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 #include "hns_dsaf_mac.h"
26 
27 #define NIC_MAX_Q_PER_VF 16
28 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
29 
30 #define SERVICE_TIMER_HZ (1 * HZ)
31 
32 #define NIC_TX_CLEAN_MAX_NUM 256
33 #define NIC_RX_CLEAN_MAX_NUM 64
34 
35 #define RCB_IRQ_NOT_INITED 0
36 #define RCB_IRQ_INITED 1
37 #define HNS_BUFFER_SIZE_2048 2048
38 
39 #define BD_MAX_SEND_SIZE 8191
40 #define SKB_TMP_LEN(SKB) \
41 	(((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
42 
43 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
44 			 int size, dma_addr_t dma, int frag_end,
45 			 int buf_num, enum hns_desc_type type, int mtu)
46 {
47 	struct hnae_desc *desc = &ring->desc[ring->next_to_use];
48 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
49 	struct iphdr *iphdr;
50 	struct ipv6hdr *ipv6hdr;
51 	struct sk_buff *skb;
52 	__be16 protocol;
53 	u8 bn_pid = 0;
54 	u8 rrcfv = 0;
55 	u8 ip_offset = 0;
56 	u8 tvsvsn = 0;
57 	u16 mss = 0;
58 	u8 l4_len = 0;
59 	u16 paylen = 0;
60 
61 	desc_cb->priv = priv;
62 	desc_cb->length = size;
63 	desc_cb->dma = dma;
64 	desc_cb->type = type;
65 
66 	desc->addr = cpu_to_le64(dma);
67 	desc->tx.send_size = cpu_to_le16((u16)size);
68 
69 	/* config bd buffer end */
70 	hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
71 	hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
72 
73 	/* fill port_id in the tx bd for sending management pkts */
74 	hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
75 		       HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
76 
77 	if (type == DESC_TYPE_SKB) {
78 		skb = (struct sk_buff *)priv;
79 
80 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
81 			skb_reset_mac_len(skb);
82 			protocol = skb->protocol;
83 			ip_offset = ETH_HLEN;
84 
85 			if (protocol == htons(ETH_P_8021Q)) {
86 				ip_offset += VLAN_HLEN;
87 				protocol = vlan_get_protocol(skb);
88 				skb->protocol = protocol;
89 			}
90 
91 			if (skb->protocol == htons(ETH_P_IP)) {
92 				iphdr = ip_hdr(skb);
93 				hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
94 				hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
95 
96 				/* check for tcp/udp header */
97 				if (iphdr->protocol == IPPROTO_TCP &&
98 				    skb_is_gso(skb)) {
99 					hnae_set_bit(tvsvsn,
100 						     HNSV2_TXD_TSE_B, 1);
101 					l4_len = tcp_hdrlen(skb);
102 					mss = skb_shinfo(skb)->gso_size;
103 					paylen = skb->len - SKB_TMP_LEN(skb);
104 				}
105 			} else if (skb->protocol == htons(ETH_P_IPV6)) {
106 				hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
107 				ipv6hdr = ipv6_hdr(skb);
108 				hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
109 
110 				/* check for tcp/udp header */
111 				if (ipv6hdr->nexthdr == IPPROTO_TCP &&
112 				    skb_is_gso(skb) && skb_is_gso_v6(skb)) {
113 					hnae_set_bit(tvsvsn,
114 						     HNSV2_TXD_TSE_B, 1);
115 					l4_len = tcp_hdrlen(skb);
116 					mss = skb_shinfo(skb)->gso_size;
117 					paylen = skb->len - SKB_TMP_LEN(skb);
118 				}
119 			}
120 			desc->tx.ip_offset = ip_offset;
121 			desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
122 			desc->tx.mss = cpu_to_le16(mss);
123 			desc->tx.l4_len = l4_len;
124 			desc->tx.paylen = cpu_to_le16(paylen);
125 		}
126 	}
127 
128 	hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
129 
130 	desc->tx.bn_pid = bn_pid;
131 	desc->tx.ra_ri_cs_fe_vld = rrcfv;
132 
133 	ring_ptr_move_fw(ring, next_to_use);
134 }
135 
136 static const struct acpi_device_id hns_enet_acpi_match[] = {
137 	{ "HISI00C1", 0 },
138 	{ "HISI00C2", 0 },
139 	{ },
140 };
141 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
142 
143 static void fill_desc(struct hnae_ring *ring, void *priv,
144 		      int size, dma_addr_t dma, int frag_end,
145 		      int buf_num, enum hns_desc_type type, int mtu)
146 {
147 	struct hnae_desc *desc = &ring->desc[ring->next_to_use];
148 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
149 	struct sk_buff *skb;
150 	__be16 protocol;
151 	u32 ip_offset;
152 	u32 asid_bufnum_pid = 0;
153 	u32 flag_ipoffset = 0;
154 
155 	desc_cb->priv = priv;
156 	desc_cb->length = size;
157 	desc_cb->dma = dma;
158 	desc_cb->type = type;
159 
160 	desc->addr = cpu_to_le64(dma);
161 	desc->tx.send_size = cpu_to_le16((u16)size);
162 
163 	/*config bd buffer end */
164 	flag_ipoffset |= 1 << HNS_TXD_VLD_B;
165 
166 	asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
167 
168 	if (type == DESC_TYPE_SKB) {
169 		skb = (struct sk_buff *)priv;
170 
171 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
172 			protocol = skb->protocol;
173 			ip_offset = ETH_HLEN;
174 
175 			/*if it is a SW VLAN check the next protocol*/
176 			if (protocol == htons(ETH_P_8021Q)) {
177 				ip_offset += VLAN_HLEN;
178 				protocol = vlan_get_protocol(skb);
179 				skb->protocol = protocol;
180 			}
181 
182 			if (skb->protocol == htons(ETH_P_IP)) {
183 				flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
184 				/* check for tcp/udp header */
185 				flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
186 
187 			} else if (skb->protocol == htons(ETH_P_IPV6)) {
188 				/* ipv6 has not l3 cs, check for L4 header */
189 				flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
190 			}
191 
192 			flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
193 		}
194 	}
195 
196 	flag_ipoffset |= frag_end << HNS_TXD_FE_B;
197 
198 	desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
199 	desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
200 
201 	ring_ptr_move_fw(ring, next_to_use);
202 }
203 
204 static void unfill_desc(struct hnae_ring *ring)
205 {
206 	ring_ptr_move_bw(ring, next_to_use);
207 }
208 
209 static int hns_nic_maybe_stop_tx(
210 	struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
211 {
212 	struct sk_buff *skb = *out_skb;
213 	struct sk_buff *new_skb = NULL;
214 	int buf_num;
215 
216 	/* no. of segments (plus a header) */
217 	buf_num = skb_shinfo(skb)->nr_frags + 1;
218 
219 	if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
220 		if (ring_space(ring) < 1)
221 			return -EBUSY;
222 
223 		new_skb = skb_copy(skb, GFP_ATOMIC);
224 		if (!new_skb)
225 			return -ENOMEM;
226 
227 		dev_kfree_skb_any(skb);
228 		*out_skb = new_skb;
229 		buf_num = 1;
230 	} else if (buf_num > ring_space(ring)) {
231 		return -EBUSY;
232 	}
233 
234 	*bnum = buf_num;
235 	return 0;
236 }
237 
238 static int hns_nic_maybe_stop_tso(
239 	struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
240 {
241 	int i;
242 	int size;
243 	int buf_num;
244 	int frag_num;
245 	struct sk_buff *skb = *out_skb;
246 	struct sk_buff *new_skb = NULL;
247 	struct skb_frag_struct *frag;
248 
249 	size = skb_headlen(skb);
250 	buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
251 
252 	frag_num = skb_shinfo(skb)->nr_frags;
253 	for (i = 0; i < frag_num; i++) {
254 		frag = &skb_shinfo(skb)->frags[i];
255 		size = skb_frag_size(frag);
256 		buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
257 	}
258 
259 	if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
260 		buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
261 		if (ring_space(ring) < buf_num)
262 			return -EBUSY;
263 		/* manual split the send packet */
264 		new_skb = skb_copy(skb, GFP_ATOMIC);
265 		if (!new_skb)
266 			return -ENOMEM;
267 		dev_kfree_skb_any(skb);
268 		*out_skb = new_skb;
269 
270 	} else if (ring_space(ring) < buf_num) {
271 		return -EBUSY;
272 	}
273 
274 	*bnum = buf_num;
275 	return 0;
276 }
277 
278 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
279 			  int size, dma_addr_t dma, int frag_end,
280 			  int buf_num, enum hns_desc_type type, int mtu)
281 {
282 	int frag_buf_num;
283 	int sizeoflast;
284 	int k;
285 
286 	frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
287 	sizeoflast = size % BD_MAX_SEND_SIZE;
288 	sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
289 
290 	/* when the frag size is bigger than hardware, split this frag */
291 	for (k = 0; k < frag_buf_num; k++)
292 		fill_v2_desc(ring, priv,
293 			     (k == frag_buf_num - 1) ?
294 					sizeoflast : BD_MAX_SEND_SIZE,
295 			     dma + BD_MAX_SEND_SIZE * k,
296 			     frag_end && (k == frag_buf_num - 1) ? 1 : 0,
297 			     buf_num,
298 			     (type == DESC_TYPE_SKB && !k) ?
