx_pkt_tc4_phy = resp.rx_pkt_tc4_phy;
3029 	apc->phy_stats.tx_pkt_tc4_phy = resp.tx_pkt_tc4_phy;
3030 	apc->phy_stats.rx_pkt_tc5_phy = resp.rx_pkt_tc5_phy;
3031 	apc->phy_stats.tx_pkt_tc5_phy = resp.tx_pkt_tc5_phy;
3032 	apc->phy_stats.rx_pkt_tc6_phy = resp.rx_pkt_tc6_phy;
3033 	apc->phy_stats.tx_pkt_tc6_phy = resp.tx_pkt_tc6_phy;
3034 	apc->phy_stats.rx_pkt_tc7_phy = resp.rx_pkt_tc7_phy;
3035 	apc->phy_stats.tx_pkt_tc7_phy = resp.tx_pkt_tc7_phy;
3036 
3037 	/* Per TC byte Counters */
3038 	apc->phy_stats.rx_byte_tc0_phy = resp.rx_byte_tc0_phy;
3039 	apc->phy_stats.tx_byte_tc0_phy = resp.tx_byte_tc0_phy;
3040 	apc->phy_stats.rx_byte_tc1_phy = resp.rx_byte_tc1_phy;
3041 	apc->phy_stats.tx_byte_tc1_phy = resp.tx_byte_tc1_phy;
3042 	apc->phy_stats.rx_byte_tc2_phy = resp.rx_byte_tc2_phy;
3043 	apc->phy_stats.tx_byte_tc2_phy = resp.tx_byte_tc2_phy;
3044 	apc->phy_stats.rx_byte_tc3_phy = resp.rx_byte_tc3_phy;
3045 	apc->phy_stats.tx_byte_tc3_phy = resp.tx_byte_tc3_phy;
3046 	apc->phy_stats.rx_byte_tc4_phy = resp.rx_byte_tc4_phy;
3047 	apc->phy_stats.tx_byte_tc4_phy = resp.tx_byte_tc4_phy;
3048 	apc->phy_stats.rx_byte_tc5_phy = resp.rx_byte_tc5_phy;
3049 	apc->phy_stats.tx_byte_tc5_phy = resp.tx_byte_tc5_phy;
3050 	apc->phy_stats.rx_byte_tc6_phy = resp.rx_byte_tc6_phy;
3051 	apc->phy_stats.tx_byte_tc6_phy = resp.tx_byte_tc6_phy;
3052 	apc->phy_stats.rx_byte_tc7_phy = resp.rx_byte_tc7_phy;
3053 	apc->phy_stats.tx_byte_tc7_phy = resp.tx_byte_tc7_phy;
3054 
3055 	/* Per TC pause Counters */
3056 	apc->phy_stats.rx_pause_tc0_phy = resp.rx_pause_tc0_phy;
3057 	apc->phy_stats.tx_pause_tc0_phy = resp.tx_pause_tc0_phy;
3058 	apc->phy_stats.rx_pause_tc1_phy = resp.rx_pause_tc1_phy;
3059 	apc->phy_stats.tx_pause_tc1_phy = resp.tx_pause_tc1_phy;
3060 	apc->phy_stats.rx_pause_tc2_phy = resp.rx_pause_tc2_phy;
3061 	apc->phy_stats.tx_pause_tc2_phy = resp.tx_pause_tc2_phy;
3062 	apc->phy_stats.rx_pause_tc3_phy = resp.rx_pause_tc3_phy;
3063 	apc->phy_stats.tx_pause_tc3_phy = resp.tx_pause_tc3_phy;
3064 	apc->phy_stats.rx_pause_tc4_phy = resp.rx_pause_tc4_phy;
3065 	apc->phy_stats.tx_pause_tc4_phy = resp.tx_pause_tc4_phy;
3066 	apc->phy_stats.rx_pause_tc5_phy = resp.rx_pause_tc5_phy;
3067 	apc->phy_stats.tx_pause_tc5_phy = resp.tx_pause_tc5_phy;
3068 	apc->phy_stats.rx_pause_tc6_phy = resp.rx_pause_tc6_phy;
3069 	apc->phy_stats.tx_pause_tc6_phy = resp.tx_pause_tc6_phy;
3070 	apc->phy_stats.rx_pause_tc7_phy = resp.rx_pause_tc7_phy;
3071 	apc->phy_stats.tx_pause_tc7_phy = resp.tx_pause_tc7_phy;
3072 }
3073 
3074 static int mana_init_port(struct net_device *ndev)
3075 {
3076 	struct mana_port_context *apc = netdev_priv(ndev);
3077 	struct gdma_dev *gd = apc->ac->gdma_dev;
