xref: /linux/drivers/net/ethernet/hisilicon/hns/hns_enet.c (revision cc622420798c4bcf093785d872525087a7798db9)
1 /*
2  * Copyright (c) 2014-2015 Hisilicon Limited.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  */
9 
10 #include <linux/clk.h>
11 #include <linux/cpumask.h>
12 #include <linux/etherdevice.h>
13 #include <linux/if_vlan.h>
14 #include <linux/interrupt.h>
15 #include <linux/io.h>
16 #include <linux/ip.h>
17 #include <linux/ipv6.h>
18 #include <linux/module.h>
19 #include <linux/phy.h>
20 #include <linux/platform_device.h>
21 #include <linux/skbuff.h>
22 
23 #include "hnae.h"
24 #include "hns_enet.h"
25 
26 #define NIC_MAX_Q_PER_VF 16
27 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
28 
29 #define SERVICE_TIMER_HZ (1 * HZ)
30 
31 #define NIC_TX_CLEAN_MAX_NUM 256
32 #define NIC_RX_CLEAN_MAX_NUM 64
33 
34 #define RCB_IRQ_NOT_INITED 0
35 #define RCB_IRQ_INITED 1
36 #define HNS_BUFFER_SIZE_2048 2048
37 
38 #define BD_MAX_SEND_SIZE 8191
39 #define SKB_TMP_LEN(SKB) \
40 	(((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
41 
42 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
43 			 int size, dma_addr_t dma, int frag_end,
44 			 int buf_num, enum hns_desc_type type, int mtu)
45 {
46 	struct hnae_desc *desc = &ring->desc[ring->next_to_use];
47 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
48 	struct iphdr *iphdr;
49 	struct ipv6hdr *ipv6hdr;
50 	struct sk_buff *skb;
51 	__be16 protocol;
52 	u8 bn_pid = 0;
53 	u8 rrcfv = 0;
54 	u8 ip_offset = 0;
55 	u8 tvsvsn = 0;
56 	u16 mss = 0;
57 	u8 l4_len = 0;
58 	u16 paylen = 0;
59 
60 	desc_cb->priv = priv;
61 	desc_cb->length = size;
62 	desc_cb->dma = dma;
63 	desc_cb->type = type;
64 
65 	desc->addr = cpu_to_le64(dma);
66 	desc->tx.send_size = cpu_to_le16((u16)size);
67 
68 	/* config bd buffer end */
69 	hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
70 	hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
71 
72 	/* fill port_id in the tx bd for sending management pkts */
73 	hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
74 		       HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
75 
76 	if (type == DESC_TYPE_SKB) {
77 		skb = (struct sk_buff *)priv;
78 
79 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
80 			skb_reset_mac_len(skb);
81 			protocol = skb->protocol;
82 			ip_offset = ETH_HLEN;
83 
84 			if (protocol == htons(ETH_P_8021Q)) {
85 				ip_offset += VLAN_HLEN;
86 				protocol = vlan_get_protocol(skb);
87 				skb->protocol = protocol;
88 			}
89 
90 			if (skb->protocol == htons(ETH_P_IP)) {
91 				iphdr = ip_hdr(skb);
92 				hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
93 				hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
94 
95 				/* check for tcp/udp header */
96 				if (iphdr->protocol == IPPROTO_TCP &&
97 				    skb_is_gso(skb)) {
98 					hnae_set_bit(tvsvsn,
99 						     HNSV2_TXD_TSE_B, 1);
100 					l4_len = tcp_hdrlen(skb);
101 					mss = skb_shinfo(skb)->gso_size;
102 					paylen = skb->len - SKB_TMP_LEN(skb);
103 				}
104 			} else if (skb->protocol == htons(ETH_P_IPV6)) {
105 				hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
106 				ipv6hdr = ipv6_hdr(skb);
107 				hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
108 
109 				/* check for tcp/udp header */
110 				if (ipv6hdr->nexthdr == IPPROTO_TCP &&
111 				    skb_is_gso(skb) && skb_is_gso_v6(skb)) {
112 					hnae_set_bit(tvsvsn,
113 						     HNSV2_TXD_TSE_B, 1);
114 					l4_len = tcp_hdrlen(skb);
115 					mss = skb_shinfo(skb)->gso_size;
116 					paylen = skb->len - SKB_TMP_LEN(skb);
117 				}
118 			}
119 			desc->tx.ip_offset = ip_offset;
120 			desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
121 			desc->tx.mss = cpu_to_le16(mss);
122 			desc->tx.l4_len = l4_len;
123 			desc->tx.paylen = cpu_to_le16(paylen);
124 		}
125 	}
126 
127 	hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
128 
129 	desc->tx.bn_pid = bn_pid;
130 	desc->tx.ra_ri_cs_fe_vld = rrcfv;
131 
132 	ring_ptr_move_fw(ring, next_to_use);
133 }
134 
135 static void fill_desc(struct hnae_ring *ring, void *priv,
136 		      int size, dma_addr_t dma, int frag_end,
137 		      int buf_num, enum hns_desc_type type, int mtu)
138 {
139 	struct hnae_desc *desc = &ring->desc[ring->next_to_use];
140 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
141 	struct sk_buff *skb;
142 	__be16 protocol;
143 	u32 ip_offset;
144 	u32 asid_bufnum_pid = 0;
145 	u32 flag_ipoffset = 0;
146 
147 	desc_cb->priv = priv;
148 	desc_cb->length = size;
149 	desc_cb->dma = dma;
150 	desc_cb->type = type;
151 
152 	desc->addr = cpu_to_le64(dma);
153 	desc->tx.send_size = cpu_to_le16((u16)size);
154 
155 	/*config bd buffer end */
156 	flag_ipoffset |= 1 << HNS_TXD_VLD_B;
157 
158 	asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
159 
160 	if (type == DESC_TYPE_SKB) {
161 		skb = (struct sk_buff *)priv;
162 
163 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
164 			protocol = skb->protocol;
165 			ip_offset = ETH_HLEN;
166 
167 			/*if it is a SW VLAN check the next protocol*/
168 			if (protocol == htons(ETH_P_8021Q)) {
169 				ip_offset += VLAN_HLEN;
170 				protocol = vlan_get_protocol(skb);
171 				skb->protocol = protocol;
172 			}
173 
174 			if (skb->protocol == htons(ETH_P_IP)) {
175 				flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
176 				/* check for tcp/udp header */
177 				flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
178 
179 			} else if (skb->protocol == htons(ETH_P_IPV6)) {
180 				/* ipv6 has not l3 cs, check for L4 header */
181 				flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
182 			}
183 
184 			flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
185 		}
186 	}
187 
188 	flag_ipoffset |= frag_end << HNS_TXD_FE_B;
189 
190 	desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
191 	desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
192 
193 	ring_ptr_move_fw(ring, next_to_use);
194 }
195 
196 static void unfill_desc(struct hnae_ring *ring)
197 {
198 	ring_ptr_move_bw(ring, next_to_use);
199 }
200 
201 static int hns_nic_maybe_stop_tx(
202 	struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
203 {
204 	struct sk_buff *skb = *out_skb;
205 	struct sk_buff *new_skb = NULL;
206 	int buf_num;
207 
208 	/* no. of segments (plus a header) */
209 	buf_num = skb_shinfo(skb)->nr_frags + 1;
210 
211 	if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
212 		if (ring_space(ring) < 1)
213 			return -EBUSY;
214 
215 		new_skb = skb_copy(skb, GFP_ATOMIC);
216 		if (!new_skb)
217 			return -ENOMEM;
218 
219 		dev_kfree_skb_any(skb);
220 		*out_skb = new_skb;
221 		buf_num = 1;
222 	} else if (buf_num > ring_space(ring)) {
223 		return -EBUSY;
224 	}
225 
226 	*bnum = buf_num;
227 	return 0;
228 }
229 
230 static int hns_nic_maybe_stop_tso(
231 	struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
232 {
233 	int i;
234 	int size;
235 	int buf_num;
236 	int frag_num;
237 	struct sk_buff *skb = *out_skb;
238 	struct sk_buff *new_skb = NULL;
239 	struct skb_frag_struct *frag;
240 
241 	size = skb_headlen(skb);
242 	buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
243 
244 	frag_num = skb_shinfo(skb)->nr_frags;
245 	for (i = 0; i < frag_num; i++) {
246 		frag = &skb_shinfo(skb)->frags[i];
247 		size = skb_frag_size(frag);
248 		buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
249 	}
250 
251 	if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
252 		buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
253 		if (ring_space(ring) < buf_num)
254 			return -EBUSY;
255 		/* manual split the send packet */
256 		new_skb = skb_copy(skb, GFP_ATOMIC);
257 		if (!new_skb)
258 			return -ENOMEM;
259 		dev_kfree_skb_any(skb);
260 		*out_skb = new_skb;
261 
262 	} else if (ring_space(ring) < buf_num) {
263 		return -EBUSY;
264 	}
265 
266 	*bnum = buf_num;
267 	return 0;
268 }
269 
270 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
271 			  int size, dma_addr_t dma, int frag_end,
272 			  int buf_num, enum hns_desc_type type, int mtu)
273 {
274 	int frag_buf_num;
275 	int sizeoflast;
276 	int k;
277 
278 	frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
279 	sizeoflast = size % BD_MAX_SEND_SIZE;
280 	sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
281 
282 	/* when the frag size is bigger than hardware, split this frag */
283 	for (k = 0; k < frag_buf_num; k++)
284 		fill_v2_desc(ring, priv,
285 			     (k == frag_buf_num - 1) ?