299 					DESC_TYPE_SKB : DESC_TYPE_PAGE,
300 			     mtu);
301 }
302 
303 int hns_nic_net_xmit_hw(struct net_device *ndev,
304 			struct sk_buff *skb,
305 			struct hns_nic_ring_data *ring_data)
306 {
307 	struct hns_nic_priv *priv = netdev_priv(ndev);
308 	struct device *dev = priv->dev;
309 	struct hnae_ring *ring = ring_data->ring;
310 	struct netdev_queue *dev_queue;
311 	struct skb_frag_struct *frag;
312 	int buf_num;
313 	int seg_num;
314 	dma_addr_t dma;
315 	int size, next_to_use;
316 	int i;
317 
318 	switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
319 	case -EBUSY:
320 		ring->stats.tx_busy++;
321 		goto out_net_tx_busy;
322 	case -ENOMEM:
323 		ring->stats.sw_err_cnt++;
324 		netdev_err(ndev, "no memory to xmit!\n");
325 		goto out_err_tx_ok;
326 	default:
327 		break;
328 	}
329 
330 	/* no. of segments (plus a header) */
331 	seg_num = skb_shinfo(skb)->nr_frags + 1;
332 	next_to_use = ring->next_to_use;
333 
334 	/* fill the first part */
335 	size = skb_headlen(skb);
336 	dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
337 	if (dma_mapping_error(dev, dma)) {
338 		netdev_err(ndev, "TX head DMA map failed\n");
339 		ring->stats.sw_err_cnt++;
340 		goto out_err_tx_ok;
341 	}
342 	priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
343 			    buf_num, DESC_TYPE_SKB, ndev->mtu);
344 
345 	/* fill the fragments */
346 	for (i = 1; i < seg_num; i++) {
347 		frag = &skb_shinfo(skb)->frags[i - 1];
348 		size = skb_frag_size(frag);
349 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
350 		if (dma_mapping_error(dev, dma)) {
351 			netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
352 			ring->stats.sw_err_cnt++;
353 			goto out_map_frag_fail;
354 		}
355 		priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
356 				    seg_num - 1 == i ? 1 : 0, buf_num,
357 				    DESC_TYPE_PAGE, ndev->mtu);
358 	}
359 
360 	/*complete translate all packets*/
361 	dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
362 	netdev_tx_sent_queue(dev_queue, skb->len);
363 
364 	wmb(); /* commit all data before submit */
365 	assert(skb->queue_mapping < priv->ae_handle->q_num);
366 	hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
367 	ring->stats.tx_pkts++;
368 	ring->stats.tx_bytes += skb->len;
369 
370 	return NETDEV_TX_OK;
371 
372 out_map_frag_fail:
373 
374 	while (ring->next_to_use != next_to_use) {
375 		unfill_desc(ring);
376 		if (ring->next_to_use != next_to_use)
377 			dma_unmap_page(dev,
378 				       ring->desc_cb[ring->next_to_use].dma,
379 				       ring->desc_cb[ring->next_to_use].length,
380 				       DMA_TO_DEVICE);
381 		else
382 			dma_unmap_single(dev,
383 					 ring->desc_cb[next_to_use].dma,
384 					 ring->desc_cb[next_to_use].length,
385 					 DMA_TO_DEVICE);
386 	}
387 
388 out_err_tx_ok:
389 
390 	dev_kfree_skb_any(skb);
391 	return NETDEV_TX_OK;
392 
393 out_net_tx_busy:
394 
395 	netif_stop_subqueue(ndev, skb->queue_mapping);
396 
397 	/* Herbert's original patch had:
398 	 *  smp_mb__after_netif_stop_queue();
399 	 * but since that doesn't exist yet, just open code it.
400 	 */
401 	smp_mb();
402 	return NETDEV_TX_BUSY;
403 }
404 
405 /**
406  * hns_nic_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
407  * @data: pointer to the start of the headers
408  * @max: total length of section to find headers in
409  *
410  * This function is meant to determine the length of headers that will
411  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
412  * motivation of doing this is to only perform one pull for IPv4 TCP
413  * packets so that we can do basic things like calculating the gso_size
414  * based on the average data per packet.
415  **/
416 static unsigned int hns_nic_get_headlen(unsigned char *data, u32 flag,
417 					unsigned int max_size)
418 {
419 	unsigned char *network;
420 	u8 hlen;
421 
422 	/* this should never happen, but better safe than sorry */
423 	if (max_size < ETH_HLEN)
424 		return max_size;
425 
426 	/* initialize network frame pointer */
427 	network = data;
428 
429 	/* set first protocol and move network header forward */
430 	network += ETH_HLEN;
431 
432 	/* handle any vlan tag if present */
433 	if (hnae_get_field(flag, HNS_RXD_VLAN_M, HNS_RXD_VLAN_S)
434 		== HNS_RX_FLAG_VLAN_PRESENT) {
435 		if ((typeof(max_size))(network - data) > (max_size - VLAN_HLEN))
436 			return max_size;
437 
438 		network += VLAN_HLEN;
439 	}
440 
441 	/* handle L3 protocols */
442 	if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
443 		== HNS_RX_FLAG_L3ID_IPV4) {
444 		if ((typeof(max_size))(network - data) >
445 		    (max_size - sizeof(struct iphdr)))
446 			return max_size;
447 
448 		/* access ihl as a u8 to avoid unaligned access on ia64 */
449 		hlen = (network[0] & 0x0F) << 2;
450 
451 		/* verify hlen meets minimum size requirements */
452 		if (hlen < sizeof(struct iphdr))
453 			return network - data;
454 
455 		/* record next protocol if header is present */
456 	} else if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
457 		== HNS_RX_FLAG_L3ID_IPV6) {
458 		if ((typeof(max_size))(network - data) >
459 		    (max_size - sizeof(struct ipv6hdr)))
460 			return max_size;
461 
462 		/* record next protocol */
463 		hlen = sizeof(struct ipv6hdr);
464 	} else {
465 		return network - data;
466 	}
467 
468 	/* relocate pointer to start of L4 header */
469 	network += hlen;
470 
471 	/* finally sort out TCP/UDP */
472 	if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
473 		== HNS_RX_FLAG_L4ID_TCP) {
474 		if ((typeof(max_size))(network - data) >
475 		    (max_size - sizeof(struct tcphdr)))
476 			return max_size;
477 
478 		/* access doff as a u8 to avoid unaligned access on ia64 */
479 		hlen = (network[12] & 0xF0) >> 2;
480 
481 		/* verify hlen meets minimum size requirements */
482 		if (hlen < sizeof(struct tcphdr))
483 			return network - data;
484 
485 		network += hlen;
486 	} else if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
487 		== HNS_RX_FLAG_L4ID_UDP) {
488 		if ((typeof(max_size))(network - data) >
489 		    (max_size - sizeof(struct udphdr)))
490 			return max_size;
491 
492 		network += sizeof(struct udphdr);
493 	}
494 
495 	/* If everything has gone correctly network should be the
496 	 * data section of the packet and will be the end of the header.
497 	 * If not then it probably represents the end of the last recognized
498 	 * header.
499 	 */
500 	if ((typeof(max_size))(network - data) < max_size)
501 		return network - data;
502 	else
503 		return max_size;
504 }
505 
506 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
507 			       struct hnae_ring *ring, int pull_len,
508 			       struct hnae_desc_cb *desc_cb)
509 {
510 	struct hnae_desc *desc;
511 	int truesize, size;
512 	int last_offset;
513 	bool twobufs;
514 
515 	twobufs = ((PAGE_SIZE < 8192) && hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
516 
517 	desc = &ring->desc[ring->next_to_clean];
518 	size = le16_to_cpu(desc->rx.size);
519 
520 	if (twobufs) {
521 		truesize = hnae_buf_size(ring);
522 	} else {
523 		truesize = ALIGN(size, L1_CACHE_BYTES);
524 		last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
525 	}
526 
527 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
528 			size - pull_len, truesize - pull_len);
529 
530 	 /* avoid re-using remote pages,flag default unreuse */
531 	if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
532 		return;
533 
534 	if (twobufs) {
535 		/* if we are only owner of page we can reuse it */
536 		if (likely(page_count(desc_cb->priv) == 1)) {
537 			/* flip page offset to other buffer */
538 			desc_cb->page_offset ^= truesize;
539 
540 			desc_cb->reuse_flag = 1;
541 			/* bump ref count on page before it is given*/
542 			get_page(desc_cb->priv);
543 		}
544 		return;
545 	}
546 
547 	/* move offset up to the next cache line */
548 	desc_cb->page_offset += truesize;
549 
550 	if (desc_cb->page_offset <= last_offset) {
551 		desc_cb->reuse_flag = 1;
552 		/* bump ref count on page before it is given*/
553 		get_page(desc_cb->priv);
554 	}
555 }
556 
557 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
558 {
559 	*out_bnum = hnae_get_field(bnum_flag,
560 				   HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
561 }
562 
563 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
564 {
565 	*out_bnum = hnae_get_field(bnum_flag,
566 				   HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
567 }
568 
569 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
570 				struct sk_buff *skb, u32 flag)
571 {
572 	struct net_device *netdev = ring_data->napi.dev;
573 	u32 l3id;
574 	u32 l4id;
575 
576 	/* check if RX checksum offload is enabled */
577 	if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
578 		return;
579 
580 	/* In hardware, we only support checksum for the following protocols:
581 	 * 1) IPv4,
582 	 * 2) TCP(over IPv4 or IPv6),
583 	 * 3) UDP(over IPv4 or IPv6),
584 	 * 4) SCTP(over IPv4 or IPv6)
585 	 * but we support many L3(IPv4, IPv6, MPLS, PPPoE etc) and L4(TCP,
586 	 * UDP, GRE, SCTP, IGMP, ICMP etc.) protocols.