3078 	u32 max_txq, max_rxq, max_queues;
3079 	int port_idx = apc->port_idx;
3080 	struct gdma_context *gc;
3081 	char vport[32];
3082 	int err;
3083 
3084 	err = mana_init_port_context(apc);
3085 	if (err)
3086 		return err;
3087 
3088 	gc = gd->gdma_context;
3089 
3090 	err = mana_query_vport_cfg(apc, port_idx, &max_txq, &max_rxq,
3091 				   &apc->indir_table_sz);
3092 	if (err) {
3093 		netdev_err(ndev, "Failed to query info for vPort %d\n",
3094 			   port_idx);
3095 		goto reset_apc;
3096 	}
3097 
3098 	max_queues = min_t(u32, max_txq, max_rxq);
3099 	if (apc->max_queues > max_queues)
3100 		apc->max_queues = max_queues;
3101 
3102 	if (apc->num_queues > apc->max_queues)
3103 		apc->num_queues = apc->max_queues;
3104 
3105 	eth_hw_addr_set(ndev, apc->mac_addr);
3106 	sprintf(vport, "vport%d", port_idx);
3107 	apc->mana_port_debugfs = debugfs_create_dir(vport, gc->mana_pci_debugfs);
3108 	return 0;
3109 
3110 reset_apc:
3111 	mana_cleanup_port_context(apc);
3112 	return err;
3113 }
3114 
3115 int mana_alloc_queues(struct net_device *ndev)
3116 {
3117 	struct mana_port_context *apc = netdev_priv(ndev);
3118 	struct gdma_dev *gd = apc->ac->gdma_dev;
3119 	int err;
3120 
3121 	err = mana_create_vport(apc, ndev);
3122 	if (err) {
3123 		netdev_err(ndev, "Failed to create vPort %u : %d\n", apc->port_idx, err);
3124 		return err;
3125 	}
3126 
3127 	err = netif_set_real_num_tx_queues(ndev, apc->num_queues);
3128 	if (err) {
3129 		netdev_err(ndev,
3130 			   "netif_set_real_num_tx_queues () failed for ndev with num_queues %u : %d\n",
3131 			   apc->num_queues, err);
3132 		goto destroy_vport;
3133 	}
3134 
3135 	err = mana_add_rx_queues(apc, ndev);
3136 	if (err)
3137 		goto destroy_vport;
3138 
3139 	apc->rss_state = apc->num_queues > 1 ? TRI_STATE_TRUE : TRI_STATE_FALSE;
3140 
3141 	err = netif_set_real_num_rx_queues(ndev, apc->num_queues);
3142 	if (err) {
3143 		netdev_err(ndev,
3144 			   "netif_set_real_num_rx_queues () failed for ndev with num_queues %u : %d\n",
3145 			   apc->num_queues, err);
3146 		goto destroy_vport;
3147 	}
3148 
3149 	mana_rss_table_init(apc);
3150 
3151 	err = mana_config_rss(apc, TRI_STATE_TRUE, true, true);
3152 	if (err) {
3153 		netdev_err(ndev, "Failed to configure RSS table: %d\n", err);
3154 		goto destroy_vport;
3155 	}
3156 
3157 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode) {
3158 		err = mana_pf_register_filter(apc);
3159 		if (err)
3160 			goto destroy_vport;
3161 	}
3162 
3163 	mana_chn_setxdp(apc, mana_xdp_get(apc));
3164 
3165 	return 0;
3166 
3167 destroy_vport:
3168 	mana_destroy_vport(apc);
3169 	return err;
3170 }
3171 
3172 int mana_attach(struct net_device *ndev)
3173 {