286 					sizeoflast : BD_MAX_SEND_SIZE,
287 			     dma + BD_MAX_SEND_SIZE * k,
288 			     frag_end && (k == frag_buf_num - 1) ? 1 : 0,
289 			     buf_num,
290 			     (type == DESC_TYPE_SKB && !k) ?
291 					DESC_TYPE_SKB : DESC_TYPE_PAGE,
292 			     mtu);
293 }
294 
295 int hns_nic_net_xmit_hw(struct net_device *ndev,
296 			struct sk_buff *skb,
297 			struct hns_nic_ring_data *ring_data)
298 {
299 	struct hns_nic_priv *priv = netdev_priv(ndev);
300 	struct device *dev = priv->dev;
301 	struct hnae_ring *ring = ring_data->ring;
302 	struct netdev_queue *dev_queue;
303 	struct skb_frag_struct *frag;
304 	int buf_num;
305 	int seg_num;
306 	dma_addr_t dma;
307 	int size, next_to_use;
308 	int i;
309 
310 	switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
311 	case -EBUSY:
312 		ring->stats.tx_busy++;
313 		goto out_net_tx_busy;
314 	case -ENOMEM:
315 		ring->stats.sw_err_cnt++;
316 		netdev_err(ndev, "no memory to xmit!\n");
317 		goto out_err_tx_ok;
318 	default:
319 		break;
320 	}
321 
322 	/* no. of segments (plus a header) */
323 	seg_num = skb_shinfo(skb)->nr_frags + 1;
324 	next_to_use = ring->next_to_use;
325 
326 	/* fill the first part */
327 	size = skb_headlen(skb);
328 	dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
329 	if (dma_mapping_error(dev, dma)) {
330 		netdev_err(ndev, "TX head DMA map failed\n");
331 		ring->stats.sw_err_cnt++;
332 		goto out_err_tx_ok;
333 	}
334 	priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
335 			    buf_num, DESC_TYPE_SKB, ndev->mtu);
336 
337 	/* fill the fragments */
338 	for (i = 1; i < seg_num; i++) {
339 		frag = &skb_shinfo(skb)->frags[i - 1];
340 		size = skb_frag_size(frag);
341 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
342 		if (dma_mapping_error(dev, dma)) {
343 			netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
344 			ring->stats.sw_err_cnt++;
345 			goto out_map_frag_fail;
346 		}
347 		priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
348 				    seg_num - 1 == i ? 1 : 0, buf_num,
349 				    DESC_TYPE_PAGE, ndev->mtu);
350 	}
351 
352 	/*complete translate all packets*/
353 	dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
354 	netdev_tx_sent_queue(dev_queue, skb->len);
355 
356 	wmb(); /* commit all data before submit */
357 	assert(skb->queue_mapping < priv->ae_handle->q_num);
358 	hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
359 	ring->stats.tx_pkts++;
360 	ring->stats.tx_bytes += skb->len;
361 
362 	return NETDEV_TX_OK;
363 
364 out_map_frag_fail:
365 
366 	while (ring->next_to_use != next_to_use) {
367 		unfill_desc(ring);
368 		if (ring->next_to_use != next_to_use)
369 			dma_unmap_page(dev,
370 				       ring->desc_cb[ring->next_to_use].dma,
371 				       ring->desc_cb[ring->next_to_use].length,
372 				       DMA_TO_DEVICE);
373 		else
374 			dma_unmap_single(dev,
375 					 ring->desc_cb[next_to_use].dma,
376 					 ring->desc_cb[next_to_use].length,
377 					 DMA_TO_DEVICE);
378 	}
379 
380 out_err_tx_ok:
381 
382 	dev_kfree_skb_any(skb);
383 	return NETDEV_TX_OK;
384 
385 out_net_tx_busy:
386 
387 	netif_stop_subqueue(ndev, skb->queue_mapping);
388 
389 	/* Herbert's original patch had:
390 	 *  smp_mb__after_netif_stop_queue();
391 	 * but since that doesn't exist yet, just open code it.
392 	 */
393 	smp_mb();
394 	return NETDEV_TX_BUSY;
395 }
396 
397 /**
398  * hns_nic_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
399  * @data: pointer to the start of the headers
400  * @max: total length of section to find headers in
401  *
402  * This function is meant to determine the length of headers that will
403  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
404  * motivation of doing this is to only perform one pull for IPv4 TCP
405  * packets so that we can do basic things like calculating the gso_size
406  * based on the average data per packet.
407  **/
408 static unsigned int hns_nic_get_headlen(unsigned char *data, u32 flag,
409 					unsigned int max_size)
410 {
411 	unsigned char *network;
412 	u8 hlen;
413 
414 	/* this should never happen, but better safe than sorry */
415 	if (max_size < ETH_HLEN)
416 		return max_size;
417 
418 	/* initialize network frame pointer */
419 	network = data;
420 
421 	/* set first protocol and move network header forward */
422 	network += ETH_HLEN;
423 
424 	/* handle any vlan tag if present */
425 	if (hnae_get_field(flag, HNS_RXD_VLAN_M, HNS_RXD_VLAN_S)
426 		== HNS_RX_FLAG_VLAN_PRESENT) {
427 		if ((typeof(max_size))(network - data) > (max_size - VLAN_HLEN))
428 			return max_size;
429 
430 		network += VLAN_HLEN;
431 	}
432 
433 	/* handle L3 protocols */
434 	if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
435 		== HNS_RX_FLAG_L3ID_IPV4) {
436 		if ((typeof(max_size))(network - data) >
437 		    (max_size - sizeof(struct iphdr)))
438 			return max_size;
439 
440 		/* access ihl as a u8 to avoid unaligned access on ia64 */
441 		hlen = (network[0] & 0x0F) << 2;
442 
443 		/* verify hlen meets minimum size requirements */
444 		if (hlen < sizeof(struct iphdr))
445 			return network - data;
446 
447 		/* record next protocol if header is present */
448 	} else if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
449 		== HNS_RX_FLAG_L3ID_IPV6) {
450 		if ((typeof(max_size))(network - data) >
451 		    (max_size - sizeof(struct ipv6hdr)))
452 			return max_size;
453 
454 		/* record next protocol */
455 		hlen = sizeof(struct ipv6hdr);
456 	} else {
457 		return network - data;
458 	}
459 
460 	/* relocate pointer to start of L4 header */
461 	network += hlen;
462 
463 	/* finally sort out TCP/UDP */
464 	if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
465 		== HNS_RX_FLAG_L4ID_TCP) {
466 		if ((typeof(max_size))(network - data) >
467 		    (max_size - sizeof(struct tcphdr)))
468 			return max_size;
469 
470 		/* access doff as a u8 to avoid unaligned access on ia64 */
471 		hlen = (network[12] & 0xF0) >> 2;
472 
473 		/* verify hlen meets minimum size requirements */
474 		if (hlen < sizeof(struct tcphdr))
475 			return network - data;
476 
477 		network += hlen;
478 	} else if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
479 		== HNS_RX_FLAG_L4ID_UDP) {
480 		if ((typeof(max_size))(network - data) >
481 		    (max_size - sizeof(struct udphdr)))
482 			return max_size;
483 
484 		network += sizeof(struct udphdr);
485 	}
486 
487 	/* If everything has gone correctly network should be the
488 	 * data section of the packet and will be the end of the header.
489 	 * If not then it probably represents the end of the last recognized
490 	 * header.