587 	 *
588 	 * Hardware limitation:
589 	 * Our present hardware RX Descriptor lacks L3/L4 checksum "Status &
590 	 * Error" bit (which usually can be used to indicate whether checksum
591 	 * was calculated by the hardware and if there was any error encountered
592 	 * during checksum calculation).
593 	 *
594 	 * Software workaround:
595 	 * We do get info within the RX descriptor about the kind of L3/L4
596 	 * protocol coming in the packet and the error status. These errors
597 	 * might not just be checksum errors but could be related to version,
598 	 * length of IPv4, UDP, TCP etc.
599 	 * Because there is no-way of knowing if it is a L3/L4 error due to bad
600 	 * checksum or any other L3/L4 error, we will not (cannot) convey
601 	 * checksum status for such cases to upper stack and will not maintain
602 	 * the RX L3/L4 checksum counters as well.
603 	 */
604 
605 	l3id = hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S);
606 	l4id = hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S);
607 
608 	/*  check L3 protocol for which checksum is supported */
609 	if ((l3id != HNS_RX_FLAG_L3ID_IPV4) && (l3id != HNS_RX_FLAG_L3ID_IPV6))
610 		return;
611 
612 	/* check for any(not just checksum)flagged L3 protocol errors */
613 	if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
614 		return;
615 
616 	/* we do not support checksum of fragmented packets */
617 	if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
618 		return;
619 
620 	/*  check L4 protocol for which checksum is supported */
621 	if ((l4id != HNS_RX_FLAG_L4ID_TCP) &&
622 	    (l4id != HNS_RX_FLAG_L4ID_UDP) &&
623 	    (l4id != HNS_RX_FLAG_L4ID_SCTP))
624 		return;
625 
626 	/* check for any(not just checksum)flagged L4 protocol errors */
627 	if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
628 		return;
629 
630 	/* now, this has to be a packet with valid RX checksum */
631 	skb->ip_summed = CHECKSUM_UNNECESSARY;
632 }
633 
634 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
635 			       struct sk_buff **out_skb, int *out_bnum)
636 {
637 	struct hnae_ring *ring = ring_data->ring;
638 	struct net_device *ndev = ring_data->napi.dev;
639 	struct hns_nic_priv *priv = netdev_priv(ndev);
640 	struct sk_buff *skb;
641 	struct hnae_desc *desc;
642 	struct hnae_desc_cb *desc_cb;
643 	unsigned char *va;
644 	int bnum, length, i;
645 	int pull_len;
646 	u32 bnum_flag;
647 
648 	desc = &ring->desc[ring->next_to_clean];
649 	desc_cb = &ring->desc_cb[ring->next_to_clean];
650 
651 	prefetch(desc);
652 
653 	va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
654 
655 	/* prefetch first cache line of first page */
656 	prefetch(va);
657 #if L1_CACHE_BYTES < 128
658 	prefetch(va + L1_CACHE_BYTES);
659 #endif
660 
661 	skb = *out_skb = napi_alloc_skb(&ring_data->napi,
662 					HNS_RX_HEAD_SIZE);
663 	if (unlikely(!skb)) {
664 		netdev_err(ndev, "alloc rx skb fail\n");
665 		ring->stats.sw_err_cnt++;
666 		return -ENOMEM;
667 	}
668 
669 	prefetchw(skb->data);
670 	length = le16_to_cpu(desc->rx.pkt_len);
671 	bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
672 	priv->ops.get_rxd_bnum(bnum_flag, &bnum);
673 	*out_bnum = bnum;
674 
675 	if (length <= HNS_RX_HEAD_SIZE) {
676 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
677 
678 		/* we can reuse buffer as-is, just make sure it is local */
679 		if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
680 			desc_cb->reuse_flag = 1;
681 		else /* this page cannot be reused so discard it */
682 			put_page(desc_cb->priv);
683 
684 		ring_ptr_move_fw(ring, next_to_clean);
685 
686 		if (unlikely(bnum != 1)) { /* check err*/
687 			*out_bnum = 1;
688 			goto out_bnum_err;
689 		}
690 	} else {
691 		ring->stats.seg_pkt_cnt++;
692 
693 		pull_len = hns_nic_get_headlen(va, bnum_flag, HNS_RX_HEAD_SIZE);
694 		memcpy(__skb_put(skb, pull_len), va,
695 		       ALIGN(pull_len, sizeof(long)));
696 
697 		hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
698 		ring_ptr_move_fw(ring, next_to_clean);
699 
700 		if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
701 			*out_bnum = 1;
702 			goto out_bnum_err;
703 		}
704 		for (i = 1; i < bnum; i++) {
705 			desc = &ring->desc[ring->next_to_clean];
706 			desc_cb = &ring->desc_cb[ring->next_to_clean];
707 
708 			hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
709 			ring_ptr_move_fw(ring, next_to_clean);
710 		}
711 	}
712 
713 	/* check except process, free skb and jump the desc */
714 	if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
715 out_bnum_err:
716 		*out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
717 		netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
718 			   bnum, ring->max_desc_num_per_pkt,
719 			   length, (int)MAX_SKB_FRAGS,
720 			   ((u64 *)desc)[0], ((u64 *)desc)[1]);
721 		ring->stats.err_bd_num++;
722 		dev_kfree_skb_any(skb);
723 		return -EDOM;
724 	}
725 
726 	bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
727 
728 	if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
729 		netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
730 			   ((u64 *)desc)[0], ((u64 *)desc)[1]);
731 		ring->stats.non_vld_descs++;
732 		dev_kfree_skb_any(skb);
733 		return -EINVAL;
734 	}
735 
736 	if (unlikely((!desc->rx.pkt_len) ||
737 		     hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
738 		ring->stats.err_pkt_len++;
739 		dev_kfree_skb_any(skb);
740 		return -EFAULT;
741 	}
742 
743 	if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
744 		ring->stats.l2_err++;
745 		dev_kfree_skb_any(skb);
746 		return -EFAULT;
747 	}
748 
749 	ring->stats.rx_pkts++;
750 	ring->stats.rx_bytes += skb->len;
751 
752 	/* indicate to upper stack if our hardware has already calculated
753 	 * the RX checksum
754 	 */
755 	hns_nic_rx_checksum(ring_data, skb, bnum_flag);
756 
757 	return 0;
758 }
759 
760 static void
761 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
762 {
763 	int i, ret;
764 	struct hnae_desc_cb res_cbs;
765 	struct hnae_desc_cb *desc_cb;
766 	struct hnae_ring *ring = ring_data->ring;
767 	struct net_device *ndev = ring_data->napi.dev;
768 
769 	for (i = 0; i < cleand_count; i++) {
770 		desc_cb = &ring->desc_cb[ring->next_to_use];
771 		if (desc_cb->reuse_flag) {
772 			ring->stats.reuse_pg_cnt++;
773 			hnae_reuse_buffer(ring, ring->next_to_use);
774 		} else {
775 			ret = hnae_reserve_buffer_map(ring, &res_cbs);
776 			if (ret) {
777 				ring->stats.sw_err_cnt++;
778 				netdev_err(ndev, "hnae reserve buffer map failed.\n");
779 				break;
780 			}
781 			hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
782 		}
783 
784 		ring_ptr_move_fw(ring, next_to_use);
785 	}
786 
787 	wmb(); /* make all data has been write before submit */
788 	writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
789 }
790 
791 /* return error number for error or number of desc left to take
792  */
793 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
794 			      struct sk_buff *skb)
795 {
796 	struct net_device *ndev = ring_data->napi.dev;
797 
798 	skb->protocol = eth_type_trans(skb, ndev);
799 	(void)napi_gro_receive(&ring_data->napi, skb);
800 	ndev->last_rx = jiffies;
801 }
802 
803 static int hns_desc_unused(struct hnae_ring *ring)
804 {
805 	int ntc = ring->next_to_clean;
806 	int ntu = ring->next_to_use;
807 
808 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
809 }
810 
811 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
812 			       int budget, void *v)
813 {
814 	struct hnae_ring *ring = ring_data->ring;
815 	struct sk_buff *skb;
816 	int num, bnum;
817 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
818 	int recv_pkts, recv_bds, clean_count, err;
819 	int unused_count = hns_desc_unused(ring);
820 
821 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
822 	rmb(); /* make sure num taken effect before the other data is touched */
823 
824 	recv_pkts = 0, recv_bds = 0, clean_count = 0;
825 	num -= unused_count;
826 
827 	while (recv_pkts < budget && recv_bds < num) {
828 		/* reuse or realloc buffers */
829 		if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
830 			hns_nic_alloc_rx_buffers(ring_data,
831 						 clean_count + unused_count);
832 			clean_count = 0;