3174 	struct mana_port_context *apc = netdev_priv(ndev);
3175 	int err;
3176 
3177 	ASSERT_RTNL();
3178 
3179 	err = mana_init_port(ndev);
3180 	if (err)
3181 		return err;
3182 
3183 	if (apc->port_st_save) {
3184 		err = mana_alloc_queues(ndev);
3185 		if (err) {
3186 			mana_cleanup_port_context(apc);
3187 			return err;
3188 		}
3189 	}
3190 
3191 	apc->port_is_up = apc->port_st_save;
3192 
3193 	/* Ensure port state updated before txq state */
3194 	smp_wmb();
3195 
3196 	netif_device_attach(ndev);
3197 
3198 	return 0;
3199 }
3200 
3201 static int mana_dealloc_queues(struct net_device *ndev)
3202 {
3203 	struct mana_port_context *apc = netdev_priv(ndev);
3204 	unsigned long timeout = jiffies + 120 * HZ;
3205 	struct gdma_dev *gd = apc->ac->gdma_dev;
3206 	struct mana_txq *txq;
3207 	struct sk_buff *skb;
3208 	int i, err;
3209 	u32 tsleep;
3210 
3211 	if (apc->port_is_up)
3212 		return -EINVAL;
3213 
3214 	mana_chn_setxdp(apc, NULL);
3215 
3216 	if (gd->gdma_context->is_pf && !apc->ac->bm_hostmode)
3217 		mana_pf_deregister_filter(apc);
3218 
3219 	/* No packet can be transmitted now since apc->port_is_up is false.
3220 	 * There is still a tiny chance that mana_poll_tx_cq() can re-enable
3221 	 * a txq because it may not timely see apc->port_is_up being cleared
3222 	 * to false, but it doesn't matter since mana_start_xmit() drops any
3223 	 * new packets due to apc->port_is_up being false.
3224 	 *
3225 	 * Drain all the in-flight TX packets.
3226 	 * A timeout of 120 seconds for all the queues is used.
3227 	 * This will break the while loop when h/w is not responding.
3228 	 * This value of 120 has been decided here considering max
3229 	 * number of queues.
3230 	 */
3231 
3232 	for (i = 0; i < apc->num_queues; i++) {
3233 		txq = &apc->tx_qp[i].txq;
3234 		tsleep = 1000;
3235 		while (atomic_read(&txq->pending_sends) > 0 &&
3236 		       time_before(jiffies, timeout)) {
3237 			usleep_range(tsleep, tsleep + 1000);
3238 			tsleep <<= 1;
3239 		}
3240 		if (atomic_read(&txq->pending_sends)) {
3241 			err = pcie_flr(to_pci_dev(gd->gdma_context->dev));
3242 			if (err) {
3243 				netdev_err(ndev, "flr failed %d with %d pkts pending in txq %u\n",
3244 					   err, atomic_read(&txq->pending_sends),
3245 					   txq->gdma_txq_id);
3246 			}
3247 			break;
3248 		}
3249 	}
3250 
3251 	for (i = 0; i < apc->num_queues; i++) {
3252 		txq = &apc->tx_qp[i].txq;
3253 		while ((skb = skb_dequeue(&txq->pending_skbs))) {
3254 			mana_unmap_skb(skb, apc);
3255 			dev_kfree_skb_any(skb);
3256 		}
3257 		atomic_set(&txq->pending_sends, 0);
3258 	}
3259 	/* We're 100% sure the queues can no longer be woken up, because
3260 	 * we're sure now mana_poll_tx_cq() can't be running.