491 	 */
492 	if ((typeof(max_size))(network - data) < max_size)
493 		return network - data;
494 	else
495 		return max_size;
496 }
497 
498 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
499 			       struct hnae_ring *ring, int pull_len,
500 			       struct hnae_desc_cb *desc_cb)
501 {
502 	struct hnae_desc *desc;
503 	int truesize, size;
504 	int last_offset;
505 	bool twobufs;
506 
507 	twobufs = ((PAGE_SIZE < 8192) && hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
508 
509 	desc = &ring->desc[ring->next_to_clean];
510 	size = le16_to_cpu(desc->rx.size);
511 
512 	if (twobufs) {
513 		truesize = hnae_buf_size(ring);
514 	} else {
515 		truesize = ALIGN(size, L1_CACHE_BYTES);
516 		last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
517 	}
518 
519 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
520 			size - pull_len, truesize - pull_len);
521 
522 	 /* avoid re-using remote pages,flag default unreuse */
523 	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
524 		return;
525 
526 	if (twobufs) {
527 		/* if we are only owner of page we can reuse it */
528 		if (likely(page_count(desc_cb->priv) == 1)) {
529 			/* flip page offset to other buffer */
530 			desc_cb->page_offset ^= truesize;
531 
532 			desc_cb->reuse_flag = 1;
533 			/* bump ref count on page before it is given*/
534 			get_page(desc_cb->priv);
535 		}
536 		return;
537 	}
538 
539 	/* move offset up to the next cache line */
540 	desc_cb->page_offset += truesize;
541 
542 	if (desc_cb->page_offset <= last_offset) {
543 		desc_cb->reuse_flag = 1;
544 		/* bump ref count on page before it is given*/
545 		get_page(desc_cb->priv);
546 	}
547 }
548 
549 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
550 {
551 	*out_bnum = hnae_get_field(bnum_flag,
552 				   HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
553 }
554 
555 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
556 {
557 	*out_bnum = hnae_get_field(bnum_flag,
558 				   HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
559 }
560 
561 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
562 			       struct sk_buff **out_skb, int *out_bnum)
563 {
564 	struct hnae_ring *ring = ring_data->ring;
565 	struct net_device *ndev = ring_data->napi.dev;
566 	struct hns_nic_priv *priv = netdev_priv(ndev);
567 	struct sk_buff *skb;
568 	struct hnae_desc *desc;
569 	struct hnae_desc_cb *desc_cb;
570 	struct ethhdr *eh;
571 	unsigned char *va;
572 	int bnum, length, i;
573 	int pull_len;
574 	u32 bnum_flag;
575 
576 	desc = &ring->desc[ring->next_to_clean];
577 	desc_cb = &ring->desc_cb[ring->next_to_clean];
578 
579 	prefetch(desc);
580 
581 	va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
582 
583 	/* prefetch first cache line of first page */
584 	prefetch(va);
585 #if L1_CACHE_BYTES < 128
586 	prefetch(va + L1_CACHE_BYTES);
587 #endif
588 
589 	skb = *out_skb = napi_alloc_skb(&ring_data->napi,
590 					HNS_RX_HEAD_SIZE);
591 	if (unlikely(!skb)) {
592 		netdev_err(ndev, "alloc rx skb fail\n");
593 		ring->stats.sw_err_cnt++;
594 		return -ENOMEM;
595 	}
596 
597 	prefetchw(skb->data);
598 	length = le16_to_cpu(desc->rx.pkt_len);
599 	bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
600 	priv->ops.get_rxd_bnum(bnum_flag, &bnum);
601 	*out_bnum = bnum;
602 
603 	if (length <= HNS_RX_HEAD_SIZE) {
604 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
605 
606 		/* we can reuse buffer as-is, just make sure it is local */
607 		if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
608 			desc_cb->reuse_flag = 1;
609 		else /* this page cannot be reused so discard it */
610 			put_page(desc_cb->priv);
611 
612 		ring_ptr_move_fw(ring, next_to_clean);
613 
614 		if (unlikely(bnum != 1)) { /* check err*/
615 			*out_bnum = 1;
616 			goto out_bnum_err;
617 		}
618 	} else {
619 		ring->stats.seg_pkt_cnt++;
620 
621 		pull_len = hns_nic_get_headlen(va, bnum_flag, HNS_RX_HEAD_SIZE);
622 		memcpy(__skb_put(skb, pull_len), va,
623 		       ALIGN(pull_len, sizeof(long)));
624 
625 		hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
626 		ring_ptr_move_fw(ring, next_to_clean);
627 
628 		if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
629 			*out_bnum = 1;
630 			goto out_bnum_err;
631 		}
632 		for (i = 1; i < bnum; i++) {
633 			desc = &ring->desc[ring->next_to_clean];
634 			desc_cb = &ring->desc_cb[ring->next_to_clean];
635 
636 			hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
637 			ring_ptr_move_fw(ring, next_to_clean);
638 		}
639 	}
640 
641 	/* check except process, free skb and jump the desc */
642 	if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
643 out_bnum_err:
644 		*out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
645 		netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
646 			   bnum, ring->max_desc_num_per_pkt,
647 			   length, (int)MAX_SKB_FRAGS,
648 			   ((u64 *)desc)[0], ((u64 *)desc)[1]);
649 		ring->stats.err_bd_num++;
650 		dev_kfree_skb_any(skb);
651 		return -EDOM;
652 	}
653 
654 	bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
655 
656 	if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
657 		netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
658 			   ((u64 *)desc)[0], ((u64 *)desc)[1]);
659 		ring->stats.non_vld_descs++;
660 		dev_kfree_skb_any(skb);
661 		return -EINVAL;
662 	}
663 
664 	if (unlikely((!desc->rx.pkt_len) ||
665 		     hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
666 		ring->stats.err_pkt_len++;
667 		dev_kfree_skb_any(skb);
668 		return -EFAULT;
669 	}
670 
671 	if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
672 		ring->stats.l2_err++;
673 		dev_kfree_skb_any(skb);
674 		return -EFAULT;
675 	}
676 
677 	/* filter out multicast pkt with the same src mac as this port */
678 	eh = eth_hdr(skb);
679 	if (unlikely(is_multicast_ether_addr(eh->h_dest) &&
680 		     ether_addr_equal(ndev->dev_addr, eh->h_source))) {
681 		dev_kfree_skb_any(skb);
682 		return -EFAULT;
683 	}
684 
685 	ring->stats.rx_pkts++;
686 	ring->stats.rx_bytes += skb->len;
687 
688 	if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L3E_B) ||
689 		     hnae_get_bit(bnum_flag, HNS_RXD_L4E_B))) {
690 		ring->stats.l3l4_csum_err++;
691 		return 0;
692 	}
693 
694 	skb->ip_summed = CHECKSUM_UNNECESSARY;
695 
696 	return 0;
697 }
698 
699 static void
700 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
701 {
702 	int i, ret;
703 	struct hnae_desc_cb res_cbs;
704 	struct hnae_desc_cb *desc_cb;
705 	struct hnae_ring *ring = ring_data->ring;
706 	struct net_device *ndev = ring_data->napi.dev;
707 
708 	for (i = 0; i < cleand_count; i++) {
709 		desc_cb = &ring->desc_cb[ring->next_to_use];
710 		if (desc_cb->reuse_flag) {
711 			ring->stats.reuse_pg_cnt++;
712 			hnae_reuse_buffer(ring, ring->next_to_use);
713 		} else {
714 			ret = hnae_reserve_buffer_map(ring, &res_cbs);
715 			if (ret) {
716 				ring->stats.sw_err_cnt++;
717 				netdev_err(ndev, "hnae reserve buffer map failed.\n");
718 				break;
719 			}
720 			hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
721 		}
722 
723 		ring_ptr_move_fw(ring, next_to_use);
724 	}
725 
726 	wmb(); /* make all data has been write before submit */
727 	writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
728 }
729 
730 /* return error number for error or number of desc left to take
731  */
732 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
733 			      struct sk_buff *skb)
734 {
735 	struct net_device *ndev = ring_data->napi.dev;
736 
737 	skb->protocol = eth_type_trans(skb, ndev);
738 	(void)napi_gro_receive(&ring_data->napi, skb);
739 	ndev->last_rx = jiffies;
740 }
741 
742 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
743 			       int budget, void *v)