833 			unused_count = hns_desc_unused(ring);
834 		}
835 
836 		/* poll one pkt */
837 		err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
838 		if (unlikely(!skb)) /* this fault cannot be repaired */
839 			goto out;
840 
841 		recv_bds += bnum;
842 		clean_count += bnum;
843 		if (unlikely(err)) {  /* do jump the err */
844 			recv_pkts++;
845 			continue;
846 		}
847 
848 		/* do update ip stack process*/
849 		((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
850 							ring_data, skb);
851 		recv_pkts++;
852 	}
853 
854 out:
855 	/* make all data has been write before submit */
856 	if (clean_count + unused_count > 0)
857 		hns_nic_alloc_rx_buffers(ring_data,
858 					 clean_count + unused_count);
859 
860 	return recv_pkts;
861 }
862 
863 static void hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
864 {
865 	struct hnae_ring *ring = ring_data->ring;
866 	int num = 0;
867 
868 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
869 
870 	/* for hardware bug fixed */
871 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
872 
873 	if (num > 0) {
874 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
875 			ring_data->ring, 1);
876 
877 		napi_schedule(&ring_data->napi);
878 	}
879 }
880 
881 static void hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
882 {
883 	struct hnae_ring *ring = ring_data->ring;
884 	int num = 0;
885 
886 	num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
887 
888 	if (num == 0)
889 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
890 			ring, 0);
891 	else
892 		napi_schedule(&ring_data->napi);
893 }
894 
895 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
896 					    int *bytes, int *pkts)
897 {
898 	struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
899 
900 	(*pkts) += (desc_cb->type == DESC_TYPE_SKB);
901 	(*bytes) += desc_cb->length;
902 	/* desc_cb will be cleaned, after hnae_free_buffer_detach*/
903 	hnae_free_buffer_detach(ring, ring->next_to_clean);
904 
905 	ring_ptr_move_fw(ring, next_to_clean);
906 }
907 
908 static int is_valid_clean_head(struct hnae_ring *ring, int h)
909 {
910 	int u = ring->next_to_use;
911 	int c = ring->next_to_clean;
912 
913 	if (unlikely(h > ring->desc_num))
914 		return 0;
915 
916 	assert(u > 0 && u < ring->desc_num);
917 	assert(c > 0 && c < ring->desc_num);
918 	assert(u != c && h != c); /* must be checked before call this func */
919 
920 	return u > c ? (h > c && h <= u) : (h > c || h <= u);
921 }
922 
923 /* netif_tx_lock will turn down the performance, set only when necessary */
924 #ifdef CONFIG_NET_POLL_CONTROLLER
925 #define NETIF_TX_LOCK(ndev) netif_tx_lock(ndev)
926 #define NETIF_TX_UNLOCK(ndev) netif_tx_unlock(ndev)
927 #else
928 #define NETIF_TX_LOCK(ndev)
929 #define NETIF_TX_UNLOCK(ndev)
930 #endif
931 /* reclaim all desc in one budget
932  * return error or number of desc left
933  */
934 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
935 			       int budget, void *v)
936 {
937 	struct hnae_ring *ring = ring_data->ring;
938 	struct net_device *ndev = ring_data->napi.dev;
939 	struct netdev_queue *dev_queue;
940 	struct hns_nic_priv *priv = netdev_priv(ndev);
941 	int head;
942 	int bytes, pkts;
943 
944 	NETIF_TX_LOCK(ndev);
945 
946 	head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
947 	rmb(); /* make sure head is ready before touch any data */
948 
949 	if (is_ring_empty(ring) || head == ring->next_to_clean) {
950 		NETIF_TX_UNLOCK(ndev);
951 		return 0; /* no data to poll */
952 	}
953 
954 	if (!is_valid_clean_head(ring, head)) {
955 		netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
956 			   ring->next_to_use, ring->next_to_clean);
957 		ring->stats.io_err_cnt++;
958 		NETIF_TX_UNLOCK(ndev);
959 		return -EIO;
960 	}
961 
962 	bytes = 0;
963 	pkts = 0;
964 	while (head != ring->next_to_clean) {
965 		hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
966 		/* issue prefetch for next Tx descriptor */
967 		prefetch(&ring->desc_cb[ring->next_to_clean]);
968 	}
969 
970 	NETIF_TX_UNLOCK(ndev);
971 
972 	dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
973 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
974 
975 	if (unlikely(priv->link && !netif_carrier_ok(ndev)))
976 		netif_carrier_on(ndev);
977 
978 	if (unlikely(pkts && netif_carrier_ok(ndev) &&
979 		     (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
980 		/* Make sure that anybody stopping the queue after this
981 		 * sees the new next_to_clean.
982 		 */
983 		smp_mb();
984 		if (netif_tx_queue_stopped(dev_queue) &&
985 		    !test_bit(NIC_STATE_DOWN, &priv->state)) {
986 			netif_tx_wake_queue(dev_queue);
987 			ring->stats.restart_queue++;
988 		}
989 	}
990 	return 0;
991 }
992 
993 static void hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
994 {
995 	struct hnae_ring *ring = ring_data->ring;
996 	int head;
997 
998 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
999 
1000 	head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1001 
1002 	if (head != ring->next_to_clean) {
1003 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1004 			ring_data->ring, 1);
1005 
1006 		napi_schedule(&ring_data->napi);
1007 	}
1008 }
1009 
1010 static void hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1011 {
1012 	struct hnae_ring *ring = ring_data->ring;
1013 	int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1014 
1015 	if (head == ring->next_to_clean)
1016 		ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1017 			ring, 0);
1018 	else
1019 		napi_schedule(&ring_data->napi);
1020 }
1021 
1022 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1023 {
1024 	struct hnae_ring *ring = ring_data->ring;
1025 	struct net_device *ndev = ring_data->napi.dev;
1026 	struct netdev_queue *dev_queue;
1027 	int head;
1028 	int bytes, pkts;
1029 
1030 	NETIF_TX_LOCK(ndev);
1031 
1032 	head = ring->next_to_use; /* ntu :soft setted ring position*/
1033 	bytes = 0;
1034 	pkts = 0;
1035 	while (head != ring->next_to_clean)
1036 		hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1037 
1038 	NETIF_TX_UNLOCK(ndev);
1039 
1040 	dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1041 	netdev_tx_reset_queue(dev_queue);
1042 }
1043 
1044 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1045 {
1046 	struct hns_nic_ring_data *ring_data =
1047 		container_of(napi, struct hns_nic_ring_data, napi);
1048 	int clean_complete = ring_data->poll_one(
1049 				ring_data, budget, ring_data->ex_process);
1050 
1051 	if (clean_complete >= 0 && clean_complete < budget) {
1052 		napi_complete(napi);
1053 		ring_data->fini_process(ring_data);
1054 		return 0;
1055 	}
1056 
1057 	return clean_complete;
1058 }
1059 
1060 static irqreturn_t hns_irq_handle(int irq, void *dev)
1061 {
1062 	struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1063 
1064 	ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1065 		ring_data->ring, 1);
1066 	napi_schedule(&ring_data->napi);
1067 
1068 	return IRQ_HANDLED;
1069 }
1070 
1071 /**
1072  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1073  *@ndev: net device
1074  */
1075 static void hns_nic_adjust_link(struct net_device *ndev)
1076 {
1077 	struct hns_nic_priv *priv = netdev_priv(ndev);
1078 	struct hnae_handle *h = priv->ae_handle;
1079 	int state = 1;
1080 
1081 	if (ndev->phydev) {
1082 		h->dev->ops->adjust_link(h, ndev->phydev->speed,
1083 					 ndev->phydev->duplex);
1084 		state = ndev->phydev->link;
1085 	}
1086 	state = state && h->dev->ops->get_status(h);
1087 
1088 	if (state != priv->link) {
1089 		if (state) {
1090 			netif_carrier_on(ndev);
1091 			netif_tx_wake_all_queues(ndev);
1092 			netdev_info(ndev, "link up\n");
1093 		} else {
1094 			netif_carrier_off(ndev);
1095 			netdev_info(ndev, "link down\n");
1096 		}
1097 		priv->link = state;
1098 	}
1099 }
1100 
1101 /**
1102  *hns_nic_init_phy - init phy
1103  *@ndev: net device
1104  *@h: ae handle
1105  * Return 0 on success, negative on failure
1106  */
1107 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1108 {
1109 	struct phy_device *phy_dev = h->phy_dev;
1110 	int ret;
1111 
1112 	if (!h->phy_dev)
1113 		return 0;
1114 
1115 	if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1116 		phy_dev->dev_flags = 0;