3261 	 */
3262 
3263 	apc->rss_state = TRI_STATE_FALSE;
3264 	err = mana_config_rss(apc, TRI_STATE_FALSE, false, false);
3265 	if (err && mana_en_need_log(apc, err))
3266 		netdev_err(ndev, "Failed to disable vPort: %d\n", err);
3267 
3268 	/* Even in err case, still need to cleanup the vPort */
3269 	mana_destroy_vport(apc);
3270 
3271 	return 0;
3272 }
3273 
3274 int mana_detach(struct net_device *ndev, bool from_close)
3275 {
3276 	struct mana_port_context *apc = netdev_priv(ndev);
3277 	int err;
3278 
3279 	ASSERT_RTNL();
3280 
3281 	apc->port_st_save = apc->port_is_up;
3282 	apc->port_is_up = false;
3283 
3284 	/* Ensure port state updated before txq state */
3285 	smp_wmb();
3286 
3287 	netif_tx_disable(ndev);
3288 
3289 	if (apc->port_st_save) {
3290 		err = mana_dealloc_queues(ndev);
3291 		if (err) {
3292 			netdev_err(ndev, "%s failed to deallocate queues: %d\n", __func__, err);
3293 			return err;
3294 		}
3295 	}
3296 
3297 	if (!from_close) {
3298 		netif_device_detach(ndev);
3299 		mana_cleanup_port_context(apc);
3300 	}
3301 
3302 	return 0;
3303 }
3304 
3305 static int mana_probe_port(struct mana_context *ac, int port_idx,
3306 			   struct net_device **ndev_storage)
3307 {
3308 	struct gdma_context *gc = ac->gdma_dev->gdma_context;
3309 	struct mana_port_context *apc;
3310 	struct net_device *ndev;
3311 	int err;
3312 
3313 	ndev = alloc_etherdev_mq(sizeof(struct mana_port_context),
3314 				 gc->max_num_queues);
3315 	if (!ndev)
3316 		return -ENOMEM;
3317 
3318 	*ndev_storage = ndev;
3319 
3320 	apc = netdev_priv(ndev);
3321 	apc->ac = ac;
3322 	apc->ndev = ndev;
3323 	apc->max_queues = gc->max_num_queues;
3324 	apc->num_queues = gc->max_num_queues;
3325 	apc->tx_queue_size = DEF_TX_BUFFERS_PER_QUEUE;
3326 	apc->rx_queue_size = DEF_RX_BUFFERS_PER_QUEUE;
3327 	apc->port_handle = INVALID_MANA_HANDLE;
3328 	apc->pf_filter_handle = INVALID_MANA_HANDLE;
3329 	apc->port_idx = port_idx;
3330 
3331 	mutex_init(&apc->vport_mutex);
3332 	apc->vport_use_count = 0;
3333 
3334 	ndev->netdev_ops = &mana_devops;
3335 	ndev->ethtool_ops = &mana_ethtool_ops;
3336 	ndev->mtu = ETH_DATA_LEN;
3337 	ndev->max_mtu = gc->adapter_mtu - ETH_HLEN;
3338 	ndev->min_mtu = ETH_MIN_MTU;
3339 	ndev->needed_headroom = MANA_HEADROOM;
3340 	ndev->dev_port = port_idx;
3341 	/* Recommended timeout based on HW FPGA re-config scenario. */
3342 	ndev->watchdog_timeo = 15 * HZ;
3343 	SET_NETDEV_DEV(ndev, gc->dev);
3344 
3345 	netif_set_tso_max_size(ndev, GSO_MAX_SIZE);
3346 
3347 	netif_carrier_off(ndev);
3348 
3349 	netdev_rss_key_fill(apc->hashkey, MANA_HASH_KEY_SIZE);
3350 
3351 	err = mana_init_port(ndev);
3352 	if (err)
3353 		goto free_net;
3354 
3355 	err = mana_rss_table_alloc(apc);
3356 	if (err)
3357 		goto reset_apc;
3358 
3359 	/* Initialize the per port queue reset work.*/
3360 	INIT_WORK(&apc->queue_reset_work,
3361 		  mana_per_port_queue_reset_work_handler);
3362 
3363 	netdev_lockdep_set_classes(ndev);
3364 
3365 	ndev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3366 	ndev->hw_features |= NETIF_F_RXCSUM;