744 {
745 	struct hnae_ring *ring = ring_data->ring;
746 	struct sk_buff *skb;
747 	int num, bnum, ex_num;
748 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
749 	int recv_pkts, recv_bds, clean_count, err;
750 
751 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
752 	rmb(); /* make sure num taken effect before the other data is touched */
753 
754 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
755 recv:
756 	while (recv_pkts < budget && recv_bds < num) {
757 		/* reuse or realloc buffers*/
758 		if (clean_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
759 			hns_nic_alloc_rx_buffers(ring_data, clean_count);
760 			clean_count = 0;
761 		}
762 
763 		/* poll one pkg*/
764 		err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
765 		if (unlikely(!skb)) /* this fault cannot be repaired */
766 			break;
767 
768 		recv_bds += bnum;
769 		clean_count += bnum;
770 		if (unlikely(err)) {  /* do jump the err */
771 			recv_pkts++;
772 			continue;
773 		}
774 
775 		/* do update ip stack process*/
776 		((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
777 							ring_data, skb);
778 		recv_pkts++;
779 	}
780 
781 	/* make all data has been write before submit */
782 	if (recv_pkts < budget) {
783 		ex_num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
784 
785 		if (ex_num > clean_count) {
786 			num += ex_num - clean_count;
787 			rmb(); /*complete read rx ring bd number*/
788 			goto recv;
789 		}
790 	}
791 
792 	/* make all data has been write before submit */
793 	if (clean_count > 0)
794 		hns_nic_alloc_rx_buffers(ring_data, clean_count);
795 
796 	return recv_pkts;
797 }
798 
799 static void hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
800 {
801 	struct hnae_ring *ring = ring_data->ring;
802 	int num = 0;
803 
804 	/* for hardware bug fixed */
805 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
806 
807 	if (num > 0) {
808 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
809 			ring_data->ring, 1);
810 
811 		napi_schedule(&ring_data->napi);
812 	}
813 }
814 
815 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
816 					    int *bytes, int *pkts)
817 {
818 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
819 
820 	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
821 	(*bytes) += desc_cb->length;
822 	/* desc_cb will be cleaned, after hnae_free_buffer_detach*/
823 	hnae_free_buffer_detach(ring, ring->next_to_clean);
824 
825 	ring_ptr_move_fw(ring, next_to_clean);
826 }
827 
828 static int is_valid_clean_head(struct hnae_ring *ring, int h)
829 {
830 	int u = ring->next_to_use;
831 	int c = ring->next_to_clean;
832 
833 	if (unlikely(h > ring->desc_num))
834 		return 0;
835 
836 	assert(u > 0 && u < ring->desc_num);
837 	assert(c > 0 && c < ring->desc_num);
838 	assert(u != c && h != c); /* must be checked before call this func */
839 
840 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
841 }
842 
843 /* netif_tx_lock will turn down the performance, set only when necessary */
844 #ifdef CONFIG_NET_POLL_CONTROLLER
845 #define NETIF_TX_LOCK(ndev) netif_tx_lock(ndev)
846 #define NETIF_TX_UNLOCK(ndev) netif_tx_unlock(ndev)
847 #else
848 #define NETIF_TX_LOCK(ndev)
849 #define NETIF_TX_UNLOCK(ndev)
850 #endif
851 /* reclaim all desc in one budget
852  * return error or number of desc left
853  */
854 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
855 			       int budget, void *v)
856 {
857 	struct hnae_ring *ring = ring_data->ring;
858 	struct net_device *ndev = ring_data->napi.dev;
859 	struct netdev_queue *dev_queue;
860 	struct hns_nic_priv *priv = netdev_priv(ndev);
861 	int head;
862 	int bytes, pkts;
863 
864 	NETIF_TX_LOCK(ndev);
865 
866 	head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
867 	rmb(); /* make sure head is ready before touch any data */
868 
869 	if (is_ring_empty(ring) || head == ring->next_to_clean) {
870 		NETIF_TX_UNLOCK(ndev);
871 		return 0; /* no data to poll */
872 	}
873 
874 	if (!is_valid_clean_head(ring, head)) {
875 		netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
876 			   ring->next_to_use, ring->next_to_clean);
877 		ring->stats.io_err_cnt++;
878 		NETIF_TX_UNLOCK(ndev);
879 		return -EIO;
880 	}
881 
882 	bytes = 0;
883 	pkts = 0;
884 	while (head != ring->next_to_clean) {
885 		hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
886 		/* issue prefetch for next Tx descriptor */
887 		prefetch(&ring->desc_cb[ring->next_to_clean]);
888 	}
889 
890 	NETIF_TX_UNLOCK(ndev);
891 
892 	dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
893 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
894 
895 	if (unlikely(priv->link && !netif_carrier_ok(ndev)))
896 		netif_carrier_on(ndev);
897 
898 	if (unlikely(pkts && netif_carrier_ok(ndev) &&
899 		     (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
900 		/* Make sure that anybody stopping the queue after this
901 		 * sees the new next_to_clean.
902 		 */
903 		smp_mb();
904 		if (netif_tx_queue_stopped(dev_queue) &&
905 		    !test_bit(NIC_STATE_DOWN, &priv->state)) {
906 			netif_tx_wake_queue(dev_queue);
907 			ring->stats.restart_queue++;
908 		}
909 	}
910 	return 0;
911 }
912 
913 static void hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
914 {
915 	struct hnae_ring *ring = ring_data->ring;
916 	int head = ring->next_to_clean;
917 
918 	/* for hardware bug fixed */
919 	head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
920 
921 	if (head != ring->next_to_clean) {
922 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
923 			ring_data->ring, 1);
924 
925 		napi_schedule(&ring_data->napi);
926 	}
927 }
928 
929 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
930 {
931 	struct hnae_ring *ring = ring_data->ring;
932 	struct net_device *ndev = ring_data->napi.dev;
933 	struct netdev_queue *dev_queue;
934 	int head;
935 	int bytes, pkts;
936 
937 	NETIF_TX_LOCK(ndev);
938 
939 	head = ring->next_to_use; /* ntu :soft setted ring position*/
940 	bytes = 0;
941 	pkts = 0;
942 	while (head != ring->next_to_clean)
943 		hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
944 
945 	NETIF_TX_UNLOCK(ndev);
946 
947 	dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
948 	netdev_tx_reset_queue(dev_queue);
949 }
950 
951 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
952 {
953 	struct hns_nic_ring_data *ring_data =
954 		container_of(napi, struct hns_nic_ring_data, napi);
955 	int clean_complete = ring_data->poll_one(
956 				ring_data, budget, ring_data->ex_process);
957 
958 	if (clean_complete >= 0 && clean_complete < budget) {
959 		napi_complete(napi);
960 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
961 			ring_data->ring, 0);
962 
963 		ring_data->fini_process(ring_data);
964 		return 0;
965 	}
966 
967 	return clean_complete;
968 }
969 
970 static irqreturn_t hns_irq_handle(int irq, void *dev)
971 {
972 	struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
973 
974 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
975 		ring_data->ring, 1);
976 	napi_schedule(&ring_data->napi);
977 
978 	return IRQ_HANDLED;
979 }
980 
981 /**
982  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
983  *@ndev: net device
984  */
985 static void hns_nic_adjust_link(struct net_device *ndev)
986 {
987 	struct hns_nic_priv *priv = netdev_priv(ndev);
988 	struct hnae_handle *h = priv->ae_handle;
989 
990 	h->dev->ops->adjust_link(h, ndev->phydev->speed, ndev->phydev->duplex);
991 }
992 
993 /**
994  *hns_nic_init_phy - init phy
995  *@ndev: net device
996  *@h: ae handle
997  * Return 0 on success, negative on failure
998  */
999 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1000 {
1001 	struct hns_nic_priv *priv = netdev_priv(ndev);