1117 
1118 		ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1119 					 h->phy_if);
1120 	} else {
1121 		ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1122 	}
1123 	if (unlikely(ret))
1124 		return -ENODEV;
1125 
1126 	phy_dev->supported &= h->if_support;
1127 	phy_dev->advertising = phy_dev->supported;
1128 
1129 	if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1130 		phy_dev->autoneg = false;
1131 
1132 	return 0;
1133 }
1134 
1135 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1136 {
1137 	struct hns_nic_priv *priv = netdev_priv(netdev);
1138 	struct hnae_handle *h = priv->ae_handle;
1139 
1140 	napi_enable(&priv->ring_data[idx].napi);
1141 
1142 	enable_irq(priv->ring_data[idx].ring->irq);
1143 	h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1144 
1145 	return 0;
1146 }
1147 
1148 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1149 {
1150 	struct hns_nic_priv *priv = netdev_priv(ndev);
1151 	struct hnae_handle *h = priv->ae_handle;
1152 	struct sockaddr *mac_addr = p;
1153 	int ret;
1154 
1155 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1156 		return -EADDRNOTAVAIL;
1157 
1158 	ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1159 	if (ret) {
1160 		netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1161 		return ret;
1162 	}
1163 
1164 	memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1165 
1166 	return 0;
1167 }
1168 
1169 void hns_nic_update_stats(struct net_device *netdev)
1170 {
1171 	struct hns_nic_priv *priv = netdev_priv(netdev);
1172 	struct hnae_handle *h = priv->ae_handle;
1173 
1174 	h->dev->ops->update_stats(h, &netdev->stats);
1175 }
1176 
1177 /* set mac addr if it is configed. or leave it to the AE driver */
1178 static void hns_init_mac_addr(struct net_device *ndev)
1179 {
1180 	struct hns_nic_priv *priv = netdev_priv(ndev);
1181 
1182 	if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1183 		eth_hw_addr_random(ndev);
1184 		dev_warn(priv->dev, "No valid mac, use random mac %pM",
1185 			 ndev->dev_addr);
1186 	}
1187 }
1188 
1189 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1190 {
1191 	struct hns_nic_priv *priv = netdev_priv(netdev);
1192 	struct hnae_handle *h = priv->ae_handle;
1193 
1194 	h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1195 	disable_irq(priv->ring_data[idx].ring->irq);
1196 
1197 	napi_disable(&priv->ring_data[idx].napi);
1198 }
1199 
1200 static void hns_set_irq_affinity(struct hns_nic_priv *priv)
1201 {
1202 	struct hnae_handle *h = priv->ae_handle;
1203 	struct hns_nic_ring_data *rd;
1204 	int i;
1205 	int cpu;
1206 	cpumask_t mask;
1207 
1208 	/*diffrent irq banlance for 16core and 32core*/
1209 	if (h->q_num == num_possible_cpus()) {
1210 		for (i = 0; i < h->q_num * 2; i++) {
1211 			rd = &priv->ring_data[i];
1212 			if (cpu_online(rd->queue_index)) {
1213 				cpumask_clear(&mask);
1214 				cpu = rd->queue_index;
1215 				cpumask_set_cpu(cpu, &mask);
1216 				(void)irq_set_affinity_hint(rd->ring->irq,
1217 							    &mask);
1218 			}
1219 		}
1220 	} else {
1221 		for (i = 0; i < h->q_num; i++) {
1222 			rd = &priv->ring_data[i];
1223 			if (cpu_online(rd->queue_index * 2)) {
1224 				cpumask_clear(&mask);
1225 				cpu = rd->queue_index * 2;
1226 				cpumask_set_cpu(cpu, &mask);
1227 				(void)irq_set_affinity_hint(rd->ring->irq,
1228 							    &mask);
1229 			}
1230 		}
1231 
1232 		for (i = h->q_num; i < h->q_num * 2; i++) {
1233 			rd = &priv->ring_data[i];
1234 			if (cpu_online(rd->queue_index * 2 + 1)) {
1235 				cpumask_clear(&mask);
1236 				cpu = rd->queue_index * 2 + 1;
1237 				cpumask_set_cpu(cpu, &mask);
1238 				(void)irq_set_affinity_hint(rd->ring->irq,
1239 							    &mask);
1240 			}
1241 		}
1242 	}
1243 }
1244 
1245 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1246 {
1247 	struct hnae_handle *h = priv->ae_handle;
1248 	struct hns_nic_ring_data *rd;
1249 	int i;
1250 	int ret;
1251 
1252 	for (i = 0; i < h->q_num * 2; i++) {
1253 		rd = &priv->ring_data[i];
1254 
1255 		if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1256 			break;
1257 
1258 		snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1259 			 "%s-%s%d", priv->netdev->name,
1260 			 (i < h->q_num ? "tx" : "rx"), rd->queue_index);
1261 
1262 		rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1263 
1264 		ret = request_irq(rd->ring->irq,
1265 				  hns_irq_handle, 0, rd->ring->ring_name, rd);
1266 		if (ret) {
1267 			netdev_err(priv->netdev, "request irq(%d) fail\n",
1268 				   rd->ring->irq);
1269 			return ret;
1270 		}
1271 		disable_irq(rd->ring->irq);
1272 		rd->ring->irq_init_flag = RCB_IRQ_INITED;
1273 	}
1274 
1275 	/*set cpu affinity*/
1276 	hns_set_irq_affinity(priv);
1277 
1278 	return 0;
1279 }
1280 
1281 static int hns_nic_net_up(struct net_device *ndev)
1282 {
1283 	struct hns_nic_priv *priv = netdev_priv(ndev);
1284 	struct hnae_handle *h = priv->ae_handle;
1285 	int i, j;
1286 	int ret;
1287 
1288 	ret = hns_nic_init_irq(priv);
1289 	if (ret != 0) {
1290 		netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1291 		return ret;
1292 	}
1293 
1294 	for (i = 0; i < h->q_num * 2; i++) {
1295 		ret = hns_nic_ring_open(ndev, i);
1296 		if (ret)
1297 			goto out_has_some_queues;
1298 	}
1299 
1300 	ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1301 	if (ret)
1302 		goto out_set_mac_addr_err;
1303 
1304 	ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1305 	if (ret)
1306 		goto out_start_err;
1307 
1308 	if (ndev->phydev)
1309 		phy_start(ndev->phydev);
1310 
1311 	clear_bit(NIC_STATE_DOWN, &priv->state);
1312 	(void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1313 
1314 	return 0;
1315 
1316 out_start_err:
1317 	netif_stop_queue(ndev);
1318 out_set_mac_addr_err:
1319 out_has_some_queues:
1320 	for (j = i - 1; j >= 0; j--)
1321 		hns_nic_ring_close(ndev, j);
1322 
1323 	set_bit(NIC_STATE_DOWN, &priv->state);
1324 
1325 	return ret;
1326 }
1327 
1328 static void hns_nic_net_down(struct net_device *ndev)
1329 {
1330 	int i;
1331 	struct hnae_ae_ops *ops;
1332 	struct hns_nic_priv *priv = netdev_priv(ndev);
1333 
1334 	if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1335 		return;
1336 
1337 	(void)del_timer_sync(&priv->service_timer);
1338 	netif_tx_stop_all_queues(ndev);
1339 	netif_carrier_off(ndev);
1340 	netif_tx_disable(ndev);
1341 	priv->link = 0;
1342 
1343 	if (ndev->phydev)
1344 		phy_stop(ndev->phydev);
1345 
1346 	ops = priv->ae_handle->dev->ops;
1347 
1348 	if (ops->stop)
1349 		ops->stop(priv->ae_handle);
1350 
1351 	netif_tx_stop_all_queues(ndev);
1352 
1353 	for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1354 		hns_nic_ring_close(ndev, i);
1355 		hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1356 
1357 		/* clean tx buffers*/
1358 		hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1359 	}
1360 }
1361 
1362 void hns_nic_net_reset(struct net_device *ndev)
1363 {
1364 	struct hns_nic_priv *priv = netdev_priv(ndev);
1365 	struct hnae_handle *handle = priv->ae_handle;
1366 
1367 	while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1368 		usleep_range(1000, 2000);
1369 
1370 	(void)hnae_reinit_handle(handle);
1371 
1372 	clear_bit(NIC_STATE_RESETTING, &priv->state);
1373 }
1374 
1375 void hns_nic_net_reinit(struct net_device *netdev)
1376 {
1377 	struct hns_nic_priv *priv = netdev_priv(netdev);
1378 
1379 	netif_trans_update(priv->netdev);
1380 	while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1381 		usleep_range(1000, 2000);
1382 
1383 	hns_nic_net_down(netdev);
1384 	hns_nic_net_reset(netdev);
1385 	(void)hns_nic_net_up(netdev);
1386 	clear_bit(NIC_STATE_REINITING, &priv->state);
1387 }
1388 
1389 static int hns_nic_net_open(struct net_device *ndev)
1390 {
1391 	struct hns_nic_priv *priv = netdev_priv(ndev);
1392 	struct hnae_handle *h = priv->ae_handle;
1393 	int ret;
1394 
1395 	if (test_bit(NIC_STATE_TESTING, &priv->state))
1396 		return -EBUSY;
1397 
1398 	priv->link = 0;
1399 	netif_carrier_off(ndev);
1400 
1401 	ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1402 	if (ret < 0) {
1403 		netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1404 			   ret);
1405 		return ret;
1406 	}
1407 