3367 	ndev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3368 	ndev->hw_features |= NETIF_F_RXHASH;
3369 	ndev->features = ndev->hw_features | NETIF_F_HW_VLAN_CTAG_TX |
3370 			 NETIF_F_HW_VLAN_CTAG_RX;
3371 	ndev->vlan_features = ndev->features;
3372 	xdp_set_features_flag(ndev, NETDEV_XDP_ACT_BASIC |
3373 			      NETDEV_XDP_ACT_REDIRECT |
3374 			      NETDEV_XDP_ACT_NDO_XMIT);
3375 
3376 	err = register_netdev(ndev);
3377 	if (err) {
3378 		netdev_err(ndev, "Unable to register netdev.\n");
3379 		goto free_indir;
3380 	}
3381 
3382 	netif_carrier_on(ndev);
3383 
3384 	debugfs_create_u32("current_speed", 0400, apc->mana_port_debugfs, &apc->speed);
3385 
3386 	return 0;
3387 
3388 free_indir:
3389 	mana_cleanup_indir_table(apc);
3390 reset_apc:
3391 	mana_cleanup_port_context(apc);
3392 free_net:
3393 	*ndev_storage = NULL;
3394 	netdev_err(ndev, "Failed to probe vPort %d: %d\n", port_idx, err);
3395 	free_netdev(ndev);
3396 	return err;
3397 }
3398 
3399 static void adev_release(struct device *dev)
3400 {
3401 	struct mana_adev *madev = container_of(dev, struct mana_adev, adev.dev);
3402 
3403 	kfree(madev);
3404 }
3405 
3406 static void remove_adev(struct gdma_dev *gd)
3407 {
3408 	struct auxiliary_device *adev = gd->adev;
3409 	int id = adev->id;
3410 
3411 	auxiliary_device_delete(adev);
3412 	auxiliary_device_uninit(adev);
3413 
3414 	mana_adev_idx_free(id);
3415 	gd->adev = NULL;
3416 }
3417 
3418 static int add_adev(struct gdma_dev *gd, const char *name)
3419 {
3420 	struct auxiliary_device *adev;
3421 	struct mana_adev *madev;
3422 	int ret;
3423 
3424 	madev = kzalloc(sizeof(*madev), GFP_KERNEL);
3425 	if (!madev)
3426 		return -ENOMEM;
3427 
3428 	adev = &madev->adev;
3429 	ret = mana_adev_idx_alloc();
3430 	if (ret < 0)
3431 		goto idx_fail;
3432 	adev->id = ret;
3433 
3434 	adev->name = name;
3435 	adev->dev.parent = gd->gdma_context->dev;
3436 	adev->dev.release = adev_release;
3437 	madev->mdev = gd;
3438 
3439 	ret = auxiliary_device_init(adev);
3440 	if (ret)
3441 		goto init_fail;
3442 
3443 	/* madev is owned by the auxiliary device */
3444 	madev = NULL;
3445 	ret = auxiliary_device_add(adev);
3446 	if (ret)
3447 		goto add_fail;
3448 
3449 	gd->adev = adev;
3450 	dev_dbg(gd->gdma_context->dev,
3451 		"Auxiliary device added successfully\n");
3452 	return 0;
3453 
3454 add_fail:
3455 	auxiliary_device_uninit(adev);
3456 
3457 init_fail:
3458 	mana_adev_idx_free(adev->id);
3459 
3460 idx_fail:
3461 	kfree(madev);
3462 
3463 	return ret;
3464 }
3465 
3466 static void mana_rdma_service_handle(struct work_struct *work)
3467 {
3468 	struct mana_service_work *serv_work =
3469 		container_of(work, struct mana_service_work, work);
3470 	struct gdma_dev *gd = serv_work->gdma_dev;
3471 	struct device *dev = gd->gdma_context->dev;
3472 	int ret;
3473 
3474 	if (READ_ONCE(gd->rdma_teardown))
3475 		goto out;
3476 
3477 	switch (serv_work->event) {
3478 	case GDMA_SERVICE_TYPE_RDMA_SUSPEND:
3479 		if (!gd->adev || gd->is_suspended)
3480 			break;
3481 
3482 		remove_adev(gd);
3483 		gd->is_suspended = true;
3484 		break;
3485 
3486 	case GDMA_SERVICE_TYPE_RDMA_RESUME:
3487 		if (!gd->is_suspended)
3488 			break;
3489 
3490 		ret = add_adev(gd, "rdma");
3491 		if (ret)
3492 			dev_err(dev, "Failed to add adev on resume: %d\n", ret);
3493 		else
3494 			gd->is_suspended = false;
3495 		break;
3496 
3497 	default:
3498 		dev_warn(dev, "unknown adev service event %u\n",
3499 			 serv_work->event);
3500 		break;
3501 	}
3502 
3503 out:
3504 	kfree(serv_work);
3505 }
3506 
3507 int mana_rdma_service_event(struct gdma_context *gc, enum gdma_service_type event)
3508 {
3509 	struct gdma_dev *gd = &gc->mana_ib;
3510 	struct mana_service_work *serv_work;
3511 
3512 	if (gd->dev_id.type != GDMA_DEVICE_MANA_IB) {
3513 		/* RDMA device is not detected on pci */
3514 		return 0;
3515 	}
3516 
3517 	serv_work = kzalloc(sizeof(*serv_work), GFP_ATOMIC);
3518 	if (!serv_work)
3519 		return -ENOMEM;
3520 
3521 	serv_work->event = event;
3522 	serv_work->gdma_dev = gd;
3523 
3524 	INIT_WORK(&serv_work->work, mana_rdma_service_handle);
3525 	queue_work(gc->service_wq, &serv_work->work);
3526 
3527 	return 0;
3528 }
3529 
3530 #define MANA_GF_STATS_PERIOD (2 * HZ)
3531 
3532 static void mana_gf_stats_work_handler(struct work_struct *work)
3533 {
3534 	struct mana_context *ac =
3535 		container_of(to_delayed_work(work), struct mana_context, gf_stats_work);
3536 	int err;
3537 
3538 	err = mana_query_gf_stats(ac);
3539 	if (err == -ETIMEDOUT) {
3540 		/* HWC timeout detected - reset stats and stop rescheduling */
3541 		ac->hwc_timeout_occurred = true;
3542 		memset(&ac->hc_stats, 0, sizeof(ac->hc_stats));
3543 		return;
3544 	}
3545 	schedule_delayed_work(&ac->gf_stats_work, MANA_GF_STATS_PERIOD);
3546 }
3547 
3548 int mana_probe(struct gdma_dev *gd, bool resuming)
3549 {
3550 	struct gdma_context *gc = gd->gdma_context;
3551 	struct mana_context *ac = gd->driver_data;
3552 	struct mana_port_context *apc = NULL;
3553 	struct device *dev = gc->dev;
3554 	u8 bm_hostmode = 0;
3555 	u16 num_ports = 0;
3556 	int err;
3557 	int i;
3558 
3559 	dev_info(dev,
3560 		 "Microsoft Azure Network Adapter protocol version: %d.%d.%d\n",
3561 		 MANA_MAJOR_VERSION, MANA_MINOR_VERSION, MANA_MICRO_VERSION);
3562 
3563 	err = mana_gd_register_device(gd);
3564 	if (err)
3565 		return err;
3566 
3567 	if (!resuming) {
3568 		ac = kzalloc(sizeof(*ac), GFP_KERNEL);
3569 		if (!ac)
3570 			return -ENOMEM;
3571 
3572 		ac->gdma_dev = gd;
3573 		gd->driver_data = ac;
3574 	}
3575 
3576 	err = mana_create_eq(ac);
3577 	if (err) {
3578 		dev_err(dev, "Failed to create EQs: %d\n", err);
3579 		goto out;
3580 	}
3581 
3582 	err = mana_query_device_cfg(ac, MANA_MAJOR_VERSION, MANA_MINOR_VERSION,
3583 				    MANA_MICRO_VERSION, &num_ports, &bm_hostmode);
3584 	if (err)
3585 		goto out;
3586 
3587 	ac->bm_hostmode = bm_hostmode;
3588 
3589 	if (!resuming) {
3590 		ac->num_ports = num_ports;
3591 
3592 		INIT_WORK(&ac->link_change_work, mana_link_state_handle);
3593 	} else {
3594 		if (ac->num_ports != num_ports) {
3595 			dev_err(dev, "The number of vPorts changed: %d->%d\n",
3596 				ac->num_ports, num_ports);
3597 			err = -EPROTO;