1002 	struct phy_device *phy_dev = NULL;
1003 
1004 	if (!h->phy_node)
1005 		return 0;
1006 
1007 	if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1008 		phy_dev = of_phy_connect(ndev, h->phy_node,
1009 					 hns_nic_adjust_link, 0, h->phy_if);
1010 	else
1011 		phy_dev = of_phy_attach(ndev, h->phy_node, 0, h->phy_if);
1012 
1013 	if (unlikely(!phy_dev) || IS_ERR(phy_dev))
1014 		return !phy_dev ? -ENODEV : PTR_ERR(phy_dev);
1015 
1016 	phy_dev->supported &= h->if_support;
1017 	phy_dev->advertising = phy_dev->supported;
1018 
1019 	if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1020 		phy_dev->autoneg = false;
1021 
1022 	priv->phy = phy_dev;
1023 
1024 	return 0;
1025 }
1026 
1027 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1028 {
1029 	struct hns_nic_priv *priv = netdev_priv(netdev);
1030 	struct hnae_handle *h = priv->ae_handle;
1031 
1032 	napi_enable(&priv->ring_data[idx].napi);
1033 
1034 	enable_irq(priv->ring_data[idx].ring->irq);
1035 	h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1036 
1037 	return 0;
1038 }
1039 
1040 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1041 {
1042 	struct hns_nic_priv *priv = netdev_priv(ndev);
1043 	struct hnae_handle *h = priv->ae_handle;
1044 	struct sockaddr *mac_addr = p;
1045 	int ret;
1046 
1047 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1048 		return -EADDRNOTAVAIL;
1049 
1050 	ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1051 	if (ret) {
1052 		netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1053 		return ret;
1054 	}
1055 
1056 	memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1057 
1058 	return 0;
1059 }
1060 
1061 void hns_nic_update_stats(struct net_device *netdev)
1062 {
1063 	struct hns_nic_priv *priv = netdev_priv(netdev);
1064 	struct hnae_handle *h = priv->ae_handle;
1065 
1066 	h->dev->ops->update_stats(h, &netdev->stats);
1067 }
1068 
1069 /* set mac addr if it is configed. or leave it to the AE driver */
1070 static void hns_init_mac_addr(struct net_device *ndev)
1071 {
1072 	struct hns_nic_priv *priv = netdev_priv(ndev);
1073 	struct device_node *node = priv->dev->of_node;
1074 	const void *mac_addr_temp;
1075 
1076 	mac_addr_temp = of_get_mac_address(node);
1077 	if (mac_addr_temp && is_valid_ether_addr(mac_addr_temp)) {
1078 		memcpy(ndev->dev_addr, mac_addr_temp, ndev->addr_len);
1079 	} else {
1080 		eth_hw_addr_random(ndev);
1081 		dev_warn(priv->dev, "No valid mac, use random mac %pM",
1082 			 ndev->dev_addr);
1083 	}
1084 }
1085 
1086 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1087 {
1088 	struct hns_nic_priv *priv = netdev_priv(netdev);
1089 	struct hnae_handle *h = priv->ae_handle;
1090 
1091 	h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1092 	disable_irq(priv->ring_data[idx].ring->irq);
1093 
1094 	napi_disable(&priv->ring_data[idx].napi);
1095 }
1096 
1097 static void hns_set_irq_affinity(struct hns_nic_priv *priv)
1098 {
1099 	struct hnae_handle *h = priv->ae_handle;
1100 	struct hns_nic_ring_data *rd;
1101 	int i;
1102 	int cpu;
1103 	cpumask_t mask;
1104 
1105 	/*diffrent irq banlance for 16core and 32core*/
1106 	if (h->q_num == num_possible_cpus()) {
1107 		for (i = 0; i < h->q_num * 2; i++) {
1108 			rd = &priv->ring_data[i];
1109 			if (cpu_online(rd->queue_index)) {
1110 				cpumask_clear(&mask);
1111 				cpu = rd->queue_index;
1112 				cpumask_set_cpu(cpu, &mask);
1113 				(void)irq_set_affinity_hint(rd->ring->irq,
1114 							    &mask);
1115 			}
1116 		}
1117 	} else {
1118 		for (i = 0; i < h->q_num; i++) {
1119 			rd = &priv->ring_data[i];
1120 			if (cpu_online(rd->queue_index * 2)) {
1121 				cpumask_clear(&mask);
1122 				cpu = rd->queue_index * 2;
1123 				cpumask_set_cpu(cpu, &mask);
1124 				(void)irq_set_affinity_hint(rd->ring->irq,
1125 							    &mask);
1126 			}
1127 		}
1128 
1129 		for (i = h->q_num; i < h->q_num * 2; i++) {
1130 			rd = &priv->ring_data[i];
1131 			if (cpu_online(rd->queue_index * 2 + 1)) {
1132 				cpumask_clear(&mask);
1133 				cpu = rd->queue_index * 2 + 1;
1134 				cpumask_set_cpu(cpu, &mask);
1135 				(void)irq_set_affinity_hint(rd->ring->irq,
1136 							    &mask);
1137 			}
1138 		}
1139 	}
1140 }
1141 
1142 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1143 {
1144 	struct hnae_handle *h = priv->ae_handle;
1145 	struct hns_nic_ring_data *rd;
1146 	int i;
1147 	int ret;
1148 
1149 	for (i = 0; i < h->q_num * 2; i++) {
1150 		rd = &priv->ring_data[i];
1151 
1152 		if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1153 			break;
1154 
1155 		snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1156 			 "%s-%s%d", priv->netdev->name,
1157 			 (i < h->q_num ? "tx" : "rx"), rd->queue_index);
1158 
1159 		rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1160 
1161 		ret = request_irq(rd->ring->irq,
1162 				  hns_irq_handle, 0, rd->ring->ring_name, rd);
1163 		if (ret) {
1164 			netdev_err(priv->netdev, "request irq(%d) fail\n",
1165 				   rd->ring->irq);
1166 			return ret;
1167 		}
1168 		disable_irq(rd->ring->irq);
1169 		rd->ring->irq_init_flag = RCB_IRQ_INITED;
1170 	}
1171 
1172 	/*set cpu affinity*/
1173 	hns_set_irq_affinity(priv);
1174 
1175 	return 0;
1176 }
1177 
1178 static int hns_nic_net_up(struct net_device *ndev)
1179 {
1180 	struct hns_nic_priv *priv = netdev_priv(ndev);
1181 	struct hnae_handle *h = priv->ae_handle;
1182 	int i, j, k;
1183 	int ret;
1184 
1185 	ret = hns_nic_init_irq(priv);
1186 	if (ret != 0) {
1187 		netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1188 		return ret;
1189 	}
1190 
1191 	for (i = 0; i < h->q_num * 2; i++) {
1192 		ret = hns_nic_ring_open(ndev, i);
1193 		if (ret)
1194 			goto out_has_some_queues;
1195 	}
1196 
1197 	for (k = 0; k < h->q_num; k++)
1198 		h->dev->ops->toggle_queue_status(h->qs[k], 1);
1199 
1200 	ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1201 	if (ret)
1202 		goto out_set_mac_addr_err;
1203 
1204 	ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1205 	if (ret)
1206 		goto out_start_err;
1207 
1208 	if (priv->phy)
1209 		phy_start(priv->phy);
1210 
1211 	clear_bit(NIC_STATE_DOWN, &priv->state);
1212 	(void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1213 
1214 	return 0;
1215 
1216 out_start_err:
1217 	netif_stop_queue(ndev);
1218 out_set_mac_addr_err:
1219 	for (k = 0; k < h->q_num; k++)
1220 		h->dev->ops->toggle_queue_status(h->qs[k], 0);
1221 out_has_some_queues:
1222 	for (j = i - 1; j >= 0; j--)
1223 		hns_nic_ring_close(ndev, j);
1224 
1225 	set_bit(NIC_STATE_DOWN, &priv->state);
1226 
1227 	return ret;
1228 }
1229 
1230 static void hns_nic_net_down(struct net_device *ndev)
1231 {
1232 	int i;
1233 	struct hnae_ae_ops *ops;
1234 	struct hns_nic_priv *priv = netdev_priv(ndev);
1235 
1236 	if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1237 		return;
1238 
1239 	(void)del_timer_sync(&priv->service_timer);
1240 	netif_tx_stop_all_queues(ndev);
1241 	netif_carrier_off(ndev);
1242 	netif_tx_disable(ndev);
1243 	priv->link = 0;
1244 
1245 	if (priv->phy)
1246 		phy_stop(priv->phy);
1247 
1248 	ops = priv->ae_handle->dev->ops;
1249 
1250 	if (ops->stop)
1251 		ops->stop(priv->ae_handle);
1252 
1253 	netif_tx_stop_all_queues(ndev);
1254 
1255 	for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1256 		hns_nic_ring_close(ndev, i);
1257 		hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1258 
1259 		/* clean tx buffers*/
1260 		hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1261 	}
1262 }
1263 
1264 void hns_nic_net_reset(struct net_device *ndev)
1265 {
1266 	struct hns_nic_priv *priv = netdev_priv(ndev);
1267 	struct hnae_handle *handle = priv->ae_handle;
1268 
1269 	while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1270 		usleep_range(1000, 2000);