1408 	ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1409 	if (ret < 0) {
1410 		netdev_err(ndev,
1411 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1412 		return ret;
1413 	}
1414 
1415 	ret = hns_nic_net_up(ndev);
1416 	if (ret) {
1417 		netdev_err(ndev,
1418 			   "hns net up fail, ret=%d!\n", ret);
1419 		return ret;
1420 	}
1421 
1422 	return 0;
1423 }
1424 
1425 static int hns_nic_net_stop(struct net_device *ndev)
1426 {
1427 	hns_nic_net_down(ndev);
1428 
1429 	return 0;
1430 }
1431 
1432 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1433 static void hns_nic_net_timeout(struct net_device *ndev)
1434 {
1435 	struct hns_nic_priv *priv = netdev_priv(ndev);
1436 
1437 	hns_tx_timeout_reset(priv);
1438 }
1439 
1440 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1441 			    int cmd)
1442 {
1443 	struct phy_device *phy_dev = netdev->phydev;
1444 
1445 	if (!netif_running(netdev))
1446 		return -EINVAL;
1447 
1448 	if (!phy_dev)
1449 		return -ENOTSUPP;
1450 
1451 	return phy_mii_ioctl(phy_dev, ifr, cmd);
1452 }
1453 
1454 /* use only for netconsole to poll with the device without interrupt */
1455 #ifdef CONFIG_NET_POLL_CONTROLLER
1456 void hns_nic_poll_controller(struct net_device *ndev)
1457 {
1458 	struct hns_nic_priv *priv = netdev_priv(ndev);
1459 	unsigned long flags;
1460 	int i;
1461 
1462 	local_irq_save(flags);
1463 	for (i = 0; i < priv->ae_handle->q_num * 2; i++)
1464 		napi_schedule(&priv->ring_data[i].napi);
1465 	local_irq_restore(flags);
1466 }
1467 #endif
1468 
1469 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1470 				    struct net_device *ndev)
1471 {
1472 	struct hns_nic_priv *priv = netdev_priv(ndev);
1473 	int ret;
1474 
1475 	assert(skb->queue_mapping < ndev->ae_handle->q_num);
1476 	ret = hns_nic_net_xmit_hw(ndev, skb,
1477 				  &tx_ring_data(priv, skb->queue_mapping));
1478 	if (ret == NETDEV_TX_OK) {
1479 		netif_trans_update(ndev);
1480 		ndev->stats.tx_bytes += skb->len;
1481 		ndev->stats.tx_packets++;
1482 	}
1483 	return (netdev_tx_t)ret;
1484 }
1485 
1486 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1487 {
1488 	struct hns_nic_priv *priv = netdev_priv(ndev);
1489 	struct hnae_handle *h = priv->ae_handle;
1490 	int ret;
1491 
1492 	if (!h->dev->ops->set_mtu)
1493 		return -ENOTSUPP;
1494 
1495 	if (netif_running(ndev)) {
1496 		(void)hns_nic_net_stop(ndev);
1497 		msleep(100);
1498 
1499 		ret = h->dev->ops->set_mtu(h, new_mtu);
1500 		if (ret)
1501 			netdev_err(ndev, "set mtu fail, return value %d\n",
1502 				   ret);
1503 
1504 		if (hns_nic_net_open(ndev))
1505 			netdev_err(ndev, "hns net open fail\n");
1506 	} else {
1507 		ret = h->dev->ops->set_mtu(h, new_mtu);
1508 	}
1509 
1510 	if (!ret)
1511 		ndev->mtu = new_mtu;
1512 
1513 	return ret;
1514 }
1515 
1516 static int hns_nic_set_features(struct net_device *netdev,
1517 				netdev_features_t features)
1518 {
1519 	struct hns_nic_priv *priv = netdev_priv(netdev);
1520 
1521 	switch (priv->enet_ver) {
1522 	case AE_VERSION_1:
1523 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1524 			netdev_info(netdev, "enet v1 do not support tso!\n");
1525 		break;
1526 	default:
1527 		if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1528 			priv->ops.fill_desc = fill_tso_desc;
1529 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1530 			/* The chip only support 7*4096 */
1531 			netif_set_gso_max_size(netdev, 7 * 4096);
1532 		} else {
1533 			priv->ops.fill_desc = fill_v2_desc;
1534 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1535 		}
1536 		break;
1537 	}
1538 	netdev->features = features;
1539 	return 0;
1540 }
1541 
1542 static netdev_features_t hns_nic_fix_features(
1543 		struct net_device *netdev, netdev_features_t features)
1544 {
1545 	struct hns_nic_priv *priv = netdev_priv(netdev);
1546 
1547 	switch (priv->enet_ver) {
1548 	case AE_VERSION_1:
1549 		features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1550 				NETIF_F_HW_VLAN_CTAG_FILTER);
1551 		break;
1552 	default:
1553 		break;
1554 	}
1555 	return features;
1556 }
1557 
1558 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1559 {
1560 	struct hns_nic_priv *priv = netdev_priv(netdev);
1561 	struct hnae_handle *h = priv->ae_handle;
1562 
1563 	if (h->dev->ops->add_uc_addr)
1564 		return h->dev->ops->add_uc_addr(h, addr);
1565 
1566 	return 0;
1567 }
1568 
1569 static int hns_nic_uc_unsync(struct net_device *netdev,
1570 			     const unsigned char *addr)
1571 {
1572 	struct hns_nic_priv *priv = netdev_priv(netdev);
1573 	struct hnae_handle *h = priv->ae_handle;
1574 
1575 	if (h->dev->ops->rm_uc_addr)
1576 		return h->dev->ops->rm_uc_addr(h, addr);
1577 
1578 	return 0;
1579 }
1580 
1581 /**
1582  * nic_set_multicast_list - set mutl mac address
1583  * @netdev: net device
1584  * @p: mac address
1585  *
1586  * return void
1587  */
1588 void hns_set_multicast_list(struct net_device *ndev)
1589 {
1590 	struct hns_nic_priv *priv = netdev_priv(ndev);
1591 	struct hnae_handle *h = priv->ae_handle;
1592 	struct netdev_hw_addr *ha = NULL;
1593 
1594 	if (!h)	{
1595 		netdev_err(ndev, "hnae handle is null\n");
1596 		return;
1597 	}
1598 
1599 	if (h->dev->ops->clr_mc_addr)
1600 		if (h->dev->ops->clr_mc_addr(h))
1601 			netdev_err(ndev, "clear multicast address fail\n");
1602 
1603 	if (h->dev->ops->set_mc_addr) {
1604 		netdev_for_each_mc_addr(ha, ndev)
1605 			if (h->dev->ops->set_mc_addr(h, ha->addr))
1606 				netdev_err(ndev, "set multicast fail\n");
1607 	}
1608 }
1609 
1610 void hns_nic_set_rx_mode(struct net_device *ndev)
1611 {
1612 	struct hns_nic_priv *priv = netdev_priv(ndev);
1613 	struct hnae_handle *h = priv->ae_handle;
1614 
1615 	if (h->dev->ops->set_promisc_mode) {
1616 		if (ndev->flags & IFF_PROMISC)
1617 			h->dev->ops->set_promisc_mode(h, 1);
1618 		else
1619 			h->dev->ops->set_promisc_mode(h, 0);
1620 	}
1621 
1622 	hns_set_multicast_list(ndev);
1623 
1624 	if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1625 		netdev_err(ndev, "sync uc address fail\n");
1626 }
1627 
1628 struct rtnl_link_stats64 *hns_nic_get_stats64(struct net_device *ndev,
1629 					      struct rtnl_link_stats64 *stats)
1630 {
1631 	int idx = 0;
1632 	u64 tx_bytes = 0;
1633 	u64 rx_bytes = 0;
1634 	u64 tx_pkts = 0;
1635 	u64 rx_pkts = 0;
1636 	struct hns_nic_priv *priv = netdev_priv(ndev);
1637 	struct hnae_handle *h = priv->ae_handle;
1638 
1639 	for (idx = 0; idx < h->q_num; idx++) {
1640 		tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1641 		tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1642 		rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1643 		rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1644 	}
1645 
1646 	stats->tx_bytes = tx_bytes;
1647 	stats->tx_packets = tx_pkts;
1648 	stats->rx_bytes = rx_bytes;
1649 	stats->rx_packets = rx_pkts;
1650 
1651 	stats->rx_errors = ndev->stats.rx_errors;
1652 	stats->multicast = ndev->stats.multicast;
1653 	stats->rx_length_errors = ndev->stats.rx_length_errors;
1654 	stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1655 	stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1656 
1657 	stats->tx_errors = ndev->stats.tx_errors;
1658 	stats->rx_dropped = ndev->stats.rx_dropped;
1659 	stats->tx_dropped = ndev->stats.tx_dropped;
1660 	stats->collisions = ndev->stats.collisions;
1661 	stats->rx_over_errors = ndev->stats.rx_over_errors;
1662 	stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1663 	stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1664 	stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1665 	stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1666 	stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1667 	stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1668 	stats->tx_window_errors = ndev->stats.tx_window_errors;
1669 	stats->rx_compressed = ndev->stats.rx_compressed;
1670 	stats->tx_compressed = ndev->stats.tx_compressed;
1671 
1672 	return stats;
1673 }
1674 
1675 static u16
1676 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1677 		     void *accel_priv, select_queue_fallback_t fallback)
1678 {
1679 	struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1680 	struct hns_nic_priv *priv = netdev_priv(ndev);