3598 			goto out;
3599 		}
3600 
3601 		enable_work(&ac->link_change_work);
3602 	}
3603 
3604 	if (ac->num_ports == 0)
3605 		dev_err(dev, "Failed to detect any vPort\n");
3606 
3607 	if (ac->num_ports > MAX_PORTS_IN_MANA_DEV)
3608 		ac->num_ports = MAX_PORTS_IN_MANA_DEV;
3609 
3610 	ac->per_port_queue_reset_wq =
3611 		create_singlethread_workqueue("mana_per_port_queue_reset_wq");
3612 	if (!ac->per_port_queue_reset_wq) {
3613 		dev_err(dev, "Failed to allocate per port queue reset workqueue\n");
3614 		err = -ENOMEM;
3615 		goto out;
3616 	}
3617 
3618 	if (!resuming) {
3619 		for (i = 0; i < ac->num_ports; i++) {
3620 			err = mana_probe_port(ac, i, &ac->ports[i]);
3621 			/* we log the port for which the probe failed and stop
3622 			 * probes for subsequent ports.
3623 			 * Note that we keep running ports, for which the probes
3624 			 * were successful, unless add_adev fails too
3625 			 */
3626 			if (err) {
3627 				dev_err(dev, "Probe Failed for port %d\n", i);
3628 				break;
3629 			}
3630 		}
3631 	} else {
3632 		for (i = 0; i < ac->num_ports; i++) {
3633 			rtnl_lock();
3634 			apc = netdev_priv(ac->ports[i]);
3635 			enable_work(&apc->queue_reset_work);
3636 			err = mana_attach(ac->ports[i]);
3637 			rtnl_unlock();
3638 			/* we log the port for which the attach failed and stop
3639 			 * attach for subsequent ports
3640 			 * Note that we keep running ports, for which the attach
3641 			 * were successful, unless add_adev fails too
3642 			 */
3643 			if (err) {
3644 				dev_err(dev, "Attach Failed for port %d\n", i);
3645 				break;
3646 			}
3647 		}
3648 	}
3649 
3650 	err = add_adev(gd, "eth");
3651 
3652 	INIT_DELAYED_WORK(&ac->gf_stats_work, mana_gf_stats_work_handler);
3653 	schedule_delayed_work(&ac->gf_stats_work, MANA_GF_STATS_PERIOD);
3654 
3655 out:
3656 	if (err) {
3657 		mana_remove(gd, false);
3658 	} else {
3659 		dev_dbg(dev, "gd=%p, id=%u, num_ports=%d, type=%u, instance=%u\n",
3660 			gd, gd->dev_id.as_uint32, ac->num_ports,
3661 			gd->dev_id.type, gd->dev_id.instance);
3662 		dev_dbg(dev, "%s succeeded\n", __func__);
3663 	}
3664 
3665 	return err;
3666 }
3667 
3668 void mana_remove(struct gdma_dev *gd, bool suspending)
3669 {
3670 	struct gdma_context *gc = gd->gdma_context;
3671 	struct mana_context *ac = gd->driver_data;
3672 	struct mana_port_context *apc;
3673 	struct device *dev = gc->dev;
3674 	struct net_device *ndev;
3675 	int err;
3676 	int i;
3677 
3678 	disable_work_sync(&ac->link_change_work);
3679 	cancel_delayed_work_sync(&ac->gf_stats_work);
3680 
3681 	/* adev currently doesn't support suspending, always remove it */
3682 	if (gd->adev)
3683 		remove_adev(gd);
3684 
3685 	for (i = 0; i < ac->num_ports; i++) {
3686 		ndev = ac->ports[i];
3687 		if (!ndev) {
3688 			if (i == 0)
3689 				dev_err(dev, "No net device to remove\n");
3690 			goto out;
3691 		}
3692 
3693 		apc = netdev_priv(ndev);
3694 		disable_work_sync(&apc->queue_reset_work);
3695 
3696 		/* All cleanup actions should stay after rtnl_lock(), otherwise
3697 		 * other functions may access partially cleaned up data.