1271 
1272 	(void)hnae_reinit_handle(handle);
1273 
1274 	clear_bit(NIC_STATE_RESETTING, &priv->state);
1275 }
1276 
1277 void hns_nic_net_reinit(struct net_device *netdev)
1278 {
1279 	struct hns_nic_priv *priv = netdev_priv(netdev);
1280 
1281 	priv->netdev->trans_start = jiffies;
1282 	while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1283 		usleep_range(1000, 2000);
1284 
1285 	hns_nic_net_down(netdev);
1286 	hns_nic_net_reset(netdev);
1287 	(void)hns_nic_net_up(netdev);
1288 	clear_bit(NIC_STATE_REINITING, &priv->state);
1289 }
1290 
1291 static int hns_nic_net_open(struct net_device *ndev)
1292 {
1293 	struct hns_nic_priv *priv = netdev_priv(ndev);
1294 	struct hnae_handle *h = priv->ae_handle;
1295 	int ret;
1296 
1297 	if (test_bit(NIC_STATE_TESTING, &priv->state))
1298 		return -EBUSY;
1299 
1300 	priv->link = 0;
1301 	netif_carrier_off(ndev);
1302 
1303 	ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1304 	if (ret < 0) {
1305 		netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1306 			   ret);
1307 		return ret;
1308 	}
1309 
1310 	ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1311 	if (ret < 0) {
1312 		netdev_err(ndev,
1313 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1314 		return ret;
1315 	}
1316 
1317 	ret = hns_nic_net_up(ndev);
1318 	if (ret) {
1319 		netdev_err(ndev,
1320 			   "hns net up fail, ret=%d!\n", ret);
1321 		return ret;
1322 	}
1323 
1324 	return 0;
1325 }
1326 
1327 static int hns_nic_net_stop(struct net_device *ndev)
1328 {
1329 	hns_nic_net_down(ndev);
1330 
1331 	return 0;
1332 }
1333 
1334 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1335 static void hns_nic_net_timeout(struct net_device *ndev)
1336 {
1337 	struct hns_nic_priv *priv = netdev_priv(ndev);
1338 
1339 	hns_tx_timeout_reset(priv);
1340 }
1341 
1342 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1343 			    int cmd)
1344 {
1345 	struct hns_nic_priv *priv = netdev_priv(netdev);
1346 	struct phy_device *phy_dev = priv->phy;
1347 
1348 	if (!netif_running(netdev))
1349 		return -EINVAL;
1350 
1351 	if (!phy_dev)
1352 		return -ENOTSUPP;
1353 
1354 	return phy_mii_ioctl(phy_dev, ifr, cmd);
1355 }
1356 
1357 /* use only for netconsole to poll with the device without interrupt */
1358 #ifdef CONFIG_NET_POLL_CONTROLLER
1359 void hns_nic_poll_controller(struct net_device *ndev)
1360 {
1361 	struct hns_nic_priv *priv = netdev_priv(ndev);
1362 	unsigned long flags;
1363 	int i;
1364 
1365 	local_irq_save(flags);
1366 	for (i = 0; i < priv->ae_handle->q_num * 2; i++)
1367 		napi_schedule(&priv->ring_data[i].napi);
1368 	local_irq_restore(flags);
1369 }
1370 #endif
1371 
1372 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1373 				    struct net_device *ndev)
1374 {
1375 	struct hns_nic_priv *priv = netdev_priv(ndev);
1376 	int ret;
1377 
1378 	assert(skb->queue_mapping < ndev->ae_handle->q_num);
1379 	ret = hns_nic_net_xmit_hw(ndev, skb,
1380 				  &tx_ring_data(priv, skb->queue_mapping));
1381 	if (ret == NETDEV_TX_OK) {
1382 		ndev->trans_start = jiffies;
1383 		ndev->stats.tx_bytes += skb->len;
1384 		ndev->stats.tx_packets++;
1385 	}
1386 	return (netdev_tx_t)ret;
1387 }
1388 
1389 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1390 {
1391 	struct hns_nic_priv *priv = netdev_priv(ndev);
1392 	struct hnae_handle *h = priv->ae_handle;
1393 	int ret;
1394 
1395 	/* MTU < 68 is an error and causes problems on some kernels */
1396 	if (new_mtu < 68)
1397 		return -EINVAL;
1398 
1399 	if (!h->dev->ops->set_mtu)
1400 		return -ENOTSUPP;
1401 
1402 	if (netif_running(ndev)) {
1403 		(void)hns_nic_net_stop(ndev);
1404 		msleep(100);
1405 
1406 		ret = h->dev->ops->set_mtu(h, new_mtu);
1407 		if (ret)
1408 			netdev_err(ndev, "set mtu fail, return value %d\n",
1409 				   ret);
1410 
1411 		if (hns_nic_net_open(ndev))
1412 			netdev_err(ndev, "hns net open fail\n");
1413 	} else {
1414 		ret = h->dev->ops->set_mtu(h, new_mtu);
1415 	}
1416 
1417 	if (!ret)
1418 		ndev->mtu = new_mtu;
1419 
1420 	return ret;
1421 }
1422 
1423 static int hns_nic_set_features(struct net_device *netdev,
1424 				netdev_features_t features)
1425 {
1426 	struct hns_nic_priv *priv = netdev_priv(netdev);
1427 	struct hnae_handle *h = priv->ae_handle;
1428 
1429 	switch (priv->enet_ver) {
1430 	case AE_VERSION_1:
1431 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1432 			netdev_info(netdev, "enet v1 do not support tso!\n");
1433 		break;
1434 	default:
1435 		if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1436 			priv->ops.fill_desc = fill_tso_desc;
1437 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1438 			/* The chip only support 7*4096 */
1439 			netif_set_gso_max_size(netdev, 7 * 4096);
1440 			h->dev->ops->set_tso_stats(h, 1);
1441 		} else {
1442 			priv->ops.fill_desc = fill_v2_desc;
1443 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1444 			h->dev->ops->set_tso_stats(h, 0);
1445 		}
1446 		break;
1447 	}
1448 	netdev->features = features;
1449 	return 0;
1450 }
1451 
1452 static netdev_features_t hns_nic_fix_features(
1453 		struct net_device *netdev, netdev_features_t features)
1454 {
1455 	struct hns_nic_priv *priv = netdev_priv(netdev);
1456 
1457 	switch (priv->enet_ver) {
1458 	case AE_VERSION_1:
1459 		features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1460 				NETIF_F_HW_VLAN_CTAG_FILTER);
1461 		break;
1462 	default:
1463 		break;
1464 	}
1465 	return features;
1466 }
1467 
1468 /**
1469  * nic_set_multicast_list - set mutl mac address
1470  * @netdev: net device
1471  * @p: mac address
1472  *
1473  * return void
1474  */
1475 void hns_set_multicast_list(struct net_device *ndev)
1476 {
1477 	struct hns_nic_priv *priv = netdev_priv(ndev);
1478 	struct hnae_handle *h = priv->ae_handle;
1479 	struct netdev_hw_addr *ha = NULL;
1480 
1481 	if (!h)	{
1482 		netdev_err(ndev, "hnae handle is null\n");
1483 		return;
1484 	}
1485 
1486 	if (h->dev->ops->set_mc_addr) {
1487 		netdev_for_each_mc_addr(ha, ndev)
1488 			if (h->dev->ops->set_mc_addr(h, ha->addr))
1489 				netdev_err(ndev, "set multicast fail\n");
1490 	}
1491 }
1492 
1493 void hns_nic_set_rx_mode(struct net_device *ndev)
1494 {
1495 	struct hns_nic_priv *priv = netdev_priv(ndev);
1496 	struct hnae_handle *h = priv->ae_handle;
1497 
1498 	if (h->dev->ops->set_promisc_mode) {
1499 		if (ndev->flags & IFF_PROMISC)
1500 			h->dev->ops->set_promisc_mode(h, 1);
1501 		else
1502 			h->dev->ops->set_promisc_mode(h, 0);
1503 	}
1504 
1505 	hns_set_multicast_list(ndev);
1506 }
1507 
1508 struct rtnl_link_stats64 *hns_nic_get_stats64(struct net_device *ndev,
1509 					      struct rtnl_link_stats64 *stats)
1510 {
1511 	int idx = 0;
1512 	u64 tx_bytes = 0;
1513 	u64 rx_bytes = 0;
1514 	u64 tx_pkts = 0;
1515 	u64 rx_pkts = 0;
1516 	struct hns_nic_priv *priv = netdev_priv(ndev);
1517 	struct hnae_handle *h = priv->ae_handle;
1518 
1519 	for (idx = 0; idx < h->q_num; idx++) {
1520 		tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1521 		tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1522 		rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1523 		rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1524 	}
1525 
1526 	stats->tx_bytes = tx_bytes;
1527 	stats->tx_packets = tx_pkts;
1528 	stats->rx_bytes = rx_bytes;
1529 	stats->rx_packets = rx_pkts;
1530 
1531 	stats->rx_errors = ndev->stats.rx_errors;
1532 	stats->multicast = ndev->stats.multicast;
1533 	stats->rx_length_errors = ndev->stats.rx_length_errors;
1534 	stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1535 	stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1536 
1537 	stats->tx_errors = ndev->stats.tx_errors;
1538 	stats->rx_dropped = ndev->stats.rx_dropped;