1681 
1682 	/* fix hardware broadcast/multicast packets queue loopback */
1683 	if (!AE_IS_VER1(priv->enet_ver) &&
1684 	    is_multicast_ether_addr(eth_hdr->h_dest))
1685 		return 0;
1686 	else
1687 		return fallback(ndev, skb);
1688 }
1689 
1690 static const struct net_device_ops hns_nic_netdev_ops = {
1691 	.ndo_open = hns_nic_net_open,
1692 	.ndo_stop = hns_nic_net_stop,
1693 	.ndo_start_xmit = hns_nic_net_xmit,
1694 	.ndo_tx_timeout = hns_nic_net_timeout,
1695 	.ndo_set_mac_address = hns_nic_net_set_mac_address,
1696 	.ndo_change_mtu = hns_nic_change_mtu,
1697 	.ndo_do_ioctl = hns_nic_do_ioctl,
1698 	.ndo_set_features = hns_nic_set_features,
1699 	.ndo_fix_features = hns_nic_fix_features,
1700 	.ndo_get_stats64 = hns_nic_get_stats64,
1701 #ifdef CONFIG_NET_POLL_CONTROLLER
1702 	.ndo_poll_controller = hns_nic_poll_controller,
1703 #endif
1704 	.ndo_set_rx_mode = hns_nic_set_rx_mode,
1705 	.ndo_select_queue = hns_nic_select_queue,
1706 };
1707 
1708 static void hns_nic_update_link_status(struct net_device *netdev)
1709 {
1710 	struct hns_nic_priv *priv = netdev_priv(netdev);
1711 
1712 	struct hnae_handle *h = priv->ae_handle;
1713 
1714 	if (h->phy_dev) {
1715 		if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1716 			return;
1717 
1718 		(void)genphy_read_status(h->phy_dev);
1719 	}
1720 	hns_nic_adjust_link(netdev);
1721 }
1722 
1723 /* for dumping key regs*/
1724 static void hns_nic_dump(struct hns_nic_priv *priv)
1725 {
1726 	struct hnae_handle *h = priv->ae_handle;
1727 	struct hnae_ae_ops *ops = h->dev->ops;
1728 	u32 *data, reg_num, i;
1729 
1730 	if (ops->get_regs_len && ops->get_regs) {
1731 		reg_num = ops->get_regs_len(priv->ae_handle);
1732 		reg_num = (reg_num + 3ul) & ~3ul;
1733 		data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1734 		if (data) {
1735 			ops->get_regs(priv->ae_handle, data);
1736 			for (i = 0; i < reg_num; i += 4)
1737 				pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1738 					i, data[i], data[i + 1],
1739 					data[i + 2], data[i + 3]);
1740 			kfree(data);
1741 		}
1742 	}
1743 
1744 	for (i = 0; i < h->q_num; i++) {
1745 		pr_info("tx_queue%d_next_to_clean:%d\n",
1746 			i, h->qs[i]->tx_ring.next_to_clean);
1747 		pr_info("tx_queue%d_next_to_use:%d\n",
1748 			i, h->qs[i]->tx_ring.next_to_use);
1749 		pr_info("rx_queue%d_next_to_clean:%d\n",
1750 			i, h->qs[i]->rx_ring.next_to_clean);
1751 		pr_info("rx_queue%d_next_to_use:%d\n",
1752 			i, h->qs[i]->rx_ring.next_to_use);
1753 	}
1754 }
1755 
1756 /* for resetting subtask */
1757 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
1758 {
1759 	enum hnae_port_type type = priv->ae_handle->port_type;
1760 
1761 	if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
1762 		return;
1763 	clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1764 
1765 	/* If we're already down, removing or resetting, just bail */
1766 	if (test_bit(NIC_STATE_DOWN, &priv->state) ||
1767 	    test_bit(NIC_STATE_REMOVING, &priv->state) ||
1768 	    test_bit(NIC_STATE_RESETTING, &priv->state))
1769 		return;
1770 
1771 	hns_nic_dump(priv);
1772 	netdev_info(priv->netdev, "try to reset %s port!\n",
1773 		    (type == HNAE_PORT_DEBUG ? "debug" : "service"));
1774 
1775 	rtnl_lock();
1776 	/* put off any impending NetWatchDogTimeout */
1777 	netif_trans_update(priv->netdev);
1778 
1779 	if (type == HNAE_PORT_DEBUG) {
1780 		hns_nic_net_reinit(priv->netdev);
1781 	} else {
1782 		netif_carrier_off(priv->netdev);
1783 		netif_tx_disable(priv->netdev);
1784 	}
1785 	rtnl_unlock();
1786 }
1787 
1788 /* for doing service complete*/
1789 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
1790 {
1791 	WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
1792 
1793 	smp_mb__before_atomic();
1794 	clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
1795 }
1796 
1797 static void hns_nic_service_task(struct work_struct *work)
1798 {
1799 	struct hns_nic_priv *priv
1800 		= container_of(work, struct hns_nic_priv, service_task);
1801 	struct hnae_handle *h = priv->ae_handle;
1802 
1803 	hns_nic_update_link_status(priv->netdev);
1804 	h->dev->ops->update_led_status(h);
1805 	hns_nic_update_stats(priv->netdev);
1806 
1807 	hns_nic_reset_subtask(priv);
1808 	hns_nic_service_event_complete(priv);
1809 }
1810 
1811 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
1812 {
1813 	if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
1814 	    !test_bit(NIC_STATE_REMOVING, &priv->state) &&
1815 	    !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
1816 		(void)schedule_work(&priv->service_task);
1817 }
1818 
1819 static void hns_nic_service_timer(unsigned long data)
1820 {
1821 	struct hns_nic_priv *priv = (struct hns_nic_priv *)data;
1822 
1823 	(void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1824 
1825 	hns_nic_task_schedule(priv);
1826 }
1827 
1828 /**
1829  * hns_tx_timeout_reset - initiate reset due to Tx timeout
1830  * @priv: driver private struct
1831  **/
1832 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
1833 {
1834 	/* Do the reset outside of interrupt context */
1835 	if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
1836 		set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1837 		netdev_warn(priv->netdev,
1838 			    "initiating reset due to tx timeout(%llu,0x%lx)\n",
1839 			    priv->tx_timeout_count, priv->state);
1840 		priv->tx_timeout_count++;
1841 		hns_nic_task_schedule(priv);
1842 	}
1843 }
1844 
1845 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
1846 {
1847 	struct hnae_handle *h = priv->ae_handle;
1848 	struct hns_nic_ring_data *rd;
1849 	bool is_ver1 = AE_IS_VER1(priv->enet_ver);
1850 	int i;
1851 
1852 	if (h->q_num > NIC_MAX_Q_PER_VF) {
1853 		netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
1854 		return -EINVAL;
1855 	}
1856 
1857 	priv->ring_data = kzalloc(h->q_num * sizeof(*priv->ring_data) * 2,
1858 				  GFP_KERNEL);
1859 	if (!priv->ring_data)
1860 		return -ENOMEM;
1861 
1862 	for (i = 0; i < h->q_num; i++) {
1863 		rd = &priv->ring_data[i];
1864 		rd->queue_index = i;
1865 		rd->ring = &h->qs[i]->tx_ring;
1866 		rd->poll_one = hns_nic_tx_poll_one;
1867 		rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
1868 			hns_nic_tx_fini_pro_v2;
1869 
1870 		netif_napi_add(priv->netdev, &rd->napi,
1871 			       hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
1872 		rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1873 	}
1874 	for (i = h->q_num; i < h->q_num * 2; i++) {
1875 		rd = &priv->ring_data[i];
1876 		rd->queue_index = i - h->q_num;
1877 		rd->ring = &h->qs[i - h->q_num]->rx_ring;
1878 		rd->poll_one = hns_nic_rx_poll_one;
1879 		rd->ex_process = hns_nic_rx_up_pro;
1880 		rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
1881 			hns_nic_rx_fini_pro_v2;
1882 
1883 		netif_napi_add(priv->netdev, &rd->napi,
1884 			       hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
1885 		rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1886 	}
1887 
1888 	return 0;
1889 }
1890 
1891 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
1892 {
1893 	struct hnae_handle *h = priv->ae_handle;
1894 	int i;
1895 
1896 	for (i = 0; i < h->q_num * 2; i++) {
1897 		netif_napi_del(&priv->ring_data[i].napi);
1898 		if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
1899 			(void)irq_set_affinity_hint(
1900 				priv->ring_data[i].ring->irq,
1901 				NULL);
1902 			free_irq(priv->ring_data[i].ring->irq,
1903 				 &priv->ring_data[i]);
1904 		}
1905 
1906 		priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
1907 	}
1908 	kfree(priv->ring_data);
1909 }
1910 
1911 static void hns_nic_set_priv_ops(struct net_device *netdev)
1912 {
1913 	struct hns_nic_priv *priv = netdev_priv(netdev);
1914 	struct hnae_handle *h = priv->ae_handle;
1915 
1916 	if (AE_IS_VER1(priv->enet_ver)) {
1917 		priv->ops.fill_desc = fill_desc;
1918 		priv->ops.get_rxd_bnum = get_rx_desc_bnum;
1919 		priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1920 	} else {
1921 		priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
1922 		if ((netdev->features & NETIF_F_TSO) ||
1923 		    (netdev->features & NETIF_F_TSO6)) {
1924 			priv->ops.fill_desc = fill_tso_desc;