3698 		 */
3699 		rtnl_lock();
3700 
3701 		err = mana_detach(ndev, false);
3702 		if (err)
3703 			netdev_err(ndev, "Failed to detach vPort %d: %d\n",
3704 				   i, err);
3705 
3706 		if (suspending) {
3707 			/* No need to unregister the ndev. */
3708 			rtnl_unlock();
3709 			continue;
3710 		}
3711 
3712 		unregister_netdevice(ndev);
3713 		mana_cleanup_indir_table(apc);
3714 
3715 		rtnl_unlock();
3716 
3717 		free_netdev(ndev);
3718 	}
3719 
3720 	mana_destroy_eq(ac);
3721 out:
3722 	if (ac->per_port_queue_reset_wq) {
3723 		destroy_workqueue(ac->per_port_queue_reset_wq);
3724 		ac->per_port_queue_reset_wq = NULL;
3725 	}
3726 
3727 	mana_gd_deregister_device(gd);
3728 
3729 	if (suspending)
3730 		return;
3731 
3732 	gd->driver_data = NULL;
3733 	gd->gdma_context = NULL;
3734 	kfree(ac);
3735 	dev_dbg(dev, "%s succeeded\n", __func__);
3736 }
3737 
3738 int mana_rdma_probe(struct gdma_dev *gd)
3739 {
3740 	int err = 0;
3741 
3742 	if (gd->dev_id.type != GDMA_DEVICE_MANA_IB) {
3743 		/* RDMA device is not detected on pci */
3744 		return err;
3745 	}
3746 
3747 	err = mana_gd_register_device(gd);
3748 	if (err)
3749 		return err;
3750 
3751 	err = add_adev(gd, "rdma");
3752 	if (err)
3753 		mana_gd_deregister_device(gd);
3754 
3755 	return err;
3756 }
3757 
3758 void mana_rdma_remove(struct gdma_dev *gd)
3759 {
3760 	struct gdma_context *gc = gd->gdma_context;
3761 
3762 	if (gd->dev_id.type != GDMA_DEVICE_MANA_IB) {
3763 		/* RDMA device is not detected on pci */
3764 		return;
3765 	}
3766 
3767 	WRITE_ONCE(gd->rdma_teardown, true);
3768 	flush_workqueue(gc->service_wq);
3769 
3770 	if (gd->adev)
3771 		remove_adev(gd);
3772 
3773 	mana_gd_deregister_device(gd);
3774 }
3775 
3776 struct net_device *mana_get_primary_netdev(struct mana_context *ac,
3777 					   u32 port_index,
3778 					   netdevice_tracker *tracker)
3779 {
3780 	struct net_device *ndev;
3781 
3782 	if (port_index >= ac->num_ports)
3783 		return NULL;
3784 
3785 	rcu_read_lock();
3786 
3787 	/* If mana is used in netvsc, the upper netdevice should be returned. */
3788 	ndev = netdev_master_upper_dev_get_rcu(ac->ports[port_index]);
3789 
3790 	/* If there is no upper device, use the parent Ethernet device */
3791 	if (!ndev)
3792 		ndev = ac->ports[port_index];
3793 
3794 	netdev_hold(ndev, tracker, GFP_ATOMIC);
3795 	rcu_read_unlock();
3796 
3797 	return ndev;
3798 }
3799 EXPORT_SYMBOL_NS(mana_get_primary_netdev, "NET_MANA");
3800