1539 	stats->tx_dropped = ndev->stats.tx_dropped;
1540 	stats->collisions = ndev->stats.collisions;
1541 	stats->rx_over_errors = ndev->stats.rx_over_errors;
1542 	stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1543 	stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1544 	stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1545 	stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1546 	stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1547 	stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1548 	stats->tx_window_errors = ndev->stats.tx_window_errors;
1549 	stats->rx_compressed = ndev->stats.rx_compressed;
1550 	stats->tx_compressed = ndev->stats.tx_compressed;
1551 
1552 	return stats;
1553 }
1554 
1555 static const struct net_device_ops hns_nic_netdev_ops = {
1556 	.ndo_open = hns_nic_net_open,
1557 	.ndo_stop = hns_nic_net_stop,
1558 	.ndo_start_xmit = hns_nic_net_xmit,
1559 	.ndo_tx_timeout = hns_nic_net_timeout,
1560 	.ndo_set_mac_address = hns_nic_net_set_mac_address,
1561 	.ndo_change_mtu = hns_nic_change_mtu,
1562 	.ndo_do_ioctl = hns_nic_do_ioctl,
1563 	.ndo_set_features = hns_nic_set_features,
1564 	.ndo_fix_features = hns_nic_fix_features,
1565 	.ndo_get_stats64 = hns_nic_get_stats64,
1566 #ifdef CONFIG_NET_POLL_CONTROLLER
1567 	.ndo_poll_controller = hns_nic_poll_controller,
1568 #endif
1569 	.ndo_set_rx_mode = hns_nic_set_rx_mode,
1570 };
1571 
1572 static void hns_nic_update_link_status(struct net_device *netdev)
1573 {
1574 	struct hns_nic_priv *priv = netdev_priv(netdev);
1575 
1576 	struct hnae_handle *h = priv->ae_handle;
1577 	int state = 1;
1578 
1579 	if (priv->phy) {
1580 		if (!genphy_update_link(priv->phy))
1581 			state = priv->phy->link;
1582 		else
1583 			state = 0;
1584 	}
1585 	state = state && h->dev->ops->get_status(h);
1586 
1587 	if (state != priv->link) {
1588 		if (state) {
1589 			netif_carrier_on(netdev);
1590 			netif_tx_wake_all_queues(netdev);
1591 			netdev_info(netdev, "link up\n");
1592 		} else {
1593 			netif_carrier_off(netdev);
1594 			netdev_info(netdev, "link down\n");
1595 		}
1596 		priv->link = state;
1597 	}
1598 }
1599 
1600 /* for dumping key regs*/
1601 static void hns_nic_dump(struct hns_nic_priv *priv)
1602 {
1603 	struct hnae_handle *h = priv->ae_handle;
1604 	struct hnae_ae_ops *ops = h->dev->ops;
1605 	u32 *data, reg_num, i;
1606 
1607 	if (ops->get_regs_len && ops->get_regs) {
1608 		reg_num = ops->get_regs_len(priv->ae_handle);
1609 		reg_num = (reg_num + 3ul) & ~3ul;
1610 		data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1611 		if (data) {
1612 			ops->get_regs(priv->ae_handle, data);
1613 			for (i = 0; i < reg_num; i += 4)
1614 				pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1615 					i, data[i], data[i + 1],
1616 					data[i + 2], data[i + 3]);
1617 			kfree(data);
1618 		}
1619 	}
1620 
1621 	for (i = 0; i < h->q_num; i++) {
1622 		pr_info("tx_queue%d_next_to_clean:%d\n",
1623 			i, h->qs[i]->tx_ring.next_to_clean);
1624 		pr_info("tx_queue%d_next_to_use:%d\n",
1625 			i, h->qs[i]->tx_ring.next_to_use);
1626 		pr_info("rx_queue%d_next_to_clean:%d\n",
1627 			i, h->qs[i]->rx_ring.next_to_clean);
1628 		pr_info("rx_queue%d_next_to_use:%d\n",
1629 			i, h->qs[i]->rx_ring.next_to_use);
1630 	}
1631 }
1632 
1633 /* for resetting suntask*/
1634 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
1635 {
1636 	enum hnae_port_type type = priv->ae_handle->port_type;
1637 
1638 	if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
1639 		return;
1640 	clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1641 
1642 	/* If we're already down, removing or resetting, just bail */
1643 	if (test_bit(NIC_STATE_DOWN, &priv->state) ||
1644 	    test_bit(NIC_STATE_REMOVING, &priv->state) ||
1645 	    test_bit(NIC_STATE_RESETTING, &priv->state))
1646 		return;
1647 
1648 	hns_nic_dump(priv);
1649 	netdev_info(priv->netdev, "try to reset %s port!\n",
1650 		    (type == HNAE_PORT_DEBUG ? "debug" : "service"));
1651 
1652 	rtnl_lock();
1653 	/* put off any impending NetWatchDogTimeout */
1654 	priv->netdev->trans_start = jiffies;
1655 
1656 	if (type == HNAE_PORT_DEBUG) {
1657 		hns_nic_net_reinit(priv->netdev);
1658 	} else {
1659 		netif_carrier_off(priv->netdev);
1660 		netif_tx_disable(priv->netdev);
1661 	}
1662 	rtnl_unlock();
1663 }
1664 
1665 /* for doing service complete*/
1666 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
1667 {
1668 	WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
1669 
1670 	smp_mb__before_atomic();
1671 	clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
1672 }
1673 
1674 static void hns_nic_service_task(struct work_struct *work)
1675 {
1676 	struct hns_nic_priv *priv
1677 		= container_of(work, struct hns_nic_priv, service_task);
1678 	struct hnae_handle *h = priv->ae_handle;
1679 
1680 	hns_nic_update_link_status(priv->netdev);
1681 	h->dev->ops->update_led_status(h);
1682 	hns_nic_update_stats(priv->netdev);
1683 
1684 	hns_nic_reset_subtask(priv);
1685 	hns_nic_service_event_complete(priv);
1686 }
1687 
1688 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
1689 {
1690 	if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
1691 	    !test_bit(NIC_STATE_REMOVING, &priv->state) &&
1692 	    !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
1693 		(void)schedule_work(&priv->service_task);
1694 }
1695 
1696 static void hns_nic_service_timer(unsigned long data)
1697 {
1698 	struct hns_nic_priv *priv = (struct hns_nic_priv *)data;
1699 
1700 	(void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1701 
1702 	hns_nic_task_schedule(priv);
1703 }
1704 
1705 /**
1706  * hns_tx_timeout_reset - initiate reset due to Tx timeout
1707  * @priv: driver private struct
1708  **/
1709 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
1710 {
1711 	/* Do the reset outside of interrupt context */
1712 	if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
1713 		set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1714 		netdev_warn(priv->netdev,
1715 			    "initiating reset due to tx timeout(%llu,0x%lx)\n",
1716 			    priv->tx_timeout_count, priv->state);
1717 		priv->tx_timeout_count++;
1718 		hns_nic_task_schedule(priv);
1719 	}
1720 }
1721 
1722 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
1723 {
1724 	struct hnae_handle *h = priv->ae_handle;
1725 	struct hns_nic_ring_data *rd;
1726 	int i;
1727 
1728 	if (h->q_num > NIC_MAX_Q_PER_VF) {
1729 		netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
1730 		return -EINVAL;
1731 	}
1732 
1733 	priv->ring_data = kzalloc(h->q_num * sizeof(*priv->ring_data) * 2,
1734 				  GFP_KERNEL);
1735 	if (!priv->ring_data)
1736 		return -ENOMEM;
1737 
1738 	for (i = 0; i < h->q_num; i++) {
1739 		rd = &priv->ring_data[i];
1740 		rd->queue_index = i;
1741 		rd->ring = &h->qs[i]->tx_ring;
1742 		rd->poll_one = hns_nic_tx_poll_one;
1743 		rd->fini_process = hns_nic_tx_fini_pro;
1744 
1745 		netif_napi_add(priv->netdev, &rd->napi,
1746 			       hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
1747 		rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1748 	}
1749 	for (i = h->q_num; i < h->q_num * 2; i++) {
1750 		rd = &priv->ring_data[i];
1751 		rd->queue_index = i - h->q_num;
1752 		rd->ring = &h->qs[i - h->q_num]->rx_ring;
1753 		rd->poll_one = hns_nic_rx_poll_one;
1754 		rd->ex_process = hns_nic_rx_up_pro;
1755 		rd->fini_process = hns_nic_rx_fini_pro;
1756 
1757 		netif_napi_add(priv->netdev, &rd->napi,
1758 			       hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
1759 		rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1760 	}
1761 
1762 	return 0;
1763 }
1764 
1765 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
1766 {
1767 	struct hnae_handle *h = priv->ae_handle;