1925 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1926 			/* This chip only support 7*4096 */
1927 			netif_set_gso_max_size(netdev, 7 * 4096);
1928 		} else {
1929 			priv->ops.fill_desc = fill_v2_desc;
1930 			priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1931 		}
1932 		/* enable tso when init
1933 		 * control tso on/off through TSE bit in bd
1934 		 */
1935 		h->dev->ops->set_tso_stats(h, 1);
1936 	}
1937 }
1938 
1939 static int hns_nic_try_get_ae(struct net_device *ndev)
1940 {
1941 	struct hns_nic_priv *priv = netdev_priv(ndev);
1942 	struct hnae_handle *h;
1943 	int ret;
1944 
1945 	h = hnae_get_handle(&priv->netdev->dev,
1946 			    priv->fwnode, priv->port_id, NULL);
1947 	if (IS_ERR_OR_NULL(h)) {
1948 		ret = -ENODEV;
1949 		dev_dbg(priv->dev, "has not handle, register notifier!\n");
1950 		goto out;
1951 	}
1952 	priv->ae_handle = h;
1953 
1954 	ret = hns_nic_init_phy(ndev, h);
1955 	if (ret) {
1956 		dev_err(priv->dev, "probe phy device fail!\n");
1957 		goto out_init_phy;
1958 	}
1959 
1960 	ret = hns_nic_init_ring_data(priv);
1961 	if (ret) {
1962 		ret = -ENOMEM;
1963 		goto out_init_ring_data;
1964 	}
1965 
1966 	hns_nic_set_priv_ops(ndev);
1967 
1968 	ret = register_netdev(ndev);
1969 	if (ret) {
1970 		dev_err(priv->dev, "probe register netdev fail!\n");
1971 		goto out_reg_ndev_fail;
1972 	}
1973 	return 0;
1974 
1975 out_reg_ndev_fail:
1976 	hns_nic_uninit_ring_data(priv);
1977 	priv->ring_data = NULL;
1978 out_init_phy:
1979 out_init_ring_data:
1980 	hnae_put_handle(priv->ae_handle);
1981 	priv->ae_handle = NULL;
1982 out:
1983 	return ret;
1984 }
1985 
1986 static int hns_nic_notifier_action(struct notifier_block *nb,
1987 				   unsigned long action, void *data)
1988 {
1989 	struct hns_nic_priv *priv =
1990 		container_of(nb, struct hns_nic_priv, notifier_block);
1991 
1992 	assert(action == HNAE_AE_REGISTER);
1993 
1994 	if (!hns_nic_try_get_ae(priv->netdev)) {
1995 		hnae_unregister_notifier(&priv->notifier_block);
1996 		priv->notifier_block.notifier_call = NULL;
1997 	}
1998 	return 0;
1999 }
2000 
2001 static int hns_nic_dev_probe(struct platform_device *pdev)
2002 {
2003 	struct device *dev = &pdev->dev;
2004 	struct net_device *ndev;
2005 	struct hns_nic_priv *priv;
2006 	u32 port_id;
2007 	int ret;
2008 
2009 	ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2010 	if (!ndev)
2011 		return -ENOMEM;
2012 
2013 	platform_set_drvdata(pdev, ndev);
2014 
2015 	priv = netdev_priv(ndev);
2016 	priv->dev = dev;
2017 	priv->netdev = ndev;
2018 
2019 	if (dev_of_node(dev)) {
2020 		struct device_node *ae_node;
2021 
2022 		if (of_device_is_compatible(dev->of_node,
2023 					    "hisilicon,hns-nic-v1"))
2024 			priv->enet_ver = AE_VERSION_1;
2025 		else
2026 			priv->enet_ver = AE_VERSION_2;
2027 
2028 		ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2029 		if (IS_ERR_OR_NULL(ae_node)) {
2030 			ret = PTR_ERR(ae_node);
2031 			dev_err(dev, "not find ae-handle\n");
2032 			goto out_read_prop_fail;
2033 		}
2034 		priv->fwnode = &ae_node->fwnode;
2035 	} else if (is_acpi_node(dev->fwnode)) {
2036 		struct acpi_reference_args args;
2037 
2038 		if (acpi_dev_found(hns_enet_acpi_match[0].id))
2039 			priv->enet_ver = AE_VERSION_1;
2040 		else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2041 			priv->enet_ver = AE_VERSION_2;
2042 		else
2043 			return -ENXIO;
2044 
2045 		/* try to find port-idx-in-ae first */
2046 		ret = acpi_node_get_property_reference(dev->fwnode,
2047 						       "ae-handle", 0, &args);
2048 		if (ret) {
2049 			dev_err(dev, "not find ae-handle\n");
2050 			goto out_read_prop_fail;
2051 		}
2052 		priv->fwnode = acpi_fwnode_handle(args.adev);
2053 	} else {
2054 		dev_err(dev, "cannot read cfg data from OF or acpi\n");
2055 		return -ENXIO;
2056 	}
2057 
2058 	ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2059 	if (ret) {
2060 		/* only for old code compatible */
2061 		ret = device_property_read_u32(dev, "port-id", &port_id);
2062 		if (ret)
2063 			goto out_read_prop_fail;
2064 		/* for old dts, we need to caculate the port offset */
2065 		port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2066 			: port_id - HNS_SRV_OFFSET;
2067 	}
2068 	priv->port_id = port_id;
2069 
2070 	hns_init_mac_addr(ndev);
2071 
2072 	ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2073 	ndev->priv_flags |= IFF_UNICAST_FLT;
2074 	ndev->netdev_ops = &hns_nic_netdev_ops;
2075 	hns_ethtool_set_ops(ndev);
2076 
2077 	ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2078 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2079 		NETIF_F_GRO;
2080 	ndev->vlan_features |=
2081 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2082 	ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2083 
2084 	/* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2085 	ndev->min_mtu = MAC_MIN_MTU;
2086 	switch (priv->enet_ver) {
2087 	case AE_VERSION_2:
2088 		ndev->features |= NETIF_F_TSO | NETIF_F_TSO6;
2089 		ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2090 			NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2091 			NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2092 		ndev->max_mtu = MAC_MAX_MTU_V2 -
2093 				(ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2094 		break;
2095 	default:
2096 		ndev->max_mtu = MAC_MAX_MTU -
2097 				(ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2098 		break;
2099 	}
2100 
2101 	SET_NETDEV_DEV(ndev, dev);
2102 
2103 	if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2104 		dev_dbg(dev, "set mask to 64bit\n");
2105 	else
2106 		dev_err(dev, "set mask to 64bit fail!\n");
2107 
2108 	/* carrier off reporting is important to ethtool even BEFORE open */
2109 	netif_carrier_off(ndev);
2110 
2111 	setup_timer(&priv->service_timer, hns_nic_service_timer,
2112 		    (unsigned long)priv);
2113 	INIT_WORK(&priv->service_task, hns_nic_service_task);
2114 
2115 	set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2116 	clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2117 	set_bit(NIC_STATE_DOWN, &priv->state);
2118 
2119 	if (hns_nic_try_get_ae(priv->netdev)) {
2120 		priv->notifier_block.notifier_call = hns_nic_notifier_action;
2121 		ret = hnae_register_notifier(&priv->notifier_block);
2122 		if (ret) {
2123 			dev_err(dev, "register notifier fail!\n");
2124 			goto out_notify_fail;
2125 		}
2126 		dev_dbg(dev, "has not handle, register notifier!\n");
2127 	}
2128 
2129 	return 0;
2130 
2131 out_notify_fail:
2132 	(void)cancel_work_sync(&priv->service_task);
2133 out_read_prop_fail:
2134 	free_netdev(ndev);
2135 	return ret;
2136 }
2137 
2138 static int hns_nic_dev_remove(struct platform_device *pdev)
2139 {
2140 	struct net_device *ndev = platform_get_drvdata(pdev);
2141 	struct hns_nic_priv *priv = netdev_priv(ndev);
2142 
2143 	if (ndev->reg_state != NETREG_UNINITIALIZED)
2144 		unregister_netdev(ndev);
2145 
2146 	if (priv->ring_data)
2147 		hns_nic_uninit_ring_data(priv);
2148 	priv->ring_data = NULL;
2149 
2150 	if (ndev->phydev)
2151 		phy_disconnect(ndev->phydev);
2152 
2153 	if (!IS_ERR_OR_NULL(priv->ae_handle))
2154 		hnae_put_handle(priv->ae_handle);
2155 	priv->ae_handle = NULL;
2156 	if (priv->notifier_block.notifier_call)
2157 		hnae_unregister_notifier(&priv->notifier_block);
2158 	priv->notifier_block.notifier_call = NULL;
2159 
2160 	set_bit(NIC_STATE_REMOVING, &priv->state);
2161 	(void)cancel_work_sync(&priv->service_task);
2162 
2163 	free_netdev(ndev);
2164 	return 0;
2165 }
2166 
2167 static const struct of_device_id hns_enet_of_match[] = {
2168 	{.compatible = "hisilicon,hns-nic-v1",},
2169 	{.compatible = "hisilicon,hns-nic-v2",},
2170 	{},
2171 };
2172 
2173 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2174 
2175 static struct platform_driver hns_nic_dev_driver = {
2176 	.driver = {
2177 		.name = "hns-nic",
2178 		.of_match_table = hns_enet_of_match,
2179 		.acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2180 	},
2181 	.probe = hns_nic_dev_probe,
2182 	.remove = hns_nic_dev_remove,
2183 };
2184 
2185 module_platform_driver(hns_nic_dev_driver);
2186 
2187 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2188 MODULE_AUTHOR("Hisilicon, Inc.");
2189 MODULE_LICENSE("GPL");
2190 MODULE_ALIAS("platform:hns-nic");
2191