1768 	int i;
1769 
1770 	for (i = 0; i < h->q_num * 2; i++) {
1771 		netif_napi_del(&priv->ring_data[i].napi);
1772 		if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
1773 			(void)irq_set_affinity_hint(
1774 				priv->ring_data[i].ring->irq,
1775 				NULL);
1776 			free_irq(priv->ring_data[i].ring->irq,
1777 				 &priv->ring_data[i]);
1778 		}
1779 
1780 		priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1781 	}
1782 	kfree(priv->ring_data);
1783 }
1784 
1785 static void hns_nic_set_priv_ops(struct net_device *netdev)
1786 {
1787 	struct hns_nic_priv *priv = netdev_priv(netdev);
1788 	struct hnae_handle *h = priv->ae_handle;
1789 
1790 	if (AE_IS_VER1(priv->enet_ver)) {
1791 		priv->ops.fill_desc = fill_desc;
1792 		priv->ops.get_rxd_bnum = get_rx_desc_bnum;
1793 		priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1794 	} else {
1795 		priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
1796 		if ((netdev->features & NETIF_F_TSO) ||
1797 		    (netdev->features & NETIF_F_TSO6)) {
1798 			priv->ops.fill_desc = fill_tso_desc;
1799 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1800 			/* This chip only support 7*4096 */
1801 			netif_set_gso_max_size(netdev, 7 * 4096);
1802 			h->dev->ops->set_tso_stats(h, 1);
1803 		} else {
1804 			priv->ops.fill_desc = fill_v2_desc;
1805 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1806 		}
1807 	}
1808 }
1809 
1810 static int hns_nic_try_get_ae(struct net_device *ndev)
1811 {
1812 	struct hns_nic_priv *priv = netdev_priv(ndev);
1813 	struct hnae_handle *h;
1814 	int ret;
1815 
1816 	h = hnae_get_handle(&priv->netdev->dev,
1817 			    priv->ae_node, priv->port_id, NULL);
1818 	if (IS_ERR_OR_NULL(h)) {
1819 		ret = PTR_ERR(h);
1820 		dev_dbg(priv->dev, "has not handle, register notifier!\n");
1821 		goto out;
1822 	}
1823 	priv->ae_handle = h;
1824 
1825 	ret = hns_nic_init_phy(ndev, h);
1826 	if (ret) {
1827 		dev_err(priv->dev, "probe phy device fail!\n");
1828 		goto out_init_phy;
1829 	}
1830 
1831 	ret = hns_nic_init_ring_data(priv);
1832 	if (ret) {
1833 		ret = -ENOMEM;
1834 		goto out_init_ring_data;
1835 	}
1836 
1837 	hns_nic_set_priv_ops(ndev);
1838 
1839 	ret = register_netdev(ndev);
1840 	if (ret) {
1841 		dev_err(priv->dev, "probe register netdev fail!\n");
1842 		goto out_reg_ndev_fail;
1843 	}
1844 	return 0;
1845 
1846 out_reg_ndev_fail:
1847 	hns_nic_uninit_ring_data(priv);
1848 	priv->ring_data = NULL;
1849 out_init_phy:
1850 out_init_ring_data:
1851 	hnae_put_handle(priv->ae_handle);
1852 	priv->ae_handle = NULL;
1853 out:
1854 	return ret;
1855 }
1856 
1857 static int hns_nic_notifier_action(struct notifier_block *nb,
1858 				   unsigned long action, void *data)
1859 {
1860 	struct hns_nic_priv *priv =
1861 		container_of(nb, struct hns_nic_priv, notifier_block);
1862 
1863 	assert(action == HNAE_AE_REGISTER);
1864 
1865 	if (!hns_nic_try_get_ae(priv->netdev)) {
1866 		hnae_unregister_notifier(&priv->notifier_block);
1867 		priv->notifier_block.notifier_call = NULL;
1868 	}
1869 	return 0;
1870 }
1871 
1872 static int hns_nic_dev_probe(struct platform_device *pdev)
1873 {
1874 	struct device *dev = &pdev->dev;
1875 	struct net_device *ndev;
1876 	struct hns_nic_priv *priv;
1877 	struct device_node *node = dev->of_node;
1878 	int ret;
1879 
1880 	ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
1881 	if (!ndev)
1882 		return -ENOMEM;
1883 
1884 	platform_set_drvdata(pdev, ndev);
1885 
1886 	priv = netdev_priv(ndev);
1887 	priv->dev = dev;
1888 	priv->netdev = ndev;
1889 
1890 	if (of_device_is_compatible(node, "hisilicon,hns-nic-v1"))
1891 		priv->enet_ver = AE_VERSION_1;
1892 	else
1893 		priv->enet_ver = AE_VERSION_2;
1894 
1895 	priv->ae_node = (void *)of_parse_phandle(node, "ae-handle", 0);
1896 	if (IS_ERR_OR_NULL(priv->ae_node)) {
1897 		ret = PTR_ERR(priv->ae_node);
1898 		dev_err(dev, "not find ae-handle\n");
1899 		goto out_read_prop_fail;
1900 	}
1901 
1902 	ret = of_property_read_u32(node, "port-id", &priv->port_id);
1903 	if (ret)
1904 		goto out_read_prop_fail;
1905 
1906 	hns_init_mac_addr(ndev);
1907 
1908 	ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
1909 	ndev->priv_flags |= IFF_UNICAST_FLT;
1910 	ndev->netdev_ops = &hns_nic_netdev_ops;
1911 	hns_ethtool_set_ops(ndev);
1912 
1913 	ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1914 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1915 		NETIF_F_GRO;
1916 	ndev->vlan_features |=
1917 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
1918 	ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
1919 
1920 	switch (priv->enet_ver) {
1921 	case AE_VERSION_2:
1922 		ndev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1923 		ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1924 			NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
1925 			NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
1926 		break;
1927 	default:
1928 		break;
1929 	}
1930 
1931 	SET_NETDEV_DEV(ndev, dev);
1932 
1933 	if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
1934 		dev_dbg(dev, "set mask to 64bit\n");
1935 	else
1936 		dev_err(dev, "set mask to 32bit fail!\n");
1937 
1938 	/* carrier off reporting is important to ethtool even BEFORE open */
1939 	netif_carrier_off(ndev);
1940 
1941 	setup_timer(&priv->service_timer, hns_nic_service_timer,
1942 		    (unsigned long)priv);
1943 	INIT_WORK(&priv->service_task, hns_nic_service_task);
1944 
1945 	set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
1946 	clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
1947 	set_bit(NIC_STATE_DOWN, &priv->state);
1948 
1949 	if (hns_nic_try_get_ae(priv->netdev)) {
1950 		priv->notifier_block.notifier_call = hns_nic_notifier_action;
1951 		ret = hnae_register_notifier(&priv->notifier_block);
1952 		if (ret) {
1953 			dev_err(dev, "register notifier fail!\n");
1954 			goto out_notify_fail;
1955 		}
1956 		dev_dbg(dev, "has not handle, register notifier!\n");
1957 	}
1958 
1959 	return 0;
1960 
1961 out_notify_fail:
1962 	(void)cancel_work_sync(&priv->service_task);
1963 out_read_prop_fail:
1964 	free_netdev(ndev);
1965 	return ret;
1966 }
1967 
1968 static int hns_nic_dev_remove(struct platform_device *pdev)
1969 {
1970 	struct net_device *ndev = platform_get_drvdata(pdev);
1971 	struct hns_nic_priv *priv = netdev_priv(ndev);
1972 
1973 	if (ndev->reg_state != NETREG_UNINITIALIZED)
1974 		unregister_netdev(ndev);
1975 
1976 	if (priv->ring_data)
1977 		hns_nic_uninit_ring_data(priv);
1978 	priv->ring_data = NULL;
1979 
1980 	if (priv->phy)
1981 		phy_disconnect(priv->phy);
1982 	priv->phy = NULL;
1983 
1984 	if (!IS_ERR_OR_NULL(priv->ae_handle))
1985 		hnae_put_handle(priv->ae_handle);
1986 	priv->ae_handle = NULL;
1987 	if (priv->notifier_block.notifier_call)
1988 		hnae_unregister_notifier(&priv->notifier_block);
1989 	priv->notifier_block.notifier_call = NULL;
1990 
1991 	set_bit(NIC_STATE_REMOVING, &priv->state);
1992 	(void)cancel_work_sync(&priv->service_task);
1993 
1994 	free_netdev(ndev);
1995 	return 0;
1996 }
1997 
1998 static const struct of_device_id hns_enet_of_match[] = {
1999 	{.compatible = "hisilicon,hns-nic-v1",},
2000 	{.compatible = "hisilicon,hns-nic-v2",},
2001 	{},
2002 };
2003 
2004 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2005 
2006 static struct platform_driver hns_nic_dev_driver = {
2007 	.driver = {
2008 		.name = "hns-nic",
2009 		.of_match_table = hns_enet_of_match,
2010 	},
2011 	.probe = hns_nic_dev_probe,
2012 	.remove = hns_nic_dev_remove,
2013 };
2014 
2015 module_platform_driver(hns_nic_dev_driver);
2016 
2017 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2018 MODULE_AUTHOR("Hisilicon, Inc.");
2019 MODULE_LICENSE("GPL");
2020 MODULE_ALIAS("platform:hns-nic");
2021